Handbook of Research Methods in

Tourism
Quantitative and Qualitative Approaches

Edited by

Larry Dwyer
University of New South Wales, Australia

Alison Gill
Simon Fraser University, Canada

Neelu Seetaram
Bournemouth University, UK

Edward Elgar
Cheltenham, UK Northampton, MA, USA

DWYER 9781781001288 Part I.indd iii 25/07/2012 11:06

 Larry Dwyer, Alison Gill and Neelu Seetaram 2012

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or
transmitted in any form or by any means, electronic, mechanical or photocopying, recording, or
otherwise without the prior permission of the publisher.

Published by
Edward Elgar Publishing Limited
The Lypiatts
15 Lansdown Road
Cheltenham
Glos GL50 2JA
UK

Edward Elgar Publishing, Inc.

William Pratt House
9 Dewey Court
Northampton
Massachusetts 01060
USA

A catalogue record for this book

is available from the British Library

Library of Congress Control Number: 2012935315

ISBN 978 1 78100 128 8 (cased)

Typeset by Servis Filmsetting Ltd, Stockport, Cheshire

Printed and bound by MPG Books Group, UK
01

DWYER 9781781001288 Part I.indd iv 25/07/2012 11:06

Contents

List of contributors vii

Preface xvi

PART I QUANTITATIVE RESEARCH METHODS

1 Statistical testing techniques 13

Gang Li
2 Regression analysis 31
Jaume Rossell
3 Time series analysis 47
Shuang Cang and Neelu Seetaram
4 Demand modeling and forecasting 71
Grace Bo Peng, Haiyan Song and Stephen F. Witt
5 Structural equation modeling 91
Jenny (Jiyeon) Lee and Gerard Kyle
6 Discrete choice analysis and experimental design 113
Clive Morley
7 Panel data analysis 127
Neelu Seetaram and Sylvain Petit
8 The almost ideal demand system 145
Sarath Divisekera
9 Hedonic price analysis 170
Andreas Papatheodorou, Zheng Lei and Alexandros Apostolakis
10 Factor analysis 183
Lindsay W. Turner and Chau Vu
11 Cluster analysis 212
Liz Fredline
12 Inputoutput and SAM models 227
Clemente Polo and Elisabeth Valle
13 CGE modeling 261
Larry Dwyer and Tien Duc Pham
14 Costbenefit analysis 290
Larry Dwyer

PART II QUALITATIVE RESEARCH METHODS

Gayle R. Jennings

15 Grounded theory 325

Olga Junek and Les Killion

v
 vi Handbook of research methods in tourism

16 Ethnographic methods 339

Kathleen M. Adams
17 Focus groups 352
Carl Cater and Tiffany Low
18 Interview techniques 365
Nancy Gard McGehee
19 Participant observation 377
Nuno F. Ribeiro and Eric W. Foemmel
20 Cross-cultural approaches 392
Carla Almeida Santos and Changsup Shim
21 Archival research 403
Dallen J. Timothy
22 Community case study research 417
Dianne Dredge and Rob Hales

PART III MIXED METHODS

23 Content analysis 443

Svetlana Stepchenkova
24 Meta-analyses of tourism research 459
Ulrike Gretzel and Heather Kennedy-Eden
25 Network analysis 472
Ehsan Ahmed

Index 495

DWYER 9781781001288 Part I.indd vi 25/07/2012 11:06

Contributors

Kathleen M. Adams is Professor of Anthropology at Loyola University Chicago and

Adjunct Curator at the Field Museum of Natural History. Her book, Art as Politics:
Recrafting Identities, Tourism and Power in Tana Toraja, Indonesia (2006), received the
2009 Alpha Sigma Nu National Book Award. She also co-edited Home and Hegemony:
Domestic Work and Identity Politics in South and Southeast Asia (2000) and Everyday
Life in Southeast Asia (2011). Her articles on tourism and ethnic identity, travel and
nation-building, cultural representations and politics of arts appear in Annals of Tourism
Research, Tourist Studies, American Ethnologist, Museum Anthropology and other jour-
nals.
Ehsan Ahmed is an associate lecturer in the School of Marketing, Australian School
of Business at the University of New South Wales. His research is based on inter-
organizational relationships and networks. He has a particular interest in how the
relationships between tourism stakeholders influence their behaviour and business deci-
sions. Ehsan is currently researching his PhD at the University of New South Wales. In
his thesis, Ehsan uses quantitative network analysis to examine how the relationships
among tourism stakeholders influence socially responsible business practices such as
saving energy, contributing to the local community and keeping a long-term balance
between economic gain and non-economic stewardship.
Alexandros Apostolakis is a senior lecturer at the Department of Economics, University
of Portsmouth. The research interests of Dr Apostolakis focus on the economics of
tourism, economics of culture and cultural resources, database analysis, the analysis
of discrete choice experiments, marketing and management of tourism and tourism
businesses and the economics of defence. Dr Apostolakis has published in leading aca-
demic journals and has conducted extensive advisory work. He has also been a reviewer
for several international journals such as the Annals of Tourism Research, Tourism
Management, Journal of Environmental Management, Oxford Economic Papers and
Tourism Economics.
Shuang Cang is a senior lecturer in statistics and quantitative analysis at the School of
Tourism, Bournemouth University (UK). She gained her PhD from the University of
Abertay Dundee (UK), MSc with distinction from Kings College London (UK) and
BSc with first class from Heilongjiang University (China). She has expertise and wide
experience in data mining, artificial intelligence, pattern recognition and multivari-
ate statistics. Her tourism-related research interests cover tourism demand forecasting
model, tourist impact studies and tourism segmentation. She has served as a reviewer for
tourism journals including the Annals of Tourism Research and Tourism Management.
Carl Cater is a lecturer in tourism at Aberystwyth University, Wales. His research
centres on the experiential turn in tourism and the subsequent growth of special inter-
est sectors, particularly adventure tourism and ecotourism. He has undertaken field

vii
viii Handbook of research methods in tourism

research, supervision and teaching worldwide. He is a fellow of the Royal Geographical

Society, a qualified pilot, diver, lifesaver, mountain and tropical forest leader, and main-
tains an interest in both the practice and pursuit of sustainable outdoor tourism activity.
He has written over 20 papers and book chapters, is co-author (with Dr Erlet Cater) of
Marine Ecotourism: Between the Devil and the Deep Blue Sea (CABI, 2007), and is an
editorial board member of Tourism Geographies, Journal of Ecotourism and Tourism in
Marine Environments.
Sarath Divisekera is an Associate Professor in Economics at Victoria University,
Melbourne, Australia. Sarath has a Bachelor of Economics (First-Class Honors),
a Masters in Development Economics and a PhD in the economics of travel and
tourism. His recent research is mainly in the areas of tourism demand modelling and
consumption analysis, tourism policy and taxation, and tourism and climate change.
He has authored over 50 peer-reviewed articles and monographs, and has published
in most top-tier tourism journals. He is the recipient of the Vice Chancellors Citation
Award for Excellence in Research (2008) and the Faculty of Business and Law
Staff Research Award (2011). He was a visiting Professor at the National Centre
for Tourism Policy Studies at Limerick University, Ireland, and FH JOANNEUM
University of Applied Sciences, Austria. Sarath is an appointed member of the edito-
rial board of Tourism Economics.
Dianne Dredge is an associate professor in the School of Tourism and Hospitality
Management, Southern Cross University, Australia. She has 20 years experience as a
tourism and environmental planner in Queensland, New South Wales, Canada, Mexico
and China. Her work has included conceptual design and site analyses of large-scale
integrated resort proposals; integration of tourism in strategic and local area plans;
comparative analyses of competitive destinations; studies into the re-imaging of desti-
nations in crisis; and assessment of the environmental impacts of tourism. She has also
been involved in tourism capacity-building activities in local governments and tourism
organizations, including stakeholder audits and community consultation. Diannes
current research is in local government tourism management, place-based planning and
management of tourism destinations, capacity building in tourism organizations and
tourism planning and policy.
Larry Dwyer, PhD, is Professor of Travel and Tourism Economics in the Australian
School of Business at the University of New South Wales. Larry publishes widely
in the areas of tourism economics, management and policy, and has been awarded
numerous research grants to contribute to tourism knowledge. Larry is President of the
International Academy for Study of Tourism, President of the International Association
for Tourism Economics and a member of the International Advisory Board of the
Business Enterprises for Sustainable Tourism Education Network (BESTEN). He is an
appointed member of the editorial boards of 21 international tourism journals.
Eric W. Foemmel, PhD, is an independent researcher and writer living in California. Dr
Foemmel is a graduate of the Recreation, Park and Tourism Management Department
at Pennsylvania State University with a doctoral minor in anthropology. His primary
interest is the study of the American tattoo culture, and he has conducted ethnographic
research at Venice Beach and Traditional Ink in Los Angeles, California, to explore
 Contributors ix

the intrinsic motivations behind tattooing and its impact on social norms. He has also
collected the oral life histories of venerated tattooers who have heavily influenced the
American tattoo industry. Dr Foemmel has also established a research and publishing
company, Uptown Research, LLC.
Liz Fredline is a senior lecturer in the Department of Tourism, Leisure, Hotel and Sport
Management at Griffith University, Queensland, Australia. She gained her doctorate
at Griffith University. Her main area of research interest is the impacts of tourism and
events on host communities. She is also interested in quantitative research design and
analysis.
Alison Gill, PhD, is a professor at Simon Fraser University in Vancouver, British
Columbia, Canada, where she holds a joint appointment in the Department of
Geography and the School of Resource and Environmental Management. Alison has
published extensively on issues of tourism and the transformation of place. For many
years her research has focused on aspects of growth and change associated with tourism
in mountain resort communities. She serves on a number of editorial boards for both
geography and tourism journals and is a Fellow of the International Academy for the
Study of Tourism.
Ulrike Gretzel is Associate Professor for Marketing at the Institute for Innovation
in Business and Social Research, University of Wollongong, and Director of the
Laboratory for Intelligent Systems in Tourism (LIST). She received her PhD in com-
munications from the University of Illinois at Urbana-Champaign. Her research focuses
on persuasion in humantechnology interaction, experience design, use of technology
for interpretation, adoption and use of social media, interorganizational information
systems and other issues related to the development and use of intelligent systems in
tourism. She uses qualitative and quantitative research methods and has a particular
interest in network analysis and text mining.
Rob Hales is a lecturer in the Department of Tourism, Leisure, Hotel and Sport
Management, Griffith University, Australia. His research interests focus on social science
issues in a range of contexts that include sustainable tourism, ecotourism, outdoor rec-
reation, social movement studies and indigenous studies. His current research interest
is a community-based project aiming to understand urban and peri-urban Indigenous
Peoples vulnerability and adaptive capacity to climate change. The central theme
running through all his research projects is an emphasis on social and environmental
justice issues. His background in environmental science, eco- and adventure guiding as
well as outdoor environmental education informs his research and teaching.
Gayle R. Jennings, PhD, is the Director of Research, Imagine Consulting Group
International. Her research agenda focuses on practical and applied research for busi-
ness and industry, research training and education, qualitative methodologies, and
quality tourism experiences. Gayle is also Adjunct Professor of Tourism Management,
Department of Tourism, Leisure, Hotel and Sport Management, Griffith University, Gold
Coast Campus. She has sole authored and edited a number of books, written book chapters
and journal articles across a range of topics relating to theoretical paradigms that inform
research processes, education, waterbased tourism and quality tourism experiences.
x Handbook of research methods in tourism

Olga Junek is a lecturer in tourism and events management at Victoria University,

Melbourne. Her lengthy overseas work experience in education and tourism has given
her a broad international perspective, which permeates her teaching and her research
interests. She teaches and researches in a number of tourism and events-related areas,
including event and tourism education, internationalization of the curriculum and the
well-being of international students. She is currently undertaking her PhD in the area of
international students and their leisure and travel behaviour.
Heather Kennedy-Eden is a PhD candidate in the Institute for Innovation in Business
and Social Research at the University of Wollongong and a research assistant in the
Laboratory for Intelligent Systems in Tourism (LIST). She has 15-years experience
working in the tourism industry, specifically event management, hotel operations, envi-
ronment interpretation and the travel industry. Her research focuses on the influence of
smart phones on family bonding and family vacation, adoption and use of social media,
technology adoption in event management and other areas related to technology use in
tourism and involves qualitative as well as quantitative research methods.
Les Killion is Associate Professor in Tourism at Central Queensland University,
Australia, where he has served as Head of School of Marketing and Tourism. He has
undertaken research consultancies for regional tourism authorities and destination
marketing organizations in New South Wales and Central Queensland. His research
interests, founded primarily on qualitative approaches and paradigms, are in the areas
of tourism policy, impacts assessment, marketing and development and events manage-
ment.
Gerard Kyle is a professor within the Department of Recreation, Park and Tourism
Sciences at Texas A&M University. His research is informed by theory rooted in envi-
ronmental/conservation psychology and explores humans interactions and response to
nature and wildland areas. This work primarily relies on survey research methods and
multivariate modelling tools to construct models of human behaviour in varied contexts.
Applications have examined an array of human dimensions-related issues within the
United States and Australia related to wildfire management, coastal and inland fisheries
management, climate change, invasive species and other threats to parks and protected
areas. His peer-reviewed articles have been published in the Journal of Environmental
Psychology, Environment & Behavior, Society & Natural Resources and Leisure Sciences.
Jenny (Jiyeon) Lee is a lecturer of School of Marketing at the University of New South
Wales, Sydney, Australia. She received her doctoral degree in Recreation, Park, and
Tourism Sciences from the Texas A&M University. Her research interest lies in tourism
marketing, focusing on place meaning/attachment, tourist behaviours and psychol-
ogy and structural equation modelling. Her research has spanned a wide domain to
address managerial and marketing issues, both the public (i.e., park services) and private
sectors (i.e., travel agencies, festivals/events, and destination marketing organizations).
Her work is published in Journal of Travel Research, Journal of Travel and Tourism
Marketing, and Journal of Park and Recreation Administration.
Zheng Lei is a lecturer in the Department of Air Transport at Cranfield University, UK.
He received his doctorate from Surrey University. His main research interests are avia-
 Contributors xi

tion policy, business strategy and regional competitiveness. He is a specialist in panel

data analysis and has also served as a reviewer for several international journals such
as Annals of Tourism Research. He is Assistant Editor of the Journal of Air Transport
Studies.
Gang Li is a senior lecturer in economics in the School of Hospitality and Tourism
Management at the University of Surrey in the UK, where he also received his PhD in
2004. Before then, he studied economics and statistics in China. Dr Lis research inter-
ests include econometric modelling and forecasting of tourism demand and quantitative
studies of tourist behaviour. His research is published in leading tourism and forecasting
journals such as Annals of Tourism Research, Tourism Management, Journal of Travel
Research and International Journal of Forecasting. He is a co-author of the book The
Advanced Econometrics of Tourism Demand (2009).
Tiffany Low is a researcher based at the Institute for Tourism Research at the University
of Bedfordshire. She completed double degrees in Business and Hotel Management with
a specialism in financial economics. She is currently undertaking research for a PhD
examining how customers value ethical consumption in the hotel sector. Tiffany has
worked as a research assistant at universities in Australia and has experience project
managing research projects funded under the Commonwealth Governments Australian
Research Council; one involving the conceptualization and measurement of derived cus-
tomer value in relation to the timeshare industry; and the second in relation to the needs
of customers and employees following service failures.
Nancy Gard McGehee is an associate professor and J. Willard and Alice S. Marriott
Junior Faculty Fellow of Hospitality in the Hospitality and Tourism Management
Department at Virginia Tech, Blacksburg, VA, USA. She is a sociologist whose primary
research interests fall under the socio-cultural impacts of tourism development. Nancy
has explored this in a number of contexts, including tourism-related entrepreneurship,
the role of tourism in the cultivation of various forms of community capital, and volun-
teer tourism. While she has conducted research using a wide range of methods, Nancys
passion lies in qualitative research, as it allows for a more critical and interpretive per-
spective.
Clive Morley is Professor of Quantitative Analysis and Deputy Head (Learning and
Teaching) in the Graduate School of Business and Law at RMIT, where he teaches in the
MBA. His expertise lies in the areas of applied quantitative data analysis, tourism eco-
nomics, forecasting and strategic analysis techniques. He has published a series of papers
in leading journals on the theory and practice of tourism demand modelling and on the
impacts of global airline alliances on tourism. Other research has considered the entre-
preneurial climate in Australia, the (mis)use of performance indicators in Australian call
centres, the career progression of women in accounting, Australian football attendances
and why business academics remain in academia.
Andreas Papatheodorou is Associate Professor in Industrial and Spatial Economics
with Emphasis on Tourism at the School of Business Administration, University of the
Aegean, Greece. He is also an external examiner at Cranfield University. He received
his doctorate from the University of Oxford and commenced his academic career at the
xii Handbook of research methods in tourism

University of Surrey, UK. His research interests are related to competition, competitive-
ness and regional development in tourism with focus on air transport. He is also the
Editor-in-Chief of the Journal of Air Transport Studies and a resource editor for Annals
of Tourism Research.
Grace Bo Peng graduated from the School of Economics, Xiamen University, with
a Bachelor degree in finance and law. She was an exchange student with Feng Chia
University in Taiwan for one semester during her undergraduate study and joined the
School of Hotel and Tourism Management at the Hong Kong Polytechnic University as
an MPhil student in August 2010. Her MPhil research topic is A meta-analysis of inter-
national tourism demand elasticities and forecasting accuracy.
Sylvain Petit is an assistant professor at the Facult de Droit, dEconomie et de Gestion,
University of Valenciennes et Hainaut-Cambrsis (France) where he teaches international
trade, econometrics and history of economics. He is the manager of the Mobility and
Sustainable Development branch for the laboratory of the Institut de Dveloppement et
de Prospective (Institute of Development and Prospects) at University of Valenciennes.
He is an executive member of the board of the International Association for Tourism
Economics (IATE). He obtained his PhD from the University of Lille I. With co-authors
Dr J-J. Nowak and Dr M. Sahli, he was awarded best paper prize at the 3rd conference
of IATE in 2011. His research areas are tourism economics, international trade and
applied econometrics.
Tien Duc Pham, PhD, is a lecturer at the School of Tourism, Queensland University.
He has a strong background in impact analysis using the CGE modelling technique in
general, and has applied the technique to tourism research in recent years. He has been
heavily involved in the production of tourism satellite account at the state and sub-state
levels of the Australian Economy. Dr Pham has been awarded several research grants
from Sustainable Tourism Cooperative Research Centre, Australia. He also provides
consultancy advice on tourism issues to Australian government departments.
Clemente Polo is Professor of Economics at the Universidad Autnoma de Barcelona
since 1992. He obtained his BA in economics at the Universidad Complutense de Madrid
in 1972 and his PhD at the Universidad Autnoma de Barcelona in 1984. He served as
economic adviser to the President of Spain from 1990 until 1993. He has taught at other
Spanish universities. His research has focused on the study of the effects of fiscal poli-
cies, tourism shocks and public capital accumulation using inputoutput, SAM, CGE
and econometric models. His work has been published in prestigious international and
national journals and books.
Nuno F. Ribeiro, PhD, is a Research Associate with the Indigenous Peoples Health
Research Centre in Saskatchewan, Canada. Dr Ribeiros research interests are pri-
marily related to the comparative study of culture and behaviour within tourism and
leisure contexts, with parallel interests in the development and application of innovative
research methodologies. Dr Ribeiros educational background is in tourism planning
and development and cultural anthropology and he holds graduate degrees from the
Pennsylvania State University in University Park, United States in tourism manage-
ment and anthropology. He has been awarded several research grants, keeps an active
 Contributors xiii

research and publication agenda, and has presented in numerous scholarly conferences
and symposia.
Jaume Rossell gained his PhD in economics and business at the University of the
Balearic Islands (Spain) in 2001. He is Associate Professor at the Department of
Applied Economics and Associate Researcher at the Economic Research Center, and
teaches microeconomics and tourism economics at the same university. He is director
of the Master in Tourism and Environmental Economics. His main research interests
include tourism demand modelling, environmental economics and climate change. In
2009 he was awarded with the honour of Emerging Scholar of Distinction from the
International Academy for the Study of Tourism.
Carla Almeida Santos is an associate professor and Director of Graduate Studies in
the Department of Recreation, Sport and Tourism at the University of Illinois. Her
research programme is focused on the examination of communicative practices as a
means of addressing the socio-political and cultural impact of tourism on the worlds
people and cultures. She currently serves on the editorial boards of Annals of Tourism
Research, Journal of Travel Research, Journal of Hospitality and Tourism Management
and Tourism, Culture and Communication.
Neelu Seetaram is a lecturer in tourism development and impacts at Bournemouth
University, UK. She is a member of the editorial board of the Journal of Travel Research
and co-editor of a special issue of the Journal Tourism Economics. She is an executive
member of the Board of the International Association for Tourism Economics (IATE)
and a member of the German Aviation Research Society. She obtained her PhD at
Monash University (Australia) and recipient of the Hamburg Airport prize for best
PhD paper in 2009. Neelu has an MSc in economics and econometrics from Nottingham
University, UK, and a BSc (Hons) in economics from the University of Mauritius. Her
research interests are focused on the areas of tourism and airline economics, demand
modeling and economic impact studies.
Changsup Shim is a PhD candidate in the Department of Recreation, Sport and
Tourism at the University of Illinois. He completed a masters degree in urban and
regional planning at the Seoul National University in South Korea and has been
involved in multiple tourism research projects. His research is focused on the impactof
globalization, commercialism and urban development on the concept of place and
local identity. He approaches this issue through tourism and leisure phenomena that
take place in contemporary metropolitan areas. His research interests include: urban
tourism, cultural sustainability, authenticity, urban regeneration and postmodernism.
Haiyan Song is Chair Professor of Tourism and Associate Dean of the School of Hotel
and Tourism Management at the Hong Kong Polytechnic University. His research
interests include tourism demand modelling and forecasting, tourist impact assess-
ment, tourism supply chain management and tourist satisfaction studies. He has pub-
lished widely in such journals as Annals of Tourism Research, Tourism Management,
and Journal of Travel Research. Haiyan is a Fellow and First Vice President of the
International Academy for the Study of Tourism.
xiv Handbook of research methods in tourism

Svetlana Stepchenkova, PhD, is an assistant professor at the Department of Tourism,

Recreation and Sport Management at the University of Florida and an associate direc-
tor of the Eric Friedheim Tourism Institute. Her research interests include destination
management, marketing and branding, with a focus on quantitative assessment of des-
tination image using qualitative data. Svetlana studies the influence of media messages
on image formation, and destination image as a factor in explaining destination choice.
Svetlana is also interested in applications of information technologies in travel and
tourism, particularly virtual travel communities, destination websites and user-generated
content as a means of obtaining a competitive advantage in destination marketing and
management.
Dallen J. Timothy is Professor of Community Resources and Development and Senior
Sustainability Scientist at Arizona State University. He is also Visiting Professor of
Heritage Tourism at the University of Sunderland (UK) and Adjunct Professor of
Geography at Indiana University. Professor Timothy is the editor of the Journal of
Heritage Tourism and serves on the editorial boards of 11 international journals. His
tourism-related research interests include intangible heritage, religious tourism, visual
images, responsible tourism, geopolitics, supranational and globalization, and devel-
oping region dynamics. He currently has ongoing research projects in Israel/Palestine,
Western Europe, Central America, Mexico and South Asia.
Lindsay W. Turner is Professor of Econometrics in the School of International Business
at Victoria University, Australia. He is also a professorial research fellow at the Central
University of Finance and Economics, Beijing and Senior Research Fellow in the
Institute of the Chinese Economy, Peking University, Beijing. Lindsay is an internation-
ally recognized researcher and a specialist in tourism forecasting and tourism economics.
He is author of the Pacific Asia Travel Association (PATA) Forecasts published annu-
ally. Lindsay is a Fellow of the International Academy for the Study of Tourism and the
International Association of Scientific Experts in Tourism, special advisor on the edito-
rial board of the journal Tourism Economics, editorial board member of the Journal of
Travel Research and the China Tourism Research Journal.
Elisabeth Valle is Doctor Assistant at the Department of Applied Economics at the
University of the Balearic Islands (UIB). She has a Bachelors Degree in economics
from UIB (1998), and in 2004 she got her PhD from the same university with the thesis
Multisectorial models applied to the Balearic economy, a project awarded with the II
International Prize of Tourist Studies Gabriel Escarrer 2005. Her research is focused on
inputoutput tables, systems of social accounting and general equilibrium models.
Chau Vu is a senior lecturer in the Faculty of Business and Law, School of Accounting
and Finance at Victoria University, Melbourne. Her research interests include tourism
economics and forecasting, applied quantitative data analysis and financial economet-
rics. She has published in journals such as the Journal of Tourism Research, Tourism and
Hospitality Research and Tourism Economics.
Stephen F. Witt is Emeritus Professor (Tourism Forecasting) in the Faculty of Business,
Economics and Law at the University of Surrey, UK, and a consultant to the Pacific Asia
Travel Association (PATA) based in Bangkok, Thailand. His main research interests
 Contributors xv

are econometric modelling of international tourism demand and assessment of the accu-
racy of different forecasting methods within the tourism context. He has published 150
journal articles and book chapters as well as 30 books. The latter include: The Tourism
Marketing and Management Handbook; The Management of International Tourism;
Modeling and Forecasting Demand in Tourism; The Advanced Econometrics of Tourism
Demand; Tourism Forecasts for Europe 20012005; and Asia Pacific Tourism Forecasts
20112013.

DWYER 9781781001288 Part I.indd xv 25/07/2012 11:06

 Preface

The chapters in this book address the most important established and emerging qualita-
tive and quantitative research methods in tourism. There are many books on research
methods, including research methods in tourism. The question thus arises: why another
book? Our response to this question is that we are not aware of any book that structures
the material on research methods in the way that the authors have done for this book.
Each chapter is structured to provide detailed overview of the nature of the research
method, its use in tourism, its advantages and limitations and future directions for
research.
The process began with a determination of the major research methods that are used
by tourism researchers, quantitative, qualitative and mixed. While we believe that we
have selected the major research methods in current use, we are aware that some readers
will be disappointed that one or other research method has not been included. To this
we must plead that any book such as this has limitations on length. While we acknowl-
edge that the book does not address all research methods used by tourism researchers,
the content does capture the methods that are most used at the present time.
The next step was to invite authors to contribute to the book. All are active researchers
in tourism and all have international standing in the discipline. All have published works
that use the technique that they write about. The 42 authors are based in 30 universities
in eight countries giving the book a truly international perspective. A biography for each
contributor is given at the beginning of the book.
An important element of the book is that all of the authors were required to structure
their chapter in the same way.

1. Nature of the technique and its evolution

2. Background and types of problems that the technique is designed to handle
3. Applications of the technique to tourism, including discussion of studies that have
used the technique and their findings
4. Advantages and limitations of technique conceptually and for policy formulation
5. Further developments and applications of the technique in tourism research

We believe that this structure makes the contents of each chapter much more informative
to the reader, providing a comprehensive discussion of the technique itself as well as its
application in tourism and related contexts. We are confident that the imposed structure
will result in the individual contributions making an important contribution to tourism
studies, ensuring that each will be highly informative and widely referenced in the litera-
ture. The chapters are constructed in a way that they provide a detailed overview of the
different techniques irrespective of their tourism applications. In this way the volume
should appeal to social scientists in general and not just to researchers in tourism. We
would like to take this opportunity to thank each of the contributors for the professional
way that they approached their tasks and for the high quality of their contributions.

xvi

DWYER 9781781001288 Part I.indd xvi 25/07/2012 11:06

 PART I

QUANTITATIVE RESEARCH
METHODS

In the last two decades the application of quantitative techniques in the study of the
tourism phenomenon has gained momentum. This can be traced back to several factors.
The most prominent is perhaps the fact that, as the reliance of destinations on tourism
has grown and the industry expanded globally, more resources have been devoted to the
collection of quantitative data and the maintenance of tourism data sets. This may have
encouraged researchers interested in quantitative data analysis to give higher priority
to the tourism industry in their research agenda. At the same time stakeholders in the
industry, including destination managers and local and federal governments, keen to
make more informed decisions, by devising better policies and evaluating existing ones
are paying more attention to results from quantitative research. Consequently, the appli-
cation of quantitative methods in tourism data analysis has become more pronounced in
academic and non-academic research enriching the literature.
Ideally, this section would have addressed all the various quantitative techniques used
in tourism research. However, limited by space and time, the contributions in this section
of the book do not address the full range of quantitative research methods in tourism.
They do, however, cover the most important ones used by tourism researchers. This
introduction does not do justice to the detailed arguments and comprehensive treat-
ment of the issues that comprise the content of the different contributions. It is intended,
hopefully, to whet the readers appetite for the much more detailed discussion in each
chapter.
As highlighted by Gang Li in Chapter 1, Statistical testing techniques are key in
quantitative tourism research. A statistical hypothesis test is a method of making deci-
sions using data, whether from a controlled experiment or an observational study (not
controlled). In statistics, a result is called statistically significant if it is unlikely to have
occurred by chance alone, according to a pre-determined threshold probability, the
significance level. Given different assumptions concerning the distribution of the popula-
tion, statistical techniques can be broadly divided into parametric and non-parametric
groups. Both parametric and non-parametric tests have their advantages as well as limi-
tations. Tourism researchers are interested in examining the behavioural characteristics
of consumers in a variety of tourism contexts, such as tourists perception of destination
images, travel motivation, behavioural intention, perceived service performance and
satisfaction. In addition, some researchers are interested in the perceptions, attitudes and
behaviours of residents in tourist destinations. Statistical testing is regularly performed

DWYER 9781781001288 Part I.indd 1 25/07/2012 11:06

 2 Handbook of research methods in tourism

in these tourism studies in order to obtain robust findings and conclusions. Li argues
that, in general, parametric tests are more popular than their non-parametric alternatives
in tourist behaviour studies, mainly because they are relatively more robust and capable
of handling complicated research designs. By introducing more and more newly devel-
oped, robust, non-parametric techniques into future tourism research, he claims that the
accuracy and reliability of research findings can be further improved.
As Jaume Rossell indicates in Chapter 2, Regression analysis includes many tech-
niques for modeling and analysing several variables, when the focus is on the relation-
ship between a dependent variable and one or more independent variables. Regression
analysis, which is being increasingly used in both demand and supply side analyses in
tourism, helps one understand how the typical value of the dependent variable changes
when any one of the independent variables is varied, while the other independent vari-
ables are held fixed. When a causal relationship between a dependent variable and one
or more independent variables is hypothesized, regression analysis comes to the fore as
a technique for use in circumstances where a particular phenomenon is controlled and
determined by certain factors, and the aim is to try to explain and quantify variations
in the dependent variable through variations in the independent ones. Most commonly,
regression analysis estimates the conditional expectation of the dependent variable given
the independent variables that is, the average value of the dependent variable when the
independent variables are held fixed. Less commonly, the focus is on a quantile, or other
location parameter of the conditional distribution of the dependent variable given the
independent variables. In all cases, the estimation target is a function of the independ-
ent variables called the regression function. In regression analysis, it is also of interest
to characterize the variation of the dependent variable around the regression function,
which can be described by a probability distribution. As Rossell emphasizes, however,
correlation does not imply causation. This is because a spurious relationship is always
possible between variables where some degree of covariance can be found.
In Chapter 3, Cang and Seetaram provide an expos of Time series analysis as applied
in the tourism context. They first discuss the properties of time series data, such as sta-
tionarity, seasonality and existence of structural breaks in the data. These properties are
important as they provide information on how the data behave under different circum-
stances. For example, the effect of a shock, such as a terrorist attack on tourists, will have
a permanent effect on a destination if the data on international arrivals is non-stationary.
These properties also assist the researchers in their choice of modelling techniques. The
basis for time series modelling is that the current data is dependent on their historical
values. The aim of time series modelling is to predict the future values of the data with as
low margin of error as possible.
Following this, the authors discuss model selection to predict future trends in the data
set, as well as estimation techniques. The review of literature on time series techniques
shows that newer models are better at prediction but there is no single class of models
which dominates in terms of accuracy in forecasting.
This argument is further expanded by Grace Bo Peng, Haiyan Song and Stephen Witt
in Demand Modeling and Forecasting, the topic of Chapter 4. Tourism demand is nor-
mally measured by either tourist arrivals in a destination, tourists expenditure when they
visit a destination or tourist nights stayed in the destination. The variables that have been
generally accepted as the main determinants of international tourism demand comprise

DWYER 9781781001288 Part I.indd 2 25/07/2012 11:06

 Part I Quantitative research methods 3

potential tourists income levels, the relative price of tourism products in the origin and
destination countries, substitute prices of tourism products in alternative foreign desti-
nations, transportation cost, population of origin country, exchange rates, marketing
expenditure by the destination in the origin country and one-off events (which can have
a positive or negative effect). Quantitative forecasting methods organize past tourism
demand information by mathematical rules and there are three main subcategories: time
series models, econometric approaches and artificial intelligence methods. Within these
categories there are numerous different approaches as outlined by the authors. Peng,
Song and Witt note that past empirical findings show that no single forecasting method
is superior to others in all situations. The forecasting performance of the models varies
considerably depending on the forecast error measure, the forecasting horizon, data
frequency, destinationorigin country pairs and the competitors included in a compari-
son. An ongoing objective of tourism forecasting is therefore to find optimal forecasting
models according to the data characteristics. Future research directions identified by the
authors include use of meta analysis to synthesize the research findings of the previous
studies over the past four to five decades with a view to generalizing the research findings
based on the characteristics of the data used in the tourism forecasting literature. The
authors also recommend a forecasting competition based on a wide range of forecasting
methods used in the past literature with the involvement of the experts who have pub-
lished extensively using the quantitative methods highlighted above. In conclusion, the
integration of judgmental (qualitative) forecasts and quantitative forecasts would also
be a useful direction for future research as this research topic has been relatively scarce
in the tourism context.
In Chapter 5, Jiyeon (Jenny) Lee and Gerard Kyle discuss Structural equation
modeling, a statistical technique for testing and estimating causal relations using a
combination of statistical data and qualitative causal assumptions. Structural equa-
tion models (SEM) combine both multiple regression and factor analysis (see Chapter
10). Structural equation models go beyond ordinary regression models to incorporate
multiple independent and dependent variables as well as latent constructs that cluster
the observed variables they are hypothesized to represent. They also provide a way to
test a specified set of relationships among observed and latent variables as a whole, and
allow theory testing even when experiments are not possible. As a result, these methods
have become ubiquitous in all the social and behavioural sciences. Factor analysis, path
analysis and regression all represent special cases of SEM. With an advance of statisti-
cal software with graphical user interfaces, several user-friendly SEM programs have
become accessible to tourism researchers. Of these, the most widely used programs are
LISREL (Linear Structural Relationships), AMOS (Analysis of Moment Structures),
EQS (Equations), Mplus, and PLS (Partial Least Squares). In tourism research, the
technique has primarily been limited to: (1) the use of confirmatory (and sometimes
combined with exploratory) factor analysis to determine scale structure, psychometric
properties, and for scale purification; (2) testing causal relationships among latent (and
sometimes manifest) variables; and (3) testing for measurement and structural model
invariance across select groups. Lee and Kyle identify several advantages of SEM over
other analytical methodologies for hypothesis testing. Given the analysis can be applied
to both non-experimental and experimental data, it provides researchers with modeling
flexibility; an array of analytical capabilities (i.e., multi-group invariance testing); and

DWYER 9781781001288 Part I.indd 3 25/07/2012 11:06

 4 Handbook of research methods in tourism

the ability to account for measurement error. Limitations of SEM are also highlighted
including that SEM requires a priori specifications, which implies that model develop-
ment and modification should be directed by theory and empirical evidence making
it inappropriate for exploratory research where the measurement structure is not well
defined or where guiding theory and empirical evidence underlying patterns of construct
relationships is not well established. Another limitation of the use of SEM and the use
of the maximum likelihood estimation is associated with its inherent assumption of
multivariate normality and asymptotic theory that requires large sample sizes. Further
developments within SEM are identified and discussed by the authors to determine their
significance for tourism research.
Clive Morley discusses Discrete choice analysis and experimental design in Chapter 6.
Choice modelling attempts to model the decision process of an individual or segment in a
particular context. Choice modelling is regarded as an accurate and general purpose tool
available for making probabilistic predictions about certain human decision-making
behaviour. The methods of discrete choice analysis are sometimes referred to as random
utility methods (as the underlying econometric theory uses a random utility specifica-
tion) or limited dependent variable analysis (the dependent variable in the model is
categorical). Logit, multinomial logit and probit models are particular cases of discrete
choice analysis.
Tourists make many decisions in determining their trip, and many of these decisions
are intrinsically categorical, multinomial (many options are available) and unordered.
Discrete choice theory provides an appropriate and sophisticated framework for analysis
of the data at the level of individuals evoked responses, and one that has been proven to
work well in practice. The obvious applications of discrete choice methods are in demand
modelling, and especially in assessing the impact of potentially explanatory variables on
demand. This then gives useful results for marketing, forecasting, planning and policy
evaluation and development, and other decision making by operators. For example,
the technique has been used to estimate visitors willingness to pay for attractions and
substitution rates, that is the rate at which people will trade off less of one attribute for
more of another attribute, for a particular destination. Other applications range widely
from analysing various price effects on demand for a single destination from one through
wider investigations of demand across combinations of origins and destinations (and
types of destinations), to particular tourism preferences. Morley provides many different
examples of the use of this attractive and valuable technique.
Overall the combination of experimental design, stated preference data and discrete
choice models form a powerful body of techniques for analysis of tourism demand. The
potential explanatory variables used are driven by the purpose of the investigation, and
vary widely in the reported studies The more sophisticated choice models derived from
discrete choice theory quantify the impact of variables for detailed policy analysis, stra-
tegic market planning and economic analysis.
As indicated by Neelu Seetaram and Sylvain Petit in Chapter 7, Panel data analysis is
an increasingly popular form of longitudinal data analysis among social and behavioural
science researchers. A panel is a cross section or group of people who are surveyed peri-
odically over a given time span. Panel data analysis is a method of studying a particular
subject within multiple sites, periodically observed over a defined time frame. Panel
analysis has enabled researchers to undertake longitudinal analyses in a wide variety of

DWYER 9781781001288 Part I.indd 4 25/07/2012 11:06

 Part I Quantitative research methods 5

fields in the social sciences. With repeated observations of enough cross sections, panel
analysis permits the researcher to study the dynamics of change with short time series.
Panel data analysis endows regression analysis with both a spatial and temporal dimen-
sion. The spatial dimension pertains to a set of cross-sectional units of observation.
These could be countries, states, counties, firms, commodities, groups of people or even
individuals. The temporal dimension pertains to periodic observations of a set of vari-
ables characterizing these cross-sectional units over a particular time span.
Seetaram and Petit divide the empirical literature based on the panel data methods
and applied to the tourism sector, into two broad areas. The first concerns the deter-
minants of tourism demand. The study of international tourism arrivals and receipts
is a vibrant area of research and it cannot be confined to gravity models due to factors
such as seasonality, cultural links or time constraints which are particular to the con-
sumption of tourism products. The second area consists mainly of analysis of the link
between tourism and economic growth. Since, several destinations have simultaneously
experienced economic development and expansion of their tourism sector. While panel
data techniques are powerful, and generally yield reliable estimates, they nevertheless
suffer from some weaknesses. Estimations can be complex and data requirements are
fairly large, especially for dynamic panel models. The main limitation, however, is that
the panel data survey design is inevitably subject to problems related to attrition. The
latter drawback, however, does not seem to have affected tourism research as the existing
studies are based on secondary data or macro data where the occurrence is either easier
to deal with or less frequent. It is expected that future research may be based on survey
data when researchers seek to find answers to the behaviour of tourists using micro-level
data. Seetaram and Petit conclude that the use of panel data modelling techniques is
becoming more common in the analysis of tourism data. In less than a decade, the lit-
erature has progressed from the development of standard static fixed effect and random
effect models to the more sophisticated dynamic panel data cointegration models. So
far, the application of this technique has been restricted to tourism demand studies and
others which have explored the relationship between the expansion of tourism sectors
and economic growth or international trade. It can be extended to other topics of
research within the tourism context, including supply side studies.
In Chapter 8, Sarath Divisekera reviews the specification and applications of The
almost ideal demand system (AIDS). Since its introduction in 1980 as a technique to
analyse consumer behaviour, AIDS has become an essential aspect of demand theory
and has been widely used in numerous empirical studies. The attraction of AIDS is that
it gives an arbitrary first-order approximation to any demand system; it satisfies the
axioms of choice; it aggregates perfectly without invoking the assumption of parallel
linear Engel curves; and it has a functional form which is consistent with known house-
hold budget data. As Divisekera states, the AIDS model has been the principal and most
widely used established demand system employed by researchers in the tourism field.
The available empirical studies may be broadly classified into two categories: destination
choice or demand for tourist destinations; and studies that seek to analyse the demand
for tourism goods and services. The former group of studies attempts to evaluate the
effects of incomes and prices as determinants of tourism demand for different destina-
tions including domestic destinations. This has been the central focus of most of the
early empirical studies of tourism demand. The latter group of studies investigates those

DWYER 9781781001288 Part I.indd 5 25/07/2012 11:06

 6 Handbook of research methods in tourism

aspects of tourism demands that focus on the demand for destination-specific tourism
goods and services. More specifically, they attempt to analyse tourists consumption
behaviour or allocations of tourists spending among various goods and services at the
destinations they visited. Most attention among researchers in recent times appears to be
on keeping up with the changing econometric methodology, rather than on improving
the modelling process. For example, few studies have attempted to incorporate the cost
of transport into modelling international tourism demand. Similarly, most studies con-
tinue to use the CPI (adjusted for exchange rates) as a proxy for tourism prices. This is a
major drawback, as this grossly overestimates or underestimates the real cost of tourism
goods and services. Thus, there is a need to develop appropriate tourism price indices as
inputs for demand model estimation. Future studies need to concentrate on improving
databases and developing reasonable price measures that capture both the costs of travel
and tourism goods which are the key determinants of tourism demands.
Hedonic price analysis (HPA) is the topic of Chapter 9. Andreas Papatheodorou,
Zheng Lei and Alexandros Apostolakis note that while pricing is a strategic choice for all
firms, pricing decisions are often complex. From a managerial perspective, it is critically
important to understand the consumer perceptions of each of the attributes associated
with the characteristics that a customer is willing to make an extra payment for and those
which are irrelevant in the determination of consumer choices and preferences.
HPA, an economic valuation technique based on revealed preferences, has its origin in
Lancasters (1971) theory of consumer demand, wherein consumer products are regarded
as essentially bundles of separable characteristics such as comfort, convenience and
reliability that can be assessed accurately by value-conscious consumers. The theory
of HPA was formally formulated by Rosen using a conventional utility-maximizing
approach to derive implicit attribute prices for multi-attribute goods under conditions of
perfect competition. The value that consumers attach to the characteristics is reflected in
the price of the differentiated product. Thus, HPA is used to estimate economic values
for attributes or product characteristics that directly affect prices for tourism products
and services in the marketplace.
HPA aims to disentangle the impact of the various attributes on implicit prices where
positive attributes are expected to boost the overall price and have a positive effect on
individual utility levels. Correspondingly, negative attributes exert downward pressure
on the price and have a negative effect on utility levels. Seen in this way, appropriate
pricing not only generates revenue for a company to survive but can also be used as a
communicator, as a bargaining tool and a competitive weapon. The consumer can use
price as a means of comparing products, judging relative value for money or product
quality. Hence, this technique is particularly useful for managerial decision making and
evaluating individual preferences. HPA makes it easier to discern which characteristics
are valued by consumers and to what extent, providing useful insight for the strategy to
be followed in terms of market positioning, pricing and branding.
The authors argue that the methodological reliance of the technique on Lancasters
characteristics approach, make hedonic price analysis particularly adept to the study of
a multi-attribute and multi-value sector such as the tourism industry. HPA establishes
surrogate markets to place a premium to the effect of non-market based attributes on
the tourism product or service on offer. Regression techniques are then used to estimate
the implicit price for each attribute with the parameter of the hedonic price function

DWYER 9781781001288 Part I.indd 6 25/07/2012 11:06

 Part I Quantitative research methods 7

revealing the marginal value consumers place on each of the individual attributes that
comprise tourism experiences. The recent convergence of tourism and non-market envi-
ronmental resources (illustrated through the increasing popularity of the sustainability
concept in tourism operations, environmentally friendly tourism practices and ecotour-
ism resources) necessitates the need for an evaluation approach that would allow esti-
mation of the associated value placed by consumers/tourists on these non-market based
characteristics of the tourism product or service on offer. Again, HPA is well suited to
accommodate this need. The technique is based on the construction of surrogate markets
to evaluate the effect of each product or service attribute on implicit prices consumers/
tourists are willing to pay. In this respect, and despite some caveats, HPA offers a valid
alternative method of examining demand patterns compared to other (market-based)
methods.
As stated by Lindsay Turner and Chau Vu in Chapter 10, Factor analysis is a complex
methodology that provides an objective way of determining latent (hidden) concepts
within social data. Principal Component Analysis (PCA) is the most common use of
factor analysis because it is descriptive, providing significant insight into the latent struc-
ture of data that can often be used for further analysis, for example in structural equation
modelling (SEM). Confirmatory Factor Analysis (CFA) is used to determine whether a
particular set of hypotheses defining the number and type of factors applies to particu-
lar data. The hypotheses require specific theoretical definition presumably justified by
previous research, or theoretical construction such as a conceptual framework or defini-
tion from PCA. At the very least the factors must be named in advance on the basis of
theory, and possibly the variables defining these names at least suggested. As such, CFA
is potentially more powerful in deriving conclusions to research problems. The method
analyses only the variance in the correlation matrix that is shared (common) with other
(not necessarily all) variables. Consequently, the number of factors extracted may be less
than the number of variables, which in research terms is more parsimonious in defining
the original data set as founded on underlying structure. The problem lies in determining
the shared variance (covariance) in the first place.
Factor analysis is an important tool for analysing tourism data, and potentially solving
tourism-related problems. It is fundamental to a wide variety of problems encountered in
tourism because, as a social science, there is significant requirement to measure unstruc-
tured concepts, as opposed to structured items with readily measurable attributes. In
tourism the concepts requiring measurement relate to social systems involving values,
attitudes, motivations, risks, satisfaction and beliefs. Turner and Vu highlight many
examples of factor analysis used in tourism research, but argue that more needs to be
done developing hypothetical structures from previous studies and confirming constant
relationships in future research. In order to aid this process there needs to be agreement
and understanding of the basic elements from analyses that are to be reported, what
statistics and survey elements should be clearly evident, and what standards of interpre-
tation are required. They conclude that factor analysis will continue as a major tourism
research tool but there is room for improvement in the use of the technique to raise the
standard of research analysis and consequent findings.
Chapter 11 addresses Cluster analysis. Clustering is the task of assigning a set of
objects into groups (called clusters) so that the objects in the same cluster are more
similar (in some sense or another) to each other than to those in other clusters. As Liz

DWYER 9781781001288 Part I.indd 7 25/07/2012 11:06

 8 Handbook of research methods in tourism

Fredline indicates, cluster analysis is a family of multivariate techniques useful for

analysing cases based on their scores on a range of measured variables. Essentially the
technique identifies cases with a comparable pattern of responses that can be regarded,
for the purposes of the analysis, as similar. In so doing a large number of individual cases
are summarized into a smaller, more manageable number of clusters. The outcome of a
successful cluster analysis is a small number of highly homogeneous clusters that are sub-
stantially different to each. Cluster membership can then be used as a variable for further
analysis aimed at understanding the clusters and the bases on which they were formed. It
is useful in any situation where the researcher wants to identify groups of cases that are
similar to each other but different from other groups, that is, segmentation of the popula-
tion. This segmentation can be based on any variables that are regarded as important for
discriminating between groups and could include simple demographic characteristics or
psychographic characteristics such as attitudes, perceptions, and motivations. Fredline
emphasizes that cluster analysis is an interdependence rather than a dependence tech-
nique. This means that it does not aim to assess the relationship between independent
and dependent variables; rather it is focussed on the underlying structure of variables
and the usefulness of this structure in categorizing cases. Interpretation of the results of
cluster analysis must be based on researcher knowledge and experience, as well as the
logic of the solution and the extent to which clusters demonstrate high internal homoge-
neity and high external heterogeneity.
As Fredline notes, the two most common applications of cluster analysis in tourism
are market segmentation and segmentation of host community attitudes or perceptions
toward tourism. Market segmentation analysis allows tourism researchers and opera-
tors to identify important sub-groups of the total market which have different needs and
wants, including travel motivations and therefore to modify their product to specifically
target selected markets or alternatively to attempt to meet the differing needs of all seg-
ments. Segmentation of the host community is associated with research into perceptions
of the impact of different types of tourism developments and entertainment options and
the holding of special events. The assessment of how residents of a host community feel
about tourism has seen numerous applications of cluster analysis with the aim of under-
standing the impacts of tourism on host communities.
Fredline argues that the main advantage of cluster analysis is that it provides an effec-
tive and meaningful reduction of data. All tourism customers have slightly different
needs and wants but it is impossible to treat them all as individuals. For this reason it
has been common practice to either assume they all have the same needs or to segment
them based on geographic or demographic variables. Cluster analysis provides a better
segmentation approach that allows groups to be defined by their needs and therefore
identifies groups who can be targeted in a similar way because they are genuinely similar.
One of the major disadvantages of the technique in terms of its application is that the
solution is not definitive. That is, there is no single correct solution, all possible cluster
solutions are correct, it is just that some do not explain the data in a manner that is as
easy to interpret and use.
As discussed by Clemente Polo and Elisabeth Valle in Chapter 12, Inputoutput
and SAM models are tools that economists have used for many decades to measure
sectoral interdependencies, compare the economic structure of economies, quantify
production impacts, measure structural and productivity changes, study the effects

DWYER 9781781001288 Part I.indd 8 25/07/2012 11:06

 Part I Quantitative research methods 9

of redistribution policies, calculate the energy content of commodities, estimate CO2

emissions, etc.
The inputoutput (IO) model is a simple linear model first proposed by Leontief to
determine production quantities and prices in a set up where commodities are produced
with commodities. IO models (or approaches inspired by it), incorporating multiplier
analysis, have been used since the 1960s to quantify the impact of international tourism
in large, medium and small national economies as well as in regions, counties, cities
and recreational areas worldwide. The typical study included the completion of two
surveys: one directed at local businesses to determine their cost structure and the other
to incoming tourists to estimate their expenditures. The size of multipliers depends on
three factors: the importance of import leakages, the value-added coefficients of tourism
oriented sectors and the strength of linkages between tourism-oriented sectors and other
sectors. Many of the tourism studies undertaken provide other useful pieces of informa-
tion on the economies studied of great interest for policy makers: backward linkages of
tourism sectors; the relative importance of tourism with respect to other exports; gov-
ernment revenues derived from tourism; imports required to satisfy tourists demands
and multipliers for different tourist services (accommodation types, non-real estate
renting, food, entertainment, car renting, etc.) and different tourists types (residents and
non-residents classified by origin or trip motivation). Some studies have also estimated
the production, value-added and employment shares accounted for tourists demand
and measure the weight of tourism under alternative assumptions on final demand. IO
models have been extensively criticized, however, because of their restrictive assump-
tions and disregard for capacity restrictions that lead to exaggerated economic impacts.
As Polo and Valle indicate, the extension of the IO table and Leontief s open model
into a social accounting matrix (SAM) and SAM model, respectively, was a quite natural
development. A SAM may include accounts for industries, commodities, institutional
sectors (households, non-profit institutions, corporate sector, public administrations
and foreign sectors), capital (savingsinvestment) and as many auxiliary accounts as
needed. Polo and Valle argue that SAM models have two main advantages over IO
models. First, the household(s) and capital (savingsinvestment) accounts are balanced
and the induced effects caused by income generation (consumption and savingsinvest-
ment) can be analysed in a more consistent way than in extended IO models. Second,
SAM models can analyse the distributive consequences of injections, a subject that has
no room in the IO models. There are disadvantages too. On the one hand, SAM models
require a SAM and only a few statistical offices that routinely elaborate IO tables for
national or regional economies also compile SAMs.
SAM models have been employed to study tourism impacts in national, regional and
small economies during the last two decades. Since expenditure coefficients are ratios of
SAM flows to column totals, the only requirements to estimate tourism impacts with a
SAM model are a SAM of the economy studied and a vector of tourists impacts. The
fact that many national statistical offices in developed economies publish IO tables along
with national accounts but only a few of them elaborate SAMs has hampered its use in
tourism studies. For many developing countries, SAMs have been assembled to explore
the links between growth, inequality and employment, and the associations between
poverty and savings and investment levels, balance of payments, production and distri-
bution. SAMs constructed for these purposes can be employed to quantify the role of

DWYER 9781781001288 Part I.indd 9 25/07/2012 11:06

 10 Handbook of research methods in tourism

tourism in the economy and its impact on all endogenous accounts including households.
While the extension of fix coefficients beyond the production sphere is even more ques-
tionable than in the IO model, SAM models are conceptually more satisfactory than IO
models since the households and capital accounts are balanced. The authors conclude
that all things equal SAM models are preferable to IO models. Overall, the results of
IO and SAM models can be viewed as a particular case of more general models and it is
not a coincidence that the SAM framework is also known as fix-price multiplier analysis.
Simulation results from IO and SAM models should always be interpreted with caution.
Polo and Valle conclude with some observations on the importance of IO and SAM
models to the development of tourism satellite accounts and estimates of energy con-
sumption and the environmental impacts of tourism. They see a continuing role for these
approaches in tourism research.
CGE modeling is the topic of Chapter 13 by Larry Dwyer and Tien Duc Pham.
Computable general equilibrium (CGE) models are simulations that combine the
abstract Walrasian general equilibrium structure, formalized by Arrow, Debreu and
Hahn, with realistic economic data to solve numerically for the levels of supply, demand
and price that support equilibrium across a specified set of markets. CGE modeling
involves a mathematical specification of simultaneous relationships within the economy.
Many features can be built into this framework, to tailor it to the different conditions
that characterize alternative real world circumstances. In this way, the framework pro-
vides a widely encompassing means of evaluating the effects of policy changes and exog-
enous shocks on resource allocation and can also permit assessment of the distributional
effects of such changes. CGE models treat an economy as a whole, allowing for feedback
effects of one sector on another. They represent the economy as a system of flows of
goods and services between sectors.
A CGE model is a versatile structure, which can easily be adapted to the special
features of any economy, to the questions that are asked, and to the data which are
available. Its assumptions can easily be changed so that the results of a simulation can
be interpreted, for instance, as describing a short-term equilibrium, or a medium-run
equilibrium. The majority of CGE models can be separated into two broad categories,
comparative static and dynamic. For policy analysis, results from such a model are often
interpreted as showing the reaction of the economy in some future period to some exter-
nal shock or policy change that the adjustment path is not known. Dynamic CGE models
explicitly trace each variable through time, often at annual intervals so that the adjust-
ment path of the economy can be examined.
CGE models are now increasingly used in tourism economics analysis and policy
formulation. CGE models are well suited to tourism analysis and policy, given their
multi-sectoral basis. In contrast to partial equilibrium approaches, CGE models can take
account of the interrelationships between tourism, other sectors in the domestic economy
and foreign producers and consumers. CGE modeling can be tailored to alternative con-
ditions, such as flexible or fixed prices, alternative exchange rate regimes, differences in
the degree of mobility of factors of production and different types of competition. CGE
models can be used to quantify the effects of actual policies, such as changes in taxa-
tion, subsidies or government borrowing, as well as predicting the effects of a range of
alternative policies or exogenous expenditure shocks. They can be used to estimate the
impacts of changes in tourism expenditure under a range of alternative macroeconomic

DWYER 9781781001288 Part I.indd 10 25/07/2012 11:06

 Part I Quantitative research methods 11

scenarios; tailoring them to alternative conditions such as flexible or fixed prices, various
exchange rate regimes, differences in the degree of mobility of factors of production, dif-
ferent government fiscal policy stances and different types of competition. CGE models
are helpful to tourism policy makers who seek to use them to provide guidance about
a wide variety of What if? questions, arising from a wide range of domestic or inter-
national expenditure shocks or alternative policy scenarios.
Despite its growing importance in tourism research, CGE modeling can be applied
to a much broader set of policy issues than at present. One agenda for future research
in this area should be to extend the analysis to different tourism destinations, and to
include detailed analyses of the appropriate behavioural characteristics of the economic
agents that are included in model specification and of the government policy settings
that determine the context for their behaviour. This could include the effects of foreign
direct investment in tourism, tourism productivity and competitiveness, fiscal policies
for tourism, policies within wider international groupings such as the European Union,
ASEAN and so on, policies for transportation, the environment and related externalities,
including the impacts of climate change on tourism. CGE tourism modeling provides a
versatile and effective means of examining the wide range of scenarios that can occur.
Costbenefit analysis (CBA) is discussed in Chapter 14 by Larry Dwyer. Foreshadowed
by Dupuit in the nineteenth century, CBA is the primary technique used for the economic
appraisal of actions or proposals in terms of economic efficiency. A CBA provides an
estimate of the worth of a proposal relative to an accompanying estimate of what would
happen in its absence. CBA is a systematic process for identifying and assessing all costs
and benefits of a proposal in monetary terms, as they are expected to occur through the
life of a project. CBA is concerned with measuring, the change in all sources of economic
welfare, whether occurring in markets or as implicit values. Future costs and benefits are
discounted relative to present costs and benefits in a net present value sum. The policy or
project is deemed to be socially acceptable if the sum of the benefits to society (including
private and social benefits) exceeds the sum of the costs to society (including private and
social costs).
In CBA, value or benefit is measured by willingness to pay (WTP) or willingness
to accept (WTA). Formally using the compensating variations principle, the net social
benefit is the maximum net amount that residents would be willing to pay for the pro-
posal and be just as well off with the proposal as without it. Using the equivalent varia-
tions principle, the net social benefit is the minimum amount that the community would
be willing to accept as compensation for not having the proposal. The social costs of a
project are measured in terms of opportunity costs that is, the value of the marginal
benefits foregone from the same resources in alternative uses. Using these two valua-
tion principles (for benefits and costs) the analyst can determine whether the value of
consumption gained is greater than the value of consumption that it given up. The net
benefit is the sum of all welfare benefits less costs. Maximizing net welfare is the stand-
ard policy objective implicit or explicit in costbenefit studies. By quantifying the net
benefits of projects, programmes and policies in a standard manner, CBA improves the
information base for public sector decision making, thereby assisting in the assessment
of relative priorities.
As the most comprehensive of the economic appraisal techniques, CBA is particularly
important in the context of evaluating tourism policy, programmes, regulations, projects

DWYER 9781781001288 Part I.indd 11 25/07/2012 11:06

 12 Handbook of research methods in tourism

and developments. Specific examples might include regional or local tourism plans;
rezoning of land for tourism purposes; and major tourism developments such as the
creation of tourism shopping precincts, airport development, resorts and hotels, nature
reserves and sporting facilities. As such projects generally have relatively wide economic,
environmental and social implications for a community that are not captured in the
basic financial analysis undertaken by project proponents. CBA is especially appropriate
to tourism because there is often a clear trade off between economic benefits and social
costs of some programme, policy, project, investment or proposal. Tourism proposals
that might be economically beneficial may be rejected because of their adverse environ-
mental and/or social impacts.
Dwyer argues that there is substantial scope for greater use of CBA in tourism con-
texts, including assessment of special events. CBA picks up a whole range of benefits and
costs due to non priced effects arising from the absence of markets for some goods and
services affected which would not be included in economic impact analysis. CBA is not
without its challenges as the discussion by Dwyer highlights. Each type of limitation is
being addressed in the research literature and CBA remains the primary technique for
informing policy makers of the net benefits of different strategies to achieve objectives
associated with the public good.

DWYER 9781781001288 Part I.indd 12 25/07/2012 11:06

 1 Statistical testing techniques
Gang Li

INTRODUCTION

Statistical testing is one of the key tasks in quantitative tourism research. It is based
on inferential statistics, which based on probability theory and logic, are used to
make inferences about the characteristics of a population from the characteristics of a
random sample drawn from the population (Grimm, 1993, p. 123). Statistical testing
has been used for research purposes since the early 1700s (Huberty, 1993). Over the past
300 years of the development of statistical testing, four early twentieth-century statisti-
cians, namely R.A. Fisher, J. Neyman, E.S. Pearson and K. Pearson, made the most
significant contribution toward formalizing the concept (McLean and Ernest, 1998).
Most of the current statistical testing techniques are still based on the same logic they
developed.
Specifically, the pioneers developed two different approaches to statistical testing:
Fishers significance testing approach and NeymanPearsons hypothesis testing
approach. The former involves a single (null) hypothesis with a strength-of-evidence
statistic p value, while the latter involves both a null hypothesis (H0) and an alternative
hypothesis (H1) and specifies a fixed level of the probability at which the test statistic
should be rejected. The significance testing approach does not typically consider the
region of rejection or Type I and Type II errors (to be explained below), whilst the
hypothesis testing approach does not consider the extent of support (i.e., the p value). The
hypothesis testing approach is most often discussed in statistical methods for behavioral
sciences. Although the two approaches are based on different philosophical beliefs, they
share the same basic foundation: probability. Statistical testing is not deterministic, but
probabilistic. In more recent development of statistical testing, Huberty (1987) proposed
a hybrid approach which used hypothesis testing notions in designing the study and then
used significance testing as a basis for making statistical inferences. An advantage of
this hybrid approach is to take account of the two types of potential inferential decision
errors and an alternative hypothesis characterization of interest (Huberty, 1987).
Statistical testing techniques, as with many other quantitative research methods,
are commonly applied in tourism research, particularly in relation to tourist behavior.
Based on general consumer behavior models such as the Andreasen (1965) model, the
Nicosia (1966) model and the Howard and Sheth (1969) model, tourism researchers
are interested in examining the behavioral characteristics of consumers in a variety of
tourism contexts, such as tourists perception of destination images, travel motivation,
behavioral intention, perceived service performance, and satisfaction. In addition, some
researchers are interested in the perceptions, attitudes, and behaviors of residents in
tourist destinations. Statistical testing is regularly performed in these tourism studies in
order to obtain robust findings and conclusions.
The rest of this chapter will introduce some commonly used statistical testing tech-

13

DWYER 9781781001288 Part I.indd 13 25/07/2012 11:06

 14 Handbook of research methods in tourism

niques in quantitative tourism studies. Given different assumptions concerning the

distribution of the population, these statistical techniques can be broadly divided into
parametric and non-parametric groups. Parametric techniques, such as the t-test, are
based on the assumption that the sampled populations have approximately normal dis-
tribution with equal variances, while nonparametric tests, sometimes called distribution-
free statistics, such as the c2 test, require less stringent assumptions about the nature of
the probability distributions of the populations. Another important reason for using
non-parametric tests is that they allow for the analysis of categorical and ranked data,
such as the relationships between peoples holiday preference rankings and their demo-
graphic characteristics including gender, age, educational background, marital status,
and so on.

PROCEDURES OF STATISTICAL TESTING

Most statistical tests follow the same logical sequence. Following NeymanPearsons
hypothesis testing approach, the procedural steps for a statistical test are: (1) to
formulate the null and alternative hypothesis; (2) to determine a significance level
and rejection region; (3) to specify the test statistic and referent (null) distribu-
tion; (4)to calculate the test statistic value; and (5) to choose between the null and
alternative hypothesis. The steps will be explained briefly before individual tests are
introduced.

Hypothesis Formulation

A common feature of statistical testing techniques is the concept of the null hypothesis. It
usually proposes no difference between two observed values or no relationship between
two variables. The alternative hypothesis, on the other hand, proposes a significant differ-
ence or relationship. It is usually the alternative hypothesis that is directly related to the
research question which a researcher is interested in. Hypothesis formation is based on
a relevant theoretical framework and is related to a deductive research approach (Veal,
2006).
The alternative hypothesis can be formulated with or without a predetermined direc-
tion, depending on the prior knowledge about the research question. If the null hypoth-
esis states that there is no statistical difference between observed values A and B, then
the alternative hypothesis can be A B, A > B or A < B. Clearly A B includes the
possibility of both directions of A > B and A < B, and is related to a two-tailed test,
while A > B or A < B relates to a one-tailed test. In another case, if the null hypothesis
states that there is no relationship between variables X and Y, the alternative hypothesis
may refer to either a positive or negative relationship (i.e., a one-tailed test) or both (i.e.,
a two-tailed test). For example, Mok (1990) investigated a research question: whether
destination advertising increased tourist flows based on one-tailed hypothesis testing,
because the alternative hypothesis referred to only a positive relationship between des-
tination advertising and tourist flows. One-tailed and two-tailed tests are further related
to the definition of the rejection region of a hypothesis test.

DWYER 9781781001288 Part I.indd 14 25/07/2012 11:06

 Statistical testing techniques 15

Table 1.1 Consequences of decisions in hypothesis testing

Decision True state of nature

H0 is true H1 is true
Accept H0 and reject H1 Right decision Type II error
(probability 5 1 2 a) (probability 5 b)
Reject H0 and accept H1 Type error Right decision
(probability 5 a) (probability 5 1 2 b)

Significance Level and Rejection Region

To reject the null hypothesis, a statistical standard needs to be established as the basis.
The level of significance is the minimum acceptable probability that a value actually
comes from the population (Waters, 2001). If a 5 percent significance level is chosen,
the null hypothesis is rejected only when there is a probability of less than 5 percent that
the value comes from the hypothesized population. This implies that the probability of
rejecting the null hypothesis is 5 percent, when it is actually true. This error is called a
type I error, and its probability is always equal to the level of significance, designated by
a. Thus the probability of accepting a true null hypothesis is (1 2 a).
In contrast to a type I error, the error of accepting a false null hypothesis is called a
type II error, and its probability is denoted b. Therefore, the probability of rejecting a
false null hypothesis is (1 2 b), which is called the power of the test. The probability of
a type II error is often difficult to determine precisely. It depends on the actual alternate
value of the population mean when the null hypothesis is false. When the test statistic
does not fall in the rejection area, instead of concluding with the acceptance of H0, it
is usually preferable to state that there is insufficient sample evidence to reject the null
hypothesis at the significance level of a (McClave et al., 1998). Table 1.1 summarizes four
possible consequences of a hypothesis-testing decision.
As mentioned above, the rejection region is related to the formation of the alternative
hypothesis or, in other words, whether a one-tailed or two-tailed test is to be adopted.
Figure 1.1 illustrates the relationship between the type of a hypothesis test and the rejec-
tion region.

Test Statistic and Reference Distribution

The test statistic is the value, based on a sample, used to determine whether the null
hypothesis should be rejected or not (Kazmier and Pohl, 1987). For instance, the value
of a sample mean can be used as the test statistic for the hypothesized value of the popu-
lation mean. The general practice is to transform a sample statistic into a value on the
appropriate standard distribution and then use this transformed value as the test statis-
tics. Given that the sample distribution of the mean can be assumed to follow normal
distribution, the sample mean can be transformed into a z value: z 5 X 2 m0 /sX , where
X is the sample mean, m0 is the population mean value specified in the null hypothesis,
sX is the standard error of the mean, assuming the populations standard deviation s is
known.

DWYER 9781781001288 Part I.indd 15 25/07/2012 11:06

 16 Handbook of research methods in tourism

One-tailed rejection region

/2 /2

Two-tailed rejection region(s)

Source: Adapted from Marusteri and Bacarea (2010, p. 28).

Figure 1.1 Rejection regions for one- and two-tailed tests

Once the test statistic is determined, the critical value(s) can be established. A critical
value defines the rejection region at the designated level of significance. Given a normal
distribution, critical values are typically specified in terms of z values, and they also
depend on whether one-tailed or two tailed tests are used. The frequently used critical
values of z are listed in Table 1.2. The test statistic will then be calculated and compared
to the critical values in order to decide whether the null hypothesis will be rejected or not.

Table 1.2 Critical values of z at different levels of significance

Level of significance One-tailed test Two-tailed test

10% 1.28 or 21.28 61.645
5% 1.645 or 21.645 61.96
1% 2.33 or 22.33 62.58

P-values

As discussed earlier, Fishers significance testing approach is concerned with a strength-

of-evidence statistic p value instead of the rejection region, and Huberty (1987) recom-

DWYER 9781781001288 Part I.indd 16 25/07/2012 11:06

 Statistical testing techniques 17

p-value

Test statistic
z
One-tailed test

p/2 p/2

Test statistic Test statistic

z z
Two-tailed test

p-value

Test statistic
z

Source: Adapted from Marusteri and Bacarea (2010, p. 28).

Figure 1.2 P-values for one-tailed and two-tailed tests

mended both should be considered in his proposed hybrid approach. Therefore, it is

worth introducing the p value here. The p-value is the probability associated with the
occurrence of the test statistic, given that the null hypothesis is true (Kazmier and Pohl,
1987). For example, in a one-tailed test where H1 : m , m0, the computed test statistic
is z 5 22.50. Therefore, the p-value 5 P(z 22.50). If the alternative hypothesis is
H1 : m . m0 and z 5 2.50, then the p-value 5 P(z 2.50). As seen in Figure 1.2, the
p-values in these two cases of one-tailed tests are the areas under the standard normal
curve to the left of z 5 22.50 and to the right of z 5 2.50, respectively. Referring to the
standard normal probability distribution table (where the z score of 2.5 corresponds to a
probability value of 0.4938, i.e., the area between 0 and 2.5), we can observe that p-value
5 0.5 2 0.4938 5 0.0062. In the case of a two-tailed test, the p-value includes the areas
in both directions (see Figure 1.2). In other words, the p-value is doubled as in one-tailed
tests: p-value 5 0.0062 2 5 0.0124. Following the p-value approach of hypothesis
testing, the calculated p-value is compared to the level of significance in order to decide
to reject or not reject the null hypothesis. If the p-value < a, then the null hypothesis is
rejected. In the above example, if a 5 0.05 is selected, then the null hypothesis is rejected
in both one-tailed and two-tailed tests.

DWYER 9781781001288 Part I.indd 17 25/07/2012 11:06

 18 Handbook of research methods in tourism

PARAMETRIC TESTS
There are a number of parametric tests for hypothesis testing. In tourism research,
t-tests, correlation tests, and the analyses of variance and covariance are applied most
often and are thus introduced here. To apply a parametric test some general assumptions
need to be met: (1) the scores obtained from the population are based on random sam-
pling; (2) the population from which a sample is taken follows normal distribution; (3)
the scores in a sample are independent of each other; (4) samples are taken from popula-
tions of equal variances, i.e., the variability of scores for each of the groups is similar.
It is also called homogeneity of variance (Pallant, 2005). With regard to the normal dis-
tribution assumption, Gravetter and Wallnau (2000, p. 302) and Stevens (1996, p. 242)
argue that most of these statistical tests are reasonably robust with large sample sizes
(above 30 observations in each of the groups to be analysed). In these cases, violation of
normality should not cause major problems. In comparison, homogeneity of variance is
more important and needs to be tested before conducting a t-test or analysis of variance.

T-tests of Hypotheses About Population Means

In tourism research the population standard deviation (s) is usually unknown or

the sample size is relatively small. In these cases, the standard normal distribution
and z-scores cannot be used to make inference about population means. Instead, the
t-distribution should be adopted and t-scores need to be calculated. The t-distribution
has a similar shape as the normal distribution but with heavier tails, which means it is
more prone to producing values that fall far from its mean score. The t-score is calcu-
lated as t(n 21) 5 (X 2 m0 /sx) /"n, where X and sx are the sample mean and standard
deviation respectively, m0 is the hypothesized population mean value, n is the sample size,
and (n 2 1) is the degrees of freedom of the t- distribution.
There are two types of t-tests: the independent samples t-test and the paired samples
t-test. The former is used to compare the mean scores of two different groups of people
or conditions (e.g., the mean scores of perceived performance evaluated by male and
female guests), while the latter is for comparing the mean scores for the same group of
people on two different occasions (e.g., the mean scores of expectation and perceived
performance of a hotels services, evaluated by the same group of hotel guests). Hence
the independent t-test involves a continuous variable (usually measured on an interval
or ratio scale) such as the perceived performance of hotel services and a categorical vari-
able with two categories being concerned, such as gender. A paired t-test requires both
variables to be on interval or ratio scales (continuous), such as service expectation and
service performance.
The test statistic for an independent samples t-test is:
(x1 2 x2) 2 (m1 2 m2)
t5 (1.1)
s21 s22
n1
1
n2

where x1 and x2 are the mean scores of two independent samples, m1 and m2 are the hypothe-
sized mean values of the two populations and the null hypothesis is usually H0: m1 5 m2, s1 and

DWYER 9781781001288 Part I.indd 18 25/07/2012 11:06

 Statistical testing techniques 19

s2 are standard deviations of the two samples, and n1 and n2 are the sample sizes. If the equal
variance assumption holds, then s1 5 s2 5 " [ (n1 2 1) s21 1 (n2 2 1) s22 ] / (n1 1 n2 2 2) .
In this case, the degrees of freedom of the test are (n1 1 n2 2 2) . However, if the equal
variance assumption does not hold, the degrees of freedom are more complicated and can
be calculated by the following formulae:

(s21 /n1 1 s22 /n2) 2

(1.2)
(s21 /n1) 2 / (n1 2 1) 1 (s22 /n2) 2 / (n2 2 1)

For a paired samples t-test, the test statistic is given by:

d
tn 21 5 (1.3)
a d 2 nd
2 2

n (n 2 1)
where d is the difference between each pair of values of two variables in the paired sample,
d is the mean difference, and n is the sample size. The degrees of freedom of this test are (n
2 1). The paired samples t-test has an additional assumption that the difference between
the paired scores (d) follows normal distribution. However, with a sample size above 30,
violation of this assumption is unlikely to cause serious problems (Pallant, 2005).
Wang et al. (2005) used an independent-samples t-test to analyse the effects of resi-
dents employment status and income on their attitudes towards tourism development
and tourism impacts. They found that high-income residents had significantly more
positive attitudes towards socio-cultural impacts of tourism (t 5 2.213, p 5 0.028) and
residents who worked in (or residents whose relatives worked in) the tourism industry
rated the economic benefits of tourism development significantly higher (t 5 1.999, p 5
0.047). Li et al. (2009) used a paired-samples t-test to investigate the impact of online
information searches on tourists destination image development. By comparing a group
of participants perceived image of China before and after an online information search,
they found the participants overall image and affective image about China experi-
enced significant and positive changes after an online search, while the cognitive image
remained the same.

Correlation Analysis

Correlation analysis examines the strength of association between two variables by cal-
culating a correlation coefficient. The Pearson productmoment correlation is the most
commonly used parametric method for correlation analysis. It analyses the association
between two continuous variables (measured on an interval or ratio scale), such as
service quality and tourist satisfaction, or perceived destination image and visit inten-
tion. Based on a sample to estimate the population correlation, Pearsons correlation
coefficient (r) can be computed as follows:

Np ( a XY) 2 ( a X) ( a Y)
r5 (1.4)
" [ NP ( a X2) 2 ( a X) 2 ] [ NP ( a Y2) 2 ( a Y) 2 ]

DWYER 9781781001288 Part I.indd 19 25/07/2012 11:06

 20 Handbook of research methods in tourism

where X and Y are two continuous variables and Np is the number of pairs of observa-
tions. A correlation coefficient ranges from 21 to 1, suggesting a perfect negative rela-
tionship to a perfect positive relationship. The higher the coefficients absolute value,
the stronger the association between the two variables. Although there is no agreement
on the cut-off values of r for different levels of association, it is generally accepted that
|r| < 0.3 would suggest the relationship is weak (or small), and |r| > 0.7 would imply a
strong (or large) association between the two variables (see, for example, Grimm, 1993,
p. 377).
The correlation coefficient r computed based on a sample is an estimator of the popu-
lation correlation coefficient r. An r score only shows the level of association between
two variables, but it does not suggest whether this relationship is statistically significant
or not. Therefore, a statistical test of the significance of the correlation needs to be
carried out. The appropriate test is based on the t distribution, which is a transformation
of sampling distributions of r of differing sample sizes. The test statistic t is written as:

Np 2 2
1 2 r2
t5r (1.5)

where the degrees of freedom equal Np 2 2. Depending on the specification of the alter-
native hypothesis, either H1: r 2 0 or H1: r . 0 (or H1: r , 0), a two-tailed or one-tailed
t-test will be used, respectively. If the direction of the association between two variables
is clear, for example, prices are most likely to have a negative association with tourism
demand, a one-tailed t-test should be used instead of a two-tailed one.
In addition to finding out whether there is a significant relationship between variables
X and Y, researchers may also want to know if the strength of this relationship is signifi-
cantly different between groups. To test statistical significance of the difference between
correlation coefficients, the r values computed from different groups need to be con-
verted into z scores and the observed value of z (zobs) as a test statistic must be calculated
following the formulae below:

z1 2 z2
zobs 5 (1.6)
1 1
N1 2 3
1
N2 2 3
where z1 and z2 are converted z scores from r values in relation to groups 1 and 2, respec-
tively; and N1 and N2 are the numbers of observations in the two groups. The calculated
zobs will be compared with critical values of z (see Table 1.2) to decide if the correlation
coefficients are statistically significantly different.
In an example of testing the relationship between destination image and travel inten-
tion, if it is suspected that the tourists past visit experience (Z, measured by a continuous
variable: the number of past visits) is likely to be related to both their perceived destina-
tion image (X) and their travel intention (Y). In this case, the computed Pearsons cor-
relation coefficient between the perceived destination image and travel intention without
controlling the influence of tourists past visit experience, may be inflated. Therefore, a
more accurate correlation analysis should control the effect of other variables that may
be associated with both variables concerned. This is called partial correlation analysis.

DWYER 9781781001288 Part I.indd 20 25/07/2012 11:06

 Statistical testing techniques 21

It follows the same concept of the aforementioned Pearson correlation analysis, but the
computation of the partial correlation coefficient (rXY # Z) is revised as follows:

rXY 2 (rXZ) (rYZ)

rXY # Z 5 (1.7)
" (1 2 r2XZ) (1 2 r2YZ)

where rXY , rXZ and rYZ are Pearson correlation coefficients between any two of the three
variables.
It should be noted that the Pearson correction and partial correlation tests only detect
linear relationships between variables. A low and non-significant Pearson correlation
coefficient between variables X and Y does not necessarily mean there is no associa-
tion between them. A non-linear relationship may still exist. It is necessary to plot each
subjects X and Y scores on a graph (i.e., to draw a scatter plot) and visually examine
the shape of the plot to decide whether a linear correlation analysis is appropriate.
Moreover, the above correction analysis aims to detect linear association between X and
Y, but not a causal relationship. So a significant correlation coefficient between X and
Y does not necessarily mean that either X causes Y or Y causes X. Hence the result of a
correlation test should be interpreted with caution. In addition, the calculated correla-
tion coefficient is sensitive to outliers, especially in small samples. A misleading r will
be derived from such a sample. Therefore, it is necessary to detect and remove outliers
before running a correlation test.
Li and Petrick (2008) used Pearsons productmoment correlation test to detect the
relationships between consumers attitudinal loyalty and each of its three antecedents:
satisfaction, quality of alternatives, and brand investment in the context of cruising.
They found that the cruise passengers attitudinal loyalty to a cruise lines brand was
weakened by the quality of alternative options (r 5 20.22) but strengthened by their
satisfaction with the brand and investment in it (r 5 0.720 and 0.601, respectively).
In another study, Snepenger et al. (2004) employed Pearsons correlation analysis to
examine whether different tourism places had shared, independent, or divergent norma-
tive meanings. The results suggested Downtown and Yellowstone shared many mean-
ings (r > 0, p < 0.05), but Downtown had very different meanings from big-box stores
(r< 0, p < 0.05).

Analysis of Variance and Covariance

Analysis of variance (ANOVA) can be regarded as an extension of the independent

samples t-test. It compares means of more than two groups at a time. Assuming there are
g groups of subjects in a sample and we are comparing their mean scores of a continuous
variable (e.g., comparing destination satisfaction levels of tourists from g countries of
origin), the null and alternative hypotheses are: H0: m1 5 m2 5 . . . 5 mg, H1: the means
of the g groups are not all equal. In the above example, m1, m2, . . . , mg are the population
mean values of destination satisfaction of tourists from these g countries. ANOVA draws
inferences about population means by analysing sample variances. There are two sources
of variability in scores (i.e., variance): the variation between the groups and the variation
within each of the groups. The test statistic of ANOVA is an F ratio which represents the
variance between the groups divided by the variance with the groups and is defined as:

DWYER 9781781001288 Part I.indd 21 25/07/2012 11:06

 22 Handbook of research methods in tourism

between-group variance
F5 5
within-group variance

d ^ (g 2 1)
( a X1) 2 ( a X2) 2 ( a Xg) 2 ( a X) 2
c 1 1...1 2
n1 n2 ng N
(1.8)
d f ^ (N 2 g)
( a X1) 2 ( a X2) 2 ( a Xg) 2 ( a X) 2
e aX 2 c 2
1 1...1 2
n1 n2 ng N

where SXg is the sum of scores in Group g, ng is the number of subjects in Group g, SX
is the sum of all the scores in the study, N is the total number of subjects in the study and
SX2 is the sum of all squared scores in the study. The F-statistic follows the F-distribution
with (g 2 1, N 2 g) degrees of freedom under the null hypothesis.
A significant F-value means the null hypothesis is rejected, indicating that at least two
of the population means are not equal. However, a significant F-value does not show
which means are significantly different from one another. To find this out, pairwise com-
parisons among all the means need to be controlled carefully and carried out with the
Type I error. This analysis is called a post-hoc test, such as Tukeys honestly significant
different (HSD) test (Tukey, 1953[1984]) and modified Bonferroni procedures (Olejnik
et al., 1997). Due to space constraints, any discussion of these tests is beyond the scope
of this chapter.
The above explanation of ANOVA refers to one factor (or independent variable)
that separates a sample into a few groups (e.g., nationalities of tourists at a destination)
and one continuous variable (i.e., dependent variable, such as destination satisfaction).
This ANOVA is called one-way between-groups ANOVA. One-way ANOVA can also be
applied to the same subjects under different conditions. This is referred to as a repeated-
measures ANOVA or a within-subjects design. Often researchers are interested in inves-
tigating the individual or joint effect of two independent variables on one dependent
variable, such as comparing destination satisfaction levels across tourists nationalities
and age groups. This analysis is called two-way between-groups ANOVA. It involves
testing the effects of each of the two independent variables (i.e., the main effects) and
the effect of the interaction between the two independent variables, all on the depend-
ent variable concerned. All of these three tests are based on F-tests, as with a one-way
ANOVA.
In both one-way and two-way ANOVA analyses above, only one independent vari-
able is concerned. In some research situations, however, researchers are interested in
comparing groups on a range of related dependent variables such as comparing, across
tourists of different nationalities, their satisfaction with four dimensions of destination
attributes: services, prices, infrastructure, and environment. This analysis is called a
multivariate analysis of variance (or MANOVA), which is an extension of the ANOVA:
it includes more than one related dependent variable. As with ANOVA, there are one-
way and two-way MANOVA, depending on the number of independent variables being
considered.
In comparison to running a series of ANOVAs, it is more effective and robust to run
only one MANOVA. This is because to run a number of ANOVAs separately increases
the risk of an inflated Type I error (Tabachnik and Fidell, 2007), while MANOVA

DWYER 9781781001288 Part I.indd 22 25/07/2012 11:06

 Statistical testing techniques 23

takes this risk into consideration. Moreover, MANOVA takes the intercorrelations
between the dependent variables into account, which can only be achieved if these
dependent variables are considered together in one test. MANOVA has two additional
assumptions in relation to multiple dependent variables. First, the significant test is
based on multivariate normal distribution; in other words, any linear combination of the
dependent variables should be distributed normally. Second, the matrices of the covari-
ances, i.e., the variance shared between any two dependent variables, have to be equal
across all levels of the independent variable(s). The test is reasonably robust to modest
violations of the multivariate normality assumption as long as a minimum sample size of
20 is ensured for each sub-group that the whole sample is divided into by the independent
variable(s) (Tabachnik and Fidell, 2007).
The latest development of multivariate analysis is to test the effects of one or two
independent variables on one or more dependent variables while controlling for an
additional continuous variable, which is called a covariate and believed to have some
influence on the dependent variable(s) (Pallant, 2005). This type of analysis is called
analysis of covariance or ANCOVA. By controlling the effect of the covariate, ANCOVA
increases the power of the F-test, so that it is more likely to detect differences among
groups. For example, tourists past visit experiences may affect their destination satisfac-
tion evaluation. Therefore, to control the effect of this factor (measured by the number
of past visits to the destination), it is easier to find satisfaction differences across tourists
of different nationalities, different age groups, and so on. By introducing a covariate into
one-way ANOVA, two-way ANOVA, and MANOVA, one-way ANCOVA, two-way
ANCOVA, and MANCOVA can be exercised, respectively.
Various analyses of variance can be seen in tourism research. For instance, Simpson
and Siguaw (2008) applied one-way ANOVA to the investigation of the effects of
tourist type on place satisfaction, identity salience and place promotion. The results
were all statistically significant (F > 18, p < 0.001 for all). In subsequent post-hoc tests,
the authors found that winter tourists were more likely to be satisfied with the place
and to promote it via positive word-of-mouth recommendation than any other tourist
segment. Rewtrakunphaiboon and Oppewal (2008) adopted two-way ANOVA to study
the impact of destination image (more favorable and less favorable) and information
format (destination name as a package heading, price as a package heading, and desti-
nation name only) on the intention to visit a destination. They found that destinations
with a more favorable image received a significantly higher intention to visit rating
than destinations with a less favorable image, while the difference between information
format conditions was not significant. The results were that the interaction between
image and format did not have a significant effect on the intention to visit either. At a
more advanced level of statistical testing, Tsaur and Wang (2009) employed a two-way
between-groups MANOVA to detect the effects of the tip-collection strategy (tipping by
participants and tip included) and the type of service guarantee (none, attribute-specific,
and full satisfaction) on consumer evaluations of a group package tour, in terms of
expected service quality, financial risk, performance risk, and willingness to buy. This
study found significant main effects of both independent variables, but no significant
interaction effect between them. Subsequently, a number of one-way between-groups
ANOVAs were performed on each of the dependent variables, followed by post-hoc
tests.

DWYER 9781781001288 Part I.indd 23 25/07/2012 11:06

 24 Handbook of research methods in tourism

NON-PARAMETRIC STATISTICAL TESTS

The above parametric statistical tests have some stringent assumptions and normally
require large sample sizes. If the assumptions cannot be met, or the sample size is
too small (less than 20), or the data are measured on normal (categorical) or ordinal
(ranked) scales, non-parametric statistical tests should be considered. A number of non-
parametric tests have been developed to respond to their parametric counterparts. Given
the space constraints, this chapter focuses on the most often used techniques in tourism
studies. Explanation of other non-parametric techniques is available from Gibbons and
Chakraborti (2003).

Chi-square Tests

There are two types of chi-square test: for one-way designs and for two-way designs. In
one-way designs, the chi-square test is the categorical counterpart of a one-way ANOVA
with two or more groups. It is concerned with the frequency distribution of a nominal
variable and used with a sample to make an inference about the frequency distribution
of the population. It is also called the goodness-of-fit test. It requires a specification of
the population frequency distribution (i.e., expected frequencies) as the null hypothesis
and the alternative hypothesis states that the population distribution of frequencies does
not follow an expected manner. The goodness-of-fit test is used to statistically determine
if the observed frequencies obtained from a sample are significantly different from the
expected frequencies. For example, it is expected that the percentages of male and female
tourists in a holiday resort are equal, but the observed gender distribution of tourists in a
given period may deviate slightly from this balance. The goodness-of-fit test can be used
to find out whether the deviation is statically significant. The chi-square statistic (c2) is
computed as:

(fo 2 fe) 2
c2 5 a (1.9)
fe

where fo and fe are observed and expected frequencies, respectively. The degrees of
freedom of the goodness-of-fit test are the number of groups, minus one. The test is
based on the c2 distribution. As the degrees of freedom increase, the c2 distribution
approximates the shape of a normal distribution, but all c2 values are positive. The criti-
cal values become larger as the degrees of freedom increase.
The chi-square test can also be used in two-way designs, to detect whether there is a
relationship between two categorical variables or if they are independent, so this chi-
square test is called the chi-square test for independence. For example, researchers may
want to know if the employment status of residents in a tourist destination (i.e., working
in a tourism sector or not) is related to their opinion on a new tourism policy (i.e., in
favor of the new policy or not). The null hypothesis of the chi-square test for independ-
ence states that the two categorical variables are independent (or have no relationship)
and the alternative hypothesis states that these two variables are related. Based on
the presentation of this two-way design in a frequency or cross-tabulation table, the
chi-square test for independence examines whether the frequency distribution for one

DWYER 9781781001288 Part I.indd 24 25/07/2012 11:06

 Statistical testing techniques 25

categorical variable is different, depending on the classification of the other categori-

cal variables (Grimm, 1993, p. 442). The test statistic (c2) is computed using the same
formulae as with the one-way design, while the expected frequency for each cell of the
cross-tabulation table has to be calculated in advance, based on the relevant observed
frequencies, following the formulae below:

fc fr
fe 5 (1.10)
N

where fe refers to the expected frequency of a cell in the cross-tabulation table, fc and
fr are the totaled frequencies for the relevant column and row of this cell, respectively,
and N is the total number of subjects. The degrees of freedom of this chi-square test are:
df 5 (C 2 1) (R 2 1) , where C and R are the numbers of columns and rows respectively,
i.e., the number of categories of these two categorical variables. This test requires the
expected frequency for any cell to be a minimum of 5, i.e., fe $ 5. In a 2 3 2 design (i.e.,
C 5 R 5 2), violation of this assumption means a different test; Fishers exact probability
test, should be considered instead (Pallant, 2005).
Compared with the applications of parametric statistical testing techniques, the
applications of their non-parametric alternatives are much fewer. The variety of
non-parametric techniques being applied in tourism research is relatively limited.
Li (2010) applied the chi-square test for independence in his study of the asso-
ciation of loyalty (low and high categories) and brand size (small, medium, and big).
This study found that attitudinal loyalty and brand size were marginally related
(c2 5 5.894, p 5 0.052) , but market size did affect respondents behavioral loyalty sig-
nificantly (c2 5 35.174, p , 0.001) .

Other Non-parametric Tests

Chi-square tests are useful when handling nominal variables, but often a study design
involves measurement at the ordinal level, such as levels of education, income and
involvement in some activities (never, seldom, sometimes, often and always). On other
occasions, some variables originally designed to be measured at interval scales fail to
meet one or more assumptions for parametric tests. A number of other non-parametric
tests have been developed to meet research needs where the conditions of using paramet-
ric alternatives are not met. Table 1.3 summarizes these tests, along with their parametric
counterparts.

Table 1.3 Summary of non-parametric tests and their parametric alternative

Non-parametric test Parametric alternative

Wilcoxon signed-rank test Paired-samples t-test
MannWhitney test Independent-samples t-test
KruskalWallis test One-way between-groups ANOVA
Friedman test One-way repeated-measures ANOVA
Spearman rank test Pearsons productmoment correlation

DWYER 9781781001288 Part I.indd 25 25/07/2012 11:06

 26 Handbook of research methods in tourism

The Wilcoxon signed-rank test is used for two repeated measures at different times or
conditions, and is the non-parametric alternative to the repeated measures, or paired
samples t-test. Instead of comparing means, the Wilcoxon test converts scores to ranks
and compares them between Treatments 1 and 2 (or at Times 1 and 2). The null hypoth-
esis states that there is no treatment effect. The test statistic T is found by comparing the
sum of positive ranks (i.e., in favor of Treatment 1) and the sum of negative ranks (i.e.,
in favor of Treatment 2). T is the smaller sum of the two. The degrees of freedom of this
test are the number of pairs in the two treatments.
The MannWhitney test is the non-parametric alternative to the independent-samples
t-test. As with the Wilcoxon signed-rank test, the MannWhitney U test compares and
converts scores to ranks and evaluates whether the ranks for the two groups are signifi-
cantly different. The test statistic is U, which can be calculated directly, if the number of
scores is small, by counting the number of X1 less than X2 and the number of X2 less than
X1. U equals the smaller of the two numbers. When the number of scores is a bit larger,
U can be calculated by using the following formulae:

ni (ni 1 1)
Ui 5 n1n2 1 2 Ri (1.11)
2

where n1 and n2 are the numbers of scores in group 1 and group 2, Ri is the sum of
ranksfor group i (i51 or 2), U5smaller of U1 and U2, and the degrees of freedom are
(n1, n2) .
When the sample size is large (above 50 pairs for the Wilcoxon signed-rank test or
above 20 in either group for the MannWhitney U test), the sampling distributions of T
and U approximate a normal distribution. The T and U values can then be transformed
to z scores and compared to the desired critical value of z to decide whether the null
hypothesis is rejected or not.
The KruskalWallis test is the non-parametric alternative to the one-way between-
groups ANOVA. The test statistic H can be calculated by:
k R2
H5 c d 2 3 (N 1 1)
12 i
a (1.12)
N N 1 1 i51 i
( ) n

where k is the number of independent groups, ni is the number of observations

ingroupi,N is the total number of observations, and Ri is the sum of the ranks in group
i.
Wang and Pfister (2008) applied the MannWhitney U test and KruskalWallis test
to the analysis of the effects of gender, Civil Club membership, educational levels, and
length of residence on residents perception of each of the personal benefits from tourism.
The reason for choosing these tests instead of the independent-sample t-test and one-way
ANOVA was because of violation of the normality assumption. Significant test results
indicated that female residents were more likely to perceive that they had obtained ben-
efits from downtown revitalization and tourism activities with arts and cultural features
than male residents were.
The Friedman test is the non-parametric alternative to the one-way repeated measures
ANOVA. The test statistic Fr is calculated as follows:

DWYER 9781781001288 Part I.indd 26 25/07/2012 11:06

 Statistical testing techniques 27

Fr 5 c R2 d 2 3N (k 1 1)
k
12
a (1.13)
Nk (k 1 1) i51 i

where N, k and Ri are defined in the same way as in the Kruskal-Wallis H test.
When the null hypothesis is true, in large samples the sampling distributions of H and
Fr are closely approximated by the chi-squared distribution, with the degrees of freedom
equal to k 2 1.
In a study comparing different methodologies of measuring service quality in a
tour operating context, Hudson et al. (2004) ranked the ratings of 13 dimensions of
servicequality based on each of the four methodologies. The Friedman test was then
applied to the four ranked variables (each with 13 ranked observations) to determine
if there were any statistically significant differences between the four approaches to
measuring service quality. The result was not statistically significant. To substantiate
the finding, each methodology was individually compared with the other three, using
the Wilcoxon signed-ranks test. Once again, no statistically significant difference was
found.
The Spearmans rank order correlation is the non-parametric equivalent of the
Pearsons product-moment correlation. The correlation coefficient (rs) can be calculated
by the following formula, which can be derived from the formula for the Pearsons cor-
relation coefficient by using the ranks as the values of X and Y.

6 a d2
rs 5 1 2 (1.14)
n (n2 2 1)

where d 5 X 2 Y for each pair of ranks and n is the number of paired observations.
Given that n $ 10, the test for significance of rs can be carried out in the same way as
for the Pearson correlation coefficient. That is, the test statistic is based on the Students
t-distribution, with the degrees of freedom equal to (n 2 2) and the test statistic is:

rs
t5 (1.15)
a (1 2 rs ) / (n 2 2)
2

Horneman et al. (2002) used the Spearman rank correlation coefficient to examine the
association of senior travelers holiday preference rankings and some demographic vari-
ables such as income and education across six market segments. The results suggested
that respondents with more formal educations tended to prefer pioneer-type holidays.
Senior travelers with less formal education preferred Aussie-type holidays emphasizing
the natural environment and appealing to travelers who are active, adventurous, group/
family oriented, outward directed, older, cautious spenders.

ADVANTAGES, LIMITATIONS, AND FURTHER

DEVELOPMENTS

As discussed above, both parametric and non-parametric tests have their advantages as
well as limitations. Parametric tests have stringent assumptions and normally require

DWYER 9781781001288 Part I.indd 27 25/07/2012 11:06

 28 Handbook of research methods in tourism

large sample sizes. If these conditions are met, parametric tests tend to be more powerful
than their non-parametric counterparts, because parametric tests make more use of the
information available in the collected data. Parametric tests are based on measurement
at the interval level, while non-parametric tests use the ordinal information only. In
some cases where assumptions of parametric tests are not met, original measures at the
interval level have to be converted to ranks, by which a certain amount of information in
the original data is sacrificed. So non-parametric tests are less powerful and sometimes
they fail to detect a real statistical difference. Moreover, parametric statistical tests are
more flexible and can handle a range of complex situations (such as one-way ANCOVA,
two-way ANCOVA, and MANCOVA), while the situations non-parametric tests can
deal with are much more limited.
However, non-parametric tests have advantages in certain situations where parametric
alternatives are unsuitable or unavailable, such as hypothesis testing associated with cat-
egorical variables. In addition, some non-parametric tests can handle samples made up
of observations from different populations without making seemly unrealistic assump-
tions (Siegel and Castellan, 1988).
Often researchers face less straightforward situations for making a choice between
parametric and non-parametric tests, for instance, when scores deviate from normality to
some extent. Should the researcher use non-parametric techniques or still apply the more
powerful parametric tests? Although there is no clear-cut rule for the choice, some rules
of thumb have been recommended. For example, according to Leech and Onwuegbuzie
(2002), the standardized skewness and kurtosis coefficients of the scores lying within ;2
can be regarded as no serious departure from normality. In such cases, most parametric
tests are reasonably robust and can still be applied. Standardized skewness and kurtosis
coefficients beyond the ;3 range indicate serious deviation from normality and therefore
non-parametric tests are more appropriate.
To overcome the limitations of traditional non-parametric statistical techniques, par-
ticularly the lack of power and robustness, further developments have been seen. Robust
statistics is in the process of developing into a new subspecialty of its own. Several new,
more robust non-parametric methods have been developed which seek to control the
Type I error rate at their nominal levels and also maintain adequate statistical power,
even when data are non-normal and heteroskedastic (Wilcox, 2005). The developments
in new techniques such as bootstrapping contribute to the further development of non-
parametric statistical techniques. Bootstrapping is a computer-intensive re-sampling
technique which creates a new sampling distribution based on original observations,
when the assumption of a theoretical distribution is violated. The bootstrapped distribu-
tion is then used to compute p-values and test hypotheses (Erceg-Hurn and Mirosevich,
2008). Techniques from other statistical branches have also been incorporated into
non-parametric tests, such as the non-parametric Bayesian approach (see, for example,
Ghosal and Roy, 2009). Meanwhile, some developments also take place in the field of
parametric statistics, such as the study of outliers. With parallel developments in both
parametric and non-parametric statistics, their distinctions are narrowed and the differ-
ences are bridged.
The above new developments have been rarely seen in the tourism literature. Future
quantitative tourism studies should consider these more advanced techniques to enhance
the power of statistical testing and robustness of empirical findings.

DWYER 9781781001288 Part I.indd 28 25/07/2012 11:06

 Statistical testing techniques 29

CONCLUSIONS
This chapter discussed the foundations of statistical testing, followed by a number of
parametric and non-parametric statistical techniques and their applications in tourism
research. In general, parametric tests are more popular than their non-parametric
alternatives in tourist behavior studies, mainly because they are relatively more robust
and capable of handling complicated research designs. By introducing more and more
newly developed, robust, non-parametric techniques into future tourism research, the
accuracy and reliability of research findings can be further improved. It should be noted
that statistical significance does not necessarily mean theoretical significance. On the
other hand, statistical non-significance does not necessarily indicate theoretical non-
significance. Therefore caution should be exercised when interpreting a result of statisti-
cal testing.

REFERENCES

Andreasen, A.R. (1965), Attitudes and consumer behavior: a decision model, in L. Preston (ed.), New
Research in Marketing, Berkeley: Institute for Business and Economic Research, University of California,
pp. 116.
Erceg-Hurn, D.M. and V.M. Mirosevich (2008), Modern robust statistical methods: an easy way to maximize
the accuracy and power of your research, American Psychologist, 63 (7), 591601.
Ghosal, S. and A. Roy (2009), Bayesian nonparametric approach to multiple testing, in N.S. Narasimha
Sastry, T.S.S.R.K. Rao, M. Delampady and B. Rajeev (eds), Perspectives in Mathematical Sciences I:
Probability and Statistics, Singapore: World Scientific Publishing, pp. 139164.
Gibbons, J.D. and S. Chakraborti (2003), Nonparametric Statistical Inference: Fourth Edition, Revised and
Expanded, New York: Marcel Dekker.
Gravetter, F.J. and L.B. Wallnau (2000), Statistics for the Behavioral Sciences, 5th edition, Belmont, CA:
Wadsworth.
Grimm, L.G. (1993), Statistical Applications for the Behavioral Sciences, New York: John Wiley & Sons.
Horneman, L., R.W. Carter, S. Wei and H. Ruys (2002), Profiling the senior traveler: an Australian perspec-
tive, Journal of Travel Research, 41 (1), 2337.
Howard, J.A. and J.N. Sheth (1969), The Theory of Buyer Behavior, New York: John Wiley.
Huberty, C.J. (1987), On statistical testing, Educational Research, 16 (8), 49.
Huberty, C.J. (1993), Historical origins of statistical testing practices: the treatment of Fisher versus Neyman-
Pearson views in textbooks, The Journal of Experimental Education, 61 (4), 317333.
Hudson, S., P. Hudson and G.A. Miller (2004), The measurement of service quality in the tour operating
sector: a methodological comparison, Journal of Travel Research, 42 (3), 305312.
Kazmier, L.J. and N.F. Pohl (1987), Basic Statistics for Business and Economics, 2nd edition, Singapore:
McGraw-Hill.
Leech, N.L. and A.J. Onwuegbuzie (2002), A call for greater use of nonparametric statistics, paper presented
at the Annual Meeting of the Mid-South Educational Research Association. Retrieved from http://www.
eric.ed.gov/ERICWebPortal/contentdelivery/servlet/ERICServlet?accnoED471346.
Li, X. (2010), Loyalty regardless of brands? Examining three nonperformance effects on brand loyalty in a
tourism context, Journal of Travel Research, 49 (3), 323336.
Li, X. and J.F. Petrick (2008), Examining the antecedents of brand loyalty from an investment model perspec-
tive, Journal of Travel Research, 47 (1), 2534.
Li, X., B. Pan, L. Zhang and W.W. Smith (2009), The effect of online information search on image develop-
ment: insights from a mixed-methods study, Journal of Travel Research, 48 (1), 4557.
McClave, J.T., P.G. Benson and T. Sincich (1998), Statistic for Business and Economics, 7th edition, London:
Prentice Hall.
McLean, J.E. and J.M. Ernest (1998), The role of statistical significance testing in educational research,
Research in the Schools, 5 (2), 1522.
Marusteri, M. and V. Bacarea (2010), Comparing groups for statistical differences: how to choose the right
statistical test?, Biochemia Medica, 20 (1), 1532.

DWYER 9781781001288 Part I.indd 29 25/07/2012 11:06

 30 Handbook of research methods in tourism

Mok, H.M.K. (1990), A quasi-experimental measure of the effectiveness of destinational advertising: some
evidence from Hawaii, Journal of Travel Research, 29 (1), 3034.
Nicosia, F.M. (1966), Consumer Decision Processes: Marketing and Advertising Implications, Englewood Cliffs,
NJ: Prentice-Hall.
Olejnik, S., J. Li, S. Supattathum and C.J. Huberty (1997), Multiple testing and statistical power with modified
Bonferroni procedures, Journal of Educational and Behavioral Statistics, 22 (4), 389406.
Pallant, J. (2005), SPSS Survival Manual, 2nd edition, Maidenhead, UK: Open University Press.
Rewtrakunphaiboon, W. and H. Oppewal (2008), Effects of package holiday information presentation on
destination choice, Journal of Travel Research, 47 (2), 127136.
Siegel, S. and N.J. Castellan Jr. (1988), Nonparametric Statistics for the Behavioral Sciences, New York:
McGrawHill.
Simpson, P.M. and J.A. Siguaw (2008), Destination word of mouth: the role of traveler type, residents, and
identity salience, Journal of Travel Research, 47 (2), 167182.
Snepenger, D., L. Murphy, M. Snepenger and W. Anderson (2004), Normative meanings of experiences for a
spectrum of tourism places, Journal of Travel Research, 43 (2), 108117.
Stevens, J. (1996), Applied Multivariate Statistics for the Social Sciences, 3rd edition, Mahway, NJ: Lawrence
Erlbaum.
Tabachnik, B.G. and L.S. Fidell (2007), Using Multivariate Statistics, Boston, MA: Allyn & Bacon.
Tsaur, S.-H. and C.-H. Wang (2009), Tip-collection strategies, service guarantees, and consumer evaluations
of group package tours, Journal of Travel Research, 47 (4), 523534.
Tukey, J.W. (1953[1984]), The problem of multiple comparisons, in H.I. Braun (ed.), The Collected Works
of John W. Tukey, Volume VIII Multiple comparisons: 19481983, New York: Chapman & Hall, pp. 1300.
Veal, A.J. (2006), Research Methods for Leisure and Tourism: A Practical Guide, 3rd edition, Harlow, UK:
Prentice Hall.
Wang, Y.A. and R.E. Pfister (2008), Residents attitudes toward tourism and perceived personal benefits in a
rural community, Journal of Travel Research, 47 (1), 8493.
Wang, Y., G. Li and X. Bai (2005), A residential survey on urban tourism impacts in Harbin, China Tourism
Research, 1 (1), 116129.
Waters, D. (2001), Quantitative Methods for Business, 4th edition, Harlow, UK: Prentice Hall.
Wilcox, R.R. (2005), Introduction to Robust Estimation and Hypothesis Testing, 2nd edition, San Diego, CA:
Academic Press.

DWYER 9781781001288 Part I.indd 30 25/07/2012 11:06

 2 Regression analysis
Jaume Rossell

INTRODUCTION

Qualitative and quantitative research methods have often both been classed as useful and
legitimate by social researchers. However, during recent decades, quantitative methods
have come to prevail in the social sciences, and tourism research is not an exception.
The main role of quantitative research has typically been reduced to helping generate
and pose hypotheses that can then be tested using mathematical research methods.
Nowadays, many tourism journals reflect this bias in favor of quantitative methods, and
it can be seen that the purpose of qualitative research is usually to provide information
for developing further quantitative research (Lewis et al., 1995).
Within the social sciences and some natural sciences, statistics are the most widely
used branch of mathematics in quantitative research when hypotheses have to be tested.
Statistical methods are extensively used in fields such as economics, social sciences, and
biology. Quantitative research using statistical methods starts with the collection of data,
based on a certain hypothesis or theory. For instance, quantitative opinion surveys are
widely used in tourism, where statistics are commonly reported, such as the proportion
of respondents agreeing with a certain stance. In these opinion surveys, the respond-
ents are asked a set of structured questions and their responses are then tabulated. On
other occasions, researchers compile and compare different statistics in order to identify
some kind of correlation. In this case, when a causal relationship between a dependent
variable and one or more independent variables is hypothesized (and must be tested),
regression analysis comes to the fore as a technique for use in circumstances where a
particular phenomenon is controlled and determined by certain factors, and the aim is
to try to explain and quantify variations in the dependent variable through variations in
the independent ones. In tourism, researchers might measure and study the relationship
between the demand (measurable in terms of the number of tourists, length of stay, etc.)
and other key variables, such as price, income, marketing expenditure, etc. One funda-
mental principle in quantitative research should nonetheless be highlighted: that correla-
tion does not imply causation. This is because a spurious relationship is always possible
between variables where some degree of covariance can be found. Thus associations can
be carefully examined between any combination of continuous and categorical variables,
using statistical methods.
Because regression analysis estimates how the value of the dependent variable changes
when any one of the independent variables is modified (while the other independent
variables are held fixed), the said technique can be used in two main frameworks. The
first focuses on exploring the nature of the relationship between the dependent variable
and one (or more) independent variable(s) in order to analyse expected reactions in the
dependent variable, caused by induced changes in the independent one. The second con-
centrates on how to predict and forecast the independent variable. Using the example

31

DWYER 9781781001288 Part I.indd 31 25/07/2012 11:06

 32 Handbook of research methods in tourism

of the relationship between tourism demand and prices, income, and other determining
variables, in the first case the interest appeal of a research study might be to quantify the
effect of an economic crisis or the effect of a tourist tax on tourism demand. In the second
case, the interest appeal might be the expected number of tourist arrivals next year at a
particular destination.
Since initial applications, numerous techniques for carrying out regression analysis
have been developed. With most simple methods, such as linear regressions and ordi-
nary least square regressions, the regression function is defined in terms of a finite set
of unknown parameters, estimated from a collected data set. In fact, the performance
of a regression analysis will depend on the nature of the data-generating process and
how it relates to the regression approach that is used. Since the true nature of the data-
generating process is unknown, regression analysis depends to some extent on making
assumptions about this process. Although these assumptions are sometimes testable if
a large amount of data is available, the validity of the results will be conditional upon
the validity of these assumptions. Hence knowledge of the background technique is as
important as knowledge of implicit assumptions, so that any biases and shortcomings in
the application of the regression analysis are taken into account.
The rest of the chapter is organized as follows: the next section outlines the basic foun-
dations of the technique and gives an intuitive overview of it; then its main applications
in tourism research are reviewed; the limitations of regression analysis are considered;
and the final section contains the main conclusions.

BACKGROUND TO THE TECHNIQUE

Simple Regressions

Let us assume that the number of vacation days abroad is determined by a single
attribute: personal income. At the outset of any regression study, a hypothesis is
formulated about the relationship between the variables under analysis: in this case,
vacation days and earnings. According to economic theory (and common experi-
ence), tourists in the higher income bracket tend to make more trips abroad. Thus the
hypothesis can be posed that higher income levels (the independent variable) lead to
more vacations days abroad (the dependent variable). The problem can be expressed
analytically as:

E (Y/X) 5 f (X, b) (2.1)

where E (Y/X) represents the expected value of dependent variable Y (vacation days
abroad in our case), conditional upon independent variable X (one single independent
variable in the case of a simple regression); f is a mathematical function that must be
specified; and b are unknown parameters defining the mathematical function f to be
estimated.
To illustrate the problem, let us suppose that a data set containing the number of vaca-
tion days abroad (Y) and personal income (X) for various individuals (i) is available, and
let us plot this information for all the individuals in the sample using a two-dimensional

DWYER 9781781001288 Part I.indd 32 25/07/2012 11:06

 Regression analysis 33

Figure 2.1 XY plot

diagram, conventionally termed a scatter diagram, where each point on it represents

an individual from the sample (Figure 2.1).
Figure 2.1 suggests that higher values of X tend to yield higher values of Y, although
the relationship is not perfect. (It seems that knowledge of X does not suffice for an
entirely accurate prediction of Y.) In other words, the effect of income on vacation days
appears to differ across individuals, or else factors other than income influence vacation
days. Although omitted variables can lead to a bias problem (as pointed out later), the
general assumption is that vacation days are determined by earnings and by an aggrega-
tion of omitted factors named noise.
At this point, one of the simplest options is the assumption of linearity between the
variables; that is, each additional amount of income adds the same amount to vacation
days.1 Then the hypothetical relationship between vacation days and income may be
written as:

Y 5 b0 1 b1X 1 e (2.2)

where, b0 represents the constant term (that can be interpreted as the number of vacation
days with zero income); b1 the effect of an additional unit of income; and e the noise
term reflecting other factors that influence earnings. The task of a regression analysis is
to produce an estimate of unobservable parameters b0 and b1, based upon information
contained in the data set and upon some assumptions about the characteristics of e.
A simple regression analysis can be illustrated graphically thanks to the problems two
dimensions. Thus if the noise term (e) is ignored, Equation 2.2 becomes a line with an
intercept of b0 on the vertical axis and a slope of b1 (Figure 2.2), and so it could be
said that the task of estimating b0 and b1 is equivalent to the task of estimating where this
line is located.

DWYER 9781781001288 Part I.indd 33 25/07/2012 11:06

 34 Handbook of research methods in tourism

Y
= 1Y
X

Figure 2.2 XY plot with regression line

In order to estimate this line, it is useful to define the estimated error for each
observation as the vertical distance between the value of Y along the estimated line
Yi 5 b0 1 b1Xi (where Yi is a generated observation i of Y by plugging the actual value
of Xi into the equation). Thus estimated errors for each observation i may be calculated
as follows:

ei 5 Yi 2 Yi (2.3)

It is important to note that for each possible line that might be estimated, this will lead
to a different set of estimated errors. Thus in the most simple usual context, regression
analysis then chooses from among all possible lines by selecting the one for which the
sum of the squares of the estimated errors is at a minimum. This is termed the minimum
sum of squared errors or Ordinary Least Squares (OLS) criterion. The virtues of the OLS
criterion reside in the fact that it is very easy to use computationally and it has attractive
statistical properties under plausible noise term assumptions, as discussed later.

Multiple Regression

The number of vacation days abroad clearly seems to be affected by a variety of factors
in addition to personal income. A multiple regression is a procedure that generalizes
a simple regression by separately incorporating additional factors into the analysis.
Because the effect of each of these additional factors is estimated, it is valuable because it
quantifies the impact of various simultaneous influences on a single dependent variable.
Furthermore, multiple regressions are often indispensable, even when we are interested
in examining the effect of one of the independent variables, given the problem of the bias
caused by omitted variables in a simple regression.

DWYER 9781781001288 Part I.indd 34 25/07/2012 11:06

 Regression analysis 35

Let us now assume that the number of vacation days abroad (Y) is determined by two
attributes: personal income (X1) and the exchange rate (X2). If a linear effect between the
exchange rate and vacation days is assumed, the modified model may be written as:

Y 5 b0 1 b1X1 1 b2X2 1 e (2.4)

where b2 represents the effect of an additional unit on the exchange rate.

The task of estimating parameters b0, b1, and b2 is conceptually identical to the earlier
task of estimating just b0 and b1. The difference is that it is no longer a question of choos-
ing a line on a two-dimensional diagram. For Equation (2.4), regression analysis will
select a plane so that the sum of the squared errors is at a minimum. However, for a more
general multiple regression with more than two explanatory variables, the vision of the
problem that a graph allows is lost, although the estimation procedures remain the same.
Analytically, if it is assumed that the dependent variable (Y) is determined by n attributes
(X1, X2, ..., Xn), then the general model may be written as:

Y 5 b0 1 b1X1 1 b2X2 1 c1 bnXn 1 e (2.5)

where b0 is the constant term that indicates the estimated value of Y when all the
explanatory variables are 0 (X1 5 X2 5 ...5 Xn 5 0), and b1, b2, ..., bn is the vector
of parameters to be estimated, indicating the effect of an additional unit of each of the
independent variables (X1, X2, ..., Xn) on Y when the rest of the independent variables
remain constant. As in the case of a simple regression, the OLS criterion is an attractive
option because it is easy to compute and also due to its attractive statistical properties,
as discussed in continuation.

Essential Assumptions and Statistical Properties

The assumptions and statistical properties that are required in regression analysis are
dependent on the noise term e being considered a random variable, drawn by nature
from some probability distribution. It thus becomes clear that the estimates of b1, b2,
..., bn will also be random variables, because estimates generated by the minimum OLS
criterion (OLS estimator)2 will depend upon the particular value of e, drawn by nature
for each observation in the data set. Likewise, because there is a probability distribution
from which each ei is drawn, there must also be a probability distribution from which
each parameter estimate is drawn, with the distribution of the latter being a function of
the distributions of the former. The attraction of the statistical properties of a regres-
sion all reside in the relationship between the probability distribution of the parameter
estimates and the true values of those parameters. Thus if the mean of that probability
distribution is equal to the true value of the parameter we are trying to estimate, the
estimator is unbiased.
An estimator is termed consistent if it takes advantage of additional data to generate
more accurate estimates. In other words, a consistent estimator yields estimates that con-
verge on the true value of the underlying parameter as the sample size gets larger. Thus
the probability distribution of the estimate of any parameter will have a lower variance
as the sample size increases, and in the limit (infinite sample size), the estimate will equal

DWYER 9781781001288 Part I.indd 35 25/07/2012 11:06

 36 Handbook of research methods in tourism

the true value. At this point it should be noted that a lower variance in the probability
distribution of the estimator is clearly desirable, because it reduces the probability of an
estimate that differs greatly from the true value of the underlying parameter. In compar-
ing different unbiased estimators, the one with the lowest variance is termed efficient or
best.
The OLS estimator is characterized as being unbiased, consistent, and efficient under
certain assumptions. First, if the noise term for each observation, ei, is drawn from a
distribution that has a zero mean, the sum of squared errors criterion generates estimates
that are unbiased and consistent. Thus for each observation in the sample, a noise term
is drawn by nature from a different probability distribution. As long as each of these dis-
tributions has a zero mean, the minimum OLS criterion is unbiased and consistent. This
assumption is sufficient to ensure that each of the explanatory variables in the model is
uncorrelated with the expected value of the noise term.
Second, if the distributions from which the noise terms are drawn for each observa-
tion have the same variance, and the noise terms are statistically independent from each
other, then the sum of squared errors criterion gives the most efficient estimates avail-
able from any linear estimator. Additionally, it should be highlighted that, if this second
assumption is violated, the OLS criterion will remain unbiased and consistent, and it is
possible to decrease the variance of the estimator by considering what is known about
the noise term. The statistical procedures for dealing with this sort of problem go beyond
the scope of this chapter.3 However, when regression analysis in tourism literature is
considered in the next section, an outline will be given of some cases of special interest in
the field of tourism analysis.

APPLICATIONS OF REGRESSION ANALYSIS TO TOURISM

Tourism Demand Modeling and Forecasting

One field of research in academic tourism literature where regression analysis has been
used as a central research tool is tourism demand modeling and forecasting (Song and
Li, 2008). Frechtling (1996) argues that tourism products and the tourist industry have
different characteristics from other products or sectors as reasons why the determinants
of tourism demand must be identified for forecasting purposes, more than in any other
activity. These reasons include the fact that tourism products cannot be stored, and so
any unused supply cannot be kept for a later date when there is a higher demand. Neither
is it possible to separate the production process from consumption. This interaction
between consumers and producers means that goods and services must be offered when
there is a demand. At the same time, it must be remembered that tourist satisfaction is
largely dependent on complementary services. For example, destination loyalty or the
repeat visitation rate are not only dependent on tourism accommodation facilities, but,
more specifically, on all the other goods and services consumed by tourists directly or
indirectly during their stay. Likewise, tourism demand is extremely sensitive to natural
disasters and socio-political problems, such as wars, terrorist attacks, or natural dis-
asters. Finally, we must also add the need for long-term investment into facilities and
infrastructure able to meet the expectations of future demands.

DWYER 9781781001288 Part I.indd 36 25/07/2012 11:06

 Regression analysis 37

The literature highlights the existence of two big groups of quantitative methods of
determining and forecasting tourism demand: univariate and causal models (Witt and
Witt, 1995). Univariate models, used primarily for the purposes of forecasting, are based
on the assumption that forecasts can be made without including factors that determine
the level of the variable. Thus the only information that they require is the past evolution
of the variable to be forecasted. Within the regression analysis context, a simple example
of a univariate model can be represented by the autoregressive model

Yt 5 b0 1 b1Yt21 1 b2Yt22 1 c1 bnYt2n 1 ut (2.6)

where it is important to note that observations of tourism demand (Y) refer to time
periods t (years, quarters, months, etc) and that the explanatory variables are, in fact, the
lags of the same dependent variable. Different options have been proposed for variables
capable of capturing the concept of tourism demand, with the number of tourists being
the most common choice followed by tourist expenditures (Crouch, 1994).
Whatever the case, in regression analyses that use time series data, autocorrelation of
the errors is a frequent problem, violating the OLS assumption that the error terms are
uncorrelated. (This is why the error term is denoted by ut instead of et.) While it does
not bias the OLS coefficient estimates, standard errors tend to be underestimated when
the autocorrelations of the errors at low lags are positive. Autocorrelation of the errors,
which themselves are unobserved, can generally be detected, because it produces auto-
correlation in the observable residuals, and different alternatives have been proposed to
overcome this problem (Box et al., 1994).
Although an autoregressive model is a basic univariate model, the complexity of this
set of models (including the most illustrative problem of autocorrelation using time
series data) depends on the number of mathematical operations and implicit assump-
tions that are involved. Methods such as BoxJenkins, which are relatively complex
given the high number of estimations that are required, offer no computational prob-
lems nowadays, thanks to the numerous econometric programs that are available. In the
tourism demand literature, forecasts obtained with more complex models normally use
forecasts generated by simple univariate models to demonstrate (or reject) their superi-
ority (Kulendran and Witt, 2001; Lim and McAleer, 2001; Rossell, 2001; Smeral and
Wger, 2005).
Within the univariate context, an alternative group of models can be included that
considers time (instead of the lagged independent variable) as the only variable in deter-
mining tourism demand. Witt and Witt (1991) propose ten different functional forms
that try to explain the growth in tourism demand. Witt and Witt (1995) introduced the
use of the Gompertz curve for forecasting, based on the evolutionary stage of the tourism
product in the lifecycle. Later, Chan (1993), Chu (1998a, 1998b, 2004) and Wong (1997a)
focused on the strong seasonal component and proposed the use of the sinusoid function
for forecasting.
When a causal framework is considered in tourism demand, a simple example of a
regression model can be represented by the following equation:

Yt 5 b0 1 b1X1t 1 b2X2t 1 c1 bnXnt 1 ut (2.7)

DWYER 9781781001288 Part I.indd 37 25/07/2012 11:06

 38 Handbook of research methods in tourism

It should be noted that again the variables included in the regression analysis show t
as a subscript, denoting the time dependence of the observations included in the model
both for the dependent (Y) and independent variables (X1, X2, ..., Xn), as well as an error
term (ut) often characterized by a high level of autocorrelation. Although the economic
underpinnings of Equation (2.6) were described in Morley (1992) and Sakai (1988), the
first estimation of these models can be found in Gerakis (1965), Gray (1966) and Laber
(1969). In fact, empirical applications developed during the 1970s and the 80s did not
vary very much from these initial applications, despite the theoretical difficulties involved
in aggregating individual tourism demands (Morley, 1995) and the neglected treatment
of the ut term and use of diagnostic tests (Crouch, 1994; Lim, 1997a).
The set of independent variables (X1, X2, ..., Xn) can include consumers approximate
income, generally measured by using available family income or the per capita GDP
(Archer, 1980; Deng and Athanasopoulos, 2011; Gray, 1982; Harrop, 1973; Kulendran
and Divisekera, 2007); the price of the destination, usually specified as the ratio between
the consumer price indexes of the destination and country of origin, given the difficulty
in finding statistical series of tourism prices (Jung and Fujii, 1976; Kulendran and
Divisekera, 2007; Kwack, 1972; Rosensweig, 1988); the cost of transport from the issuing
country to the destination (Brida and Risso, 2009; Martin and Witt, 1988); the nominal
(or real) exchange rate between the origin country and the destination (Brida and Risso,
2009; Kulendran and Divisekera, 2007; Tremblay, 1989; Truett and Truett, 1987); the
price of substitute and/or complementary destinations, usually defined in the form of an
index, based on the weighted average of the prices of other destinations that compete for
the same market as the destination under consideration (Divisekera, 2003; Uysal and
Crompton, 1985; Witt and Martin, 1987, advertising or the cost of promotion; and other
types of variables such as a change in preferences (Barry and OHagan, 1972; Crouch et
al., 1992; Kulendran and Divisekera, 2007), the capital invested in infrastructure, which
defines both the tourist industrys quality and capacity (Carey, 1989; Geyikdagi, 1995,
Khadaroo and Seetanah, 2007), the stock of immigrants at the home country (Seetaram
and Dwyer, 2009) or at the destination (Dwyer et al., 2010; Seetaram, 2012), and even
meteorological variables (Barry and OHagan, 1972).
No definitive conclusions can be reached from different meta-analyses of the results of
tourism demand modeling made before the 90s (Crouch 1994, 1996; Lim 1997b, 1999)
with regard to the value of the parameters of those variables most often used in specifica-
tions. What does stand out is the infrequency with which econometric tests are used to
examine the basic hypotheses of the linear regression model, together with the absence
of serial correlation, homoscedasticity, the linear functional form, the normality of the
error term, and the exogeneity of the independent variables (Lim, 1997a).
Possibly, this is why since the 1990s studies have tried to improve on specifications.
Wong (1997b), for example, analyses the statistical properties of series of tourist arriv-
als from different countries, demonstrating an absence of stationarity in most cases.
The solution to this problem involves the use of cointegration analysis, which was not
widely applied to aggregate estimation models and tourism demand forecasting until the
1990s, with some of the most representative studies being made by Bonham and Gangnes
(1996), Dritsakis (2004), Kulendran (1996), Kulendran and King (1997), Lathiras and
Syriopoulos (1998) and Syriopoulos (1995).
Within the causal model category, demand systems or tourist expenditure systems

DWYER 9781781001288 Part I.indd 38 25/07/2012 11:06

 Regression analysis 39

should also be mentioned. These have centered their attention on the problem of consum-
ers faced with a choice of market goods and services. The formulation of simultaneous
equation systems seeks to avoid the biases that occur with uni-equation models, when
consumer decisions regarding tourism services are considered in isolation from all other
consumer goods. The first applications of this methodology can be found in Kliman
(1981), Taplin (1980) and Van Soest and Kooreman (1987). However, the appearance
of an article by Deaton and Muellbauer (1980) lent greater weight by laying the founda-
tions of consumer behavior theory, leading to the adoption of the methodology and its
application to tourism demand in Divisekera (2003), Fujii et al. (1985), Pyo et al. (1991),
and Sakai (1988). In OHagan and Harrison (1984), Smeral (1988), or Syriopoulos and
Sinclair (1993), the model is extended to include the second-stage selection of one of a
choice of destinations, so that the following third stage would correspond to the alloca-
tion of spending on goods and services at each destination. During the last decade, simul-
taneous equation systems have centered on an analysis of time series properties within
the said system (Corts-Jimnez et al. 2009; De Mello and Fortuna, 2005; Durbarry and
Sinclair, 2003; Li et al., 2004; Li et al., 2006; Mangion et al, 2005).
Nonetheless, together with aggregate models, thanks to recent developments in the
field of microeconometrics, models with microdata have been gaining ground. Since the
earliest applications by Morley (1992), Rugg (1973), and Witt (1982), based on discrete
choice models, random utility models have been used as a tool to estimate the parameters
that determine a certain choice of destination (Morley 1994a and 1994b). In reference to
regression analysis, the main problem that these models have to solve is the nature of the
dependent variable, which could be binomial (entailing making or not making a specific
choice, and so it can only take 0 or 1), multinomial (entailing making a choice from a
particular set), count data (observations can take only non-negative integer values), etc.
Different applications have attempted to tackle problems of various kinds. One of the
main goals within this new framework is the number of dependent variables (dependent
on the aim of the study) that may influence such decisions (Correia et al., 2007; Eugenio-
Martin and Campos-Soria, 2010; Huybers, 2005; Nicolau and Ms, 2005; Pestana et al.,
2008; Thrane, 2008).

Supply-oriented, Policy, and Environmental Analyses and Other Issues

Given that regression analysis is a research tool capable of finding a statistical relation-
ship between an objective (dependent) variable and a set of causal factors (independent
variables), it is not surprising that its use in tourism research has not been limited to
tourism demand analysis. In analyses of the supply, the first issue that comes to mind
is modeling the tourist supply function (Borooah, 1999), including, for instance, the
number of hotel beds as a dependent variable and other factors, such as previous demand
indicators, quality, or exchange rates, as independent ones. However, it should be noted
that the huge complexity of defining the tourism supply product, a lack of supply-related
data at most destinations and difficulty in identifying the different components that make
up the tourism product have led analyses of the supply and the use of regression analysis
to center on some specific issues.
One of the most interesting of these applications is the hedonic pricing model, where
an explanation of how prices are formed from different key characteristics is proposed.

DWYER 9781781001288 Part I.indd 39 25/07/2012 11:06

 40 Handbook of research methods in tourism

Using regression analysis as a reference, the price of a package holiday or overnight

stays at a hotel are used as the dependent variable, explained by a set of explanatory
variables, such as the destinations location, number of nights stay, hotel category,
characteristics of the services offered by the hotels, etc. Some applications of the hedonic
regression method in tourism can be found in Clewer et al. (1992), Espinet et al. (2003),
Haroutunian and Pashardes (2005), Mangion et al. (2005), Papatheodorou (2002),
Rambonilaza (2006), Rigall-i-Torrent and Fluvi (2007), Sinclair et al. (1990), and
Thrane (2005). Estimating the hedonic price function econometrically allows percentage
price differences among destinations to be derived, together with the marginal value of
an additional unit of a certain characteristic.
Davies and Downward (1996) centered their attention on the amount of sales, using
a sample of UK hotel companies during the period 198993. Using regression analysis,
variations in benefits in relation to sales were explained as a function of lagged market
share variables, the cyclical evolution of the unemployment rate and a concentration
index. Skill at capturing a marked segment, with a longer season by hotels, is analysed
in Cap et al. (2007), where the number of months the hotel stays open is explained
through a set of supply factors, using a sample of Balearic hotels. In this case, because
of the integer and limited nature of the independent variable (number of months), an
ordered logit was used as the estimation procedure. The problem of seasonality and its
determinants is also studied in Rossell et al. (2004), where the economic determinants
of seasonal patterns are analysed from a demand perspective using the Gini coefficient
as the dependent variable.
Regression analysis has not only focused on revenues and benefits from tourism. The
environmental effects of tourism have generated debate on a new concept sustainable
tourism which has been the object of growing attention by governments, international
organizations, academic institutions, and individual researchers. A good reflection of the
interest that this issue has aroused is the abundance of literature mainly motivated by
an analysis of factors that can help to achieve sustainable tourism levels. Most natural
resources related to tourism are characterized by the impossibility of their exclusion, so
the role of prices as an indicator of scarcity or as a motivating factor in the behavior of
economic agents disappears. The result is the deterioration and over-exploitation of the
environment. That is why it is necessary to convert the importance that the environment
represents in social wellbeing into monetary units and then to internalize the externalities
generated by tourism, thereby achieving a sustainable model of tourism development.
Valuation methodologies can be classified by differentiating those that produce a
direct monetary valuation and those that produce an indirect one. The first group
includes the contingent valuation method and bidding games, while the second involves
the travel cost and hedonic pricing methods.4 Both groups, especially the second, require
a regression analysis for different purposes. Contingent valuation involves asking people
directly (or indirectly) for an economic valuation of an environmental (or non-market)
good or service. Some examples of its application include Asafu-Adjaye and Tapsuwan
(2008) for scuba diving; Dutta et al. (2007) for heritage tourism; Greiner and Rolfe
(2004) for natural areas; Lee and Han (2002) for national parks, and Rulleau et al. (2011)
for recreational values. The usefulness of regression analysis in contingent valuation
techniques lies mainly (but not exclusively) in the quantification of the payment decision.
The travel cost method, which is suited to estimating the user value of a natural

DWYER 9781781001288 Part I.indd 40 25/07/2012 11:06

 Regression analysis 41

resource when it is mainly visited for recreational purposes, has been used to assess the
value of destinations specializing in nature or ecotourism and natural areas or resources
located at destinations where the main motivation for the trip is a visit to a natural
area or contact with the natural environment (Herath and Kennedy, 2004; Loomis and
Keske, 2009; Maille and Mendelsohn, 1993; Riera, 2000; Wang at al. 2009). In this case,
regression analysis can be used in the estimation of the demand for visits to a certain
destination.
Once environmental issues have been evaluated, policy measures must be considered
and applied. Although a wide range of instruments can be found in the literature on
policies for the protection of the environment, the internalization of externalities and
management of public services most commonly analysed through regression analysis
are market-based instruments. Thus a possible extension of demand studies that con-
sider prices to be a determining factor in tourism is an analysis of taxes (Aguil et al.,
2005; Blair et al., 1987; Bonham et al., 1991; Palmer et al., 2007). When a unit tax is
levied on tourism accommodation, the tax burden falls on buyers and sellers, depend-
ing on supply and demand elasticities. Generally speaking, levying an ad-valorem tax
on accommodation will reduce occupancy rates by raising prices. The price rise for
consumers, divided by the reduction in the producer price, is more or less equal to the
ratio between the supply and demand elasticities. The higher the elasticity of supply
compared with the elasticity of demand, the higher the tax burden on the buyers. In
cases where there is an infinite elasticity of supply or zero demand, the tax burden falls
exclusively on buyers.

LIMITATIONS OF REGRESSION ANALYSIS

Since the true form of the data-generating process is generally not known, regression
analysis often depends to some extent on making assumptions about this process. Two
common mistakes are (1) to reach the conclusion that there is a strong link between two
variables, while the influence of another perhaps more important one may not have been
estimated and (2) not to take into account the fact that relations between the different
explained and explanatory variables can be circular (X explains Y and Y explains X).
Other limitations that should be highlighted are sometimes (but not always) testable if
a large amount of data is available and they are linked to the assumptions of the regres-
sion analysis. First, as is evident in the name multiple linear regression, it is assumed that
the relationship between variables is linear. As a general rule, it is prudent to look always
at bivariate scatterplots of the variables under analysis. If curvature in the relation-
ships is evident, it is possible to consider either transforming the variables or explicitly
allowing for non-linear components. In practice, the assumption of linearity is difficult
to confirm and more complex functional forms require a large amount of data that is
not always available. Second, in reference to the normal distribution of the residuals,
even though most tests are quite robust with regard to violations of this assumption, it
is always a good idea, before drawing final conclusions, to review the distributions of
the major variables of interest. The recommendation here is to depict histograms for the
residuals, as well as normal probability plots, in order to inspect the distribution of the
residual values.

DWYER 9781781001288 Part I.indd 41 25/07/2012 11:06

 42 Handbook of research methods in tourism

Additional limitations concerning the amount of explanatory variables to be used and

the necessary number of observations should also be mentioned. Although one can be
seduced by including as many predictor variables as possible (and usually at least a few of
them will come out significant), it is important to be aware that the confidence intervals
are not 100 percent, so it is possible (a low probability, but higher when an additional
number of independent variables is included) to find a statistical relationship when, in
fact, there is no relationship between the variables. This problem is compounded when,
in addition, the number of observations is relatively low. Some authors recommend that,
as a general rule, one should have at least 10 to 20 times as many observations as one has
variables, otherwise parameter estimations are probably very unstable and unlikely to be
replicated if the study were conducted again.
The multicollinearity problem should also be highlighted here. Multicollinearity can
appear when two or more predictor variables in a multiple regression model are highly
correlated and, as a result, the coefficient estimates may change erratically in response
to small changes in the model or the data. Multicollinearity does not reduce the predic-
tive power or reliability of the model as a whole, at least within the sample data itself.
However, it can affect calculations regarding individual predictors. A high degree of
multicollinearity can also cause computer software packages to be unable to perform
the matrix inversion that is required for computing the regression coefficients, or it may
make the results of that inversion inaccurate.
Even though most multiple regression assumptions cannot be tested explicitly, major
violations can be detected and they should be dealt with appropriately. Error analysis
and test diagnostics can be used to detect particular problems and, in the end, to validate
results obtained through regression analysis. However, the statistical procedures for
dealing with this sort of problem go beyond the scope of this chapter.

CONCLUSIONS

Regression analysis has become a popular statistical instrument in tourism for the
investigation of relationships between variables. The investigator looks to determine
the causal effect of one or more variable upon another; for instance, the effect of a price
increase or income decrease on tourism demand. To explore such issues, the investiga-
tor collects data for the variables of interest and uses regression analysis to estimate the
quantitative effect of the causal variables on the dependent variable. Additionally the
degree of confidence of the true relationship being close to the estimated one (known as
statistical significance) is also usually assessed in order to make the exercise more cred-
ible.
Regression techniques have long been central to the field of econometrics and increas-
ingly they have become important to tourism researchers as well. In this chapter, an
overview of the technique has been given in order to show how regression analysis works,
what it assumes, and what the most typical limitations are, using different examples
from tourism literature. It has been shown that regression analysis has mainly focused
on determining tourism demand, both in order to forecast and quantify the relation-
ship between its determinants. However, its use in tourism research has not been limited
to demand analysis. Other tourism research issues such as tourism supply, analyses of

DWYER 9781781001288 Part I.indd 42 25/07/2012 11:06

 Regression analysis 43

policy measures, valuation techniques applied to environmental tourism issues, and, in

general, any study aimed at quantifying the relationship between a dependent variable
and a set of causal factors have benefited and can benefit in the future from regression
analysis.
The major conceptual limitation of all regression techniques is that you can only dis-
cover relationships, but never be sure about underlying causal mechanisms. Nonetheless,
as a bare minimum, regression analysis helps establish the existence of connections that
call for closer investigation.

NOTES
1. Although the linearity assumption is common in regression studies, it is by no means essential to the appli-
cation of the technique, and it can easily be relaxed when the researcher has reason to suppose a priori that
the relationship in question is non-linear. For instance, the popular relationship Y 5 aXb (a and b being
parameters to be estimated) can be transformed into the linear relationship lnY 5 lna 1 blogX. In this
context, b can be interpreted directly as the elasticity between Y and X.
2. Please note that other alternative criteria for generating parameter estimates would also be considered as
estimators, such as minimizing the sum of the absolute values of errors.
3. Numerous manuals deal with the issue of regression analysis. Some of the most popular among economists
are Greene (2006) and Gujarati and Porter (2009).
4. The hedonic price model is discussed in Chapter 9 of this volume.

REFERENCES

Aguil, E., A. Riera and J. Rossell (2005), The short-term price effect of a tourist tax on the demand for
tourism through a dynamic demand model, Tourism Management, 26 (3), 359365.
Archer, B.H. (1980), Forecasting demand: quantitative and intuitive techniques, International Journal of
Tourism Management, 1 (1), 512.
Asafu-Adjaye, J. and S. Tapsuwan (2008), A contingent valuation study of scuba diving benefits: case study in
Mu Ko Similan Marine National Park, Thailand, Tourism Management, 29 (6), 11221130.
Barry, K. and J. OHagan (1972), An econometric study of British tourist expenditure in Ireland, Economic
and Social Review, 3 (2), 143161.
Blair, A.R., F. Giarrantini and H. Spiro (1987), Incidence of the amusement tax, National Tax Journal, 40
(1), 6169.
Bonham, C.S. and B. Gangnes (1996), Interevention analysis with cointegrated time series: the case of Hawaii
hotel room tax, Applied Economics, 28, 12811293.
Bonham, C.S, E. Fujiii, E. Im and J. Mak (1991), The impact of the hotel tax room: an interrupted time series
approach, National Tax Journal, 45 (4), 433441.
Borooah, V. (1999), The supply of hotel rooms in Queensland, Australia, Annals of Tourism Research, 26,
9851003.
Box, G.E.P., G.M. Jenkins and G.C. Reinsel (1994), Time Series Analysis: Forecasting and Control, Upper
Saddle River, NJ: Prentice-Hall.
Brida, J.G. and W.A. Risso (2009), A dynamic panel data study of the German demand for tourism in South
Tyrol, Tourism and Hospitality Research, 9 (4), 305313.
Cap, J., A. Riera and J. Rossell (2007), Accommodation determinants of seasonal patterns, Annals of
Tourism Research, 34 (2), 422436.
Carey, K. (1989), Tourism development in LDCs: hotel capacity expansion with reference to Barbados,
World Development, 17 (1), 5967.
Chan, Y.M. (1993), Forecasting tourism: a sine wave time series regression approach, Journal of Travel
Research, 32 (2), 5860.
Chu, F.L. (1998a), Forecasting tourist arrivals: nonlinear sine wave or ARIMA?, Journal of Travel Research,
36 (3), 7984.
Chu, F.L. (1998b), Forecasting tourism: a combined approach, Tourism Management, 19 (6), 515520.

DWYER 9781781001288 Part I.indd 43 25/07/2012 11:06

 44 Handbook of research methods in tourism

Chu, F.L. (2004), Forecasting tourism demand: a cubic polynomial approach, Tourism Management, 25 (2),
209218.
Clewer, A., A. Pack and M.T. Sinclair (1992), Price competitiveness and inclusive tourim holidays in
Europeancities, in P. Johnson and B. Thomas (eds) Choice and Demand in Tourism, London: Mansell, pp.
123144.
Correia, A., C.M. Santos and C.P. Barros (2007), Tourism in Latin America. a choice analysis, Annals of
Tourism Research, 34 (3), 610629.
Corts-Jimnez, I., R. Durbarry and M. Pulina (2009), Estimation of outbound Italian tourism demand: a
monthly dynamic EC-LAIDS model, Tourism Economics, 15 (3), 547565.
Crouch, G.I. (1994), A meta-analysis of tourism demand, Annals of Tourism Research, 22 (1), 103118.
Crouch, G.I. (1996), Demand elasticities in international marketing. a meta-analytical application to tourism,
Journal of Business Research, 36, 117136.
Crouch, G.I., L. Schultz and J. Valerio (1992), Marketing international tourism to Australia: a regression
analysis, Tourism Management, 13 (2), 196208.
Davies, B. and D. Downward (1996), The structure, conduct, performance paradigm as applied to the UK
hotel industry, Tourism Economics, 2 (2), 151158.
De Mello, M.M. and N. Fortuna (2005), Testing alternative dynamic systems for modelling tourism demand,
Tourism Economics, 11, 517537.
Deaton, A. and J. Muellbauer (1980), Economics and Consumer Behaviour, New York: Cambridge University
Press.
Deng, M. and G. Athanasopoulos (2011), Modelling Australian domestic and international inbound travel: a
spatial and temporal approach, Tourism Management, 32, 10751084.
Divisekera, S. (2003), A model of demand for international tourism, Annals of Tourism Research, 30 (1),
3149.
Dritsakis, N. (2004), Cointegration analysis of German and British tourism demand for Greece, Tourism
Management, 25 (1), 111119.
Durbarry, R. and M.T. Sinclair (2003), Market shares analysis: the case of French tourism demand, Annals
of Tourism Research, 30, 927941.
Dutta, M., S. Banerjee and Z. Husain (2007), Untapped demand for heritage: a contingent valuation study of
Prinsep Ghat, Calcutta, Tourism Management, 28, 8395.
Dwyer L., P. Forsyth, B. King and N. Seetaram (2010), Migration related determinants of Australian inbound
and outbound tourism flows, report submitted to Sustainable Tourism Cooperative Research Centre
(STCRC), Australia.
Espinet, J.M., M. Saez, G. Coenders and M. Fluvi (2003), Effect on prices of the attributes of holiday hotels:
a hedonic prices approach, Tourism Economics, 9 (2), 165177.
Eugenio-Martin, J.L. and J.A. Campos-Soria (2010), Climate in the region of origin and destination choice in
outbound tourism demand, Tourism Management, 31, 744753.
Frechtling, D.C. (1996), Practical Tourism Forecasting, London: Butteworth-Heinemann.
Fujii, E.T., M. Khaled and J. Mak (1985), An almost ideal demand system for visitor expenditures, Journal
of Transport Economics and Policy, May, 161171.
Gerakis, A.S. (1965), Effects of exchange-rate devaluations and revaluations on receipts from tourism,
International Monetary Fund Staff Papers, 12 (3), 36584.
Geyikdagi, N.V. (1995), Investments in tourism development and the demand for travel, Rivista Internationale
di Scienze Economiche e Commerciali, 42 (5), 391403.
Gray, H.P. (1966), The demand for international travel by the United States and Canada, International
Economic Review, 7 (1), 8392.
Gray, H.P. (1982), The contributions of economics to tourism, Annals of Tourism Research, 9 (1), 10525.
Greene, W. (2006), Econometric Analysis. Ventura, CA: Academic Internet Publishers.
Greiner, R. and J. Rolfe (2004), Estimating consumer surplus and elasticity of demand of tourist visitation to
a region in North Queensland using contingent valuation, Tourism Economics, 10 (3), 317328.
Gujarati, D.N. and D.C. Porter (2009), Basic Econometrics, New York: McGraw-Hill.
Haroutunian, S. and P. Pashardes (2005) Using brochure information for the hedonic analysis of holiday
packages, Tourism Economics, 11 (1), 6984.
Harrop, J. (1973), On the economics of the tourist boom, Bulletin of Economic Research, (May), 5572.
Herath, G. and J. Kennedy (2004), Estimating the economic value of Mount Buffalo National Park with the
Travel Cost and Contingent Valuation Models, Tourism Economics, 10 (1), 6378.
Huybers, T. (2005), Destination choice modelling: whats in a name?, Tourism Economics, 11 (3), 329350.
Jung, J.M. and E.T. Fujii (1976), The price elasticity of demand for air travel: some new evidence, Journal of
Transport Economics and Policy, 10 (3), 257262.
Khadaroo, J. and B. Seetanah (2007), Transport infrastructure and tourism development, Annals of Tourism
Research, 34, 10211032.

DWYER 9781781001288 Part I.indd 44 25/07/2012 11:06

 Regression analysis 45

Kliman, M.L. (1981), A quantitative analysis of Canadian overseas tourism, Transportation Research A, 15
(6), 487497.
Kulendran, N. (1996), Modelling quarterly tourist flows to Australia using cointegration analysis, Tourism
Economics, 2 (3), 203222.
Kulendran, N. and S. Divisekera (2007), Measuring the economic impact of Australian tourism marketing
expenditure, Tourism Economics, 13, 261274.
Kulendran, N. and M.L. King (1997), Forecasting international quarterly tourists flows using error-correction
and time-series models, International Journal of Forecasting, 13, 319327.
Kulendran, N. and S.F. Witt (2001), Cointegration versus least squares regression, Annals of Tourism
Research, 28 (2), 291311.
Kwack, S.Y. (1972), Effects of income and prices on travel spending abroad 1960 III1967 IV, International
Economic Review, 13 (2), 245256.
Laber, G. (1969), Determinants of international travel between Canada and the United States, Geographical
Analysis, 1 (4), 329336.
Lathiras, P. and C. Siriopoulos (1998), The demand for tourism to Greece: a cointegration approach, Tourism
Economics, 4 (2), 171185.
Lee, C.K. and S. Han (2002), Estimating the use and preservation values of national parks tourism resources
using a contingent valuation method, Tourism Management, 23 (5), 531540.
Lewis, R.C., R.E. Chambers and Il. E. Chacko (1995), Marketing: Leadership in Hospitality, New York: Van
Nostrand Reinhold.
Li, G., H. Song and S.F. Witt (2004), Modeling tourism demand: a dynamic linear AIDS approach, Journal
of Travel Research, 43, 141150.
Li, G., H. Song and S.F. Witt (2006), Time varying parameter and fixed parameter linear AIDS: an applica-
tion to tourism demand forecasting, International Journal of Forecasting, 22, 5771.
Lim, C. (1997a), An econometric classification and review of international tourism models, Tourism
Economics, 3 (1), 6982.
Lim, C. (1997b), Review of international tourism demand models, Annals of Tourism Research, 24 (4),
835849.
Lim, C. (1999), A meta-analytic review of international tourism demand, Journal of Travel Research, 37 (3),
273289.
Lim, C. and M. McAleer (2001), Forecasting tourism arrivals, Annals of Tourism Research, 28 (4), 965977.
Loomis, J. and C. Keske (2009), The economic value of novel means of ascending high mountain peaks:
a travel cost demand model of Pikes Peak cog railway riders, automobile users and hikers, Tourism
Economics, 15 (2), 426436.
Maille, R. and R. Mendelsohn (1993), Valuing ecotourism in Madagascar, Journal of Environmental
Management, 38, 213218.
Mangion, M.L., R. Durbarry and M.T. Sinclair (2005), Tourism competitiveness: price and quality, Tourism
Economics, 11, 4568.
Martin, C.A. and S.F. Witt (1988), Substitute prices in models of tourism demand, Annals of Tourism
Research, 15 (2), 255268.
Morley, C.L. (1992), A microeconomic theory of international tourism demand, Annals of Tourism Research,
19, 250267.
Morley, C.L. (1994a), Discrete choice analysis of the impact of tourism prices, Journal of Travel Research,
33 (2), 814.
Morley, C.L. (1994b), Experimental destination choice analysis, Annals of Tourism Research, 21 (4), 780791.
Morley, C.L. (1995), Tourism demand: characteristics, segmentation and aggregation, Tourism Economics,
1 (4), 315328.
Nicolau, J.L. and F.J. Ms (2005), Stochastic modeling. A three-stage tourist choice process, Annals of
Tourism Research, 32 (1), 4969.
OHagan, J.W. and M.J. Harrison (1984), Market shares of US tourist expenditure in Europe: an econometric
analysis, Applied Economics, 16, 919931.
Palmer, T., A. Riera and J. Rossell (2007), Taxing tourism: the case of rental cars in Mallorca, Tourism
Management, 28 (1), 271279.
Papatheodorou, A. (2002), Exploring competitiveness in Mediterranean resorts, Tourism Economics, 8 (2),
133150.
Pestana, C., R. Butler and A. Correia (2008), Heterogeneity in destination choice. Tourism in Africa, Journal
of Travel Research, 47 (2), 235246.
Pyo, S.S., M. Uysal and R.W. Mclellan (1991), A linear expenditure model for tourism demand, Annals of
Tourism Research, 18 (3), 443454.
Rambonilaza, M. (2006), Labelling and differentiation strategy in the recreational housing rental market of
rural destinations: the French case, Tourism Economics, 12 (3), 347359.

DWYER 9781781001288 Part I.indd 45 25/07/2012 11:06

 46 Handbook of research methods in tourism

Riera, A. (2000), Mass tourism and the demand for protected natural areas: a travel cost approach, Journal
of Environmental Economics and Management, 39, 97116.
Rigall-i-Torrent, R. and M. Fluvi (2007), Public goods in tourism municipalities: formal analysis, empirical
evidence and implications for sustainable development, Tourism Economics, 13 (3), 361378.
Rosensweig, J.A. (1988), Elasticities of substitution in Caribbean tourism, Journal of Development Economics,
29 (1), 89100.
Rossell, J. (2001), Forecasting turning points in international tourist arrivals in the Balearic Islands, Tourism
Economics, 7 (4), 365380.
Rossell, J., A. Riera and A. Sans (2004), The economic determinants of seasonal patterns, Annals of
Tourism Research, 31 (3), 697711.
Rugg, D. (1973), The choice of journey destination: a theoretical and empirical analysis, Review of Economics
and Statistics, 55 (1), 6472.
Rulleau, B., J. Dehez and P. Point (2011), Recreational value, user heterogeneity and site characteristics in
contingent valuation, Tourism Management, doi:10.1016/j.tourman.2011.03.002.
Sakai, M.Y. (1988), A micro-analysis of business travel demand, Applied Economics, 20, 14811495.
Seetaram, N. (2012), Estimating demand elasticities for Australias international outbound tourism, Tourism
Economics, (in press).
Seetaram, N. and L. Dwyer (2009), Immigration and tourism demand in Australia: a panel data approach,
ANATOLIA: An International Journal of Tourism and Hospitality Research, 20 (1), 212222.
Sinclair, M.T., A. Clewer and A. Pack (1990), Hedonic prices and the marketing of package holidays: the case
of tourism resorts in Malaga, in G.J. Ashworth and B. Goodall (eds) Marketing of Tourism Places, London:
Routledge, pp. 85103.
Smeral, E. (1988), Tourism demand, economic theory and econometrics: an integrated approach, Journal of
Travel Research, 26, 3843.
Smeral, E. and M. Wger (2005), Does complexity matter? Methods for improving forecasting accuracy in
tourism: the case of Austria, Journal of Travel Research, 44 (1), 100110.
Song, H. and G. Li (2008), Tourism demand modeling and forecasting: a review of recent research, Tourism
Management, 29, 203220.
Syriopoulos, T.C. (1995), A dynamic model of demand for Mediterranean tourism, International Review of
Applied Economics, 9 (3), 318336.
Syriopoulos, T.C. and M.T. Sinclair (1993), An econometric study of tourism demand: the AIDS model of US
and European tourism in Mediterranean countries, Applied Economics, 25, 15411552.
Taplin, J.H.E. (1980), A coherence approach to estimates of price elasticities in the vacation travel market,
Journal of Transport Economics and Policy, May, 1935.
Thrane, C. (2005), Hedonic price models and sun-and-beach package tours: the Norwegian case, Journal of
Travel Research, 43 (3), 302308.
Thrane, C. (2008), The determinants of students destination choice for their summer vacation trip,
Scandinavian Journal of Hospitality and Tourism, 8 (4), 333348.
Tremblay, D. (1989), Pooling international tourism in Western Europe, Annals of Tourism Research, 16 (4),
477491.
Truett, D.B. and L.J. Truett (1987), The response of tourism to international economic conditions: Greece,
Mexico, and Spain, The Journal of Developing Areas, 21 (2), 177190.
Uysal, M. and J.L. Crompton (1985), An overview of approaches used to forecast tourism demand, Journal
of Travel Research, 23, 715.
Van Soest, A. and P. Kooreman (1987), A micro-econometric analysis of vacation behaviour, Journal of
Econometrics, 2, 215226.
Wang, E., Z. Li, B.B. Little and Y. Yang (2009), The economic impact of tourism in Xinghai Park, China: a
travel cost value analysis using count data regression models, Tourism Economics, 15 (2), 413425.
Witt, S.F. (1982), A binary choice model of foreign holiday demand, Journal of Economic Studies, 10 (1),
4659.
Witt, S.F. and C.A. Martin (1987), Deriving a relative price index for inclusion in international tourism
demand estimation models: comment, Journal of Travel Research, 25 (3), 2330.
Witt, S.F. and C.A. Witt (1991), Tourism forecasting: error magnitude, direction of change error, and trend
change error, Journal of Travel Research, 30 (2), 2633.
Witt, S.F. and C.A. Witt (1995), Forecasting tourism demand: a review of empirical research, International
Journal of Forecasting, 11 (3), 447475.
Wong, K. (1997a), The relevance of business cycles in forecasting international tourist arrivals, Tourism
Management, 18 (8), 581586.
Wong, K. (1997b), An investigation of the time series behaviour of international tourist arrivals, Tourism
Economics, 3 (2), 185200.

DWYER 9781781001288 Part I.indd 46 25/07/2012 11:06

 3 Time series analysis
Shuang Cang and Neelu Seetaram

INTRODUCTION

A time series typically consists of a set of observations made on a variable y, taken at

equally spaced internals over time (Harvey, 1993, p. 1). The study of time series data is
normally performed on two levels. Time series analysis comprises methods for extracting
meaningful statistics and identifying key characteristics of the data. Time series model-
ling is performed in order to predict the future behaviour of y using historical data. The
basis of time series modelling is that movement in y is explained solely by its past values
or by its position in relation [to] time (Harvey, 1993). Since time series models only
require historical observations of a variable, it is less costly in terms of data collection
and model estimation (Song and Li, 2008).
In tourism research, both time series analysis and time series modelling have received
great attention in the past three decades, perhaps due to the availability of relevant time
series data. Common types of tourism time series are tourist arrivals, departures, expend-
iture levels, number of nights spent at a destination, price of hotel room, airfare, tourism
contribution to the GDP and so on. These time series are mostly available on a monthly,
quarterly and/or yearly basis. Frechtling (1996, 2001) highlighted five patterns in tourism
time series: stationarity, seasonality, linear trend, non-linear trend and stepped series.
The identification of these patterns can be an end in itself as seen by several studies which
have attempted to describe the characteristics of tourism data for a number of destina-
tions. On the other hand, prior information on key features on the data set in use, assists
researchers when choosing an appropriate method for developing and estimating time
series and regression models.
As mentioned above, the main purpose of time series modelling is for forecasting. This
is done to equip stakeholders with relevant information that would assist them in their
decision-making processes and in order to devise relevant tourism-related policies for
the industry. However, according to Frechtling (1996, 2001), the time series non-causal
approach is limited by the lack of explanatory variables and it is best used for short- to
medium-term forecasting because it is assumed that the factors related to seasonality,
trend and cycle are slow to change and can be extrapolated in the short term (Cheong
and Turner, 2005).
The aim of this chapter is to review the application of time series analysis and model-
ling in the tourism literature. Note that there exists a vast number of such techniques
which have been employed to analyse tourism data. It is beyond the scope of this chapter
to discuss all. This chapter instead focuses on the most widely applied methods within
the tourism context. It is made up of three distinct sections. In the first section the charac-
teristics of time series data are discussed. The second section concentrates on the classic
time series modelling techniques. The last section discusses directions for new research
and focuses on approaches which are fairly recent such as neural networks models which

47

DWYER 9781781001288 Part I.indd 47 25/07/2012 11:06

 48 Handbook of research methods in tourism

may be expected to become more extensively utilized by tourism researchers in the near
future.

TIMES SERIES CHARACTERISTICS OF TOURISM DATA SETS

Stationarity

A stochastic process is said to be stationary if its mean and variance are constant over time and
the value of its covariance between the two time periods depends only on the distance or gap or
lag between the two time periods and not the actual time at which the covariance is computed.
(Gujarati and Porter, 2009, p. 741)

Generally researchers seek to know the stationarity status of variables for two reasons.
First, they are keen to verify that their regression results are valid and not spurious.
According to Granger and Newbold (1974), the rule of thumb for suspecting spuri-
ous results is an R2 which is higher than the Durbin Watson statistic. Second, they
wish to know whether the effect of external shocks on the variable will be temporary
or permanent. The mathematical formulae for mean, variance and covariance are
defined as follows: mean, E (yt) 5 h; variance, Var (yt) 5 E (yt 2 h) 2 5 s2; covariance,
E [ (yt 2 m) (yt1k 2 m) ] 5 lk (h is the expected value of mean, variance is a measure of
how far a set of variables (yt) is spread out and covariance is a measure of how much two
random variables change together).
A stationary process is one where the mean, variance and covariance of yt 1 k is the
same as the covariance of yt, they are time invariant and the process is said to be mean
reverting. A purely random or white noise process has zero mean, constant variance
and is serially uncorrelated. On the other hand, a series which has a time varying mean,
variance, or both, is nonstationary. An example of a nonstationary process is a random
walk. There are two types of random walks: random walks with drifts and random walks
without drifts. If

yt 5 yt21 1 et (3.1)

where et is a white noise error term (mean is zero and variance is constant) then the series
is said to be a random walk without a drift. As t increases, so does the variance while
the mean remains constant. In this case, the effect of random shocks in the data is said
to be persistent. The random walk series is said to have infinite memory as the effect of
shocks does not die out. Take the case of tourist arrivals at a destination. If a random
shock occurs, for example, a terrorist attack, the outbreak of an epidemic or a natural
disaster, the number of arrivals will not revert back to its original mean if the time series
of arrivals follow a random walk. The shock will have a permanent effect on the number
of arrivals at the destination.
A random walk with a drift is given by the equation below:

yt 5 m 1 yt21 1 et (3.2)

DWYER 9781781001288 Part I.indd 48 25/07/2012 11:06

 Time series analysis 49

m is the drift parameter. The variance and mean are dependent on time. A random walk
series without drift can be made stationary by taking the first difference. This type of
series is also referred to as a difference stationary series or a series containing unit root or
integrated of order 1. They are said to have a stochastic trend that is the slow long-run
evolution of the time series cannot be predicted (Gujarati and Porter, 2009). On the other
hand, when yt is a function of time, the series can be made stationary by subtracting the
mean of yt from yt, in which case the series is referred to as a trend stationary series. The
process of removing the influence of time from yt is called detrending. A random walk
with a drift and a time trend is given below

yt 5 b1 1 b2t 1 yt21 1 et (3.3)

Time series which need to be differenced d number of times before they turn stationary
are said to be integrated of the order (d) or yt ~ I(d). Several tests have been developed in
order to test for stationarity of time series data. The two most popular tests used in the
tourism literature are the augmented DickeyFuller (ADF) (Dickey and Fuller, 1979)
which is the extension of the DickeyFuller (DF) and the PhillipsPerron (PP) (Phillips
and Perron, 1988). These tests use the existence of a unit root as the null hypothesis.
DickeyFuller (DF) tests whether a unit root is present in an autoregressive model. It
is named after the statisticians D.A. Dickey and W.A. Fuller, who developed the test in
1979. DF test involves fitting the regression model:

yt 5 m 1 yt21 1 et (3.4)

where

et ~ N (0, s2)

and yt is the variable of interest, t is the time index, is a coefficient and et is the error
term. The hypotheses to be tested are as follows:

H0: 51
H1: 00 , 1

If H0 is rejected, the process is stationary, otherwise it is non-stationary. An augmented

DickeyFuller (ADF) test removes all the structural effects (autocorrelation) in the time
series and then tests using the same procedure.
In the PP test, the null hypothesis is that a time series is integrated of order 1. The PP
test addresses the issue that the process generating data for yt might have a higher order
of autocorrelation than is admitted in the test equation, making yt21 endogenous and
thus invalidating the DickeyFuller (DF) t-test. Whilst the ADF addresses this issue by
introducing lags of Dyt as regressors in the test equation, the PhillipsPerron test makes
a non-parametric correction to the t-test statistic. The test is robust with respect to
unspecified autocorrelation and heteroskedasticity in the disturbance process of the test
equation. Table 3.1 summarizes the different forms of unit root tests and the type of time
series data for which they are relevant.

DWYER 9781781001288 Part I.indd 49 25/07/2012 11:06

 50 Handbook of research methods in tourism

Table 3.1 Summaries of ADF and PP tests under H0: 5 0 and H1: 5 1 hypotheses

TYPE Equation Condition

r
ADF When the time series doesnt have a
Dyt 5 yt21 1 a aiDyt2i 1 et
i 51
trend and is potentially slow turning
et ~ N (0, s2) around zero
r
ADF When the time series does not have a
Dyt 5 a0 1 yt21 1 a aiDyt2i 1 et
i 51
trend and is potentially slow turning
around a non-zero value
r
ADF When the time series has a trend in it
Dyt 5 gt 1 a0 1 yt21 1 a aiDyt2i 1 et
i 51
(either up or down) and is potentially
slow turning around a trend line you
would draw through the data
PP Dyt 5 yt21 1 et Autoregressive
PP Dyt 5 m 1 yt21 1 et Autoregressive with drift
PP Dyt 5 m 1 gt 1 yt21 1 et Trend stationary

When performing these tests it is crucial to select the correct number of augmenting
lags (r). The number of lags is determined by minimizing the Schwartz Bayesian infor-
mation criterion or minimizing the Akaike information criterion or lags are dropped
until the last lag is statistically insignificant. The DickeyFuller t-statistic does not follow
a standard t-distribution as the sampling distribution of this test statistic is skewed to the
left with a long, left-hand-tail. The test statistic is t 5 ^ / SE (^ ) , where ^ and SE (^ ) are
the estimated coefficient on yt21 and its standard error from the ordinary least square
regression.
Unit root testing can be the sole focus of a study or a prerequisite for regression
analysis among researchers who wish to avoid reporting data that lack statistical robust-
ness. Testing for unit root has become fairly common in tourism demand modelling.
This practice started in the mid-1990s. Gonzles and Moralez (1995) were among the
first to check for unit roots using tourism data. They found the presence of unit root on
annual tourism expenditure in Spain. Note that since they also utilised a monthly series:
January 1979 to December 1993, they performed seasonal unit root tests. These tests are
discussed in the next section. Song and Witt (2000), use PP and ADF to pretest their data
set for unit root before specifying a demand model for outbound tourism from the UK
to Austria, Belgium, France, Germany, Italy, the Netherlands, Greece, Spain, the Irish
Republic, Switzerland, the United States and rest of the world. They found number of
holidays and income to be I(1) (integrated of order 1) while mixed results were obtained
for relative prices. Prior to model specification, Lim and McAleer (2002) conducted the
individual ADF tests for the logarithms of arrivals from Hong Kong and Singapore to
Australia. These two series were found to be trend stationary. While studying tourist
arrivals to South Korea, Song and Witt (2003) used the ADF and Perron (1989) tests
to show that the number of arrivals from Germany, Japan, UK and the United States,
income of the host countries, respective exchange rate and relevant airfares to South
Korea were I(1).

DWYER 9781781001288 Part I.indd 50 25/07/2012 11:06

 Time series analysis 51

4000
Holiday
2000

0
93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11
4000
Business
Total no. of UK inbound visits (in thousand)

2000

0
93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11
500
Study

0
93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11
4000
Visit Friend/Relative
2000

0
93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11
2000
Miscellaneous
1000

0
93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11
Years 19932011 Q2

Source: Compiled from International Passenger Survey UK National Statistics available at http://www.
statistics.gov.uk.

Figure 3.1 UK inbound tourism demand (quarterly data)

Unit Roots with Seasonality

It is a fact that seasonality is one of the most salient characteristics of the tourism
industry. Seasonality has been described as a temporal imbalance in the phenomena
of tourism, and may be expressed in terms of dimensions of such elements as number
of visitors, expenditure of visitors, traffic on highways and other forms of transporta-
tion, employment and admission to attractions (Butler, 1994, p. 332). In other words
since most destinations have peak and off-peak seasons, consumption in the tourism
industry display a predictable and regular pattern which recur every year. It occurs
for two reasons: natural factors, such as climate, and institutional, such as school
holidays (Butler, 2001). The effect of seasonality is seen in monthly or quarterly data.
Figure 3.1 shows international arrivals to the UK (UK inbound) for four categories
of visitors.
The data clearly shows that arrivals for holiday and study purposes are subject to
a high degree of seasonality, with the third quarter representing the peak time. The
pattern is repeated from 1993 to 2010. On the other hand, arrivals for business pur-
poses do not display any such patterns. Seasonality is identifiable in the more frequent

DWYER 9781781001288 Part I.indd 51 25/07/2012 11:06

 52 Handbook of research methods in tourism

104
1.5
Holiday
1

0.5
93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10
10000
Business
5000
Total no. of UK inbound visits (in thousand)

0
93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10
1000
Study
800
600

93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10
10000
Visit Friend/Relative
5000

0
93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10
3000
Miscellaneous
2000

1000
93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10
Years 19932010

Source: Compiled from International Passenger Survey UK National Statistics available at http://www.
statistics.gov.uk.

Figure 3.2 UK inbound tourism demand (yearly data)

intervals, for example, monthly and quarterly data but this is lost in annual data. This
is illustrated in Figure 3.2 where seasonality has been removed by aggregating the
data.
When data display seasonal patterns, the type of unit roots present is different too.
They can be quarterly, semi annual or annual and the relevant differencing strategy is
required to achieve stationarity (Song et al., 2009). The most commonly used test is the
Hylleberg, Engle, Granger and Yoo (1990) also known as HEGY. Below is the standard
HEGY regression applicable to quarterly data.
4
D4yt 5 a piyit21 1 et (3.5)
i51

(where pi (i 5 1, 2, 3, 4) are the coefficients)

Acceptance of the null hypothesis implies that the series contain seasonal unit root
while rejecting H0 implies that the series is stationary. If all of the values of p are not
simultaneously equal to zero, the interpretation of the test results changes. If only p1 5 0
then the series has normal unit root; if p2 5 0 then the series has semi-annual root and if

DWYER 9781781001288 Part I.indd 52 25/07/2012 11:06

 Time series analysis 53

p3 5 p4 5 0, then an annual units root is present (Song et al., 2009). The HEGY test may
be adjusted for a constant and a seasonal deterministic trend as given below
4
D4yt 5 p0 1 a piyit21 1 et
i51
4 4
(3.6)
D4yt 5 1 a piyit21 1 a biQit21 1 et
i51 i51

where Qi represents seasonal dummies. These tests have been widely applied in the
tourism context. Some examples are Alleyne (2006), Kulendran and King (1997),
Kulendran and Wong (2005) and Lim and McAleer (2002). Lim and McAleer (2002)
found tourist arrivals have unit roots at zero and semi-annual frequencies, but no annual
unit roots. They used data of quarterly tourist arrivals from Malaysia to Australia from
1975 to 1996. Kulendran and King (1997) applied HEGY to test for seasonal and non-
seasonal unit roots in the log of tourist arrivals to Australia from Japan, New Zealand,
the UK and the United States. The test was also applied to the log of quarterly income
and the log of relevant price variables. The results of the HEGY tests suggest that quar-
terly tourist arrivals from the United States and New Zealand had unit roots at zero
and biannual frequencies. The series for the UK had one non-seasonal unit root. The
Japanese series had unit roots at the zero frequency and one seasonal frequency and all
income, price and airfare series had one non-seasonal unit root. Kulendran and Wong
(2005) applied the HEGY technique to 19 series including quarterly international to
Australia from 1975(1) to 1998(4) from ten markets and quarterly departures from the
UK to nine destinations. Alleyne (2006) tested for seasonal unit root in tourist arrivals
to Jamaica using six tourist arrival series, from 1968(1) to 2001(3). He concluded that
pretesting for seasonal unit roots improves forecasting of arrivals to Jamaica.

Unit Root with Structural Breaks

Time series can contain structural breaks. Structural breaks occur when there is a change
in one of the parameters (the trend coefficient, mean or variance) of a data-generating
process (Hansen, 2001). The breaks are caused by shocks which are applied to the series.
Shocks can be positive or negative. Examples of positive shocks are the hosting of special
events such as the Olympic Games, and negative shocks that have affected the tourism
industry are the 9/11 terrorist attacks, outbreak of epidemics, war and peaks in the price
of crude oil. The effect of these shocks on the data series can be can be permanent or
temporary. Permanent in this context means that in a given sample of data, the change
resulting from a shock will still be in effect at the end of the data sample (Perron, 2005).
Thus, if the series contain a unit root, a shock will permanently change its growth path
while, if it is found to be trend-stationary shocks will have only temporary effects and
the series will eventually revert back to its original trend. The presence of structural
breaks may bias the results of standard unit root tests. A unit root test which allows for
structural breaks in the data was proposed by Perron (1989). This was an exogenous test
for structural breaks since the date of the break is estimated using a priori information.
More recent tests devised are endogenous, where the break point in the data is deter-
mined by the tests.

DWYER 9781781001288 Part I.indd 53 25/07/2012 11:06

 54 Handbook of research methods in tourism

One commonly used test in the literature is that of Zivot and Andrews (1992). This
test allows for a one-time break in the trend and/or intercept of a time series. The aim
of the ZivotAndrews test is to see if a structural break exists in a time series, by testing
the null hypothesis of a unit root against the alternative of (trend) stationarity with
structural change at some unknown point. According to Byrne and Perman (2006) and
Perron (2005), however, the weakness of the ZivotAndrews test is that under the null
hypothesis, structural breaks are not allowed, which creates a bias towards rejecting the
null hypothesis of a unit root process in favour of the alternative. Another weakness of
the ZivotAndrews test is the inability to deal with more than one structural break in
a series. Other tests for structural break in the data include those of Lee and Strazicich
(2003, 2004) and Lumsdaine and Papell (1997).
Whiltshire (2009) performed a ZivotAndrews test of the annual time series on
Australian international passenger flow from 1956 to 2008. The test identified a break
in 1974 which corresponded to the first oil crisis. The effect of the break on the data is
illustrated in Figure 3.3.
Narayan (2005) performs the ZivotAndrews (1992) and LumsdainePapell (1997),
and finds that the Rabukas military coups only had a temporary effect on tourist
expenditure in Fiji. In their study of the tourism industry of Bali, Smyth et al. (2009)
demonstrate that the effects of the recent terrorist acts on the growth path of tourist
arrivals from major markets are only transitory and that as a consequence Balis tourism
sector is sustainable in the long run.
Njegovan (2006) quantified the effect of exogenous shocks on annual passenger
numbers to Australia, Canada, Germany, the US and the UK using the ZivotAndrews
(1992) test. The evidence for the UK, Germany and Australia indicates that the series
are trend-stationary containing one structural break in both the trend and intercept in
1977. The authors conclude that the structural break is most likely a result of the OPEC
oil crises in the 1970s. His analysis of more recent demand shocks suggests that they had
only temporary effects on annual passenger growth.
Using the ZivotAndrews (1992) and LumsdainePapell (1997) unit root tests, Lee
and Chien (2008) analysed Taiwanese data on the real GDP, tourism development and
real international value of the Taiwanese currency. Their aim was to investigate whether
regime changes have broken down the stability of the long-run relationships between
tourism development and real GDP in Taiwan from 1959 to 2003. They conclude that
the variables are subject to structural breaks which corresponded to actual incidents
related to the politics economy and tourism which occurred in Taiwan.

Cointegration Test and ECM

If two or more variables are non-stationary but integrated of the same order i.e., x ~
I(d) and y ~ I(d) there may exist linear combinations of these variables which are sta-
tionary. Then there is a long-run relationship that prevents them from drifting away
from each other and they are said to be cointegrated. The idea of variables which are
non-stationary but cointegrated is credited to Granger (1981) and further developed in
Granger and Weiss (1983) and Engle and Granger (1987). These authors revolution-
ized the way non-stationary data are perceived and introduced methods for modelling
data which ensures that regression results from non-stationary data are statistically

DWYER 9781781001288 Part I.indd 54 25/07/2012 11:06

 18

17

DWYER 9781781001288 Part I.indd 55

16 Break Point identified
by Zivot and Andrews
15

14

13

55
ln (passenger numbers)
12

11

10

1956
1958
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008

Year

Source: Whiltshire (2009).

Figure 3.3 Australian international passenger numbers with trend line and structural break

25/07/2012 11:06
 56 Handbook of research methods in tourism

sound. One of the most widely applied tests is Engle and Granger (1987). Consider the
following model

yt 5 a1 1 a2xt 1 ut (3.7)

where x ~ I(d) and y ~ I(d).

Estimating Equation (3.7) will give spurious results unless yt and xt are cointegrated.
The first step to detect cointegration is to estimate the equation to obtain ut. The
second step is to test for the stationarity of ut using the ADF statistics. If ut is found to
be stationary then yt and xt are cointegrated and the equation can be expressed in the
form or an error correction model. The vector (a1 a2) is referred to as the cointegra-
tion vector. According to Engle and Granger (1987), a1 and a2 will be consistent and
efficient. a2 is the long-run coefficient. This relationship can be modelled into an error
correction. Error correction is defined by Engle and Granger as a proportion of the
disequilibrium from one period is corrected in the next period (Engle and Granger,
1987, p. 254).
r r
Dyt 5 a biDxt2i 1 a Dyit2j 1 lut21 1 et (3.8)
i50 j51

The ECM is estimated to obtain the short term coefficients. This technique is referred
to as the Engle and Granger (1987) two step method. Since the standard errors of the
parameters of the second regression are the true standard errors, the model may be used
for forecasting purposes (Song et al., 2009). An advantage of ECM is that since regres-
sors in the model are almost orthogonal, the problem of multicollinearity can be avoided
(Song et al., 2009; Syriopoulos, 1995). This method is, however, not without flaws.
According to Song and Witt (2000), the Engle and Granger method does not prove that
the relationships among the cointegrated variables are long term and the robustness of
estimations from small samples is not always achieved. Futhermore, problems arise in
identifying individual structural relationships.
In their modelling of outbound tourism from UK, Song and Witt (2000) applied
ADF and PP tests for the 12 long-run regressions and found that in every case except
France, the null hypothesis of no cointegration is rejected at least at the 10 per cent level
of significance. They used the EngleGranger (1997) technique to specify the long-run
relationship between departures from UK and income and relative prices. Furthermore,
the authors perform ex-post forecasts over a period of 6 years, on four different model
specifications and show that their ECM model outperforms the remaining three in pre-
dicting UK outbound tourism.
The aim of the study Kulendran and Wilson (2000) was to identify the economic
variables that are important in influencing business trips to Australia from four of its
major trading partners. Quarterly data from 1982(1) to 1996(4) were used. The HEGY
method was applied to examine the time series properties of the dependent and inde-
pendent variables: business travel, origin country real income, openness to trade, real
imports, relative price and holiday travel. The Full Information Maximum Likelihood
technique was used to estimate the long-run relationship between business travel and
its determinants. The authors concluded that the importance of the economic variables

DWYER 9781781001288 Part I.indd 56 25/07/2012 11:06

 Time series analysis 57

varied from country to country, although overall openness to trade and origin country
real income were important variables explaining business travel.
Kulendran and Witt (2001) study business travel to Australia from Japan, New
Zealand, the United Kingdom and the United States. They utilize quarterly data for the
period 1982(1)96(4). Real gross domestic product, trade openness and real exchange
rate were amongst the variables included in their model. After having tested for unit
root and finding that the majority of the variables have unit roots at zero frequency,
Johansens (1988) and Johansen and Juselius (1990) tests for cointegration were applied.
The results suggest that there is one significant long-run relationship for Japan, and there
are four for New Zealand and two for both the United Kingdom and the United States.
Lim and McAleer (2002) developed demand models to explain tourism arrivals to
Australia from Hong Kong and Singapore. Incomes in Hong Kong and Singapore,
tourism prices in Australia, and transportation costs and exchange rates between the
two countries and Australia were used to explain the number of arrivals to Australia.
Quarterly data from 1975(1) to 1996(4) were used for Hong Kong and 1980(4)96(4) for
Singapore. After performing an ADF test for unit roots and Johansens maximum likeli-
hood procedure for cointegration to estimate the number of cointegrating vectors, the
authors developed ECM models to explain tourism demand.
Narayan (2004) examined the short- and long-run relationships between visitor arriv-
als in Fiji, real disposable incomes and relative hotel and substitute prices for the period
19702000, using cointegration techniques and error correction models. A generalized
DickeyFuller type regression was used to test the significance of lagged levels of the var-
iables under consideration in a conditional unrestricted ECM. The results of this study
suggest that rises in income in the main markets of Fiji had a positive impact on arrivals
while relative hotel and substitute prices have negative impacts on visitor arrivals.
Algieri (2006) examined the determinants of and their relevant impact on tourism
revenues in Russia over the period 1993(12) to 2002(10). Variations in income, exchange
rates and prices, as well as unpredicted events such as relevant political changes were
identified as the relevant factors affecting demand. A cointegration analysis is performed
and the results suggested a robust and significant long-run cointegration relationship
between Russian tourism receipts, world GDP, real exchange rates and prices of air
transport.
Webber (2001) used the Johansen (1988) and the Engle and Granger (1987) methods
to test for the long-run demand for Australian outbound leisure using quarterly data
from 1983(1) to 1997(4) for nine major tourism destinations. It was found that variances
in the exchange rate are significant determinants of long-run tourism demand. Real dis-
posable income and substitute prices have inelastic long-run effects on tourism, while the
long-run relative price elasticity tends to differ widely across destinations.

TIME SERIES MODELLING

This section expands further on the different types of time series models available to
tourism researchers. There are two main types of modelling in time series analysis. The
first is the classic time series model. Here the nave model may be considered as one of
the most basic. The degree of complexity rises greatly as researchers move to modelling

DWYER 9781781001288 Part I.indd 57 25/07/2012 11:06

 58 Handbook of research methods in tourism

techniques such as the moving average (MA), autoregressive (AR), autoregressive

moving average (ARMA), autoregressive integrated moving average (ARIMA) and
seasonal ARIMA (SARIMA), exponential smoothing (ES) and Winters multiplica-
tive ES, basic structural modelling (BSM) and generalized autoregressive conditional
heteroskedasticity (GARCH). These modelling techniques have been widely applied in
the tourism literature. Often studies develop several models which are estimated and
results compared. The aim is to find the best model that fits the data and outperforms
the rest in terms of accuracy in forecasting. From the review of literature provided by
Coshall and Charlesworth (2011), it is apparent that one class of forecasting models
does not consistently dominate the other in terms of predictive accuracy. Their per-
formance depends on the data set in use, for example, destination origin pairs and
the frequency of data. The effectiveness of the different models is also conditional on
the forecasting horizon and the technique used to assess their accuracy (Coshall and
Charlesworth, 2011). Since several techniques are applied in a given study, this section
will first explain the different classic techniques before providing their application to
the tourism literature. The preferred estimation technique is often the BoxJenkins
Method.
The second type of modelling technique using time series data is based on artifi-
cial intelligence (AI), especially neural network (NN) techniques. Given that they are
fairly new in the tourism context, their application in the tourism literature is not as
widespread as the classic models. The common NN techniques that are applied in the
tourism context are the multi-layer perceptron neural networks (MLPNN) and support
vector regression neural networks (SVRNN).

Classic Time Series Model

Nave models: nave 1 and nave 2

The nave 1 model simply states that the future forecasts are equal to the recent available
value. For example, the nave 1 model operates on the assumption that there is no change
and that the number of visitors at time t, yt, is the same as the value at time t 2 4, denoted
by yt 2 4 and is described as yt 5 yt24. The nave 2 model operates on the assumption that
the number of visitors at time t, is subject to a constant growth, yt is equal to the value at
time t 2 4 times by a modification factor which includes the influence of the yt28 (long-
range value) and the nave 2 formula is yt 5 yt24 { 1 1 (yt24 2 yt28) / yt28 } . The nave
1 and 2 are frequently used as benchmarks for assessing predictive accuracy in tourism
forecasting exercises (Coshall and Charlesworth, 2001).

Moving average and autoregressive moving average (ARMA)

As a means to improve on the nave model, the patterns and trends of a time series can
be extrapolated using a moving average (MA) or smoothing model. A moving average is
taken to estimate the local mean which is used for forecasting. This method is to smooth
out peaks in the original series. An MA model is where yt, at time t, is equal to a constant
(a0) plus a moving average of current year and past year errors is given below.

yt 5 a0 1 at 1 b1at21 (3.9)

DWYER 9781781001288 Part I.indd 58 25/07/2012 11:06

 Time series analysis 59

This series is an MA(1) since it includes one lag of the error term. A more general Ma(q)
series take the form of

yt 5 at 1 b1at21 1 b2at22 1 . . . 1 bpat2p (3.10)

where at is white noise with zero mean and variance. The equation above can be rewritten
in terms of the backward shift operator B

yt 5 (1 1 b1B 1 b2B2 1 . . . 1 bqBq) at 5 Fq (B) at (3.11)

where Bat 5 at21. An autoregressive also known as vector autoregressive series, is a

multivariate series where the dependent variable is related to its past values. An autore-
gressive series of order p, also referred to as an AR(p) or VAR(p) series can be written
in the form of:

yt 5 b1yt21 1 b2yt22 1 . . . 1 bpyt2p 1 at 5 Qp (B) yt (3.12)

where at is white noise and B is a backward shift operator. A mixed series is one which
combines autoregressive and moving average terms and is known as an ARMA (Box and
Jenkins, 1970).

yt 5 b1yt21 1 b2yt22 1 . . . 1 bpyt2p 1 at 5 (1 1 b1B 1 b2B2 1 . . . 1 bqBq) at (3.13)

or

Qp (B) yt 5 Fq (B) at

An ARMA (p, q) contains p autoregressive and q moving average terms. The process is
stationary when the roots of q all exceed unity in absolute value. The process is invertible
when the roots of F all exceed unity in absolute value.

Univariate autoregressive integrated moving average (ARIMA) and seasonal ARIMA

(SARIMA)
Note that the AR(p) model is the special case ARMA(p, 0) and the MA(q) model is
the special case ARMA(0, q). If y ~ I(d) and then ARMA is applied to it. The result-
ing model is an autoregressive integrated moving average (ARIMA) (p,d,q). ARIMA
models can also be used to model seasonality. This type of models are called seasonal
autoregressive integrated moving average model or SARIMA ARIMA(p,d,q)(P,D,Q)s
(Box and Jenkins, 1970), which is presented as follows:

p (B) FP (B s) [ (1 2 B) d (1 2 B s) Dyt 2 m ] 5 qq (B) QQ (B s) at (3.14)

where B is a backward shift operator with Byt 5 yt21 and Bat 5 at21, yt is the
value to be forecasted and at is the residual at time period t, m is overall mean of
series which is a constant. p (B) 5 1 2 f1B 2 f2B2 2 . . . 2 fpBp is a non-seasonal
AR of order p, qq (B) 5 1 2 1B 2 2B2 2 . . . 2 qBq is a non-seasonal MA of

DWYER 9781781001288 Part I.indd 59 25/07/2012 11:06

 60 Handbook of research methods in tourism

order q, FP (Bs) 5 1 2 F1Bs 2 F2Bs2 2 . . . 2 FPBsP is a seasonal AR of order P,

QQ (Bs) 5 1 2 Q1Bs 2 Q2Bs2 2 . . . 2 QQBsQ is a seasonal MA polynomial of order Q. yt
as in the above equation of ARIMA(p,d,q) (P,D,Q)s, is the forecasting value and s 5 4
if it is quarterly data and s 5 12 if it is monthly data. The ARIMA(p,d,q) is as follows,
where s 5 0, P 5 D 5 Q 5 0 in SARIMA.

p (B) [ (1 2 B) dyt 2 m ] 5 qq (B) at (3.15)

According to Song and Li (2008), ARIMA models dominate the literature on tourism
time series. They have been applied to over two-thirds of post-2000 studies of tourism
demand that have employed time series forecasting techniques (Song and Li, 2008).

Exponential smoothing (ES)

Exponential smoothing models have been shown to generate accurate forecasts of
tourism demand; furthermore, when seasonality is a pertinent factor, the existence
of unit roots is not an issue in their application (Lim and McAleer, 2002). These
methods use weighted values of past observations to generate forecasts. The weights
declineexponentially over time, since the most recent data are considered to be more
influential on forecasts than are older observations. The simple (one parameter) expo-
nential smoothing model has limited application to tourism forecasting, since it requires
no trend or seasonality to be present in the time series. The use of an algorithm to
remove noise from a data set in order to highlight important patterns is referred to as
data smoothing. In exponential smoothing more recent data are given more weight by
assigning exponentially decreasing weight to observations as they get older.
It can be shown that most ES models such as simple ES, Holts ES, Browns ES,
damped-trend ES, simple seasonal ES, and Winters additive ES are all special cases of
the ARIMA model. In fact, simple ES is equivalent to an ARIMA (0, 1, 1), Holts ES is
equivalent to an ARIMA (0, 2, 2), Browns ES is equivalent to an ARIMA (0, 2, 2) with
restriction among MA parameters, damped-trend ES is equivalent to an ARIMA (1, 1,
2), simple seasonal ES is equivalent to an ARIMA (0, 1, (1, s, s 1 1))(0, 1, 0) with restric-
tions among MA parameters, and Winters additive ES is equivalent to an ARIMA (0,
1, s 1 1)(0, 1, 0) with restrictions among MA parameters.

Winters multiplicative ES
The Winters multiplicative exponential smoothing (WMES) (Douglas et al., 1990) con-
siders the effects of seasonality to be multiplicative, which is, growing (or decreasing)
over time.

yt 5 Tt 3 St 3 Lt 1 et (3.16)

where yt is the forecasting at time t, Tt represents the trend component, St represents the
seasonality, Lt is the long term cycles and et is the error. The WMES method constructs
three statistically related series used to make the actual forecast: the smoothed data
series, the seasonal index and the trend series. This method requires at least 2 years of
back data to calculate forecasting values. It is calculated by solving the following three
updating formulas (Equations (3.17)(3.19)).

DWYER 9781781001288 Part I.indd 60 25/07/2012 11:06

 Time series analysis 61
yt
Lt 5 a 1 (1 2 a) { Lt21 1 Tt21 } (3.17)
St2s
Tt 5 g { Lt 2 Lt21 } 1 (1 2 g) Tt21 (3.18)

St 5 d e f 1 (1 2 d) St2s
yt
(3.19)
Lt
yt1k 5 { Lt 1 kTt } St1k 2s (3.20)

where s is a number of periods per year, k is the period ahead forecasting. a, g and d
represent three smoothing constants with values between 0 and 1. Equation (3.20) can be
used in the forecasting.

BoxJenkins method
The Box and Jenkins (1970) method for forecasting autoregressive integrated moving
average (ARIMA) has become widely used in many fields for time series analysis includ-
ing tourism demand forecasting. This is a three-step iterative procedure used in fore-
casting. In order to specify the model, the data is tested for stationarity and presence
of seasonality is identified. Data is first differenced to achieve stationarity, if necessary
and appropriate values of p and q are fitted to the model. In the second step the model is
estimated. In the third step, diagnostic tests are performed to verify how well the model
fit the data and whether the residuals are white noise. If the results are not satisfactory,
the process is repeated from step 1.
Lim and McAleer (2002) applied various BoxJenkins autoregressive integrated
moving average (ARIMA) models which they estimate using quarterly data from 1975(1)
to 1989(4) on tourist arrivals to tourist arrivals to Australia from Hong Kong, Malaysia
and Singapore. The best-fit model was used to perform post-sample forecasts for the
period 1990(1) to 1996(4). They concluded that while the ARIMA model for Hong Kong
was effective in forecasting arrivals, it was not as accurate as the model for Singapore.
In an analysis of tourist arrivals to Durban, South Africa, from the United States,
Burger et al. (2001) considered a variety of techniques. These were nave, MA, decompo-
sition, single ES, ARIMA, multiple regression and two non traditional methods, genetic
regression and neural networks. The authors used data from 1992 to 1998 to compare
actual and the predicted number of visitors, given by each model. They found that the
neural network method performed the best.
Papatheodorou and Song (2005) studied tourism flows for the period 1960 to 2000
for the six major World Tourism Organization regions and the world. The results of the
ADF test suggested that all the series were I(1). ARIMA models were estimated using the
BoxJenkins technique to forecast international tourism arrivals, international tourism
receipt and real per capita international tourism receipt. They concluded that perform-
ance differed dramatically among the regions, with sharp fluctuations. Some regions
were even expected to experience negative tourism growth in real and per capita terms.
The aim of the study performed by Wong et al. (2007) was to assess the relative effec-
tiveness of techniques as compared to combined techniques in forecasting exercises. The
study is based on data on demand for Hong Kong tourism from its top ten markets:
mainland China, Taiwan, Japan, the United States, Macau, South Korea, Singapore,
the UK, Australia and the Philippines. Quarterly data from 1984(1) to 2004(2) were

DWYER 9781781001288 Part I.indd 61 25/07/2012 11:06

 62 Handbook of research methods in tourism

utilized. The data period 1984(1) to 1999(2) was used to estimate the individual fore-
casting models and the subsequent period for forecasting evaluation.The forecasts are
derived from four different forecasting models: the ARIMA model, the autoregressive
distributed lag model, the ECM and the vector autoregressive model.
The HEGY test was used to test for seasonal unit roots and it shows that all the
variables have non-seasonal unit roots. Tourist arrivals from Macau, the United States,
Singapore and the Philippines have annual seasonal unit roots while the series of tourist
arrivals from the UK has both annual and semi-annual seasonal unit roots. For the GDP
variables, one semi-annual seasonal unit root is found in the Taiwan series. In addi-
tion, two seasonal unit roots were identified in the China, Macau, Korea, the UK and
the Philippines series. The Engle and Granger two-stage approach was used to specify
the ECM and the seasonal ARIMA based on the standard BoxJenkins method. The
authors concluded that although the combined forecasts do not always outperform the
best single model forecasts, there is sufficient evidence to suggest that forecast combina-
tion can considerably reduce the risk of forecasting failure. More recent studies have
reached similar conclusions.
Cang and Hemmington (2010) studied inbound tourism expenditure in the UK disag-
gregated by purpose of visit: holiday, business, study, visit friends or relatives (VFR) and
miscellaneous. Data were modelled using ARIMA and WMES. The nave 2 forecast-
ing model was used as a benchmark to compare with the ARIMA and WMES models.
This study demonstrated that the ARIMA model outperforms the WMES model, but
it is not statistically superior to the WMES model. The ARIMA and WMES models
are both statistically superior to the nave 2 model for this UK inbound expenditure
data set. The ARIMA model forecasted a higher increasing trend for expenditure for
the business travel, whereas the WMES model forecasted a higher increasing trend for
expenditure for the miscellaneous group. The authors recommended combining values
from the ARIMA and the WMES models to use as forecasting values for business and
miscellaneous purposes.
Coshall and Charlesworth (2011) used four models to forecast UK outbound tourism
to Europe. They were univariate volatility models, exponential smoothing models,
a regression-based approach and the nave 2 model. The data used were seasonally
unadjusted quarterly UK outbound tourism numbers by air to the 18 most popular
destinations in Europe in 2006. Each of the models was considered individually and
in combination in terms of forecasting demand for international tourism. The authors
concluded that the models generated accurate predictions of tourism flows, but they were
most effective when combined with other models.
Brida and Risso (2011) explored the SARIMA technique to forecast tourism
demand by using monthly time series of the overnight stays in South Tyrol (Italy)
from January 1950 to December 2005. The authors tested for seasonal unit roots using
HEGY to find that the variables of interest are seasonally stationary. A DickeyFuller
(ADF) test was used to examine the null hypothesis of a stochastic trend (or a unit
root). The presence of a unit root in the logarithm of overnight stays in South Tyrol
was detected, making it necessary to first difference the series. The seasonal ARIMA
method was developed from the standard model of Box and Jenkins (1970). The
forecasting performance was assessed using a SARIMA (2,1,2)(0,1,1) and data for
January 2006 to December 2008.

DWYER 9781781001288 Part I.indd 62 25/07/2012 11:06

 Time series analysis 63

Generalized autoregressive conditional heteroskedasticity (GARCH)

Auto-regressive conditional heteroskedasticity (ARCH) (Engle, 1982) are commonly
used in modelling financial time series that exhibit time-varying volatility clustering, i.e.,
periods of swings followed by periods of relative calm. Time series also often exhibit
volatility clustering or persistence. In volatility clustering, large changes tend to follow
large changes, and small changes tend to follow small changes. The changes from one
period to the next are typically of unpredictable sign. Large disturbances, positive or
negative, become part of the information set used to construct the variance forecast of
the next periods disturbance.
In the ARCH(q) model, the error term split into a stochastic term zt and a time-
dependent standard deviation st characterizing the typical size of the term, the error
term is presented as follows:

et 5 stzt (3.21)

where zt is a random variable drawn from a Gaussian distribution centred at 0 with

standard deviation equal to 1. zt ~ N (0, 1) and s2t 5 a0 1 Sqi51aie2t2i, where a0 . 0 and
ai . 0 and Sqi51ai , 1. An ARCH(q) model can be estimated using Ordinary Least
Squares. A methodology to test for the lag length of ARCH errors using the Lagrange
multiplier test was proposed by Engle (1982). In that case, the GARCH(p, q) model,
where p is the order of the GARCH terms s2 and q is the order of the ARCH terms e2,
is given by
q p
s2t 5 a0 1 a aie2t2i 1 a bis2t2i (3.22)
i51 i51

where a0 . 0 and ai . 0 and Sqi51ai , 1 and bi . 0 and Sqi51bi , 1

The GARCH modelling technique and its variations have applied in the study of
tourism demand. For example, Shareef and McAleer (2005) modelled the volatility in
monthly international tourist arrivals and the growth rate of monthly tourist arrivals
for six small island tourist economies using GARCH(1,1). Chan et al. (2005) exam-
ined the volatility of the logarithm of the monthly tourist arrivals from Japan, New
Zealand, the UK and the United States to Australia between 1975 and 2000 using this
technique.

LIMITATIONS AND FUTURE DIRECTIONS

The most important limitations of time series analysis are that it supposes that data are
either stationary or can be made stationary by simple transformations and that data
can be accurately represented in a linear form (Rodrigues et al., 2010). However, in a
complex world data often have structures which are non linear and chaotic (Rodrigues
et al., 2010). This makes it difficult for typical time series models to fit the data with
an adequate level of accuracy and, thereby, reduces the effectiveness of the model in
forecasting. AI techniques such as artificial neural networks (ANN) can overcome this
problem since a feature of ANN is nonlinear mapping from inputs to outputs. ANN can

DWYER 9781781001288 Part I.indd 63 25/07/2012 11:06

 64 Handbook of research methods in tourism

extract meaningful patterns from data to represents relevant information and has been
proposed as a suitable alternative to classic time series methods. ANN is a computing
technique that processes information in a similar manner to the human brain (Law,
2000).
According to Rodrigues et al. (2010, p. 619), the ability to learn through examples
and to generalize the learned information is one of the most attractive characteristics of
the ANN. Due to its capacity for modelling complex non-linear relationships among
data, the ANN can be used successfully in solving time series forecasting problems.
Furthermore, Song and Li (2008, p. 413) stipulate that [t]he main advantage of AI tech-
niques is that it does not require any preliminary or additional information about data
such as distribution and probability and that the unique features of ANNs, such as the
ability to adapt to imperfect data, nonlinearity, and arbiter function mapping, make
this method a useful alternative to the classical (statistic) regression forecasting models.
AI has grown rapidly as a field of research across a variety of disciplines in recent years
(Song and Li, 2008). ANNs have three fundamental features parallel processing,
distributed memory and adaptability that provide them with a series of advantages
compared to other processing systems, such as robustness and a tolerance to error and
noise (Palmer et al. 2006, p. 782). According to Song and Li (2008) empirical evidence
shows that ANNs generally outperform the classical time-series and multiple regression
models in tourism forecasting.

Multi-layer Perceptron Neural Networks (MLPNN)

The MLPNN is a feed forward neutral network with a single-layer and proper number
of hidden units can be used to closely approximate the true relationship among the
variables. MLPNN is especially useful when this true relationship is unknown or non-
linear. MLPNN is based on a linear combination of the input variables which is trans-
formed by a nonlinear activation function, thus a linear combination of the outputs
of these activation functions is constructed for the final outputs yi. In MLPNN, the
training data is considered as a set of pairs (X(1), y(1)) , . . ., (X(i), y(i)) ,. . ., (X(N), y(N)) ,
where X(i) ( Rm denotes the input space and has a corresponding actual target value
y(i) ( R for i 5 1, 2, . . ., N, where N corresponds to the size of the data set. Vector
X(i) has m dimensions and normally uses the historical tourism demand observations
as inputs which is X(i) 5 { x1(i), x2(i), x3(i), . . ., xm(i) } 5 { yt21, yt22 , . . ., yt2m } m ( Rm, where
t 5 m 1 1, m 1 2, ..., N and i 5 t 2 m, the actual target value y(i) 5 yt ( R at time t.
For example, X(1) 5 { y4 y3 y2 y1 } , X(2) 5 { y5 y4 y3 y2 } and corresponding target values
y(1) 5 y5, y(2) 5 y6 if setting m 5 4.
For inputs dimension M vector with H hidden units of neural networks, the output yi
of the neural networks can be written as

yt 5 a wk(2)ga a wkj(1)xj(i) b
H M
(3.23)
k 51 j51

where xj(i) is the input value at time t, g is a nonlinear activation function. The common
logistic sigmoid activation function g (x) 5 1 / (1 1 exp (2x)) is used in general. The
parameters H and the weights wkj(1), wk(2) need to be determined by minimizing the error of

DWYER 9781781001288 Part I.indd 64 25/07/2012 11:06

 Time series analysis 65

output of combination model yCt and the actual true target outputyas following formula
using the training data set. Sometimes, the validation data set (apart from the training
data set) is used to determine the parameters weights and number of hidden units H in
the models in order to overcome the over training problem (Christopher, 1995).

E (W) 5 12 a { yCi 2 yi } 2 5 12 a e a wk(2)ga a wkj(1)xj(i) b 2 yi f

N N H M 2
(3.24)
i51 i51 k 51 j51

Minimizing the error function E (W) by initial W (which might for instance be chosen at
random) and then update the weight vector by moving a small distance h (learning rate
parameter) in W-space in the direction in which E (W) decreases most rapidly (the direc-
tion of 2WE). By iterating this process using the equation below, eventually the weight
W vector will converge to a point at which E is minimized. This point of the W value is
the final value that is used in yt 5 SHk51wk(2)g (SM (1) (i)
j51wkj xj ) .

0E
Wnew 5 Wold 2 h (3.25)
0W
Support Vector Regression Neural Networks (SVRNN)

The foundations of support vector regression neural networks (SVRNN) were first devel-
oped by Vapnik (Vapnik, 1995; Vapnik et al., 1996). Contrary to other methods that
seek to minimize the forecasting error term, the basis of the SVMs is to allow an insensi-
tive error range to avoid over- or the under-fitting (Pai and Hong 2005). In SVRNN, the
training data which is used to construct a forecasting model is a subset of the whole avail-
able data and is considered as a set of pairs (X(1), y(1)) , . . ., (X(i), y(i)) , . . ., (X(N), y(N)) ,
where X(i) ( Rm denotes the input space (m is the width or dimension of the inputs)
and has a corresponding actual target value y(i) ( R for i 5 1, 2, . . ., N, where N cor-
responds to the size of the data set. Vector X(i) with m dimension and actual target
value y(i) are inputs and outputs, respectively. For this study, the historical tourism
demand observations at time t, X(i) 5 { yt21 yt22 , . . ., yt2m } m ( Rm and t 5 m 1 1, m
1 2, ..., N and i 5 t 2 m. Here we set the actual target value y(i) 5 yt ( R at time t.
For example, X(1) 5 { y4 y3 y2 y1 } , X(2) 5 { y5 y4 y3 y2 } and corresponding target values
y(1) 5 y5, y(2) 5 y6 if m 5 4.
The idea of the regression problem is to determine a function that can predict future
values accurately. The generic support vector regression (SVR) estimating function, the
forecasting model, has the following general form:

y(i) 5 f (X) 5 (W # F (X)) 1 b (3.26)

where X has the form X(i), W ( Rm, b ( R are the best weights and base to be determined
using the training data set (subset of the whole data set which is used to build the fore-
casting model), and F denotes a nonlinear transformation from Rm to a high dimen-
sional space. The weight vector W can be written in terms of data points as:
N
W 5 a (ai 2 a*i ) F (X (i)) (3.27)
i51

DWYER 9781781001288 Part I.indd 65 25/07/2012 11:06

 66 Handbook of research methods in tourism

where ai, a*i are the quadratic problem Lagrange multipliers and the e-insensitive loss
function is the most widely used cost function (Mller et al. 1997). The e-insensitive loss
function function is in the form:

0 f (X) 2 y 0 2 e , f or 0 f (x) 2 y 0 $ e
G (f (X) 2 y) 5 e (3.28)
0 otherwise

By solving the quadratic optimization problem in Equation (3.29), the e-insensitive loss
function can be minimized:

1 N N

a ( a*i 2 a i) ( a*j 2 a j) F (X (i) #

) F ( X (j)
) 2 a a*i (y 2 e) 2 ai (y 1 e)
(i) (i)
(3.29)
2 i,j51 i51

subject to
N

a (ai 2 a*i ) 5 0, ai, a*i [ [ 0, C ]

i51

By substituting Equation (3.27) into Equation (3.26), the generic forecasting SVRNN
model can be rewritten as follows:
N N
y(i) 5 f (X) 5 a (ai 2 a*i ) (F (X(i)) # F (X)) 1 b 5 a (ai 2 a*i ) K (X(i), X) 1 b (3.30)
i51 i51

In this equation, the dot product can be replaced with function K (X(i),X) , known as the
kernel. The Gaussian and linear functions are commonly used as the kernel function in
SVR and are K (X(i), X) 5 exp { 2g 0 X 2 X(i) 0 2 } (g is constant) and K (X(i), X) 5 X(i)TX ,
respectively. Kernel functions enable dot product to be performed in high-dimensional
feature space using low-dimensional space data inputs without knowing the transforma-
tion F. For example, if the data set is quarterly tourism data, thus the previous one year
(inputs width m 5 4), a year and a quarter (m 5 5) up to the previous two years (m 5 8),
etc. can be used as inputs, respectively. For simplicity, the Gaussian with s 5 1 is often
used.
Although the first applications of the ANN technique can be traced back to the mid
to late 1990s, see for example, Pattie and Snyder (1996) and Law and Au (1999), it is
not until the 2000s that this technique started gaining momentum within the tourism
context. Pattie and Snyder (1996) used a back-propagation neural network model with
two hidden layers to forecast monthly overnight stays in US national park systems.
Law and Au (1999) presented a feed-forward neural network with six input and one
output nodes to forecast arrivals in Hong Kong. Cho (2003) found that the Elmans
neural network model performs well in forecasting arrivals in Hong Kong from six
countries.
Law (2000) studied tourism arrivals from 1966 to 1996 from Taiwan to Hong Kong
using several forecasting methods including back propagation ANN and a feed forward
ANN. Data from 1966 to 1991 were used for training the ANN process with data for
subsequent years used for forecasting purposes. The other methods used were multiple
regression, nave, MA and Holts ES. His results indicate that the back propagation

DWYER 9781781001288 Part I.indd 66 25/07/2012 11:06

 Time series analysis 67

ANN causal model outperforms regression models, time series models and feed-forward
networks with an exceptionally high level of accuracy.
Cheong and Turner (2005) analysed tourism arrivals to Singapore from Australia,
China, India, Japan, the UK and the United States using quarterly data. Data on total
arrivals for holiday and business were utilized. The aim of this study was to compare
the forecasting accuracy of the basic structural method (BSM) and the neural network
method. The authors also modelled their data using the nave and Holt Winters methods
for base comparisons. The results confirmed that the BSM remains a highly accurate
method and that correctly structured neural models can outperform BSM and the
simpler methods in the short term, and can also use short data series. They concluded
that their findings make neural methods a significant area for future research.
Palmer et al. (2006) apply the MLPNN technique to forecast tourism expenditure in
the Balearic Islands using quarterly data from 1986 to 2000. They show that data do
not need to be detrended or deseasonalized when MLPNN is applied and that the latter
technique outperforms classic time series methods in forecasting accuracy. Other studies
which have shown the pre-eminence of the ANN method over other classic statistical
techniques are Burger et al. (2001), Cang (2011), Law (1998), Law and Au (1999) and
Pattie and Snyder (1996).
The SVRNN method was applied by Cang (2011), in a study of UK inbound tourism
from the world for five purposes: holiday, business, study, VFR and miscellaneous.
Quarterly data from 1993(1) to 2007(4) were used. The paper investigates nine individual
models, including five support vector regression neural networks with different dimen-
sion of inputs from 4 to 8, nave 1, nave 2, SARIMA, and WMES. The author states
that since the five SVRNN models with different inputs have different time series struc-
tures, they should be treated as five different time series. Combinations of forecasting
techniques were also considered. This paper proposed a non-linear combination method
using MLPNN. The empirical results show that the proposed nonlinear MLPNN com-
bination model is robust, powerful and can provide better performance at predicting
arrivals than linear combination models.
The ANN method is not without criticism, ANN is a complex model compared to
classic time series and the good models dependent on the choice of the parameters.
This means that the determination of the parameters in the model is crucial. However,
one of advantages of ANN is that it can cope with any complex nonlinear inputs and
outputs mapping. Thus, the decision regarding the choice of whether to use classic time
series or ANN type models depend on the pattern of the data and the level of accuracy
required. Generally speaking, it is better to use simple time series models rather than
complex time series model if there is an almost regular pattern and the level of accuracy
is not a major issue. Otherwise, ANN is a very good alternative.

CONCLUSION

The aim of this chapter is to provide a review of time series techniques as applied in the
tourism context. This is achieved by explaining each of the most commonly used methods
and illustrating their application by providing relevant reviews of the literature. It is seen
that these techniques have mostly been applied to study tourism demand for different

DWYER 9781781001288 Part I.indd 67 25/07/2012 11:06

 68 Handbook of research methods in tourism

origins and destination pairs. The chapter highlights the fact that, broadly speaking, the
studies which have used these techniques can be categorized into two groups. The first
seeks to solely identify the characteristics of time series data and verify whether the series
contained unit roots, structural breaks or seasonality. The second, after having defined
the features of the data, uses these as a means for selecting models and estimation tech-
niques. The ultimate goal of the latter group is to predict future trends in the data set.
As pointed out by Song and Li (2008) and Coshall and Charlesworth (2011), newer
models are better at prediction but there is not one single class of models which domi-
nates in terms of accuracy in forecasting. The performance of the models depends on the
data set in use (Coshall and Charlesworth, 2011). ANN models have been suggested as a
suitable alternative to the class time series models. In fact, authors such as Burger et al.
(2001), Palmer at al. (2006) and Cang (2011) have shown that the neural network models
are better at prediction. There is, however, some consensus in the literature that a higher
level of accuracy is achieved by combining models for example in Cang (2011), Coshall
and Charlesworth (2011) and Wong et al. (2007). It is expected that future researchers
may choose to use combined techniques more frequently.

REFERENCES

Algieri, B. (2006), An econometric estimation of the demand for tourism: the case of Russia, Tourism
Economics, 12 (1), 520.
Alleyne, D. (2006), Can seasonal unit root testing improve the forecasting accuracy of tourist arrivals?,
Tourism Economics, 12 (1), 4564.
Box, G. and G. Jenkins (1970), Time Series Analysis: Forecasting and Control, San Francisco, CA: Holden-
Day.
Brida, J.G. and W.A. Risso (2011), Tourism demand forecasting with SARIMA models: the Case of South
Tyrol, Tourism Economics, 17 (1), 209221.
Burger, C.J.S.C., M. Dohnal, M. Kathrada and R. Law (2001), A practitioners guide to time-series methods
for tourism demand forecasting: a case study of Durban, South Africa, Tourism Management, 22, 403409.
Butler, R.W. (1994), Seasonality in tourism: issues and problems, in A.V. Seaton (ed.) Tourism: State of the
Art, Chichester, UK: Wiley, pp. 332339.
Butler, R.W. (2001), Seasonality in tourism: issues and implications, in T. Baum and S. Lundtorp (eds),
Seasonality in Tourism, Advances in Tourism Research Series, Amsterdam: Elsevier, pp. 521.
Byrne, J.P. and R. Perman (2006), Unit roots and structural breaks: a survey of the literature, Working
Papers 2006_10, University of Glasgow, UK.
Cang, S. (2011), A non-linear tourism demand forecast combination model, Tourism Economics, 17 (1), 520.
Cang, S. and N. Hemmington (2010), Forecasting UK inbound expenditure by different purposes of visit,
Journal of Hospitality and Tourism Research, 34 (3), 294309.
Chan F., C. Lim and M. McAleer (2005), Modelling multivariate international tourism demand and volatil-
ity, Tourism Management, 26, 459471.
Cheong, K.S. and L.W. Turner (2005), Neural network forecasting of tourism demand, Tourism Economics,
11 (3), 301328.
Cho, V. (2003), A comparison of three different approaches to tourist arrival forecasting, Tourism
Management, 24, 323330.
Christopher, M.B. (1995), Neural Networks for Pattern Recognition, Oxford: Oxford University Press.
Coshall, J.T. and R. Charlesworth (2011), A management orientated approach to combination forecasting of
tourism demand, Tourism Management, 32, 759769.
Dickey, D.A. and W.A. Fuller (1979), Distribution of the estimators for autoregressive time series with a unit
root, Journal of the American Statistical Association, 74, 427431.
Douglas C.M., A.J. Lynwood and S.G. John (1990), Forecasting and Time Series Analysis, 2nd edition,
Columbus, OH: McGraw-Hill, Inc.
Engle, R.F. (1982), Autoregressive conditional heteroscedasticity with estimates of the variance of United
Kingdom inflation, Econometrica, 50, 9871007.

DWYER 9781781001288 Part I.indd 68 25/07/2012 11:06

 Time series analysis 69

Engle, R.F. and C.W.J. Granger (1987), Co-integration and error-correction: representation, estimation, and
testing, Econometrica, 55, 251276.
Frechtling, D.C. (1996), Practical Tourism Forecasting, Oxford: Butterworth-Heinemann.
Frechtling, D.C. (2001), Forecasting Tourism Demand, Oxford: Butterworth-Heinemann.
Gonzlez Pilar and Paz Moralez (1995), An analysis of the international tourism demand in Spain,
International Journal of Forecasting, 11, 233251.
Granger, C.W.J. (1981), Some properties of time series data and their use in econometric model specification,
Journal of Econometrics, 16, 121130.
Granger, C.W.J. and P. Newbold (1974), Spurious regressions in econometrics, 2 (2), 111120.
Granger, C.W.J. and A.A. Weiss (1983), Time Series Analysis of Error-correction Models, New York:
Academic Press.
Gujarati, D.N. and D.C. Porter (2009), Basic Econometrics, 5th edition, New York: McGraw Hill.
Hansen, B.E. (2001), The new econometrics of structural change: dating breaks in U.S. labour productivity,
Journal of Economic Perspectives, 4, 117128.
Harvey, A.C. (1993), Time Series Models, 2nd edition, London: Harvester Wheatsheaf.
Hylleberg, S., R.F. Engle, C.W.J. Granger and B.S. Yoo (1990), Seasonal integration and cointegration,
Journal of Econometrics, 44, 215238.
Johansen, S. (1988), Statistical analysis of cointegration vectors, Journal of Economic Dynamics and Control,
12, 231254.
Johansen, S. and K. Juselius (1990), Maximum likelihood estimation and inference on cointegration with
applications to the demand of money, Oxford Bulletin of Economics and Statistics, 52, 169210.
Kulendran, N. and M.L. King (1997), Forecasting international quarterly tourist flows using error correction
and time series models, International Journal of Forecasting, 13, 319327.
Kulendran, N. and K. Wilson (2000), Modelling business tourism, Tourism Economics, 6 (1), 4759.
Kulendran, N. and S.F. Witt (2001), Cointegration versus least squares regression, Annals of Tourism
Research, 28, 291311.
Kulendran, N. and S.F. Witt (2003), Forecasting the demand for international business tourism, Journal of
Travel Research, 41, 265271.
Kulendran, N. and K.F. Wong (2005), Modeling seasonality in tourism forecasting, Journal of Travel
Research, 44, 163.
Law, R. (1998), Room occupancy rate forecasting: a neural network approach, International Journal of
Contemporary Hospitality Management, 10 (6), 234239.
Law, R. (2000), Back-propagation learning in improving the accuracy of neural network-based tourism
demand forecasting, Tourism Management, 21, 331340.
Law, R. and N. Au (1999), A neural network model to forecast Japanese demand for travel to Hong Kong,
Tourism Management, 20 (1), 8997.
Lee, C.C. and M.-S. Chien (2008), Structural breaks, tourism development, and economic growth: evidence
from Taiwan, Mathematics and Computers in Simulation, 77 (4), 358368.
Lee, J. and M.C. Strazicich (2003), Minimum LM unit root test with two structural breaks, Review of
Economics and Statistics, 63, 10821089.
Lee, J. and M.C. Strazicich (2004), Minimum LM unit root test with one structural break, Working Paper,
Department of Economics, Appalachian State University.
Lim, C. and M. McAleer (2002), Time series forecasts of international travel demand for Australia, Tourism
Management, 23, 389396.
Lumsdaine, R.L. and D.H. Papell (1997), Multiple trend breaks and the unit root hypothesis, Review of
Economics and Statistics, 79 (2), 212218.
Mller, K.R., A. Smola, G. Rtsch, B. Schlkopf, J. Kohlmorgen and V. Vapnik (1997), Predicting time series
with support vector machines, Artificial Neural NetworksICANN97, 9991004.
Narayan, P.K. (2004), Fijis tourism demand: the ARDL approach to cointegration, Tourism Economics, 10
(2), 193206.
Narayan, P.K. (2005), The structure of tourist expenditure in Fiji: evidence from unit root structural break
tests, Applied Economics, 37 (10), 11571161.
Njegovan, N. (2006), Are shocks to air passenger traffic transitory or permanent?, Journal of Transport
Economics and Policy, 40 (2), 315328.
Pai, P.F. and W.C. Hong (2005), An improved neural networks model in forecasting arrivals, Annals of
Tourism Research, 32 (4), 11381141.
Palmer, A., J.J. Montao and A. Ses (2006), Designing an artificial neural network for forecasting tourism
time series, Tourism Management, 27, 781790.
Papatheodorou, A. and H. Song (2005), International tourism forecasts: time-series analysis of world and
regional data, Tourism Economics, 11 (1), 1123.

DWYER 9781781001288 Part I.indd 69 25/07/2012 11:06

 70 Handbook of research methods in tourism

Pattie, D.C. and J. Snyder (1996), Using a neural network to forecast visitor behavior, Annals of Tourism
Research, 23 (1), 151164.
Perron, P. (1989), The Great Crash, the oil price shock and the unit root hypothesis, Econometrica, 57,
13611401.
Perron, P. (2005), Dealing with structural breaks, Palgrave Handbook of Econometrics, Vol. 1: Econometric
Theory. New York: Palgrave Macmillan.
Phillips, P.C.B. and P. Perron (1988), Testing for a unit root in time series regression, Biometrika, 75, 335346.
Rodrigues, L.J., P.S.G. de Mattos Neto, J. Albuquerque, S. Bocanegra and T.A.E. Ferreira (eds) (2010),
Forecasting chaotic and non-linear time series with artificial intelligence and statistical measures,
in Modelling Simulation and Optimization, available at http://www.intechopen.com/articles/show/title/
forecasting-chaotic-and-non-linear-time-series-with-artificial-intelligence-and-statistical-measures.
Shareef, R. and M. McAleer (2005), Modelling international tourism demand and volatility in small island
tourism economies, International Journal of Tourism Research, 7, 313333.
Smyth, R., I. Nielsen and V. Mishra (2009), Ive been to Bali too (and I will be going back): are terrorist
shocks to Balis tourist arrivals permanent or transitory?, Applied Economics, 41 (11), 13671378.
Song, H. and G. Li (2008), Tourism demanding modelling and forecasting: a review of recent research,
Tourism Management, 29, 203220.
Song, H. and S.F. Witt (2000), Tourism Demand Modelling and Forecasting: Modern Econometric Approaches,
Advances in Tourism Research Series, Oxford: Pergamon.
Song, H. and S.F. Witt (2003), Tourism forecasting: the general-to-specific approach, Journal of Travel
Research, 42, 6574.
Song, H., S.F. Witt and G. Li (2009), The Advanced Econometrics of Tourism Demand, London: Routledge.
Syriopoulos, T. (1995), A dynamic model of demand for Mediterranean tourism, International Review of
Applied Economics, 9, 318336.
Vapnik, V. (1995), The Nature of Statistical Learning Theory, New York: Springer.
Vapnik, V., S. Golowich and A. Smola (1996), Support vector machine for function approximation regression
estimation, and signal processing, Advances in Neural Information Processing Systems, 9, 281287.
Webber, A. (2001), Exchange rate volatility and cointegration in tourism demand, Journal of Travel Research,
39, 398405.
Whiltshire, T. (2009), A time-series analysis of Australian international passenger numbers, thesis submitted
to Monash University Australia, unpublished.
Wong, K.K.F., H. Song, S.F. Witt and C.D. Wu (2007), Tourism forecasting: to combine or not to combine?,
Tourism Management, 28, 10681078.
Zivot, E. and D.W.K. Andrews (1992), Further evidence on the great crash, the oil-price shock, and the unit-
root hypothesis, Journal of Business and Economic Statistics, 10 (3), 251270.

DWYER 9781781001288 Part I.indd 70 25/07/2012 11:06

 4 Demand modeling and forecasting
Grace Bo Peng, Haiyan Song and Stephen F. Witt

NATURE OF TOURISM DEMAND MODELING AND

FORECASTING

Tourism demand forecasting methods can be divided into two categories: qualitative
and quantitative methods. Quantitative forecasting methods organize past tourism
demand information by mathematical rules and there are three main subcategories: time
series models, econometric approaches, and artificial intelligence methods. According
to the complexities of the models and estimation techniques, the time series forecasting
methods can be subdivided into basic and advanced time series models. Based on their
temporal structure, econometric models can be grouped into two categories: static and
dynamic models.
Tourism demand is normally measured by either tourist arrivals in a destination,
tourists expenditure when they visit a destination, or tourist nights stayed in the desti-
nation. The variables that have been generally accepted as the main determinants of in-
ternational tourism demand comprise potential tourists income levels, the relative price
of tourism products in the origin and destination countries, substitute prices of tourism
products in alternative foreign destinations, transportation cost, population of origin
country, exchange rates, marketing expenditure by the destination in the origin country,
and one-off events (which can have a positive or negative effect). For a detailed review
of tourism demand measures and their determinants, see, for example, Martin and Witt
(1987, 1988); Song and Li (2008); Song et al. (2009); Witt and Martin (1987b); and Witt
and Witt (1992, 1995).

BASIC TIME SERIES METHODS

Time series forecasting methods (see Tables 4.1 and 4.2) are also called extrapolative
methods, and only use past data on the series to be forecast in extrapolating the future.
These models attempt to identify the patterns in the time series that cause the shifts in the
forecast variable. The advantage of time series models is that they are relatively simple to
estimate, requiring no more than one data series. Time series models may be separated
into two categories: basic time series models and advanced time series models. The basic
time series models include the nave, simple moving average, and single exponential
smoothing models. The advanced time series models include the double exponential
smoothing, autoregressive moving average, and basic structural time series models.
The Nave 1 (No Change) Model is the simplest forecasting model, which assumes that
the forecast value for this period (t) is equal to the actual value in the last period (t 2 1).
The nave 1 model has often been shown to generate the most accurate one-year-ahead
forecasts in comparison with other more sophisticated forecasting models (Martin and

71

DWYER 9781781001288 Part I.indd 71 25/07/2012 11:06

 72 Handbook of research methods in tourism

Table 4.1 Basic time series models

Method Equation
Nave 1 Ft 5 At21
Nave 2 Ft 5 At21* [ 1 1 (At21 2 At22) / At22 ]
Simple moving average Ft 5 (At21 1 At22 1 At23 1 . . . At2n) /n
Single exponential smoothing Ft 5 aAt21 1 (1 2 a) Ft21

Note: Ft 5 forecast at time t; At21 5 actual value at time t 2 1; n 5 number of lags in the moving average
process; a 5 smoothing constant (0 , a , 1) .

Table 4.2 Advanced time series models

Method Equation
Browns Yt 5 aAt21 1 (1 2 a) Yt21
Yrt 5 aYt21 1 (1 2 a) Yrt21
Ct 5 Yt 1 (Yt 2 Yrt )
Tt 5 [ (1 2 a) / a ] * (Yt 2 Yrt )
Ft1n 5 Ct 1 n * Tt
Holts Lt 5 aAt 1 (1 2 a) (Lt21 1 bt21)
bt 5 b (Lt 2 Lt21) 1 (1 2 b) bt21
Ft1n 5 Lt 1 n * bt
HoltWinters Lt 5 a (At 2 snt2h) 1 (1 2 a) (Lt21 1 bt21)
bt 5 b (Lt 2 Lt21) 1 (1 2 b) bt21
snt 5 g (At 2 Lt) 1 (1 2 g) snt2h
Ft1n 5 Lt 1 n * bt 1 snt1n2h
BoxJenkins At 5 d 1 a1At21 1 a2At22 1 . . . 1 apAt2p 2 b1et21 2 b2et22 2 . . . 2 bqet2q
Basic structural Ft 5 mt 1 gt 1 yt 1 et
time series

Note: Ft 5 forecast value at time t; At 5 actual value at time t; Yt 5 SES series at time t; Yrt 5 DES series
at time t; Ct 5 the intercept; Tt 5 the slope coefficient; n 5 the number of forecasting periods; Lt 5 smoothed
value at time t; bt5 trend estimate at time t; snt 5 seasonal variation at time t; d 5 constant; e 5 error term;
a, b, g are smoothing constants which are between 0 and 1; mt, gt, yt, et denote the trend, seasonal, cyclical
and irregular components, respectively.

Witt, 1989; Witt and Witt, 1995; Witt et al., 1994). However, the performance of the no
change model declines when dealing with sudden structural change and longer term fore-
casting (Chan et al., 1999; Witt et al., 1994).

The Nave 2 (Constant Change) Model is a widely used simple time series forecasting
model when a continuous trend is present in the data. The forecast value for period t is
obtained by multiplying the demand in period (t 2 1) by the growth rate between the
previous period (t 2 2) and the current period (t 2 1). Chan et al. (1999) used the Gulf
War as an example of a sudden shock and found that the nave 2 model performed best
compared with the ARIMA, exponential smoothing and quadratic trend curve models
in dealing with unstable data.

DWYER 9781781001288 Part I.indd 72 25/07/2012 11:06

 Demand modeling and forecasting 73

Since nave models often yield better results than econometric models (Martin and
Witt, 1989; Witt and Witt, 1995), many researchers use the nave models as benchmarks
for forecasting evaluations (Song et al., 2003a (Denmark); Turner and Witt, 2001(New
Zealand); Veloce, 2004 (Canada)).

The Simple Moving Average (SMA) Model allows the past values of a variable to deter-
mine the forecast values with equal weights assigned to the past values. The number of
observations included in the model determines the responsiveness of the model and it is
clear that the more lagged values included in the SMA model, the smoother the forecasts
become. If a time series shows wide variations around a trend, including more observa-
tions in the SMA model, will allow the model to pick up the trend better. However, the
main limitation of the SMA model is that it gives equal weight to all the lagged obser-
vations, which may not be realistic as the more recent values are likely to have a bigger
impact on the current value of the time series. Therefore, the SMA method normally
generates more accurate forecasts if the time series is less volatile (Frechtling, 1996;
Makridakis et al., 1998). Systematic errors may occur when the SMA model deals with a
time series that has a linear trend. To overcome this problem, the double moving average
method can be used to further smooth the series (Hu et al., 2004; Lim and McAleer,
2008).

The Single Exponential Smoothing (SES) Model is used to forecast a time series when
there is no trend or seasonal pattern. In a SES model, the forecast for period t is equal
to the forecast for period (t 2 1) plus a smoothing constant multiplied by the forecasting
error incurred in period (t 2 1). The smoothing constant must be between zero and one,
and is set by the forecaster, and the smaller the smoothing constant, the more weight
the method gives to the previous forecast value (Witt and Witt, 1992). Witt et al. (1992)
showed that the SES model generates forecasts with lower error magnitudes than the no
change model for domestic tourism demand, which is generally less volatile than inter-
national tourism demand.

ADVANCED TIME SERIES METHODS

Browns Double Exponential Smoothing (DES) Model was developed to deal with time
series that have a linear trend over time, either increasing or decreasing. When there is
no trend, this technique reduces to SES. Geurts and Ibrahim (1975) first applied Browns
model to forecast tourist arrivals in Hawaii and suggested that it is cheaper and easier
than the BoxJenkins approach for use in forecasting tourism demand. Sheldon (2008)
found that the DES and nave 1 models also performed well in forecasting international
tourism expenditures. However, the disadvantage of the DES model is that it does not
track non-linear trends well and often fails in picking up structural breaks in the time
series (Frechtling, 1996).

Holts Double Exponential Smoothing Model allows the trend and slope to be smoothed
using different smoothing constants. Since Browns method only uses one constant, the
estimated trend is very sensitive to random impacts, whereas Holts method is more

DWYER 9781781001288 Part I.indd 73 25/07/2012 11:06

 74 Handbook of research methods in tourism

flexible in selecting the smoothing constants (Makridakis et al., 1998). However, accord-
ing to Chen et al. (2008), Browns method outperformed Holts method based on the
mean absolute percentage error (MAPE) in forecasting tourist arrivals to US national
parks.

HoltWinters Triple Exponential Smoothing Model adds seasonal variations to Holts

model, and is appropriate when a time series has a linear trend with an additive seasonal
pattern. Since the HoltWinters model captures both the seasonal pattern and trend of
the time series, it usually outperforms other exponential smoothing methods (Lim and
McAleer, 2001). Grubb and Mason (2001) showed that the HoltWinters method with
damped trend greatly improved long-run forecasting accuracy compared with the Box
Jenkins and basic structural models in the case of UK air passenger series.

The BoxJenkins Model is the most frequently used advanced time series approach in
tourism demand forecasting. It is considered to be a relatively sophisticated technique
with the autoregressive moving average (ARMA) process as its basic form. Box
Jenkins forecasting models are identified by examining the behavior of the autocor-
relation function (AF) and partial autocorrelation function (PAF) of a stationary time
series. Stationarity of the time series is essential for the implementation of the classical
BoxJenkins model. A stationary series refers to a series that has a constant mean and
variance. If a time series is nonstationary, it should be differenced until it becomes
stationary. When the dth differences of a time series has an ARMA representation, the
time series can be written as an autoregressive integrated moving average (ARIMA)
model. The AF can help determine the number of times that the series needs to be dif-
ferenced.
Both the autoregressive (AR) and moving average (MA) models are useful forms of
the BoxJenkins model, and the AF and PAF are used to determine whether the time
series is an AR, MA, or ARMA process. When the PAF drops off to 0 after lag p, it indi-
cates that it is an AR (p) process while if AF drops off to 0 after lag q, it means that the
time series follows a MA (q) process (Chen et al., 2008).
The advantage of the BoxJenkins method is not only that these models can track the
behavior of a diverse range of time series, but also that they have fewer parameters to be
estimated in the final model than the multivariate regression approach. Such statistics
as the Akaike information criterion (AIC), Akaike final prediction error (AFPE), and
Bayesian information criterion (BIC) are used to add mathematical rigor to the judg-
ment process of identifying an appropriate ARMA or ARIMA model (De Gooijer and
Hyndman, 2006). Since seasonality is an important feature in most tourism series, sea-
sonal ARIMA (SARIMA) has gained popularity in recent years. Some key studies using
this approach include Goh and Laws (2002) study of Hong Kong tourism demand and
Kulendran and Witts (2003b) analysis of international business tourism in Australia.
To improve the explanatory power of the ARMA or ARIMA models, some research-
ers have included causal variables in the model, which are known as the ARMAX or
ARIMAX models. The causal ARMAX models, on average, outperformed both the
purely extrapolative models and simple econometric models in terms of forecasting accu-
racy, as they are developed by incorporating both the short-run dynamics as well as the

DWYER 9781781001288 Part I.indd 74 25/07/2012 11:06

 Demand modeling and forecasting 75

long-run cointegration relationships between the dependent and causal variables (Akal,
2004; Cho, 2001).

The Basic Structural Time Series Model (BSM) is constructed by decomposing a time
series into its trend, seasonal, cycle, and irregular components. A stochastic trend
captures changing consumer tastes in tourism demand, whereas stochastic seasonal-
ity allows for variations in the seasonal pattern. The irregular component represents
the transitory variations in tourism demand which are not explained by the other
components. The trend, seasonal, and cyclical components can be modeled in various
ways. Greenidge (2001) successfully applied the BSM to forecasting tourist arrivals to
Barbados. Explanatory variables can be included into the BSM to form a multivariate
structural time series model (STSM) (Gonzalez and Moral, 1995). However, according
to Turner and Witt (2001) and Kulendran and Witt (2003b), no evidence has emerged to
suggest that the inclusion of explanatory variables improves the forecasting accuracy of
the BSM. Therefore, for practitioners who are only interested in forecasting, the BSM
should be sufficient (Kim and Moosa, 2005).

ECONOMETRIC MODELS

Although time series approaches are useful tools for tourism demand forecasting, a
major limitation of these models is that the construction of the models is not based on
any economic theory that underlies tourists decision-making processes. Therefore, not
only can they not be used to analyse tourists behavior, but also they are incapable of
assisting policy makers in evaluating the effectiveness of the strategies and policies that
have been implemented in tourism development. Econometric models are superior to
time series approaches from this perspective (Song et al., 2009).

Table 4.3 Static econometric models

Method Equation
Traditional Y 5 a 1 a bixi 1 ei
regression i

Gravity model P iP j
Y5G
D2

wi 5 ai 1 a gijlogpj 1 biloga b 1 a fikdumk 1 ei

x
Static LAIDS
j P* k

logP* 5 a wilogpi
i

Note: Y 5 forecasts of tourism demand; xi 5 explanatory variables; Pj 5 population of region j; D 5 the

distance from region i to j; wi 5 budget share of the ith good, pj 5 price of the jth good; x 5 total expenditure
on all goods in the system; P* 5 aggregate price index; dumk 5 dummy variables; ei 5 disturbance term; a5
constant ; bi, G, ai, gij, bi, fik are the parameters to be estimated.

DWYER 9781781001288 Part I.indd 75 25/07/2012 11:06

 76 Handbook of research methods in tourism

Static Econometric Models (see Table 4.3) refer to those models that include the current
values of explanatory variables, but do not include lagged dependent or explanatory
variables. The main objective of static econometric models is to explore the factors that
affect tourism demand. The traditional regression approach, gravity models, and the
static almost ideal demand system (AIDS) are examples of static econometric models.

The Traditional Regression Approach generally uses ordinary least squares (OLS) as the
estimation procedure. This approach follows six main steps: (1) formulate hypotheses
based on demand theory; (2) identify the models functional form; (3) collect data; (4)
estimate the model; (5) test hypotheses; and (6) generate forecasts or evaluate the policies
(Song et al., 2009).
The advantages of the static regression approach are: (1) it explicitly addresses the
causal relationships between the demand for tourism and its influencing factors; (2) it
is useful for the assessment of political and business plans; (3) it provides several sta-
tistics to measure the accuracy and validity of the model and improve it based upon
these statistics (Frechtling, 1996). However, there are a number of limitations. The first
limitation of the static regression model is that tourism demand data tend to be trended
(nonstationary), and a regression model which contains such nonstationary series tends
to generate a spurious relationship between the dependent and independent variables
and this invalidates the diagnostic statistics of the model. The second limitation of the
static econometric model is that it cannot take dynamic changes in tourists behavior into
consideration when the model is estimated, which is rather restrictive and unnecessary.
The third limitation relates to the uncertainty involved in deriving the final model for
forecasting, as it is highly likely that different researchers armed with the same data set
may come up with completely different models. This is because no clear procedure can be
followed in terms of the model specification (Song et al., 2009). There are many examples
of the use of the traditional regression approach in the tourism forecasting literature; see
for example, Martin and Witt (1989), Witt and Martin (1987a) and Witt and Witt (1990).

The Gravity Model (GraM) examines the effects of such variables as distance and popu-
lation size on tourism demand. It assumes that the attractiveness between two countries/
regions is an inverse function of the square of their distance and is proportional to the
product of their populations. The gravity model has been widely used in international
trade research. Guo (2007) developed the gravity model to analyse the determinants
of inbound tourism demand in China, and Khadaroo and Seetanah (2008) used this
approach to investigate the role of transportation infrastructure on tourism flow.
However, the reliability of gravity models is questionable for their lack of a strong theo-
retical underpinning, which leads to the ad hoc choice of the explanatory variables (Che,
2004; Witt, 1980).

The Static Linear Almost Ideal Demand System (LAIDS) was originally introduced by
Deaton and Muellbauer (1980). It is a system of equations method, which is concerned
with long-run consumer behavior. In the static AIDS model, the budget share for good
i, which is the dependent variable, is related to the logarithms of prices and total real
expenditures. Since single equation models fail to estimate adequately the effect of a
change in tourism prices in a specific destination on the demand for traveling to other

DWYER 9781781001288 Part I.indd 76 25/07/2012 11:06

 Demand modeling and forecasting 77

destinations (substitution effect), the system of equations model was introduced to over-
come this limitation. The AIDS model is the most popular among a number of system
of equations methods, because it gives an arbitrary first-order approximation to any
demand function; and it has a functional form which complies with known household
budget data (Deaton and Muellbauer, 1980). It also has a flexible functional form which
does not impose any a priori restrictions on the elasticities of demand, that is, any good
and service in the system can be either an inferior or a normal good (Fujii et al., 1985).
Past empirical studies have shown that the static AIDS model has been a popular tech-
nique for analysing the market share of tourism demand, and also provides a range of
information about the sensitivity of tourism demand to price and expenditure changes
(De Mello et al., 2002; Han et al., 2006; Papatheodorou, 1999; Syriopoulos and Sinclair,
1993).
The static LAIDS, however, focuses on the long-run solution of the model that
assumes that consumers behave in the same manner over time and ignores the short-run
dynamics of the demand system. In reality, consumer behavior varies over time due to
change in tastes, adjustment costs, imperfect information, incorrect expectations, and
misinterpreted real price changes (Song et al., 2009).
Although static econometric models have advantages in exploring and interpreting
the elasticities of the explanatory variables, they still do not perform well in forecasting
tourism demand, as they do not consider the long run cointegration relationships and
short run dynamics in the estimation of the models (Song et al., 2009). The poor per-
formance of the static models may partly be due to the fact that they omit the word-of-
mouth effect. They ignore the dynamics of the demand system and assume that tourism
demand is not affected by previous values of the economic variables. Furthermore, the
stationarity properties of the variables are not considered in static models. As a result,
spurious regression is very likely to occur (Li, 2010).

Dynamic Econometric Models. The adoption of advanced econometric techniques, such

as the vector autoregressive (VAR) models, time varying parameter (TVP) models and
error correction models have much improved the forecasting performance of economet-
ric models. The dynamic models capture the word-of-mouth effect and the inclusion
of the causal variables increases the models explanatory power.

The Autoregressive Distributed Lag Model (ADLM) is a dynamic econometric model,

which involves a general functional form containing both the current and lagged values
of the variables. Stepwise reduction, which is also known as the general-to-specific
approach, is applied to estimate the ADLM. This process continues until all the criteria
are satisfied, including a high R2 value, statistically significant coefficients of the explana-
tory variables, and lack of autocorrelation and heteroskedasticity in the error term (Song
and Witt, 2003).
The model selection process in the ADLM involves the following steps. First, a general
demand model with a large number of explanatory variables, including the lagged
dependent and independent variables, is constructed as a general ADLM; second, the t,
F, and Wald statistics are used to test the various restrictions, in order to achieve a simple
but statistically significant specification; third, the normal diagnostic tests are carried out
to examine whether the final model is statistically acceptable; and in the last step, the

DWYER 9781781001288 Part I.indd 77 25/07/2012 11:06

 78 Handbook of research methods in tourism

final model is used for policy evaluation and forecasting (Song et al., 2009). The main
criteria for selecting an ADLM include: consistent with economic theory, data coher-
ency, parsimony, encompassing, parameter constancy and exogeneity (Thomas, 1993).
The general-to-specific methodology overcomes the disadvantages of the specific-to-
general method, which tends to end up with a highly complicated final model involving
too many variables, which may still not be able to capture the dynamic characteristics of
the demand model and leads to poor forecasting performance. Furthermore, the ADLM
does not necessarily need a priori knowledge about the integration properties of the
variables, and provides robust estimation of the model parameters even in small samples
(Narayan, 2004).
With different restrictions imposed on the parameters, an ADLM can be written into
different econometric models, such as the static model, the autoregressive model, the
growth rate model (Witt and Witt, 1992), the leading indicator model (Turner et al.,
1997; Kulendran and Witt, 2003a), the partial adjustment model (Song et al., 2003b),
the common factor model (Lee et al., 1996), the finite distributed lag model and the dead
start model (Song et al., 2003b) (see Table 4.4). These models have been widely used in
tourism forecasting in recent years, however, their performance has varied under differ-
ent conditions.
The general-to-specific methodology has been popular in tourism demand analysis,
and has been used to explore the determinants of tourism demand in specific destinations
such as Australia (Kulendran and Divisekera, 2007), Asia (Song and Lin, 2010), Korea
(Song and Witt, 2003), Denmark (Song et al., 2003a), Europe (Smeral, 2010), and Latin
America (Vanegas, 2008). Meanwhile, it has been proven to perform well in forecasting
turning points (Nadal, 2001). However, one of the possible problems with this method
is that the structure of the selected final model relies too much on the data used, even
though economic theory plays an important role in the initial form of the general model
(Song and Witt, 2003).

The Error Correction Model (ECM) expresses the current value of the dependent vari-
able as a linear function of past values of the dependent variable, current and past values
of the explanatory variables and the previous values of the error term from the cointe-
gration relationships (Kulendran and Witt, 2001). The general form of the ECM can be
written as:

Dy 5 (current and lagged Dxjt s, lagged Dyt s) 2 (1 2 1) c yt21 2 a xjxjt21 d 1 et (4.1)

k

j51

The ECM (1, 1) takes the form

Dyt 5 b0Dxt 2 (1 2 1) [ yt21 2 k0 2 k1xt21 ] 1 et (4.2)

where k0 5 a / (1 2 1) ,k1 5 (b0 1 b1) / (1 2 1) , b0 is the impact parameter, (1 2 1) is

the feedback effect, k0 and k1 are the long-run response coefficients.

Cointegration (CI) describes the relationship between a pair of nonstationary economic

variables, who share a common stochastic trend, that is, if two or more time series are

DWYER 9781781001288 Part I.indd 78 25/07/2012 11:06

 Demand modeling and forecasting 79

Table 4.4 Variations of ADLM

Model Equation

Unrestricted ADLM k p p
yt 5 a 1 a a bjixjt2i 1 a iyt2i 1 et
j 51 i 50 i 51

Static k
yt 5 a 1 a bjxjt 1 et
j 51

p
Autoregressive (AR)
yt 5 a 1 a iyt2i 1 et
i 51

Growth rate k
Dyt 5 a 1 a bjDxjt 1 et
j 51

Leading indicator k p
yt 5 a 1 a a bjixjt2i 1 et
j 51 i 51

Partial adjustment k p
yt 5 a 1 a bjxjt 1 a iyt2i 1 et
j 51 i 51

Common factor k
yt 5 a 1 a bjxjt 1 et
j 51

p
et 5 a biet2i 1 mi
i 51

Finite distributed lag k p

yt 5 a 1 a a bjxjt2i 1 et
j 50 i 50

Dead start k p p
yt 5 a 1 a a bjxjt2i 1 a iyt2i 1 et
j 51 i 51 i 51

Note: yt 5 tourism demand; xt 5 explanatory variables; k 5 number of explanatory variables; p 5 lag

length; a 5 constant; et and mi 5 error terms; bjiand i are the parameters to be estimated.

Source: Adapted from Song et al., 2009, p. 48.

individuallyintegrated, but some linear combination of them has a lower order of inte-
gration, the series are said to be cointegrated. Unit roots tests are used to examine the
CI relationship between the two series (Song et al., 2009). If the two variables are cointe-
grated, a regression model that relates the two variables to each other would not generate
spurious relationships.
The ECM and CI are a bi-directional transformation which is often called the
Granger Representation Theorem (Engle and Granger, 1987). The ECM and CI
model are useful when both long run equilibrium and short run disequilibrium relation-
ships are of interest. For example, Choyakh (2008), Daniel and Ramos (2002), Dritsakis
(2004), and Halicioglu (2010) used the ECM and CI model to examine the demand for
tourism in Tunisia, Portugal, Greece, and Turkey, respectively.

DWYER 9781781001288 Part I.indd 79 25/07/2012 11:06

 80 Handbook of research methods in tourism

The ECM has the following advantages: (1) it overcomes the spurious regression
problem by differencing the variables and it also avoids the problems of the growth rate
model, where only differenced data are used; (2) it is another form of the ADLM, which
fits in well with the general-to-specific methodology; (3) it reduces the problem of data
mining during estimation (Song et al., 2009).
In their studies on the demand for Canadian and Tunisian tourism, Veloce (2004)
and Querfelli (2008) found that ECMs provide more precise forecasts than time series
models when a differenced demand variable is concerned. Kulendran and Witt (2001)
also demonstrated that the ECM and CI models are more accurate than the traditional
econometric models in forecasting tourism demand from the UK to Germany, Greece,
the Netherlands, Portugal, Spain, and the United States.
Three main estimation methods are used to estimate ECMs: the EngleGranger
two stage (EG2S), the WickensBreusch (WB) and the ADLM procedures. However,
these methods show different levels of performance when different forecasting horizons
are considered. Wickens and Breusch (1988) suggested that in the case of the EG2S
approach, the estimation bias in the first stage may be carried over to the second stage
when the sample size is small.

The VAR Model was developed by Sims (1980), and here the variables are modeled
purely as dynamic processes, except for the deterministic variables such as trend, inter-
cept and dummy variables. In the VAR model, each variable is a linear function of the
lagged values of all variables in the system. The VAR model treats all the variables as
endogenous, and does not rely on the assumption that all the explanatory variables need
to be exogenous as in the case of the single equation models. The general VAR (p) model
can be written as:
p
Yt 5 a AiYt2i 1 BZt 1 Ut (4.3)
i51

where, Yt is a k vector of variables included in the system; Zt is a vector of deterministic

variables; Ut is a vector of regression errors that are assumed to be contemporaneously
correlated but not autocorrelated; Ai and B are matrices of parameters to be estimated.
A general VAR model, also called the unrestricted VAR model, is the standard form
of the VAR model. In the unrestricted model, lag lengths for each variable are identical
and every variable in the system is included in each equation. In a mixed VAR model,
different lag lengths are specified for each variable in each equation. A likelihood ratio
test, the Akaike Information Criterion (AIC) and Schwarz Bayesian Criterion (SBC) are
usually suggested to determine the optimal lag length (Song et al., 2009). Wong et al.
(2006) found that the Bayesian VAR model that combines priors with sample informa-
tion, overcomes the problem of overfitting and improves the forecasting performance of
the VAR models when forecasting the demand for Hong Kong tourism.
The VAR models have proved to be capable of producing accurate medium- to long-
term tourism forecasts. In a study of the demand for Macau tourism, Song and Witt
(2006) concluded that the VAR technique was found to be superior to the traditional
regression approach for the following two reasons. First, the VAR models do not require
an implicit theoretical framework in the construction and estimation of the models.

DWYER 9781781001288 Part I.indd 80 25/07/2012 11:06

 Demand modeling and forecasting 81

Second, the forecasts are easy to generate using VAR, as it does not require generat-
ing forecasts of the explanatory variables first in order to generate the forecasts of the
dependent variable. Therefore, it is easier from this aspect to produce tourism demand
forecasts using VAR models than other econometric models. Although the VAR tech-
nique has been widely and successfully used in macroeconomic forecasting, little effort
have been made to apply this method in tourism forecasting.

The Time Varying Parameter Model takes the possibility of parameter changes over time
into consideration in the modeling exercise. The TVP model overcomes the structural
instability problem caused by external shocks. According to Song and Witt (2000),
the TVP model can simulate different types of external shocks to the tourism demand
system, including policy and regime shifts, economic reforms and political uncertainties.
Besides, the TVP model performs well in capturing external influences that are gradual
and diffuse in nature, such as changes in consumer tastes and other social and psycho-
logical changes of the consumers (Song and Wong, 2003). The TVP model can be written
in a state space form:

yt 5 btxt 1 mt

bt 5 bt21 1 Rtet (4.4)

where yt is tourism demand, xt is a row vector of k explanatory variables, bt is the column

vector of k state variables, is a k 3 k matrix, Rt is a k 3 g matrix, mt is a residual with
zero mean and constant covariance, and et is a g 3 1 vector of serially uncorrelated
residuals with zero mean and constant covariance (Song et al., 2009).
The TVP method incorporates structural changes in forecasting and puts more weight
on the most recent data in estimating the demand model in order to improve forecasting
accuracy. The TVP model can be estimated using an efficient recursive algorithm called
the Kalman filter. This method is increasingly used in tourism demand analysis (for
example, Riddington, 1999; Song and Witt, 2000; Song and Wong, 2003; Wu, 2010).
According to Song et al. (2009), based on their examination of tourist arrivals in the
UK and United States, the TVP model generated the most accurate short-run forecasts,
which is consistent with previous studies on forecasting tourism demand (Song and
Wong, 2003; Song et al., 1998). In their study of the demand for Denmark tourism, Witt
et al., (2003) stated that the TVP model performs consistently well for 1-year-ahead fore-
casting for the various forecasting error criteria selected.

The Dynamic AIDS Model is the error correction form of the AIDS model. It incorpo-
rates a short-run adjustment mechanism, which is written as:

Dwi 5 diDwit21 1 a rijD log pj 1 biD log a b 1 limit21

x
(4.5)
j P*

where, mit21 is the EC term, which measures the feedback effect and is estimated from the
corresponding CI equation (Li et al., 2004).
Durbarry and Sinclair (2003) first applied the EC-AIDS approach to a tourism

DWYER 9781781001288 Part I.indd 81 25/07/2012 11:06

 82 Handbook of research methods in tourism

demand analysis of tourist expenditure in France, but they failed to include any short-
run independent variables due to insignificant coefficients. Li et al. (2004) used the
EC-AIDS model to evaluate the international tourism competitiveness of five Western
European countries and found that the dynamic linear AIDS model performed better
than the static linear AIDS model. De Mello and Fortuna (2005) studied the demand
for European tourism by UK residents and found that the dynamic AIDS model is a
data-coherent and theoretically consistent model, which provided robust estimates and
reliable forecasts.

ARTIFICIAL INTELLIGENCE METHODS

Besides the time series and econometric forecasting methods, techniques such as the
artificial intelligence (AI) model have emerged in the tourism demand forecasting litera-
ture. AI is the study and design of intelligent agents, which is a system that perceives its
environment and takes actions that maximize its chances of success. AI techniques have
rapidly developed within various research areas. According to Wang (2004), the AI fore-
casting methods, including neural networks, rough sets theory, fuzzy time series theory,
grey theory, genetic algorithms, and expert systems, tend to perform better than tradi-
tional forecasting methods. Applications of AI techniques in tourism demand analysis
include the following studies: Uysal and Roubi (1999) studied Canadian tourists expen-
ditures in the United States; Law and Au (1999) forecast the demand for Hong Kong
tourism by the Japanese; Cho (2003) analysed tourist arrivals in Hong Kong; and Wang
(2004) examined the trend of tourist arrivals in Taiwan from Hong Kong.
Not requiring preliminary information or additional data are the main advantages of
AI methods, while the lack of a theoretical underpinning and the inability to interpret
demand from an economic perspective are the limitations of these techniques (Song and
Li, 2008). These limitations restrict the practical application of AI methods in tourism
demand analysis.

The Artificial Neural Network (ANN) Model, also called Neural Network (NN) Model,
is the most widely used AI model in tourism demand forecasting. ANN is a computa-
tional model that is inspired by the structure of biological neural networks. A typical
ANN consists of a number of simple processing elements called neurons, nodes, or units.
ANN is an adaptive system that changes its structure based on external or internal infor-
mation that flows through the network during the learning phase. A typical architecture
of ANN models includes one input layer, one or two hidden layers and one output
layer (Simpson, 1989). The input layer of the ANN models comprises the explanatory
variables, and the output layer denotes the dependent variable, i.e., tourism demand.
In the hidden layers, the core part of the ANN model, different weights are given to
each explanatory variable as well as the dependent variable. The computing process is
repeated until a set of optimal weights is obtained. Each node yi in the hidden layer can
be expressed as:
1
yi 5
1 1 e2f
j

DWYER 9781781001288 Part I.indd 82 25/07/2012 11:06

 Demand modeling and forecasting 83
2
fj 5 a xiwji (4.6)
i51

And the final node Y of the output layer is obtained from:

3
Y 5 a yjwj (4.7)
j51

where, w is the weight of interconnection and x is the input variable.

The ANN model was first applied in tourism demand forecasting in the late 1990s and
incorporated some improvements in the post-2000 literature. Kim et al. (2003) studied the
senior tourist market of Western Australia. Palmer et al. (2006) forecast tourists expendi-
ture in the Balearic Islands of Spain. ANNs are useful for non-linear threshold functions,
which overcomes a restriction of multiple regression analysis. In forecasting the demand
for US and Hong Kong tourism, Cho (2003) and Uysal and Roubi (1999) separately
showed that ANNs generally outperformed the classical time series and multiple regression
models. Moreover, the ANN model can easily be updated and it performs fairly well in
one-year-ahead forecasting due to the repetitions of expected similar seasonal patterns (see
for example, Burger et al., 2001 in the case of forecasting the demand for Durban tourism).
However, the limitations of the model cannot be ignored. First, the learning process of
the hidden layers needs a large amount of data; second, the impact of explanatory vari-
ables on tourism demand cannot be generated from the model; third, the elasticities of
tourism demand cannot be calculated from ANN models (Wu, 2010).

The Rough Sets Approach was first used to forecast the demand for hotel rooms in Hong
Kong by Law and Au (1998). This model is generated from the rough sets theory and
based on data mining techniques to discover the knowledge. The advantages of the
rough sets approach are that: (1) it can model the decision processes underlying the
data in both numeric and non-numeric forms, which makes it a useful classification and
pattern recognition technique; and (2) it can generate comprehensive decision rules that
are useful for practitioners (Au and Law, 2002; Goh and Law, 2003). The only assump-
tion of this approach is that the values of the attributes can be categorized.
The rough sets approach pays much attention to such categorical variables as demo-
graphic features and psychographic variables, and forecasts tourism demand levels as
high or low instead of exact values. It analyses tourism demand from a microperspective,
which is viewed as a complementary tool to econometric forecasting models (Goh et al.,
2008; Song and Li, 2008).

The Support Vector Regression (SVR) mechanism is a novel learning machine based on
statistical learning theory and adheres to the principle of structural risk minimization
seeking to minimize an upper bound of the generalization error rather than minimize the
training error, which is the principle of ANN. It is an alternative technique which can be
used to solve classification, nonlinear regression and forecasting problems by introduc-
ing a loss function.

Chen and Wang (2007) incorporated the Genetic Algorithm (GA) technique into support
vector regression to form a GA-SVR model to improve forecasting accuracy. GA is a

DWYER 9781781001288 Part I.indd 83 25/07/2012 14:09

 84 Handbook of research methods in tourism

heuristic search algorithm that mimics natural evolution. This method is routinely used
to generate useful solutions to optimization and search problems. Using GA, Chen
and Wang (2007) forecast tourist arrivals to China, and compared the forecasting per-
formance between the GA-SVR, Back Propagation Neural Network (BPNN, the most
popular ANN model) and ARIMA techniques. The results suggested that the GA-SVR
model generated more accurate forecasts than both the BPNN and ARIMA models.

The Fuzzy Time Series Method is used to forecast a process with linguistic value observa-
tions and deals with the first order difference of the time series. The main assumption of
the fuzzy time series model is that the variation of this year is related to the variation of
the previous year and follows the trend of recent years. Therefore, if the actual variation
is considerably different from recent trends, the forecasting error is likely to be large
(Yu and Schwartz, 2006). Human judgments should be involved in the fuzzy time series
model when a decision about two parameters needs to be made. The first parameter
is the number of grades which describes the extent of variation and the second is the
window basis size referring to the number of previous observations used to generate the
prediction (Yu and Schwartz, 2006). Wang (2004) confirmed that the fuzzy time series
technique is considered to be an appropriate tool for forecasting short-term tourism
demand in Hong Kong by Taiwanese tourists. However, one of the foremost disad-
vantages is that it can hardly adapt to the shocks of the special events (Wang and Hsu,
2008).

Grey Theory is a generic theory that deals with systems with poor, incomplete, and uncer-
tain information. The grey model reduces randomness by using the accumulated generat-
ing operation where an exponential function is fitted based on a differential equation to
estimate the future trend (Yu and Schwartz, 2006). The advantage of the grey model is
that it can be constructed based on very short time series. It even can be fitted with as
few as four observations (Chiang et al., 1998). The limitations of the grey model are that
the data used must be taken at equal intervals and in consecutive order (Lin et al., 2001).

FORECASTING PERFORMANCE

Ex post and ex-ante forecasting are two different forecasting concepts. As Figure 4.1
shows, the data are available from period 1 to N, forecasts from time n to N are called
expost forecasts, and forecasts after time N are called ex-ante forecasts. In practice, only
the accuracy of ex-post forecasts can be evaluated by actual data. In terms of the ex-ante
forecasts, which are more important for practitioners, the explanatory variables have to
be forecast first before the ex-ante forecasts can be produced (Song et al., 2009).

Measures of Forecasting Accuracy

A number of forecasting evaluation criteria are available and these include forecast-
ing accuracy, time costs, money costs, and the complexity of the model. Among these
criteria, researchers are more concerned about the forecasting accuracy of the models,
as forecasting accuracy is crucial to decision making by tourism practitioners and

DWYER 9781781001288 Part I.indd 84 25/07/2012 14:09

 Demand modeling and forecasting 85

t =1 t=n t=N

model estimation period ex-post forecasts ex-ante forecasts

Time

Y1 Yn YN

Data available

Source: Song et al. (2009, p. 183).

Figure 4.1 Time horizons of forecasts

Table 4.5 Error magnitude measures

Measure Equation

Mean absolute percentage error (MAPE) 1 n 0 Yt 2 Yt 0

MAPE 5 a
n t51 Yt

Root mean square percentage error 1 n Yt 2 Yt 2

(RMSPE) RMSPE 5 aa Y b
n t51 t
Mean absolute error (MAE) 1 n
MAE 5 a 0 Yt 2 Yt 0
n t51

Note: n denotes the length of forecasting horizon; Yt is the actual value of the dependent variable; Y is the
forecast value of the dependent variable.

policy makers. In practice, the most widely used method to measure forecasting accu-
racy is error magnitude (see Table 4.5), which includes the mean absolute percentage
error (MAPE), root mean square percentage error (RMSPE) and mean absolute error
(MAE).
Besides the above commonly used error magnitude measures, some other statistical
measures can also be used to assess the performance of the forecasting models, such as
Theils U statistic, which is used to compare the performance between other forecast-
ing models and the naive no change model, the acceptable output percentage and the
normalized correlation coefficient, which are often used in measuring the forecasting
accuracy of ANN models (Law and Au, 1999); and the unbiasedness test, which is used
to test forecasting consistency (Witt et al., 2003) (see Table 4.6).
Another forecasting measure suggested by Witt and Witt (1991) is directional change,

DWYER 9781781001288 Part I.indd 85 25/07/2012 14:09

 86 Handbook of research methods in tourism

Table 4.6 Other forecasting assessment measures

Measures Equation
Theils U statistic n

a (DYt1n 2 DYt1n)
2

i 51
U5 n

a (DYt1n)
2
i 51

Acceptable output n 0 Yi 2 Yi 0
percentage a ti ti 5 1 if # 5%
i 51 Yi
Z5 3 100% for
n ti 5 0 otherwise

Normalized correlation n

coefficient a (Yi 3 Yi)

i 51
r5
n n
Y 2 3 a Y2i
a i i 51 i 51

Unbiasedness test Yt1h 5 b0 1 b1Yt1h 1 ut1h

H0 : b0 5 0, b1 5 1

Note: Yt is the actual value of the dependent variable; Yt is the forecast value of the dependent variable; and
n is the length of the forecasting horizon.

which assesses the probability of correctly forecasting directional change. Turning point
accuracy measures and trend measures are particular subsets of directional change error
measures (Witt and Witt, 1991, 1995). According to Henriksson and Merton (1981), the
probabilities of directionally correct forecasts can be written as:

p1 5 prob (DQtDQt . 0 0 DQt , 0)

p2 5 prob (DQtDQt . 0 0 DQt $ 0) (4.8)

where p1 is the probability of a directionally correct forecast on the condition that

tourism demand falls at time t, and p2 is the probability of a directionally correct fore-
cast conditional on no fall in tourism demand at time t. Suppose that n1 is the number of
observations for a decrease in tourism demand, n2 is the number of observations for no
decrease in tourism demand, m1 is the number of correct forecasts given a fall in tourism
demand, and m2 is the number of incorrect forecasts given no fall in tourism demand.
Then the estimated value of p1 1 p2 is p1 1 p2 5 m1 / n1 1 (n2 2 m2) / n2. For a forecast-
ing model which generates rational forecasts, the condition p1 1 p2 $ 1 must hold; oth-
erwise the nave no-change model is preferred (Witt et al., 2003). The directional change
error measures have been used in many other areas, but few efforts have been made to
use the tests to evaluate tourism demand forecasting models (except, for example, Witt
and Witt, 1989, 1991; Witt et al., 2003).

DWYER 9781781001288 Part I.indd 86 25/07/2012 14:09

 Demand modeling and forecasting 87

FUTURE RESEARCH DIRECTIONS

Past empirical findings show that no single forecasting method is superior to others in all
situations. The forecasting performance of the models varies considerably depending on
the forecast error measure, the forecasting horizon, data frequency, destinationorigin
country pairs, and the competitors included in a comparison. An ever-growing inter-
est in tourism forecasting is, therefore, to find optimal forecasting models according to
the data characteristics. Based on this interest, the future research directions in tourism
forecasting include the following. First, carry out a comprehensive review of the exist-
ing studies using such methods as meta-analysis to synthesize the research findings of
the previous studies over the past four to five decades with a view to generalizing the
research findings based on the characteristics of the data used in the tourism forecast-
ing literature. Second, another way of generalizing the research findings is to carry out
a forecasting competition based on a wide range of forecasting methods used in the past
literature with the involvement of the experts who have published extensively using the
quantitative methods highlighted above. Third, the integration of judgmental (qualita-
tive) forecasts and quantitative forecasts would also be a useful direction for future
research as this research topic has been relatively scarce in the tourism context.

ACKNOWLEDGEMENTS

The authors would like to acknowledge the financial support of The Hong Kong
Polytechnic University (Grant No: G-U743).

REFERENCES

Akal, M. (2004), Forecasting Turkeys tourism revenues by ARMAX Model, Tourism Management, 25 (5),
565580.
Au, N. and R. Law (2002), Categorical classification of tourism dining, Annals of Tourism Research, 29 (3),
819833.
Burger, C.J.S.C., M. Dohnal, M. Kathrada and R. Law (2001), A practitioners guide to time-series methods
for tourism demand forecasting: a case study of Durban, South Africa, Tourism Management, 22 (4):
403409.
Chan, Y., T. Hui and E. Yuen (1999), Modeling the impact of sudden environmental changes on visitor arrival
forecasts: the case of the Gulf War, Journal of Travel Research, 37 (4), 391394.
Che, Y.B. (2004), An approach to modeling regional tourist attraction, Resource Development & Market, 20
(3), 163165.
Chen, K. and C. Wang (2007), Support vector regression with genetic algorithms in forecasting tourism
demand, Tourism Management, 28 (1), 215226.
Chen R.J.C., P. Bloomfield and F.W. Cubbage (2008), Comparing forecasting models in tourism, Journal of
Hospitality & Tourism Research, 32 (1), 321.
Chiang J.S., P.L. Wu, S.D. Chiang, T.J. Chang, S.T. Chang and K.L. Wen (1998), Introduction of Grey Theory,
Taiwan: Gao-Li.
Cho, V. (2001), Tourism forecasting and its relationship with leading economic indicators, Journal of
Hospitality and Tourism Research, 25 (4), 399420.
Cho, V. (2003), A comparison of three different approaches to tourist arrival forecasting, Tourism
Management, 24 (3), 323330.
Choyakh, H. (2008), A model of tourism demand for Tunisia: inclusion of the tourism investment variable,
Tourism Economics, 14 (4), 819838.

DWYER 9781781001288 Part I.indd 87 25/07/2012 14:09

 88 Handbook of research methods in tourism

Daniel, A.C.M. and F.F.R. Ramos (2002), Modelling inbound international tourism demand to Portugal,
International Journal of Tourism Research, 4 (3), 193209.
De Gooijer, J.G. and R.J. Hyndman (2006), 25 years of time series forecasting, International Journal of
Forecasting, 22 (3), 443473.
De Mello, M.M., A. Pack and M.T. Sinclair (2002), A system of equations model of UK tourism demand in
neighbouring countries, Applied Economics, 34 (4), 509521.
De Mello, M.M. and N. Fortuna (2005), Testing alternative dynamic system for modelling tourism demand,
Tourism Economics, 11 (4), 517537.
Deaton, A. and J. Muellbauer (1980), An almost ideal demand system, The American Economic Review, 70
(3), 312326.
Dritsakis, N. (2004), Cointegration analysis of German and British tourism demand for Greece, Tourism
Management, 25 (1), 111119.
Durbarry, R. and M.T. Sinclair (2003), Market shares analysis: the case of French tourism demand, Annals
of Tourism Research, 30 (4), 927941.
Engle, R.F. and C.W.J. Granger (1987), Co-integration and error correction: representation, estimation, and
testing, Econometrica, 55 (2), 251276.
Frechtling, D.C. (1996), Practical Tourism Forecasting, Oxford: Butterworth-Heinemann.
Fujii, E.T., M. Khaled and J. Mak (1985), An almost ideal demand system for visitor expenditures, Journal
of Transport Economics and Policy, 19 (2), 161171.
Geurts, M.D. and I.B. Ibrahim (1975), Comparing the BoxJenkins approach with the exponentially
smoothed forecasting model application to Hawaii tourists, Journal of Marketing Research, 12 (2), 182188.
Goh, C. and R. Law (2002), Modeling and forecasting tourism demand for arrivals with stochastic nonsta-
tionary seasonality and intervention, Tourism Management, 23 (5), 499510.
Goh, C. and R. Law (2003), Incorporating the rough sets theory into travel demand analysis, Tourism
Management, 24 (5), 511517.
Goh, C., R. Law and H.M.K. Mok (2008), Analyzing and forecasting tourism demand: a rough sets
approach, Journal of Travel Research, 46 (3), 327338.
Gonzalez, P. and P. Moral (1995), An analysis of the international tourism demand in Spain, International
Journal of Forecasting, 11 (2), 233251.
Greenidge, K. (2001), Forecasting tourism demand: an STM approach, Annals of Tourism Research, 28 (1),
98112.
Grubb, H. and A. Mason (2001), Long lead-time forecasting of UK air passengers by Holt-Winters methods
with damped trend, International Journal of Forecasting, 17 (1), 7182.
Guo, W. (2007), Inbound tourism: an empirical research based on gravity model of international trade,
Tourism Tribune, 22 (3), 3034.
Halicioglu, F. (2010), An econometric analysis of the aggregate outbound tourism demand of Turkey,
Tourism Economics, 16 (1), 8397.
Han, Z., R. Durbarry and M.T. Sinclair (2006), Modelling US tourism demand for European destinations,
Tourism Management, 27 (1), 110.
Henriksson, R.D. and R.C. Merton (1981), On market timing and investment performance. II. Statistical
procedures for evaluating forecasting skills, The Journal of Business, 54 (4), 513533.
Hu, C., M. Chen and S.C. McChain (2004), Forecasting in short-term planning and management for a casino
buffet restaurant, Journal of Travel & Tourism Marketing, 16 (2), 7998.
Khadaroo, J. and B. Seetanah (2008), The role of transport infrastructure in international tourism develop-
ment: a gravity model approach, Tourism Management, 29 (5), 831840.
Kim, J., S. Wei and H. Ruys (2003), Segmenting the market of West Australian senior tourists using an artifi-
cial neural network, Tourism Management, 24 (1), 2534.
Kim, J.H. and I.A. Moosa (2005), Forecasting international tourist flows to Australia: a comparison between
the direct and indirect methods, Tourism Management, 26 (1), 6978.
Kulendran, N. and S. Divisekera (2007), Measuring the economic impact of Australian tourism marketing
expenditure, Tourism Economics, 13 (2), 261274.
Kulendran, N. and S.F. Witt (2001), Cointegration versus least squares regression, Annals of Tourism
Research, 28 (2), 291311.
Kulendran, N. and S.F. Witt (2003a), Leading indicator tourism forecasts, Tourism Management, 24 (5),
503510.
Kulendran, N. and S.F. Witt (2003b), Forecasting the demand for international business tourism, Journal of
Travel Research, 41, 265271.
Law, R. and N. Au (1998), A rough set approach to hotel expenditure decision rules induction, Journal of
Hospitality & Tourism Research, 22 (4), 359375.
Law, R. and N. Au (1999), A neural network model to forecast Japanese demand for travel to Hong Kong,
Tourism Management, 20, 8997.

DWYER 9781781001288 Part I.indd 88 25/07/2012 14:09

 Demand modeling and forecasting 89

Lee, C.K., T. Var and T.W. Blaine (1996), Determinants of inbound tourism expenditure, Annals of Tourism
Research, 23 (3), 527542.
Li, G. (2010), Tourism demand modelling and forecasting: a review of literature related to Greater China,
Journal of China Tourism Research, 5 (1), 240.
Li, G., H. Song and S.F. Witt (2004), Modelling tourism demand: a dynamic linear AIDS approach, Journal
of Travel Research, 43 (2), 141150.
Lim, C. and M. McAleer (2001), Forecasting tourist arrivals, Annals of Tourism Research, 28 (4), 965977.
Lim, C. and M. McAleer (2008), Analysing seasonal changes in New Zealands largest inbound market,
Tourism Recreation Research, 33 (1), 8391.
Lin, C.B., S.F. Su and Y.T. Hsu (2001), High-precision forecast using grey models, International Journal of
Systems Science, 32 (5), 609619.
Makridakis, S., S.C. Wheelwright and R.J. Hyndman (1998), Forecasting: Methods and Applications, 3rd
edition, New York: John Wiley.
Martin, C.A. and S.F. Witt (1987), Tourism demand forecasting models: choice of appropriate variable to
represent tourists cost of living, Tourism Management, 8 (3), 233246.
Martin, C.A. and S.F. Witt (1988), Substitute prices in models of tourism demand, Annals of Tourism
Research, 15 (2), 255268.
Martin, C.A. and S.F. Witt (1989), Forecasting tourism demand: a comparison of the accuracy of several
quantitative methods, International Journal of Forecasting, 5 (1), 719.
Nadal, J.R. (2001), Forecasting turning points in international visitor arrivals in the Balearic Islands, Tourism
Economics, 7 (4), 365380.
Narayan, P.K. (2004), Fijis tourism demand: the ARDL approach to cointegration, Tourism Economics, 10
(2), 193206.
Palmer, A., J.J. Montano and A. Sese (2006), Designing an artificial neural network for forecasting tourism
time series, Tourism Management, 27 (5), 781790.
Papatheodorou, A. (1999), The demand for international tourism in the Mediterranean region, Applied
Economics, 31 (5), 619630.
Querfelli, C. (2008), Co-integration analysis of quarterly European tourism demand in Tunisia, Tourism
Management, 29 (1), 127137.
Riddington, G. (1999), Forecasting ski demand: comparing learning curve and varying parameter coefficient
approaches, Journal of Forecasting, 18, 205214.
Sheldon, P.J. (2008), Forecasting tourism: expenditures versus arrivals, Journal of Travel Research, 32 (1),
1320.
Simpson, P.K. (1989), Artificial Neural Systems: Foundations, Paradigms, Applications, and Implementations,
Elmsford, NY: Pergamon.
Sims, C.A. (1980), Macroeconomics and reality, Econometrica, 48 (1), 148.
Smeral, E. (2010), Impacts of the world recession and economic crisis on tourism: forecasts and potential
risks, Journal of Travel Research, 49 (1), 3138.
Song, H. and G. Li (2008), Tourism demand modelling and forecasting: a review of recent research, Tourism
Management, 29 (2), 203220.
Song, H. and S. Lin (2010), Impacts of the financial and economic crisis on tourism in Asia, Journal of Travel
Research, 49 (1), 1630.
Song, H. and S.F. Witt (2000), Tourism Demand Modelling and Forecasting: Modern Econometric Approaches,
Oxford: Pergamon.
Song, H. and S.F. Witt (2003), Tourism forecasting: the general-to-specific approach, Journal of Travel
Research, 42 (1), 6574.
Song, H. and S.F. Witt (2006), Forecasting international tourist flows to Macau, Tourism Management, 27
(2), 214224.
Song, H. and K.K.F. Wong (2003), Tourism demand modelling: a time-varying parameter approach, Journal
of Travel Research, 42 (1), 5764.
Song, H., P. Romilly and X. Liu (1998), The UK consumption function and structural instability:
improvingforecasting performance using a time varying parameter approach, Applied Economics, 30 (7),
975983.
Song, H., S.F. Witt and T.C. Jensen (2003a), Tourism forecasting: accuracy of alternative econometric
models, International Journal of Forecasting, 19 (1), 123141.
Song, H., S.F. Witt and G. Li (2003b), Modelling and forecasting the demand for Thai tourism, Tourism
Economics, 9 (4), 363387.
Song, H., S.F. Witt and G. Li (2009), The Advanced Econometrics of Tourism Demand, NewYork: Routledge.
Syriopoulos, T.C. and M.T. Sinclair (1993), An econometric study of tourism demand: the AIDS model of US
and European tourism in Mediterranean countries, Applied Economics, 25 (12), 15411552.
Thomas, R.L. (1993), Introductory Econometrics: Theory and Applications, London: Longman.

DWYER 9781781001288 Part I.indd 89 25/07/2012 14:09

 90 Handbook of research methods in tourism

Turner, L.W. and S.F. Witt (2001), Forecasting tourism using univariate and multivariate structural time
series models, Tourism Economics, 7 (2), 135147.
Turner, L.W., N. Kulendran and H. Fernando (1997), Univariate modelling using periodic and non-periodic
analysis: inbound tourism to Japan, Australia and New Zealand compared, Tourism Economics, 3 (1),
3956.
Uysal, M. and M.S.E. Roubi (1999), Artificial neural networks versus multiple regression in tourism demand
analysis, Journal of Travel Research, 38 (2), 111118.
Vanegas, Sr M. (2008), Tourism demand response by residents of Latin American countries, International
Journal of Tourism Research, 11 (1), 1729.
Veloce, W. (2004), Forecasting inbound Canadian tourism: an evaluation of error corrections model fore-
casts, Tourism Economics, 10 (3), 262280.
Wang, C. (2004), Predicting tourism demand using fuzzy time series and hybrid grey theory, Tourism
Management, 25 (3), 367374.
Wang, C. and L. Hsu (2008), Constructing and applying an improved fuzzy time series model: taking the
tourism industry for example, Expert Systems with Applications, 34 (4), 27322738.
Wickens, M.R. and T.S. Breusch (1988), Dynamic specification, the long-run and the estimation of trans-
formed regression models, The Economic Journal, 98, 189205.
Witt, C.A. and S.F. Witt (1989), Measures of forecasting accuracy: turning point error versus size of error,
Tourism Management, 10 (3), 255260.
Witt, C.A. and S.F. Witt (1990), Appraising an econometric forecasting model, Journal of Travel Research,
18 (3), 3034.
Witt, S.F (1980), An abstract modeabstract (destination) node model of foreign holiday demand, Applied
Economics, 12 (2), 163180.
Witt, S.F. and C.A. Martin (1987a), Econometrics models for forecasting international tourism demand,
Journal of Travel Research, 25 (3), 2330.
Witt, S.F. and C.A. Martin (1987b), International tourism demand models: inclusion of marketing variables,
Tourism Management, 8 (1), 3340.
Witt, S.F. and C.A. Witt (1991), Tourism forecasting: error magnitude, direction of change error, and trend
change error, Journal of Travel Research, 30 (2), 2633.
Witt, S.F. and C.A. Witt (1992), Modeling and Forecasting Demand in Tourism, London: Academic Press Ltd.
Witt, S.F. and C.A. Witt (1995), Forecasting tourism demand: a review of empirical research, International
Journal of Forecasting, 11 (3), 447475.
Witt, S.F., G.D. Newbould and A.J. Watkins (1992), Forecasting domestic tourism demand: application to
Las Vegas arrivals data, Journal of Travel Research, 31 (1), 3641.
Witt, S.F., H. Song and P. Louvieris (2003), Statistical testing in forecasting model selection, Journal of Travel
Research, 42 (2), 151158.
Witt, S.F., C.A. Witt and N. Wilson (1994), Forecasting international tourist flows, Annals of Tourism
Research, 21 (3), 612628.
Wong, K.K.F., H. Song and K.S. Chon (2006), Bayesian models for tourism demand forecasting, Tourism
Management, 27 (5), 773780.
Wu, C. (2010), Econometric analysis of tourist expenditures, PhD Thesis, The Hong Kong Polytechnic
University.
Yu, G. and Z. Schwartz (2006), Forecasting short time-series tourism demand with artificial intelligence
models, Journal of Travel Research, 45 (2), 194203.

DWYER 9781781001288 Part I.indd 90 25/07/2012 14:09

 5 Structural equation modeling
Jenny (Jiyeon) Lee and Gerard Kyle

NATURE OF STRUCTURAL EQUATION MODELING AND ITS

EVOLUTION

Structural equation modeling (SEM) is a tool for analysing multivariate data that has
been long known in social sciences to be especially appropriate for theory testing (e.g.,
Bagozzi, 1980). It is also referred to as covariance structure analysis or covariance struc-
ture modeling. Structural equation models go beyond ordinary regression models to
incorporate multiple independent and dependent variables as well as latent constructs
that cluster the observed variables they are hypothesized to represent. They also provide
a way to test a specified set of relationships among observed and latent variables as a
whole, and allow theory testing even when experiments are not possible. As a result,
these methods have become ubiquitous in all the social and behavioral sciences (e.g.,
MacCallum and Austin, 2000).
With the advance of statistical software with graphical user interfaces, several user-
friendly SEM programs have become accessible to tourism researchers (Kline, 2005).
Of these, the most widely used programs are LISREL (Linear Structural Relationships),
AMOS (Analysis of Moment Structures), EQS (Equations), Mplus, and PLS (Partial
Least Squares). Each has idiosyncrasies and requirements for conducting analysis. These
programs enable users to either choose to write code in syntax or use a graphical editor to
generate the output by simply drawing the model. Over the past decade, these programs
have also developed in conjunction with a plethora of introductory texts that provide
insight on specific concepts, applications, and programming (e.g., LISREL (Byrne, 1998;
Jreskog and Srbom, 2004), AMOS (Arbuckle, 2009), EQS (Bentler, 2003), Mplus
(Muthn and Muthn, 19982010), PLS (Chin, 2001; Haenlein and Kaplan, 2004)).
Field-specific publications are also appearing in the tourism literature (e.g., Reisinger
and Turner, 1999).

BACKGROUND AND TYPES OF PROBLEMS ADDRESSED

Structural equation modeling combines both multiple regression and factor analysis
(Hair et al., 1998). It simultaneously tests an entire system of variables in a hypothesized
model to determine the degree to which the pattern of hypothesized relationships is con-
gruent with that observed in the data. In particular, it is has been shown to be especially
useful for testing theory where the constructs of interest are difficult to operationalize
using single indicators (Fornell and Larcker, 1981).
Testing latent variable structural models in most tourism studies involves a two-step
process (Anderson and Gerbing, 1988): (1) an examination of a measurement model
(Models A, B, and C below); and (2) the examination of a structural model (Models D

91

DWYER 9781781001288 Part I.indd 91 25/07/2012 11:06

 92 Handbook of research methods in tourism

and E; Bollen, 1989; Byrne, 1998). First, measurement models can be examined through
confirmatory factor analysis (CFA). Testing the measurement model effectively tests
hypotheses related to anticipated factor structures. In this procedure, construct valid-
ity (i.e., convergent and discriminant validity) can be verified by assessing the extent to
which the observed measures or manifest variables adequately represent each latent con-
struct. Alternately, testing of the structural model examines hypothesized relationships
among the latent constructs. This further provides insight on the predictive validity of
the latent constructs. Some researchers have also followed a four-step approach which
involves sequential testing of four nested models1 (Hayduk and Glaser, 2000; Mulaik
and Millsap, 2000): (1) a factor model (i.e., the least restricted model), (2) a confirma-
tory factor model (i.e., covariances among the factors), (3) a structural equation model
(i.e., specification of direct effects among the factors), and (4) a more constrained model
(i.e., additional constraints on the SEM model). It is intended to test nested models, thus
pinpointing problems within the SEM model such as conceptual inadequacy or measure-
ment misspecification (see Hayduk and Glaser (2000) for further discussion on the use of
the four-step procedure and its limitations).

STEPS IN PERFORMING SEM ANALYSIS

Model Specification

When SEM is used as a confirmatory technique, the model must be specified correctly,
based on the type of model the researcher is attempting to confirm. When building
the model, the researcher uses two different kinds of variables, namely exogenous and
endogenous variables. The distinction between these two types of variables is whether
the variable regresses on another variable. As in regression, the dependent variable
(DV) regresses on the independent variable (IV), meaning that the DV is being predicted
by the IV. In SEM terminology, endogenous variables regress onto exogenous vari-
ables. Exogenous variables can be recognized in a graphical version of the model as the
variables sending out arrowheads (see models depicted in Figure 5.1ae), denoting which
variable it is predicting. A variable that regresses on another variable is always an endog-
enous variable, even if this same variable is also used as a variable to be regressed upon.
Endogenous variables are recognized as the receivers of an arrowhead in the model.
For example, in Model A (Figure 5.1a), the manifest indicators (X1 through X4) are pre-
dicted by the latent exogenous construct (x1). Alternately, in Model D (Figure 5.1d), the
dependent latent construct (h1) is an exogenous variable within the measurement part of
the model (predicting Y1 through Y4), but is an endogenous variable within the structural
part of the model given that it is being predicted by x1.
When testing SEM models, often a set of plausible models are tested each reflecting
variants of guiding theory or theories. Not only must the researcher account for the theo-
retical reasons for building the model, but the researcher must also take into account the
number of data points (i.e., observed measures or manifest indicators) and the number
of parameters that the model must estimate to identify the model. An identified model
is a model where a specific parameter value uniquely identifies the model, and no other
equivalent formulation can be given by a different parameter value (Bollen, 1989). The

DWYER 9781781001288 Part I.indd 92 25/07/2012 11:06

 Structural equation modeling 93

parameter is the value of interest (all single-headed arrows in Figure 5.1ae), which
might be a regression coefficient between the exogenous and the endogenous variable or
the factor loading (regression coefficient between an indicator and its factor). If there are
fewer data points (i.e., manifest indicators) than the number of estimated parameters,
the resulting model is unidentified, since there are too few reference points to account
for all the variance in the model. The solution is to ensure enough data points have been
collected or constrain parameters within the model (or both).

Estimation of Free Parameters

Parameter estimation is accomplished by comparing the actual covariance matrix (sym-

bolized as S) representing the observed relationships between variables and the estimated
covariance matrix (symbolized as S) of the best fitting model (Brown, 2006). This is
obtained through numerical maximization of a fit criterion as provided by maximum

Model A
Latent
construct
1

x11 x21 x31 x41

Manifest
indicators
X1 X2 X3 X4

Error
1 2 3 4 terms

(a) Single factor first order measurement model

Model B
21

1 1

x11 x21 x31 x41 x52 x62 x72 x82

X1 X2 X3 X4 X5 X6 X7 X8

1 2 3 4 5 6 7 8

(b) Two factor first order measurement model

Figure 5.1 Measurement and structural models

DWYER 9781781001288 Part I.indd 93 25/07/2012 11:06

 94 Handbook of research methods in tourism

Model C

11 21
1 2

1 1

y11 y21 y31 y41 y52 y62 y72 y82

Y1 Y2 Y3 Y4 Y5 Y6 Y7 Y8

1 2 3 4 5 6 7 8

(c) Second order measurement model

Model D 1

11
1 1

x11 x21 x31 x41 y11 y21 y31 y41

X1 X2 X3 X4 Y1 Y2 Y3 Y4

1 2 3 4 1 2 3 4

(d) Structural model

Model E 1 2

11 21
1 1 1

x11 x21 x31 x41 y11 y21 y31 y41 y52 y62 y72 y82

X1 X2 X3 X4 Y1 Y2 Y3 Y4 Y1 Y2 Y3 Y4

1 2 3 4 1 2 3 4 5 6 7 8

(e) Latent variable path model

Figure 5.1 (continued)

DWYER 9781781001288 Part I.indd 94 25/07/2012 11:06

 Structural equation modeling 95

likelihood estimation (default in most SEM programs), weighted least squares, or

asymptotically distribution-free methods. This is often accomplished by using a special-
ized SEM software.

Assessment of Fit

Assessment of fit is a basic task in SEM modeling; forming the basis for accepting or
rejecting hypotheses and more often accepting one competing model over another.
The output of SEM programs includes matrices of the estimated relationships between
variables in the model. Assessment of fit essentially calculates how similar the predicted
data (reflected in the structure of the hypothesized model) are to matrices containing
the relationships in the actual data. Formal statistical tests and fit indices have been
developed for these purposes. Individual parameters of the model can also be examined
within the estimated model in order to see how well the proposed model fits the driving
theory.
As with all statistical hypothesis tests, SEM model tests are based on the assumption
that the correct and complete relevant data have been modeled. In the SEM literature,
discussion of fit has led to a variety of different recommendations on the precise applica-
tion of the various fit indices and hypothesis tests. Measures of fit differ in several ways.
Traditional approaches to modeling start from a null hypothesis, rewarding more par-
simonious models (i.e., those with fewer free parameters), to others such as the Akaike
information criterion (AIC) that focuses on how little the fitted values deviate from a
saturated model (i.e., how well they reproduce the measured values), taking into account
the number of free parameters used. Because different measures of fit capture different
elements of the fit of the model, it is appropriate to report a selection of different fit
measures.
Some of the more commonly used measures of fit include:

Chi-square (c2): A fundamental measure of fit used in the calculation of many

other fit measures. Conceptually, it is a function of the sample size and the differ-
ence between the observed covariance matrix and the model covariance matrix.
Chi-square, however, is seldom used in applied research as a sole index of model fit
owing to its sensitivity to sample size (Brown, 2006).
Akaike information criterion (AIC; Akaike, 1987): A test of relative model fit. The
preferred model is the one with the lowest AIC value.
Root mean square error of approximation (RMSEA; Steiger and Lind, 1980):
Takes into account the error of approximation in the population and asks the
question, How well would the model, with unknown but optimally chosen
parameter values, fit the population covariance matrix if it were available?
(Browne and Cudeck, 1993, pp. 137138). Good models are considered to have a
RMSEA values of 0.05 or less, acceptable models 0.08 or less, bad models of 0.1 or
higher considered unacceptable (Browne and Cudeck, 1993).
Standardized root mean residual (SRMR): The SRMR is a popular absolute fit
indicator, meaning that it does not take into account other aspects such as fit in
relation to more restricted solutions. A value of less than 0.08 is generally accept-
able (Hu and Bentler, 1999).

DWYER 9781781001288 Part I.indd 95 25/07/2012 11:07

 96 Handbook of research methods in tourism

Comparative fit index (CFI; Bentler, 1990): In examining baseline comparisons,

the CFI depends, in large part, on the average size of the correlations in the data.
If the average correlation between variables is not high, then the CFI will not be
very high. The CFI also rewards model parsimony. Values range from 0 to 1 with
values closer to 1 indicating good model fit (Hu and Bentler, 1999).
Non-normed fit index (NNFI): Also referred to as the TuckerLewis Index, the
NNFI is similar to the NFI. However, the index is lower and hence the model
is regarded as less acceptable if the model is complex. According to Marsh et al.
(1988), the NNFI is relatively independent of sample size. The NNFI is usually
lower than is the GFI but over 0.95 is considered acceptable (e.g., Hu and Bentler,
1999).
Parsimonious normed fit index (PNFI; Mulaik et al., 1989): The PNFI takes into
account model complexity in its assessment of model fit. There are now rigid rules
for interpreting PNFI values although the closer their value is to 1, the stronger
(and more parsimonious) the model.

With regard to which indices should be reported, it is not necessary or realistic to

include every index from the program output given that they may contain redundant
information. Given the plethora of fit indices, it becomes a temptation to choose those
fit indices that indicate the best fit. This should be avoided as it is essentially sweeping
important information under the carpet. In a review by McDonald and Ho (2002), it
was found that the most commonly reported fit indices are the CFI, GFI, NFI, and the
NNFI. When deciding what indices to report, going by what is most frequently used is
not necessarily good practice as some of these statistics (such as the GFI) are often relied
on purely for historical reasons, rather than for their sophistication. While there are no
golden rules for assessment of model fit, reporting a variety of indices is necessary
(Crowley and Fan, 1997) because different indices reflect different aspects of model fit.
Although the chi-square model has many problems associated with it, it is still essential
that this statistic, along with its degrees of freedom and associated p-value, be reported
(Hayduk et al., 2007; Kline, 2005). Threshold levels have been assessed by Hu and
Bentler (1999), who suggested a two-index presentation format. This includes reporting
the SRMR with the NNFI, the RMSEA, or the CFI. Alternately, Kline (2005) advocates
the use of the chi-square test, the RMSEA, the CFI, and the SRMR. Based on these
authors guidelines, most support can be found for reporting the chi-square statistic, its
degrees of freedom, and p-value, the RMSEA and its associated confidence interval, the
SRMR, the CFI, and one parsimony fit index such as the PNFI. These indices have been
chosen over other indices as they have been found to be the most insensitive to sample
size, model misspecification, and parameter estimates.

Model Modification

A hypothesized model may need to be modified in order to improve its fit to the data,
thereby estimating the most likely relationships between variables. Many programs
provide modification indices (or Lagrange multiplier indices) which report the improve-
ment in fit resulting from estimating additional parameters within the model. It is
important to note, however, that post-hoc model modifications reflect departures from

DWYER 9781781001288 Part I.indd 96 25/07/2012 11:07

 Structural equation modeling 97

the initial theory and hypotheses guiding model construction. As such, they must be
acknowledged accordingly and justified on conceptual or theoretical grounds. In the
absence of substantive justification, post-hoc model modification can reflect a trou-
blesome reaction to anomalies within the researchers data that is unlikely replicable.
Consequently, post-hoc modifications to theoretically derived models ought to be con-
sidered with caution.

Sample Size

Where the proposed SEM is the basis for a research hypothesis, ad-hoc rules of thumb
requiring the choosing of ten observations per indicator in setting a lower bound for
the adequacy of sample sizes have been widely used since their original articulation by
Nunnally (1967). Being linear in model constructs, these are easy to compute, but have
been found to result in sample sizes that are too small. Complexities which increased
information demands in structural model estimation increase with the number of poten-
tial combinations of latent variables; while the information supplied for estimation
increases with the number of measured parameters times the number of observations
in the sample size both are non-linear. Sample size in SEM can be computed through
two methods: the first as a function of the ratio of indicator variables to latent variables,
and the second as a function of minimum effect, power and significance. Software and
methods for computing both have been developed by Westland (2010).

Interpretation and Communication

Following the establishment of a model adequately fitting the data, the researcher then
moves to the interpretation of model parameters so that claims about the constructs can
be made. Caution should always be taken when making claims of causality even when
experimentation or time-ordered studies have been done. The term causal model must
be understood to mean a model that conveys causal assumptions, not necessarily
a model that produces validated causal conclusions. Collecting data at multiple time
points and using an experimental or quasi-experimental design can help rule out certain
rival hypotheses but even a randomized experiment cannot rule out all such threats to
causal inference. Good fit by a model consistent with one causal hypothesis invariably
entails equally good fit by another model consistent with an opposing causal hypothesis.
No research design, no matter how clever, can help distinguish such rival hypotheses,
save for interventional experiments. As in any science, subsequent replication and
perhaps modification will proceed from the initial finding.

APPLICATIONS IN TOURISM RESEARCH

Structural equation modelling has become one of the most commonly used data analysis
techniques among tourism researchers over the past two decades. In a review of the 1790
articles published in the major 12 tourism and hospitality journals, Palmer et al. (2005)
found that the number of articles using SEM had increased from 6 in 1998 to 21 in 2002.
As of August 2011, our review of the three major tourism journals (i.e., Annals of Tourism

DWYER 9781781001288 Part I.indd 97 25/07/2012 11:07

 98 Handbook of research methods in tourism

Research, Tourism Management, and Journal of Travel Research) identified over 300
published articles where the authors had chosen SEM as their analytical tool for scale
development and the investigation of the causal structures among constructs. In tourism
research, the technique has primarily been limited to: (1) the use of confirmatory (and
sometimes combined with exploratory) factor analysis to determine scale structure, psy-
chometric properties, and for scale purification; (2) testing causal relationships among
latent (and sometimes manifest) variables; and (3) testing for measurement and structural
model invariance across groups. Applications presented in this section provide specific
examples of how selected tourism publications address research questions using SEM.

CONFIRMATORY FACTOR ANALYSIS

Testing First-order CFA Models

Confirmatory factor analysis has been used in tourism research to develop and refine
multi-item measures of latent constructs. Most tourism researchers informally test
hypotheses about the direct relations between a theoretical latent construct and its
observed variable in first-order CFA models. These analyses evaluate the extent to
which the observed measures in a hypothesized model derived from past evidence and
theory adequately represent their underlying constructs. Tourism researchers have also
used CFA to cross-validate factorial structures across study contexts and populations. It
further allows for assessing the validity of a measurement scale of latent constructs (i.e.,
convergent and discriminant validity). This procedure usually precedes an examination
of the fit of structural relationships (Anderson and Gerbing, 1988).
Tourism researchers have examined the plausibility of multidimensional struc-
tures for a number of constructs. These include constraints (Hung and Petrick, 2010),
novelty (Petrick, 2002), perceived service quality (Albacete-Sez et al., 2007; Vogt and
Fesenmaier, 1995), leisure tourists involvement, post-visit evaluation and behaviors
(Gursoy and Gavcar, 2003; William and Soutar, 2009), and local residents attitudes
toward and perceptions of gaming as a tourism development strategy (Chen and Hsu,
2001; Kang et al., 1996). Some studies (e.g., Albacete-Sez et al., 2007; Chen and Hsu,
2001; Hung and Petrick, 2010; Kang et al., 1996) have employed exploratory factor
analysis (EFA) on selected scales that have little or no history of use in the tourism
contexts to first explore the underlying structure of a construct of interest. Using CFA,
they then confirm construct validity of the multidimensional scale emerging from EFA.
For example, Hung and Petrick (2010) first developed and validated a measurement
scale for constraints to cruising among 897 online panel members. Following Churchills
(1979) scale development guidelines, their measures were developed by first conducting
an extensive review of the literature and interviews with prospective cruise passengers.
After pre-testing the emerging scale, the authors used EFA of the pilot data for scale
purification. The final scale emerging from the EFA was then validated in CFA using an
online panel. They observed a four-dimensional structure within constraints to cruising,
including intrapersonal, interpersonal, structural, and not an option.
Similarly, Albacete-Sez et al. (2007) examined the service quality dimensions derived
from EFA within the context of Spanish rural accommodations using CFA. They found

DWYER 9781781001288 Part I.indd 98 25/07/2012 11:07

 Structural equation modeling 99

that the overall fit of a five-dimensional service quality model was considered most
adequate (i.e., personnel response, complementary offer, tourist relations, tangible ele-
ments, and empathy). In addition, Chen and Hsu (2001) and Kang et al. (1996) have
confirmed the validity of the attitude scale derived from the EFA. The former study
indicated that five dimensions of residents perceptions of the riverboat gaming effects
(i.e., free of crime, economic effects, community image, community activities, and public
services) were valid, with each dimension having different effects on support for gaming
legalization and future casino expansion. The latter study examined the validity of the
two models (i.e., the a priori model and the model derived from EFA) conceptualizing
resident perceptions of limited-stake casino development across tourism communities.
The CFA results indicated that a unidimensional model (i.e., evaluation) was applicable
to both the gambling and non-gambling town data whereas the two-factor model (i.e.,
attitudes and benefits) was applicable only to non-gambling town data.
Alternately, other researchers have primarily used CFA to test scales with an estab-
lished structure in varied tourism contexts (e.g., Gursoy and Gavcar, 2003; Petrick,
2002; Vogt and Fesenmaier, 1995; William and Soutar, 2009). Using an existing novelty
scale, Petrick (2002) investigated its reliability and validity using CFA in a golf vacation
context. His results indicated that the four dimensions of novelty (i.e., thrill, change
from routine, boredom-alleviation, and surprise) are valid and reliable when used with
golf vacationers. Vogt and Fesenmaier (1995) further refined Parasuraman et al.s (1988)
SERVQUAL scale by evaluating the factorial structure of service quality between tour-
ists and retailers in a tourism destination. While the four-dimension service quality
model (i.e., reliability, responsiveness, assurance, and access) was found to be valid for
both groups, its factor structure was different between tourists and retailers. In a similar
fashion, William and Soutar (2009) verified the construct validity of five dimensional
values (i.e., functional, emotional, social, novelty value, and value for money), satisfac-
tion, and behavioral intentions in an adventure tourism context. Gursoy and Gavcar
(2003) also examined the applicability of the consumer involvement profile (CIP) among
European leisure tourists. Their study supported the multidimensional nature of tour-
ists involvement but revealed a different factor structure (i.e., interest/pleasure, risk
probability, and risk importance) in studies with consumer goods and studies in other
recreation, leisure and tourism contexts.

Testing Hierarchical CFA Models

Some tourism studies have tested the hierarchical relations between constructs in CFA
models (see Model C in Figure 5.1c). In much the same way, first order measurement
models are developed around the hypothesis that first order latent factors account for the
variation within their manifest indicators, in hierarchical measurement models, higher-
order factors are hypothesized to account for the variation in lower-order factors. In
other words, higher-order factors are measured indirectly through the indicators of the
lower-order factors because they have no manifest variables (Kline, 2005).
Several examples of analyses of hierarchical measurement models in tourism are
provided in the literature. These includes the studies by First, Funk and Bruun (2007),
who analysed the second-order model of sport tourists motives; Konecnik and Gartner
(2007), who tested a destination customer-based brand equity (CBBETD) higher-order

DWYER 9781781001288 Part I.indd 99 25/07/2012 11:07

 100 Handbook of research methods in tourism

CFA model; and Caro and Garca (2008), who analysed the third-order model of service
quality for the travel agency industry.
First, Funk and Bruun (2007) tested a measurement model of a six first-order latent
variables relevant to sport tourists motives (i.e., running involvement, strength of
motivation, attitude toward the hosting destination, learning interest, cultural experi-
ence, and cultural learning) and their respective observed variables. They then exam-
ined the hierarchical CFA model specifying that (1) socio-psychological motivation in
the second-order factors loaded on the first three first-order factors (i.e., involvement,
motivation strength, and attitude), (2) culture-education motivation loaded on the last
three factors (i.e., attitude, learning interest, and cultural experience), and (3) cultural
learning predicted cultural experience. Various fit indices suggested that a second-order
five-factor structure of travel motives most appropriately described a running events
participants who traveled internationally. Konecnik and Gartner (2007) also examined
a higher-order CFA model of customer-based brand equity for a tourism destination
(CBBETD) using telephone interviews with potential tourists from different countries.
In their study, four first-order factors (i.e., awareness, destination image, quality, and
loyalty) were reported to be sub-dimensions of a single higher-order factor of CBBETD.
Last, Caro and Garca (2008) postulated a third-order structure for a conceptualization
of travel agency service quality. In their measurement model, seven first order factors
were considered sub-dimensions of three second order factors which were hypothesized
to load onto a single third-order factor of travel agency service quality.

Testing Structural Models

Another common use of SEM in tourism research involves the investigation of structural
relations between and/or among latent constructs and their dimensions. A broad variety
of models grounded in theory and empirical research have been tested using path analy-
sis (see Models D and E in Figure 5.1de) across various study contexts and populations.
The simultaneous testing of structural models among groups has also recently gained
interest in the literature. Specifically, the procedure is used to assess whether certain
paths in a specified causal structure are equivalent across populations. The discussion
that follows reviews some of these applications beginning with work modeling associa-
tions among single-item measures, latent and variables measures, and concluding with
multiple-group applications.

Testing Path Models Using Single-item Measures

As an alternative approach to multiple regression analysis, tourism researchers have

used path analysis (PA) to simultaneously test direct and indirect relationships among
the latent constructs and, in some instances, single-indicator models. Regardless of the
approach, when testing path models, researchers typically aim to understand: (1) which
independent variables (IVs) predict the hypothesized dependent variables (DVs), (2) how
well a combination of IVs predict hypothesized DVs and account for their variance, and
(3) the nature of each IVs influence on the DVs (i.e., effect valence).
For example, in a study of 169 countries competitiveness, Mazanec et al. (2007) used
a Competitiveness Monitor (CM) scale proposed by the World Travel and Tourism

DWYER 9781781001288 Part I.indd 100 25/07/2012 11:07

 Structural equation modeling 101

Council (WTTC). They found that both social and cultural indicators of the eight
measures explained destination competitiveness effectively. Similarly, Vargas-Snchez
et al. (2011) examined the effectiveness of several hypothesized antecedents of resi-
dents attitudes toward tourism development in a Spanish province. They observed that
residents attitude was positively, although weakly, predicted by satisfaction with public
services, their perception of personal benefits, and tourists behavior. They also found
that attitude was negatively influenced by the density of tourists within the destination
(i.e., perceived crowding) and level of perceived tourism development. In another study
employing single-item measures, Um et al. (2006) examined tourists revisit intentions
to Hong Kong. They found that perceived attractiveness of the destination, rather than
overall satisfaction, was a stronger predictor of their respondents revisit intentions.
Likewise, Oh and Hsu (2001) adopted the theory of reasoned action (Fishbein and Azjen,
1975) to develop a model of gambling behavior among gamblers in Iowa. Their results
indicated that gamblers future participation in gambling had a direct, positive relation-
ship with past behaviors and their behavioral intentions for future participation. Last,
Beerli and Martn (2004) employed PA to examine the differences in the determinants
of the image of a Spanish island among first-time and repeat tourists. Using the factor
scores emerging from an EFA, they hypothesized a path model specifying the relation-
ships among motivations, tourist experience, and perceived image. They reported that
both models for first-time and repeat tourists showed an acceptable fit to the data. They
further reported that tourist motivations related to relaxation, knowledge and past travel
experience influenced the image of the sun-based destination for both groups.

Testing Path Models Using Latent Variables

It is more common in tourism research implementing SEM to test structural models with
latent variables measured using multiple indicators. Typically, structural models are
tested following the establishment of valid and reliable measurement models through the
use of CFA (Byrne, 1998). Much of this work has been driven by a desire to understand:
(1) resident support for tourism development (Dyer et al., 2007; Gursoy and Kendall,
2006; Gursoy and Rutherford, 2004; Gursoy et al., 2002; Kang et al., 2008; Lindberg
and Johnson, 1997; Nicholas et al., 2009; Nunkoo and Ramkissoon, 2011; Yoon et al.,
2001); and (2) tourist behavior (Bign et al., 2001; Bign et al., 2005a; Bign et al., 2005b;
Bign Alcaiz et al., 2009; Connell and Meyer, 2009; del Bosque and San Martn, 2008;
Denstadli and Jacobsen, 2011; Feng and Morrison, 2007; Gross and Brown, 2008; Han
et al., 2010; Huang et al., 2010; Huang and Hsu, 2009; Hwang et al., 2005; Kaplanidou
and Vogt, 2006; Kim et al., 2011; Kolar and abkar, 2010; Lee et al., 2005; Lee et al.,
2007a; Lee et al., 2008; Martn-Ruiz et al., 2010; Nadeau et al., 2008; Qu and Lee,
2011; Reisinger and Turner, 2002; Sparks, 2007; Stokburger-Sauer, 2011; Wang and
Fesenmaier, 2004; Wu et al., 2008; Yoon and Uysal, 2005; Yuan and Jang, 2008; Yuksel
et al., 2010; abkar et al., 2010). These studies have employed multiple indicators of
various constructs to better capture the latent phenomena. With multiple indicators,
researchers are better able to assess constructs of interest. The breadth of this work has
illustrated that factors influencing tourist behavior vary along multiple dimensions.
In particular, work has shown that tourists intention to revisit specific destinations
is positively influenced by facets of (1) destination image (Bign et al., 2001; Bign

DWYER 9781781001288 Part I.indd 101 25/07/2012 11:07

 102 Handbook of research methods in tourism

Alcaiz et al., 2009; del Bosque and San Martn, 2008; Lee et al., 2008); (2) the attributes
depicted through website design (Kaplanidou and Vogt, 2006); (3) satisfaction stemming
from previous visits (Bign et al., 2001; Bign et al., 2005; Connell and Meyer, 2009; del
Bosque and San Martn, 2008; Denstadli and Jacobsen, 2011; Huang and Hsu, 2009;
Lee et al., 2005; Martn-Ruiz et al., 2010; Nadeau et al., 2008; Yoon and Uysal, 2005;
Yuan and Jang, 2008; Yuksel et al., 2010; abkar et al., 2010) and perceived perform-
ance quality (Baker and Crompton, 2000; Bign et al., 2001; abkar et al., 2010); (4) the
emotions elicited through travel experience (Bign et al., 2005); and (5) attitude toward
destination revisit (Huang and Hsu, 2009; Sparks, 2007); (6) awareness of the destination
and its attributes (Stokburger-Sauer, 2011; Yuan and Jang, 2008); and (7) travel motiva-
tions (Kolar and abkar, 2010; Yoon and Uysal, 2005).
The testing of path models in SEM also provides insight on the effects of hypothesized
processes reflected in path models. These models typically reflect causal sequences
derived from theory. Structural equation models speak directly to the tenability of these
models and the associated theory. While they do not establish causality, they do
provide insight on the tenability of the theory or theories from which they are derived.
For example, several studies have explored the mediating effect of satisfaction on the
relationship between the perceived quality and behavioral intentions in the context of
cultural festivals (Baker and Crompton, 2000; Lee, S. et al., 2007b; Yuan and Jang,
2008). In these investigations, the provision of service quality and its impact on visi-
tors intention to return is best understood through the effect of satisfaction. Baker and
Crompton (2000) found that satisfaction mediated the effect of quality on behavioral
intentions when they measured quality as the performance of the service attributes
within a community festival context. Alternately, when operationalizing quality as a
discrepancy measure (i.e., perceptions-minus-expectations), satisfaction was an insignifi-
cant mediator of the qualitybehavioral intentions relationship.

Multi-group Invariance Testing

Another application of SEM involves the testing of both the measurement and structural
model simultaneously among specified groups referred to as multigroup invariance
testing. Testing of the measurement model using the multigroup procedure involves
testing hypotheses relating to equivalence among groups in terms of: (1) the factor
structure, pattern of factor loadings, and factor variances/covariances; (2) the pattern
of structural relations among constructs across populations; (3) the constructs latent
means; and (4) the cross-validation of an instrument to establish its psychometric prop-
erties across contexts and populations.
Briefly, multigroup analysis tests the null hypothesis that covariance structures across
groups are equivalent (Byrne, 1998). Rejection of the null hypothesis suggests inequal-
ity among the groups. Subsequent procedures directed at identifying the source of the
inequality involve testing a series of increasingly restrictive hypotheses that imposes
equality constraints on sets of parameters beginning with elements of the measurement
model and continuing through the models structural parameters.
Researchers using this procedure have typically explored the moderating effect of
selected variables. These have included analyses of gender (Okazaki and Hirose, 2009),
tourist knowledge (Wong and Yeh, 2009), place attachment (Chung et al., 2011), and

DWYER 9781781001288 Part I.indd 102 25/07/2012 11:07

 Structural equation modeling 103

distance to tourist attractions (Jurowski and Gursoy, 2004). For example, Okazaki
and Hirose (2009) simultaneously tested a path model among men and women relating
Japanese consumers media use in searching for travel information. While they observed
no variation between the groups with regard to the linear relationships tested in their
path model, they did observe variation among men and women in terms of the latent
means for satisfaction with, attitude toward, and the habitual usage of mobile internet.
Women were found to be generally more positive about the technology compared to
their male counterparts. Using tourists who had taken an overseas group package tour,
Wong and Yeh (2009) also tested the invariance of a structural model to assess whether
tourist knowledge moderated the relationship between risk perception and hesitation
in travel decision making. In their analyses, they defined their two groups in terms of
tourist knowledge (i.e., low/high). They observed that, for the high knowledge group,
a higher level of knowledge decreased their level of risk perception and tendency to baulk
at purchasing a vacation.

Cross-validation of Structural Models

Another application of SEM in tourism involves the cross-validation of hypothesized

models. The strength of this approach lies in its ability to test relationships across
populations and contexts and further support or reject hypotheses grounded in specific
theory. Common in tourism research is the use of cross-sectional data. A limitation of
this form of data, however, is its ability to affirm or disconfirm theory and generalize to
broader populations. Testing among multiple (sub)populations provides a more rigor-
ous test of theory and strengthens the external validity of related findings.
The most common approach in tourism research involves model development using
a target sample and cross-validating using a second sample. The procedure investigates
whether a specified model that has established validity in one context is replicable in
another independent context. One less stringent approach adopted by tourism research-
ers has been to randomly split the sample into two groups. The first group is used to
establish the structure of the model in question and the second group is then used to
validate the emergent model. This split-half approach was employed by Oh et al. (2009),
who hypothesized that three service performance phases (i.e., sales, pre-event, and event)
would influence meeting planners overall satisfaction. Their initial model was validated
using the second sample offering strong support for their hypotheses.
Furthermore, tourism studies have performed comparative model testing to evaluate
the acceptability of hypothesized models. Some have compared the fit of the proposed
model with alternative nested models by imposing increasingly restrictive equality con-
straints (e.g., Kim and Littrell, 1999). The main goal of this particular procedure is to
find a parsimonious model. In a study examining the superiority of three nested models
relating to predictors of souvenir purchase intentions, Kim and Littrell (1999) demon-
strated that a model where tourist attitudes, shaped through certain tourism experiences
(i.e., recreational and ethnic tourism) and their associated values positively influenced
tourist souvenir purchase intentions and provided a significantly better fit to the data
compared to the other two models. Other researchers have compared fit indices and
parameter estimates of two or more non-nested competing models in order to obtain a
superior model (Bign et al., 2005a; Boo et al., 2009; Lee and Back, 2008). For instance,

DWYER 9781781001288 Part I.indd 103 25/07/2012 11:07

 104 Handbook of research methods in tourism

Bign et al. (2005a) tested two non-nested models explaining how visitors cognitive
evaluation of a theme park experience (i.e., disconfirmation) and their emotions influ-
ence their satisfaction and revisit intentions. Selected fit indices allowed them to deter-
mine model superiority and conclude the importance of emotions in response to a visit
as opposed to the role of pre-visit emotional states. Similarly, Lee and Back (2008) com-
pared the explanatory power of the three non-nested models to better understand asso-
ciation meeting attendees participation intentions. Their three competing models were
derived from the theory of reasoned action (TRA), theory of planned behavior (TPB),
and empirical evidence (meeting participation model, MPM). Their findings illustrated
that the superiority of the MPM over other competing models.

Test for a Latent Growth Model

As an extension of a structural regression model testing, the analysis of a latent growth

model (LGM) in SEM involves estimation of models with mean structures in addition
to sample covariances for longitudinal data. Specifically, it can be applied to repeated
measures data to estimate change in individual differences over time instead of treat-
ing differences in the growth trajectories of individual cases as error variance (Kline,
2005).
Latent growth modeling is analysed in the two-step procedure in order to single out
potential sources of poor model fit (see Kline, 2005, pp. 272287). The first step involves
the analysis of the baseline model that evaluates its change using the repeated measures
variables. The baseline model is structured such that: (1) each time measurement is an
indicator of two latent growth factors (i.e., initial status and linear change); (2) these
factors covary; (3) the exogenous factors are directly influenced by a mean structure; and
(4) the measurement errors of the indicators covary. In model construction, the loadings
of all indicators on the initial status factor are fixed to 1.0 while loadings on the linear
change factor are fixed to be constant to correspond to measurement occasions. Given
an acceptable change model, the subsequent procedure adds predictors to the baseline
model predicting change over time.
The study by Assaker et al. (2011) provides one of the few examples of testing LGM in
SEM within the tourism literature. They investigated the effects of novelty seeking, des-
tination image and overall satisfaction levels on revisit intentions to sun, sea, and sand
destinations over four occasions using 450 European tourists. Their findings illustrated
that satisfaction positively influenced immediate revisit intentions and that its effect on
intention decreases over time. It was also found that a higher level of novelty seeking
diminished immediate revisit intentions but increased future revisits.

ADVANTAGES AND LIMITATIONS OF STRUCTURAL

EQUATION MODELING

Advantages of SEM

Structural equation modeling has several advantages over other analytical method-
ologies for hypothesis testing. Given that the analysis can be applied to both non-

DWYER 9781781001288 Part I.indd 104 25/07/2012 11:07

 Structural equation modeling 105

experimental and experimental data, it provides researchers with (1) modeling flexibility;
(2) an array of analytical capabilities (i.e., multi-group invariance testing); and (3) the
ability to account for measurement error.
First, as noted earlier, CFA is used as a precursor to testing the structural model that
specifies relationships among latent constructs (Brown, 2006). Because the CFA model is
specified a priori, it assists researchers with obtaining a parsimonious model by offering
guidance on decisions relating to the number of factors, the pattern of factor loadings,
and the relationships among error variances. The acceptability of the specified model is
assessed by evaluating the overall goodness of fit, statistical significance, and strength of
parameter estimates. It is also a powerful technique for assessing construct validity of the
measurement scale.
Another advantage of SEM is the ability to unveil complicated construct relation-
ships. It enables researchers to test all variables of a hypothesized model simultaneously
(Byrne, 1998). Because it estimates a series of multiple regression equations in a single
comprehensive method, it increases explanatory power for model testing. For example,
in a study testing a consumer commitment model in an e-commerce travel context,
Nusair and Hua (2010) found that SEM revealed a greater number of statistically sig-
nificant relationships in the model compared to those in ordinary least squares (OLS)
regression analyses. They concluded that SEM is more effective than multiple regression
analyses for determining the superiority of theoretically derived models.
Furthermore, SEM is a powerful analytical tool for evaluating the equivalence of
measurement and structural model across population subgroups (Brown, 2006). It
allows for detecting the potential source of group differences in the factor solution and
latent causal structure (e.g., the number of factors, patterns of factor loadings, presence
of error covariances, and the means of the latent dimensions). Particularly, measurement
invariance tests can be used in scale development, as it evaluates the generalizability of
an instrument measuring specified constructs among different groups.
Last, SEM more effectively accounts for measurement error in the estimation process
(Hair et al., 1998). Unlike SEM, OLS approaches assume that all of the observed
indicators measure the phenomena of interest perfectly without error (Brown, 2006).
Consequently, it is a bold assumption that estimates derived from OLS methods are
immune from measurement error. This assumption is rarely applicable to quantita-
tive tourism research, which relies heavily on survey methods to measure variables.
Traditional methods (i.e., ANOVA and MANOVA) that assume equal and independent
error variances are also inadequate for repeated measures data (Kline, 2005). In this
regard, to many tourism researchers, it is more appealing to estimate the relationships
among variables while adjusting for measurement error.

Limitations of SEM

SEM requires a priori specifications, which implies that model development and modifi-
cation should be directed by theory and empirical evidence (Kline, 2005). Therefore, this
technique is inappropriate for exploratory research where the measurement structure is
not well defined or where guiding theory and empirical evidence underlying patterns of
construct relationships is not well-established (Brannick, 1995; Hurley et al., 1997). In
such cases, alternative techniques for data analysis such as EFA and regression analysis

DWYER 9781781001288 Part I.indd 105 25/07/2012 11:07

 106 Handbook of research methods in tourism

may be considered (Hurley et al., 1997). Due to the confirmatory nature in SEM, post-
hoc model fitting is further restricted without providing appropriate theoretical, con-
ceptual or empirical justification. Modification of an initially hypothesized model solely
based on model fit and misspecification information is considered exploratory in nature,
which should be avoided in SEM.
Another limitation of the use of SEM and the use of the maximum likelihood estima-
tion is associated with its inherent assumption of multivariate normality and asymptotic
theory that requires sufficient sample sizes (Tomarken and Waller, 2005). Structural
equation modeling is considered a valuable technique for testing hypotheses that imply
causality in the non-experimental contexts. Yet, it has been underutilized in studies using
experimental designs where it is common practice to randomly assign participants to
treatments and directly manipulate independent variables, which has the potential to
violate the multivariate normality assumptions. The inclusion of categorical variables
denoting group status in experimental designs could also raise some concerns about the
normality. Most SEM software, however, will enable researchers to address some issues
of categorical or ordinal measurement through the use of polychoric or polyserial (and
accompanying asymptotic) covariance matrices along with the weighted least squares
estimator.
Finally, as indicated earlier, most applications of SEM require large samples.
Although there is little consensus among researchers on precise sample size requirements,
consideration ought to include the complexity of the model and the type of estimation
algorithm (Kline, 2005). A small sample size increases the likelihood of encountering a
technical problem in the analysis (e.g., bias of parameter estimates, inadmissible esti-
mates, Heywood cases) and the limited statistical power (Tomarken and Waller, 2005).
Generally, larger sample sizes produce smaller standard errors, narrower confidence
intervals and, ultimately, more reliable parameter estimates.

FURTHER DEVELOPMENTS WITHIN SEM

In this final section, we discuss potential applications of SEM that have received little
attention in the tourism literature. These applications have the potential to both increase
the scientific rigor of tourism research and address a broader range of research questions.

Formative Versus Reflective Measurement

Most tourism researchers that have employed SEM have used multi-item measures of
their latent constructs. When testing their measurement model, they assume that their
indicators have been drawn from a hypothetical pool of items and that the addition or
deletion of items has little bearing of the meanings of their latent construct. The covaria-
tion among items can also be accounted for through the latent construct. That is, they
have all been developed with a narrow definition of a specific construct in mind. This
approach yields a clean factor solution with all items loading strongly on their hypothe-
sized construct and weakly on other constructs in the model. These measures are referred
to as reflective indicators. These assumptions, however, have recently been questioned
in the consumer and business management literature where researchers indicators are

DWYER 9781781001288 Part I.indd 106 25/07/2012 11:07

 Structural equation modeling 107

often developed to capture the subtleties of a specific service or business contexts (for
review, see Bollen and Lennox, 1991; Diamantopoulos and Winklhofer, 2001; Jarvis
et al., 2003). For example, measures of service quality are often developed to capture
the breadth of factors that underlie quality for the specific service type. Consequently,
attribute indicators might not necessarily covary with one another even though they
have been developed to reflect the same latent concept. Importantly, however, the
removal of any specific item from the hypothesized construct may have the potential to
drastically alter the meaning of that construct. These types of indicators are referred to
as formative measures.
The reviews by the authors cited above outline guidelines for distinguishing formative
from reflective indicators and the procedures for testing these models within SEM. For
tourism researchers, it is imperative that these distinctions be understood and established
a priori. Model misspecification has important implications for Type I error (e.g., con-
cluding a statistically significant relationship when, in fact, it does not exist) and Type
II error (e.g., concluding a statistically non-significant relationship when, in fact, it does
exist) and, ultimately, study conclusions (Jarvis et al., 2003). Researchers should note,
however, that the issue of formative measurement remains a hot topic within the SEM
literature. Concerns over construct validity, reliability, and model identification remain
problematic (for discussion, see Cenfetelli and Bassellier, 2009; Diamantopoulos and
Siguaw, 2006). While simulation and testing in applied contexts is helping to resolve
these issues and allay concern, there remain skeptics (Bagozzi, 2007; Howell et al., 2007;
Wilcox et al., 2008).

Moderator Effects

One direction for future inquiry is to test hypotheses relating to interaction effects
among latent variables. While invariance testing allows for the examination of nominal
or ordinal indicator moderation, it does not test for the effect of interactions among
latent variables (e.g., Okazaki and Hirose, 2009). Their limited use is likely associated
with the potential problems and complexities of specifying and estimating SEM models
with latent variable interactions (Tomarken and Waller, 2005). Some advantages over
multiple regression analyses (i.e., correction of bias estimates and power adjustment,
Tomarken and Waller, 2005) and the recent refinement of the technique (e.g., Kline and
Moosbrugger, 2000) will likely lead to more extensive use of SEM in testing for interac-
tions in tourism research.

Latent Growth Models

Most recently, latent growth modeling for the analysis of repeated measures data has
become a viable alternative to repeated measures ANOVA (Kline, 2005). As highlighted
by Tomarken and Waller (2005), there are various advantages associated with LGM: (1)
the flexibility of modeling random effects and residuals; (2) the ability to comparatively
evaluate growth functions and detect changes in more complex causal models; (3) the
capacity to specify time-varying covariates; (4) a better ability to analyse multiple levels
of hierarchically structured data and the multivariate change patterns across multiple
measures; and (5) a stronger statistical method for treating missing data. These benefits

DWYER 9781781001288 Part I.indd 107 25/07/2012 11:07

 108 Handbook of research methods in tourism

are relevant for tourism researchers who utilize longitudinal or other repeated measures
data (e.g., Assaker et al., 2011) although its potential has not fully explored in the exist-
ing literature. Owing to the benefits outlined above, it is likely that LGM testing using
SEM will grow within the tourism literature not only for evaluating the developmental
trajectory of constructs over time, but also for analysing multilevel models with clustered
data structures in which repeated observations are nested within individuals.

NOTE

1. Nested models refer to two or more models that are structurally the same but differ in the number of freely
estimated or constrained parameters (Brown, 2006). When the models are nested, direct statistical compar-
ison between the models is possible. It is particularly useful to obtain a parsimonious model by trimming
or building models (Kline, 2005).

REFERENCES

Akaike, H. (1987), Factor analysis and AIC, Psychometrika, 52, 317332.

Albacete-Sez, C., M. Fuentes- Fuentes and F. Llorns-Montes (2007), Service quality measurement in rural
accommodation, Annals of Tourism Research, 34 (1), 4565.
Anderson, J.C. and D.W. Gerbing (1988), Structural equation modeling in practice: a review and recom-
mended two step approach, Psychological Bulletin, 103 (3), 453460.
Arbuckle, J. (2009), AMOS 18 Users Guide, Crawfordville, FL: AMOS Development Corporation.
Assaker, G., V. Vinzi and P. OConnor (2011), Examining the effect of novelty seeking, satisfaction, and
destination image on tourists return pattern: a two factor, non-linear latent growth model, Tourism
Management, 32 (4), 890901.
Bagozzi, R. (1980), Performance and satisfaction in an industrial sales force: an examination of their anteced-
ents and simultaneity, Journal of Marketing, 44 (2), 6577.
Bagozzi, R. (2007), On the meaning of formative measurement and how it differs from reflective measurement:
comment on Howell, Breivik, and Wilcox, Psychological Methods, 12 (2), 229237.
Baker, D. and J. Crompton (2000), Quality, satisfaction and behavioral intentions, Annals of Tourism
Research, 27 (3), 785804.
Beerli, A. and J.D. Martn (2004), Tourists characteristics and the perceived image of tourist
destinations: a quantitative analysis a case study of Lanzarote, Spain, Tourism Management, 25 (5),
623636.
Bentler, P. (1990), Comparative fit indexes in structural models, Psychological Bulletin, 107 (2), 238246.
Bentler, P. (2003), EQS 6.1 for Windows [Computer Software], Encino, CA: Multivariate Software.
Bign Alcaiz, E., I. Snchez Garca and S. Sanz Blas (2009), The functionalpsychological continuum in the
cognitive image of a destination: a confirmatory analysis, Tourism Management, 30 (5), 715723.
Bign, J., L. Andreu and J. Gnoth (2005a), The theme park experience: an analysis of pleasure, arousal and
satisfaction, Tourism Management, 26 (2), 833844.
Bign, J., L. Andreu, I. Kster and A. Blesa (2005b), Quality market orientation: tourist agencies perceived
effects, Annals of Tourism Research, 32 (4), 10221038.
Bign, J., M. Snchez and J. Snchez (2001), Tourism image, evaluation variables and after purchase behav-
iour: inter-relationship, Tourism Management, 22 (6), 607616.
Bollen, K. (1989), Structural Equations with Latent Variables. New York: Wiley.
Bollen, K. and R. Lennox (1991), Conventional wisdom on measurement: a structural equation perspective,
Psychological Bulletin, 110 (2), 305314.
Boo, S., J. Busser and S. Baloglu (2009), A model of customer-based brand equity and its application to mul-
tiple destinations, Tourism Management, 30 (2), 219231.
Brannick, M.T. (1995), Critical comments on applying covariance structure modeling, Journal of
Organizational Behavior, 16 (3), 201213.
Brown, T.A. (2006), Confirmatory Factor Analysis for Applied Research. New York: The Guilford Press.
Browne, M. and R. Cudeck (1993), Alternative ways of assessing model fit, in K.A. Bollen and J.S. Long
(eds), Testing Structural Equation Models, Newbury Park, CA: Sage, pp. 136162.

DWYER 9781781001288 Part I.indd 108 25/07/2012 11:07

 Structural equation modeling 109

Byrne, B.M. (1998), Structural Equation Modeling with LISREL, PRELIS, and SIMPLIS: Basic Concepts,
Applications, and Programming, Mahwah, NJ: Lawrence Erlbaum Associates, Inc.
Caro, M.L. and M.J. Garca (2008), Developing a multidimensional and hierarchical service quality model for
the travel agency industry, Tourism Management, 29 (4), 706720.
Cenfetelli, R.T. and G. Bassellier (2009), Interpretation of formative measurement in information systems
research, MIS Quarterly, 33 (4), 689707.
Chen, J.S. and C. Hsu (2001), Developing and validating a riverboat gaming impact scale, Annals of Tourism
Research, 27 (2), 459476.
Chin, W.W. (2001), PLS-Graph Users Guide (Version 3.0), Houston, TX: Soft Modeling.
Chung, J., G. Kyle, J. Petrick and J. Absher (2011), Fairness of prices, user fee policy and willingness to pay
among visitors to a national forest, Tourism Management, 32 (5), 10381046.
Churchill, G.A. (1979), A paradigm for developing better measures of marketing constructs, Journal of
Marketing Research, 16 (1), 6473.
Connell, J. and D. Meyer (2009), Balamory revisited: an evaluation of the screen tourism destinationtourist
nexus, Tourism Management, 30 (2), 194207.
Crowley, S. and X. Fan (1997), Structural equation modeling: basic concepts and applications in personality
assessment research, Journal of Personality Assessment, 68 (3), 508531.
del Bosque, I. and H. San Martn (2008), Tourist satisfaction: a cognitive-affective model, Annals of Tourism
Research, 35 (2), 551573.
Denstadli, J. and J. Jacobsen (2011), The long and winding roads: perceived quality of scenic tourism routes,
Tourism Management, 32 (4), 780789.
Diamantopoulos, A. and J.A. Siguaw (2006), Formative versus reflective indicators in organizational
measure development: a comparison and empirical illustration, British Journal of Management, 17 (4),
263282.
Diamantopoulos, A. and H.M. Winklhofer (2001), Index construction with formative indicators: an alterna-
tive to scale development, Journal of Marketing Research, 38 (2), 269277.
Dyer, P., D. Gursoy, B. Sharma and J. Carter (2007), Structural modeling of resident perceptions of tourism
and associated development on the Sunshine Coast, Australia, Tourism Management, 28 (2), 409422.
Feng, R. and A. Morrison (2007), Quality and value network: marketing travel clubs, Annals of Tourism
Research, 34 (3), 588609.
Fishbein, M. and I. Azjen (1975), Belief, Attitude, Intention, and Behavior: An Introduction to Theory and
Research, Reading, MA: Addison-Wesley.
Fornell, C. and D.F. Larcker (1981), Evaluating structural equation models with unobserved variables and
measurement error, Journal of Marketing Research, 18 (1), 3950.
Funk, D. and T. Bruun (2007), The role of socio-psychological and culture-education motives in marketing
international sport tourism: a cross-cultural perspective, Tourism Management, 28 (3), 806819.
Gross, M. and G. Brown (2008), An empirical structural model of tourists and places: progressing involve-
ment and place attachment into tourism, Tourism Management, 29 (6), 11411151.
Gursoy, D. and E. Gavcar (2003), International leisure tourists involvement profile, Annals of Tourism
Research, 30 (4), 906926.
Gursoy, D. and K. Kendall (2006), Hosting mega events: modeling locals support, Annals of Tourism
Research, 33 (3), 603623.
Gursoy, D. and D. Rutherford (2004), Host attitudes toward tourism: an improved structural model, Annals
of Tourism Research, 31 (3), 495516.
Gursoy, D., C. Jurowski and M. Uysal (2002), Resident attitudes: a structural modeling approach, Annals of
Tourism Research, 29 (1), 79105.
Haenlein, M. and A.M. Kaplan (2004), A beginners guide to partial least squares analysis, Understanding
Statistics, 3 (4), 283297.
Hair, J.F., R.E. Anderson, R.L. Tatham and W.C. Black (1998), Multivariate Data Analysis, 5th edition,
Upper Saddle River, NJ: Prentice Hall.
Han, H., L. Hsu and C. Sheu (2010), Application of the theory of planned behavior to green hotel choice:
testing the effect of environmental friendly activities, Tourism Management, 31 (3), 325334.
Hayduk, L. and D. Glaser (2000), Jiving the four-step, waltzing around factor analysis, and other serious fun,
Structural Equation Modeling, 7 (1), 135.
Hayduk, L., G. Cummings, K. Boadu, H. Pazderka-Robinson and S. Boulianne (2007), Testing! testing! one,
two three: testing the theory in structural equation models!, Personality and Individual Differences, 42 (2),
841850.
Howell, R.D., E. Breivik and J.B. Wilcox (2007), Reconsidering formative measurement, Psychological
Methods, 12 (2), 205218.
Hu, L. and P. Bentler (1999), Cutoff criteria for fit indexes in covariance structure analysis: conventional cri-
teria versus new alternatives, Structural Equation Modeling, 6 (1), 155.

DWYER 9781781001288 Part I.indd 109 25/07/2012 11:07

 110 Handbook of research methods in tourism

Huang, C., C. Chou and P. Lin (2010), Involvement theory in constructing bloggers intention to purchase
travel products, Tourism Management, 31 (4), 513526.
Huang, S. and C. Hsu (2009), Effects of travel motivation, past experience, perceived constraint, and attitude
on revisit intention, Journal of Travel Research, 48 (1), 2944.
Hung, K. and J.F. Petrick (2010), Developing a measurement scale for constraints to cruising, Annals of
Tourism Research, 37 (1), 206228.
Hurley, A.E., T.A. Scandura, C.A. Schriesheim, M.T. Brannick, A. Seers, R.J. Vandenberg and L.J. Williams
(1997), Exploratory and confirmatory factor analysis: guidelines, issues, and alternatives, Journal of
Organizational Behavior, 18 (6), 667683.
Hwang, S., C. Lee and H. Chen (2005), The relationship among tourists involvement, place attachment and
interpretation satisfaction in Taiwans national parks, Tourism Management, 26 (2), 143156.
Jarvis, C.B., S.B. MacKenzie and P.M. Podsakoff (2003), A critical review of construct indicators and meas-
urement model misspecification in marketing and consumer research, Journal of Consumer Research, 30 (2),
199218.
Jreskog, K. and D. Srbom (2004), LISREL 8.7 for Windows [Computer Software]. Lincolnwood, IL:
Scientific Software International.
Jurowski, C. and D. Gursoy (2004), Distance effects on residents attitudes toward tourism, Annals of Tourism
Research, 31 (2), 296312.
Kang, S., C. Lee, Y. Yoon and P. Long (2008), Resident perception of the impact of limited-stakes
community-based casino gaming in mature gaming communities, Tourism Management, 29 (4), 681694.
Kang, S., P. Long and R. Perdue (1996), Resident attitudes toward legal gambling, Annals of Tourism
Research, 23 (1), 7185.
Kaplanidou, K. and C. Vogt (2006), A structural analysis of destination travel intentions as a function of Web
site features, Journal of Travel Research, 45 (2), 204216.
Kim, M., N. Chung and C. Lee (2011), The effect of perceived trust on electronic commerce: shopping online
for tourism products and services in South Korea, Tourism Management, 32 (2), 256265.
Kim, S. and M.A. Littrell (1999), Predicting souvenir purchase intentions, Journal of Travel Research, 38 (2),
153162.
Kline, A. and H. Moosbrugger (2000), Maximum likelihood estimation of latent interaction effects with the
LMS method, Psychometrika, 65 (4), 457474.
Kline, R. (2005), Principles and Practice of Structural Equation Modeling, 2nd edition, New York: The
Guilford Press.
Ko, D. and W. Stewart (2002), A structural equation model of residents attitudes for tourism development,
Tourism Management, 23 (5), 521530.
Kolar, T. and V. abkar (2010), A consumer-based model of authenticity: an oxymoron or the foundation of
cultural heritage marketing?, Tourism Management, 31 (5), 652664.
Konecnik, M. and W. Gartner (2007), Customer-based brand equity for a destination, Annals of Tourism
Research, 34 (2), 400421.
Lee, C., Y. Lee and B. Lee (2005), Koreas destination image formed by the 2002 World Cup, Annals of
Tourism Research, 32 (4), 839858.
Lee, C., Y. Yoon and S. Lee (2007a), Investigating the relationships among perceived value, satisfaction, and
recommendations: the case of the Korean DMZ, Tourism Management, 28 (1), 204214.
Lee, M. and K. Back (2008), Association meeting participation: a test of competing models, Journal of Travel
Research, 46 (3), 300310.
Lee, S., J. Petrick and J. Crompton (2007b), The roles of quality and intermediary constructs in determining
festival attendees behavioral intention, Journal of Travel Research, 45 (4), 402412.
Lee, S., D. Scott and H. Kim (2008), Celebrity fan involvement and destination perceptions, Annals of
Tourism Research, 35 (3), 809832.
Lindberg, K. and R. Johnson (1997), Modeling resident attitudes toward tourism, Annals of Tourism
Research, 24 (2), 402424.
MacCallum, R.C. and J.T. Austin (2000), Applications of structural equation modeling in psychological
research, Annual Review of Psychology, 51, 201226.
McDonald, R. and M. Ho (2002), Principles and practice in reporting structural equation analyses,
Psychological Methods, 7 (1), 6482.
Marsh, H., J. Balla and R. McDonald (1988), Goodness-of-fit indexes in confirmatory factor analysis: the
effect of sample size, Psychological Bulletin, 103 (3), 391410.
Martn-Ruiz, D., M. Castellanos-Verdugo and . Oviedo-Garca (2010), A visitors evaluation index for a
visit to an archaeological site, Tourism Management, 31 (5), 590596.
Mazanec, J., K. Wber and A. Zins (2007), Tourism destination competitiveness: from definition to explana-
tion?, Journal of Travel Research, 46 (1), 8695.
Mulaik, S. and R. Millsap (2000), Doing the four-step right, Structural Equation Modeling, 7 (1), 3673.

DWYER 9781781001288 Part I.indd 110 25/07/2012 11:07

 Structural equation modeling 111

Mulaik, S., L. James, J. Van Alstine, N. Bennett, S. Lind and C. Stillwell (1989), Evaluation of goodness-of-fit
indices for structural equation models, Psychological Bulletin, 105 (3), 430455.
Muthn, L. and B. Muthn (19982010), Mplus Users Guide, 6th edition, Los Angeles: Muthn & Muthn.
Nadeau, J., L. Heslop, N. OReilly and P. Luk (2008), Destination in a country image context, Annals of
Tourism Research, 35 (1), 84106.
Nicholas, L., B. Thapa and Y. Ko (2009), Residents perspectives of a world heritage site: the Pitons manage-
ment area, St. Lucia, Annals of Tourism Research, 36 (3), 390412.
Nunkoo, R. and H. Ramkissoon (2011), Developing a community support model for tourism, Annals of
Tourism Research, 38 (3), 964988.
Nunnally, J.C. (1967), Psychometric Theory. New York: McGraw-Hill.
Nusair, K. and N. Hua (2010), Comparative assessment of structural equation modeling and multiple regres-
sion research methodologies: e-commerce context, Tourism Management, 31 (3), 314324.
Oh, H. and C. Hsu (2001), Volitional degrees of gambling behaviors, Annals of Tourism Research, 28 (3),
618637.
Oh, H., H. Kim and K. Hong (2009), A dynamic perspective of meeting planners satisfaction: toward concep-
tualization of critical relevancy, Tourism Management, 30 (4), 471482.
Okazaki, S. and M. Hirose (2009), Does gender affect media choice in travel information search? On the use
of mobile Internet, Tourism Management, 30 (6), 794804.
Palmer, A.L., A. Ses and J.J. Montao (2005), Tourism and statistics: bibliometric study 19982002, Annals
of Tourism Research, 32 (1), 167178.
Parasuraman, A., V.A. Zeithaml and L.L. Berry (1988), SERVQUAL: a multiple-item scale for measuring
consumer perceptions of service quality, Journal of Retailing, 64 (1), 1440.
Petrick, J.F. (2002), An examination of gold vacationers novelty, Annals of Tourism Research, 29 (2),
384400.
Qu, H. and H. Lee (2011), Travelers social identification and membership behaviors in online travel commu-
nity, Tourism Management, 32 (6), 12621270.
Reisinger, Y. and L. Turner (1999), Structural equation modeling with LISREL: application in tourism,
Tourism Management, 20 (1), 7188.
Reisinger, Y. and L. Turner (2002), Cultural differences between Asian tourist markets and Australian hosts:
Part 2, Journal of Travel Research, 40 (4), 374384.
Sparks, B. (2007), Planning a wine tourism vacation? Factors that help to predict tourist behavioural inten-
tions, Tourism Management, 28 (5), 11801192.
Steiger, J.H. and J.C. Lind (1980), Statistically-based tests for the number of common factors, paper pre-
sented at the annual Spring Meeting of the Psychometric Society, Iowa City, IA.
Stokburger-Sauer, N. (2011), The relevance of visitors nation brand embeddedness and personality
congruence for nation brand identification, visit intentions and advocacy, Tourism Management, 32 (6),
12821289.
Tomarken, A.J. and N.G. Waller (2005), Structural equation modeling: strengths, limitations, and misconcep-
tions, Annual Review of Clinical Psychology, 1, 3165.
Um, S., K. Chon and Y. Ro (2006), Antecedents of revisit intention, Annals of Tourism Research, 33 (4),
11411158.
Vargas-Snchez, A., N. Porras-Bueno and M. Plaza-meja (2011), Explaining residents attitudes to tourism:
is a universal model possible?, Annals of Tourism Research, 38 (2), 460480.
Vogt, C.A. and D. Fesenmaier (1995), Tourists and retailers perceptions of services, Annals of Tourism
Research, 22 (4), 763780.
Wang, Y. and D. Fesenmaier (2004), Towards understanding members general participation in and active
contribution to an online travel community, Tourism Management, 25 (6), 709722.
Westland, J. (2010), Lower bounds on sample size in structural equation modeling, Electronic Commerce
Research and Applications, 9 (6), 476487.
Wilcox, B.J., R.D. Howell and E. Breivik (2008), Questions about formative measurement, Journal of
Business Research, 61 (12), 12191228.
William, P. and G. Soutar (2009), Value, satisfaction and behavioral intentions in an adventure tourism
context, Annals of Tourism Research, 36 (3), 413438.
Wong, J. and C. Yeh (2009), Tourist hesitation in destination decision making, Annals of Tourism Research,
36 (1), 623.
Wu, S., P. Wei and J. Chen (2008), Influential factors and relational structure of Internet banner advertising
in the tourism industry, Tourism Management, 29 (2), 221236.
Yoon, Y., D. Gursoy and J. Chen (2001), Validating a tourism development theory with structural equation
modeling, Tourism Management, 22 (4), 363372.
Yoon, Y. and M. Uysal (2005), An examination of the effects of motivation and satisfaction on destination
loyalty: a structural model, Tourism Management, 26 (1), 4556.

DWYER 9781781001288 Part I.indd 111 25/07/2012 11:07

 112 Handbook of research methods in tourism

Yuan, J. and S. Jang (2008), The effects of quality and satisfaction on awareness and behavioral intentions:
exploring the role of a wine festival, Journal of Travel Research, 46 (3), 279288.
Yuksel, A., F. Yuksel and Y. Bilim (2010), Destination attachment: effects on customer satisfaction and cogni-
tive, affective and conative loyalty, Tourism Management, 31 (2), 274284.
abkar, V., M. Breni and T. Dmitrovi (2010), Modelling perceived quality, visitor satisfaction and behav-
ioural intentions at the destination level, Tourism Management, 31 (4), 537546.

DWYER 9781781001288 Part I.indd 112 25/07/2012 11:07

 6 Discrete choice analysis and experimental design
Clive Morley

NATURE OF THE TECHNIQUE AND ITS EVOLUTION

Tourists make many decisions in determining their trip, and many of these decisions are
intrinsically categorical, multinomial (many options are available) and unordered. For
example, the choice of destination, of hotel, whether a particular attraction is visited or
not, the mode of travel. Discrete choice theory provides an appropriate and sophisti-
cated framework for analysis of the data at the level of individuals evoked responses,
and one that has been proven to work well in practice.
Forecasting for some strategic policy and marketing analysis purposes using aggregate
level demand models can be problematical. For example, a special fare offer may not be
reflected in the typical fare used in the model and the effects of changes in hotel room
rates may be very imperfectly analysed through the medium of CPI changes (which is
how tourism prices are usually measured in such models). Policy analysis can require
the use of variables that are more directly related to those the potential tourist consid-
ers in making their choices. The methodology to model choices at the individual level is
theoretically established (McFadden, 1984; Manski and McFadden, 1981; Hensher et
al., 1988) and was successfully applied in areas such as transport economics and market-
ing (for example, Hensher and Johnson, 1981) before being readily applied to tourism
demand and policy analysis (Louviere and Timmermans, 1990). Early tourism applica-
tions include Witt (1980, 1983) and Sheldon and Mak (1987).
The methods of discrete choice analysis are sometimes referred to as random utility
methods (as the underlying econometric theory uses a random utility specification) or
limited dependent variable analysis (the dependent variable in the model is categori-
cal). In a marketing context the term conjoint analysis is used for very similar methods.
Logit, multinomial logit and probit models are particular cases of discrete choice
analysis.
In the simplest application, the dependent variable is binary (takes the values 0 or 1)
denoting whether or not the particular alternative of interest was chosen. The explana-
tory variables can include various characteristics of the individual choosing and of the
options. For example, a sample of tourists in Seville could be asked if they had visited
a flamenco show, and this choice related to characteristics such as their age, gender,
income, origin and the price of the show. It is inappropriate to use ordinary least squares
(OLS) regression (for example, the dependent variable is not continuous, the errors
are not homogeneous, the errors and explanatory variables are not independent). The
binomial response theory can be readily extended to the more general multinomial case
(McFadden, 1984) with more choice options (but still a finite number of mutually exclu-
sive options). Extending to more alternatives to choose from is not enough to satisfy the
fundamental requirements for OLS regression.
The most basic and common model, the binary logit model, comes from assuming

113

DWYER 9781781001288 Part I.indd 113 25/07/2012 11:07

 114 Handbook of research methods in tourism

independent, identically distributed random errors with an extreme value type 1 distri-
bution, which makes estimation straightforward. See the next section for details of what
this involves. It is also common to further assume the usual linear function of explana-
tory variables. The upshot of these assumptions is that the probability that the particular
alternative is chosen by a particular individual can be written as:

Pi 5 exp[b9Xi]/(1 1 exp[b9Xi] (6.1)

where Xi are the explanatory variables and b is a vector of coefficients (to be estimated)
From this can be derived the expression:

ln[Pi/(1 Pi)] 5 b9Xi (6.2)

The log of the odds ratio, the left hand side of this last equation, is known as a logit and
is a source of the name for this model.
Extending the dependent variable to encompass a choice from m alternatives gives the
multinomial logit (MNL) model:

Pi 5 exp[b9Xi]/(Sexp[b9Xj]. (6.3)

5 1/(Sexp[b9Xj b9Xi]

with the sum being over all m alternatives, indexed by j (the latter form shows that values
of variables common to all alternatives are irrelevant).
The logit and MNL models are readily estimated with generally accessible statistical
modelling software. To understand the consequences of the assumptions made in arriv-
ing at these simple (but powerful and useful) models, and extensions of them, requires a
further outline of the underlying theory, as in the following section.
The data used in discrete choice analysis is at the level of individuals and includes
variables for the choice made, characteristics of the alternatives and the individual and
perhaps of the choice sets (which alternatives are available to choose from, if these vary,
as they can be made to do in experimental conditions). The choices made are observable
reflections of the preferences underlying them, so preferences can be analysed with this
methodology. The data can derive from choices actually made and observed, such as
whether or not a tourist visited France on their last trip abroad (this is known as revealed
preference data) or from experiments in which respondents are presented with various
alternatives and asked to express their choice from them (stated preference data). For
example, stated preferences could be from a survey of potential tourists in Birmingham
who are asked to choose between trips to Madrid, Paris and Brussels, with each desti-
nation having either a cheap or expensive air fare given. Typically revealed preference
data has a single choice value per respondent, whilst in stated preference experiments
respondents as potential tourists may be asked to choose a number of times in different
circumstances (that is, consider various scenarios in which the alternatives presented to
choose from, and/or their attributes, are varied). Respondents are usually asked to do
this four or more times, with different sets of defined alternatives (choice sets). So in the
Birmingham example the potential tourists might be asked to choose from Madrid with

DWYER 9781781001288 Part I.indd 114 25/07/2012 11:07

 Discrete choice 115

an expensive air fare, Paris with an expensive air fare and Brussels with a cheap air fare
(choice set A), then to choose from Madrid with a cheap air fare, Paris with an expensive
air fare and Brussels with a cheap air fare (choice set B), etc. Discrete choice analysis,
with models like the MNL outlined above (and those discussed below), is used to analyse
both revealed preference and stated preference data.

BACKGROUND AND TYPES OF PROBLEMS THE

TECHNIQUES ARE DESIGNED TO HANDLE

Discrete Choice Analysis

The microeconomic theory of tourism demand (Morley, 1992) derives from a

Lancasterian consumer theory in which the utility (or attractiveness) of a good (here
alternative) is a function of characteristics of the good. For clarity, we will henceforth
use the case of a choice of destination for a tourist visit, although the concepts just as
readily apply to other types of choices. An individuals direct utility function is a func-
tion of characteristics of the destinations available to the individual and quantities of
other goods and services that they might consume. Maximizing this utility subject to the
constraints of their money budget and time available yields the indirect utility function
Vrj, for individual r and destination j. The indirect utility is a function of characteristics
of the individual (e.g., income, the time they have available and demographic attributes
the latter possibly as proxies for taste variables), and characteristics of the tourist
destination (e.g., the fare, the transit time, attributes of the destination, the cost per
day of staying) and prices of other goods and services. Let Xrj denote the vector of these
exogenous variables.
Tour j is preferred to tour k if Vrj > Vrk, and that tour J is chosen for which VrJ 5 maxj
{Vrj}. This includes the possibility (say, j 5 0) that no tour is chosen.
In the random utility formulation Vrj 5 v(Xrj, erj) the utilities cannot be completely
characterized in terms of the observable, systematic part represented by the variables in
Xrj and so a random, unobserved element erj is incorporated. A common functional form
v(.) for the utilities is assumed to enable estimation to be operationally tractable. As in
aggregated modelling it is usual to assume a linear additive random term:

Vrj 5 Vrj* 1 erj (6.4)

The probability a particular destination (say j 5 1) is chosen, the conditional distribu-

tion of the response given the exogenous variables, is:

Pr(1) 5 P[Vr1 5 maxj {Vrj}]

5 P[Vrj* 1 erj < Vr1* 1 er1] for all j 1 (6.5)

This requires solution of an m dimensional integral (where m is the number of destina-

tions to be chosen from). Until recently this was not feasible and standard practice has
been to introduce simplifying assumptions, which will be followed here for the moment

DWYER 9781781001288 Part I.indd 115 25/07/2012 11:07

 116 Handbook of research methods in tourism

(although see later in the chapter for consideration of ways and means of testing the
validity of these assumptions and of estimating without making them).
If the random components are assumed to be independent, identically distributed with
cumulative density function F(e|Xrj), then

Pr(1) 5 PjP[erj < Vr1* 1 er1 Vrj*]

5 F9(x | Xrj)Pj 1 F[x 1 Vr1* Vrj*]dx (6.6)

where F9 is the probability density function (the derivative of F with respect to x) and the
integral is over all values of the error (x from to 1).
This expression for Pr in terms of the error distribution emphasizes its dependence on
the indirect utility function Vrj*, particularly Vr1*, and the convoluted form of relation-
ship to the distribution F. For the expression to be tractable for estimation it is common
to either adopt a distributional form for F that results in an estimable expression for Pr or
to cut through the complexity of Pr as a function of F and assume a distributional form
for Pr. The logit and probit models are widely used. The logit model derives straightfor-
wardly from Equation (6.6). If F is taken as the Weibull extreme value distribution (also
called the extreme value type I and the Gumbel distribution), then it follows that:

Pr(1) 5 exp(Vr1*) / Sjexp(Vrj*) (6.7)

Alternatively, for the probit model Pr is expressed directly as

Pr(1) 5 F(Vr1*) (6.8)

where F is the cumulative distribution function of the standard normal distribution (the
assumption of independent, identically distributed errors is unnecessary here).
The more straightforward linear probability model:

Pr(1) 5 Vr1* (6.9)

is not attractive as it does not confine the probability Pr to the interval [0,1] and restric-
tions to ensure it does so can be difficult to enforce.
A linearity assumption, Vrj* 5 b9Xrj for a vector of coefficients b, is usual, (this gives
Equation (6.3) for the MNL), although more general forms, such as second order (quad-
ratic), interaction or translog approximations, can be readily used.
Unless there are extreme values of exogenous variables or observed choice frequencies
close to zero or one in the data, there will be little empirical difference between the logit
and probit models.
Brau (2008) used a MNL model to investigate the preferences of tourists visiting
Sardinia (Italy) for stays with various proximities to the beach, risk of crowding, avail-
ability of services, cost and environmental variables. He was able to derive implicit
prices or willingness to pay values for differences in crowding, level of services, etc. On
the other hand, Chaminuka et al. (2011) applied a probit model to Kruger National Park
(South Africa) visitors choices in terms of accommodation inside or outside the park,

DWYER 9781781001288 Part I.indd 116 25/07/2012 11:07

 Discrete choice 117

price and various ecotourism possibilities. Their results showed that certain ecotourism
features were more attractive than others and that there was a willingness to pay more
than expected for them.
A strong consequence of the independent and identically distributed errors assump-
tions of the logit (and MNL) model is what is called the independence from irrelevant
alternatives (IIA) property. The IIA property means that the ratio Pr(i)/Pr(j) depends only
on the utilities of choices i and j, and not on any other alternatives (readily seen by substi-
tuting (6.7) in both terms of the ratio and cancelling out the sum terms). This means that
a change in the utility of one alternative (such as an improvement in the attractiveness
of a particular destination or a change in the taxes paid at a destination) will have a pro-
portionately equal impact on the probability of choosing all the other alternatives (i.e.,
equality of the cross-elasticities of the probabilities of choosing alternatives with respect
to the explanatory variables). This is not always plausible. The classic example is from
the situation of the choice of mode of travel between an origin and a destination, where
the traveller can choose between a bus (which happens to be blue), a car and a train. If a
new bus service is introduced into the choice (this one is red, but otherwise very similar
to the blue bus) then the new, red bus service would expect to draw patronage dispropor-
tionately from the blue bus, contrary to the IIA property. IIA implies that parameters
of the model defined on a subset of the full choice set can be unbiasedly estimated from
data on the subset. So a useful consequence of the IIA property is that estimates are valid
(consistent) even if some important alternatives are not included in the choice set.
As these restrictions are not usually theoretically plausible it is important to test for
the IIA property and this should be a standard part of the analysis of MNL models (not
all empirical studies have recognised this). The Hausman and McFadden (1984) test is
a useful one, based on the fact that, if the IIA property assumption is valid then param-
eter estimates from choice data over the full choice sets are, except for sampling error,
equal to parameter estimates from conditional choice data over choice sets restricted to
a subset of the full set of alternatives. Non-rejection of the IIA property can be taken as
strong support of the MNL model used in an analysis. Despite the stringency of the IIA
assumption empirical results often show that it is not a serious concern. But it can be,
Chaminuka et al. (2011) used a probit model because the logit model failed an IIA test.

Experimental Design

Revealed preference data has the advantage of representing actual decisions made
whereas stated preference data is of hypothetical decisions. But this means that revealed
preferences are constrained to pre-existing conditions; stated preferences allow possibili-
ties to consider extensions and hypothetical conditions, important in policy and plan-
ning studies. Revealed preference data will be deficient in information on non-chosen
alternatives and so estimating the specific effects of the explanatory variables is limited.
In stated preference studies the researcher designs the characteristics considered by the
subjects, determining and manipulating them to facilitate estimation of the variables
effects on choice outcomes. In doing so statistical principles of experimental design are
very important for efficiency and efficacy, to ensure the desired effects are well estimated
within sample size limitations. For example, in an analysis of destination hotel prices (set
at, say, high and low levels) and air fares (set at high, medium and low levels) on choice

DWYER 9781781001288 Part I.indd 117 25/07/2012 11:07

 118 Handbook of research methods in tourism

of destination, it is advisable to balance the experiment so that the high and low level
fares are equally above and below the medium, and to define the choice sets so that each
hotel price level is presented in combination with each air fare level in the options.
The key to experiments is that the researcher deliberately controls and manipulates
(pre-sets) the values of the independent (explanatory) variables and measures the
response value of the dependent variable. This enables much stronger inferences about
causality than are possible with observational data (wherein the dependent variable is
measured in response to naturally seen variations in the independent variables). The
issues with observational data are that the independent variables are not controlled, the
response values may be (in full or part) due to other variables not measured in the data,
and often multicollinearity of observed variables is present. Further, the validity of the
statistical tests and inferences usually assumes that the independent variables values are
fixed; this means that the p-values, significance and critical values from regression and
the like analyses as they are usually done in tourism (and many other economic) studies
are more a conventional guide rather than the accurate probability statements they
purport to be. The statistical inference from experimental data is more soundly based.
On the other hand, when done on human subjects, experiments suffer from concerns
about their artificiality (as in the difference between revealed and stated preferences) and
that it is not always practical or ethical to vary the values of some variables in ways the
experimenter may want. For example, including questions on illegal activities (such as
drug taking, which may be important for some tourists).
Vital to the validity of experimental data is randomization. Whilst the combinations
of values of the independent variables are set by the experimenter, the particular subjects
assigned to respond to those values need to be chosen at random. This is essential both to
provide a probabilistic basis for the statistical inference, and to randomize out the effect
of any unaccounted for variables.
Discrete choice experiments rely heavily on the use of factorial designs (these are so
called because the independent variables in experiments are commonly called factors).
Factorial designs enable large gains in efficiency, i.e., smaller sample sizes than would
otherwise be needed. Essentially, the efficiency gains come about by analysing multiple
variables effects on the dependent variable in one designed experiment, on one set of
data, designed so that each variable can be analysed on the whole data but independently
of the other variables.
A full factorial design enables the effects of all variables, and all their interactions, to
be estimated and tested. Interactions are when the effect of one variable is contingent on
the value of another variable (so that the variables are not independent and additive in
their effect on the response). In most (but not all) discrete choice experiments a fractional
factorial design is used. In fractional factorial designs certain interactions are assumed to
be negligible, so do not need to be included in the model, estimated or tested for. These
assumed zero interactions are pre-specified by the experimenter and then determine the
design. By dropping them from consideration, a smaller size of experiment, in terms
of the number of combinations of variable values set, can be determined. This can be
important in tourism discrete choice experiments, as otherwise the number of options
respondents have to choose from can be too large to be reasonable. For example, in
Albaladejo-Pina and Daz-Delfas (2009) study of tourist preferences for rural house
stays in a region of Spain, the variables included type of building (four types), location

DWYER 9781781001288 Part I.indd 118 25/07/2012 11:07

 Discrete choice 119

(town or orchard or mountain or village), number of bedrooms, price (four levels) and
ten other variables (each of two levels). There are hundreds of thousands of combina-
tions of the levels, and hence alternatives, which need to be drastically reduced to get
reasonable choice sets for presentation to respondents.
Fractional factorial designs are most useful when there are a relatively large number
of explanatory variables to be investigated and the aim is to economically consider the
effects of them all, at the expense of some ambiguity around interaction effects, as con-
trasted with a more in-depth analysis of a few variables in a scientific experiment seeking
to establish cause and effect (Cox, 1958, p. 258). They are thus appropriate for the types
of economic studies they have been used in tourism research (as in the studies referenced
in this chapter).
In the application of these designs in tourism, nearly all published studies are designed
to allow only estimation and testing of the main effects of the variables; that is all
interactions, or occasionally all but a very few interactions, are assumed negligible (an
exception is Morley (1994b)). This is due to a combination of the number of variables
(factors) it is wanted to include in the design, the number of levels (values) of these vari-
ables, and the number of times a respondent can be asked to make a choice before data
quality degrades. This latter number is often held to be quite small (most reported studies
keep it to a single digit: Morley (1994b) in his study of three price variables impacts on
the choice of Sydney as a destination of Malaysian tourists uses 12), but Hensher et al.
(2001) in an analysis of the choices of airline and ticket class for flight between Australia
and New Zealand argue that, and provide some evidence for, 16 being reasonable and
32 quite feasible.
The issue of which interactions are confounded and assumed negligible in assess-
ing main effects is not always made clear in the published reports of such experiments,
although it would be useful to readers and users of such research if this was spelt out. For
example, Lindberg et al. (1999) in their exploration of the trade-offs Danish residents
are prepared to make in tourisms impacts on them appear to use a 1/16 fraction of a
44 experiment, which would necessitate some two factor interactions being confounded
with some main effects, but which are not clarified in the paper.
Fractional designs can be designed to ensure that each alternative occurs equally often,
and equally often with each other alternative, in the data. This enables each variables
effect to be estimated independently of other variables. It is recommended practice to
always include a base option in every choice set presented to respondents. In the model,
dummy variables can indicate the presence or not of an alternative in a particular choice
set, so referring back to a common base is useful. The base alternative should be one that
is likely to be relatively unpopular, to avoid it dominating the choices and hence the data
not having sufficient frequencies for the other alternatives for efficient estimation, whilst
at the same time it is helpful if it is meaningful as a reference point.

APPLICATIONS TO TOURISM

The obvious applications of discrete choice methods are in demand modelling, and
especially in assessing the impact of potentially explanatory variables on demand. This
then gives useful results for marketing, forecasting, planning and policy evaluation and

DWYER 9781781001288 Part I.indd 119 25/07/2012 11:07

 120 Handbook of research methods in tourism

development, and other decision making by operators. For example, the technique has
been used to estimate visitors willingness to pay for attractions and substitution rates,
that is the rate at which people will trade-off less of one attribute for more of another
attribute, for a particular destination (e.g., Sardinia; Brau, 2008) or attraction (e.g.,
Yanchep caves; Tapsuwan, 2010). Other applications range widely from analysing
various price effects on demand for a single destination from one origin (e.g., demand for
Sydney from Kuala Lumpur in Morley, 1994b) through wider investigations of demand
across combinations of origins and destinations (and types of destinations), to ecotour-
ism preferences (Chaminuka et al.s 2011 study of Kruger National Park visitors). The
potential explanatory variables used are driven by the purpose of the investigation, and
vary widely in the reported studies. The papers in the tourism literature cited in this
chapter provide many different examples of the use of this attractive and valuable tech-
nique.
Revealed preference data modelling is more usual in marketing (Baltas and Doyle,
2001), but in tourism applications stated preference data from choice experiments is
more common. Revealed preference analyses in tourism are not unknown, for example
Eymann and Ronning (1997) who analysed tourists choice of destination (at a regional
level), Nicolau and Ms (2006) who considered the moderating role motivations played
in the impacts of price and distance on destination choices, and Nicolau (2010) in an
analysis of how cultural interest moderates price impacts on destination choice. The not
uncommon practice in analyses of revealed preference data of dropping from the analy-
sis infrequent visitors and less popular alternatives, due to problems in estimation on
sparse and incomplete data, will induce a bias in the results.
One common aspect of the reporting of results is worth noting. The standard R2
measure from regression is not applicable in the discrete choice case and the pseudo-R2
measures used are less intuitively informative. It is not difficult for a model to be a sig-
nificant improvement on a null model, in log-likelihood terms. It would be good practice
to report on the fitness of the model in terms of the (in-sample) percentages of predicted
choices made correctly by the model (for each actual choice category). At present this is
not common practice, but it is recommended statistically.

ADVANTAGES AND LIMITATIONS OF THE TECHNIQUES

Econometric modelling using time series data is not always able to estimate effectively
the independent effects of variables. Taking prices as an example, the observed price
series are jointly determined by underlying variables such as taxation, inflation, costs of
inputs and wages and there is therefore strong multicollinearity amongst price series. In
discrete choice experiments the data comes from a designed experiment, which allows
variations in conditions and scenarios of interest to the researcher to be put to the
respondents, even if these do not, cannot or have not been seen in reality. The responses
(choices) to these pre-determined conditions are recorded, allowing estimation of effects
not able to be estimated in other ways. For example the components of the tourist price
variables can be made to vary independently in the experimental design, and hence their
independent effects discerned, although in observed reality they are highly correlated.
Additionally, analysis of actual measured demand at a certain combination of prices

DWYER 9781781001288 Part I.indd 120 25/07/2012 11:07

 Discrete choice 121

reveals little about the demand that could be expected at the time with a different combi-
nation of prices. Experimental data on individuals responses to price variations can be
collected from an experiment designed to circumvent these problems.
Another advantageous feature of discrete choice methods is that they utilize micro-
level data on individuals, rather than aggregated demand numbers. They give data on,
and hence enable modelling of, the actual decision-making processes of potential tour-
ists. As a result a much more realistic demand model can be estimated, with independent
variables such as actual age, income and race of individuals, and both actual hotel prices
and CPI (e.g., Morley, 1994b) one closer to the actual decisions around travelling or
not, the choice of destination, etc., and closer to the actual decision-making process.
It is usual in tourism demand analysis to rely on data derived from direct observation
and measurement of actual tourism behaviour or, at one remove, from surveys asking
about tourism behaviour. There are obvious strong reasons for basing models of demand
on data of this type. However, there are serious deficiencies in analysis of observational
data for addressing certain types of questions. Analysis is restricted to variable ranges
found to have existed in the past. Parameter estimates are conditional on these ranges
and extrapolation to new conditions is of uncertain validity. For example, the impacts
of very low air fares, well below previous air fares, could not be estimated prior to some
airlines offering such fares, or the impact of a major new attraction on a destinations
attractiveness prior to it opening. Further it is not uncommon to find that some variables
of interest do not vary, or vary only minimally, over the revealed data period, so that
the impact of these variables cannot be reliably estimated. As noted, multicollinearity
amongst important variables is frequently found and this also calls into question the reli-
ability of estimation of model parameters.
Surveys of tourists at their destinations provide useful data for tourism planning.
Survey data has advantages such as its being available, often from large samples and
generally reliable, and it measures what tourists actually do. However, surveys and
experiments at destinations, and all observational data of demand, suffer from a selec-
tion bias as the information is conditional on the tourist choosing the destination and
those tourists who chose other destinations, plus those individuals who chose not to tour,
are excluded. The samples of respondents cannot be representative of potential travellers
to the destination, as those who might have chosen the destination under some circum-
stances, but did not under the existing circumstances, are excluded from the sample. The
decision-making processes of those individuals excluded are important for demand and
policy analysis. For such purposes, it is therefore necessary to go a step back and gather
data on potential tourists in their origins.
Stated preference methods allow the researcher to focus on the individuals decision-
making process directly, putting questions in a behavioural context of choice. The
context is a simulated one in which the alternatives presented are pre-determined by the
experimental design, but within that the question is put in a behavioural form such as
which would you choose, given these options?, for example, which specific airline would
you choose to go on holiday to Japan from options defined by fare, quality of food,
leg room and limousine pickup service availability. This form concentrates on discrete
choice as simulating the actual decision-making process of the potential tourist. For this
reason, and because the task facing the respondent is easier and better understood, it is
to be expected that choice data will give better results than ranking or preference ratings

DWYER 9781781001288 Part I.indd 121 25/07/2012 11:07

 122 Handbook of research methods in tourism

data (Hensher et al., 1988; Louviere, 1988). The alternatives presented in the simulations
are designed by the researcher and respondents are asked to state their choice from a
number of sets of alternatives. The researcher can control the alternatives presented for
consideration, tailoring them to address the issues important to the particular research
objectives and ensuring variables range over a suitable field. Multiple observations are
gained from each respondent, and with appropriate design an experiment can be framed
to obtain data in an efficient manner.
The major potential deficiency of stated choice data is that respondents are stating
what they would do in certain simulated, somewhat artificial conditions, and this is
perhaps not the same as what they actually do, or would do, in reality. There is evidence,
from other contexts, of bias in such data, which makes absolute demand estimation
problematical. However, as long as respondents can reasonably evaluate the alternatives
proffered for their choice the relative utility weights which are the focus of the stated
choice models can be estimated without bias. Stated choice data can be used to frame
useful models which can be straightforwardly rescaled using observational data to esti-
mate actual demand levels (Kroes and Sheldon, 1988). Discrete choice experiments have
been found to give good predictions and fits to actual market shares (Louviere, 1988;
Louviere and Woodworth, 1983) in areas such as modelling modal choices (e.g., for
long distance inter-city travel: Koppelman and Sethi, 2005), and consumer goods brand
choice (such studies abound, see as a starting point references in Baltas and Doyle, 2001).
The experimental design uses strict control over institutional rules to reduce the error
in testing well defined hypotheses framed in terms of the variables. The sample can be
randomly chosen from a defined population which fixes the external environment and
enables statistical inference from the sample to the population. Further inference from
the population sampled to other populations may be hazarded on the less scientific, but
not uncommon, basis of analogy between populations. In an appropriately designed
experiment, respondent individuals can be placed in choice contexts which satisfy
orthogonality of the factors and allow independent estimation of parameters for the
factor variables. To achieve these desirable features for the analysis it is necessary to
carefully design and conduct the experiment, in contrast to the lesser emphasis on design
considerations in observational studies.
A powerful property of logit (and probit) models is that they have been found to be
quite robust to misspecification of the error distribution, and to the omission of explana-
tory variables that are uncorrelated with the included explanatory variables (Cramer,
2007). In the latter case the estimated coefficients of the model may be biased by such a
misspecification, but the effects of the variables on the dependent variable (e.g., the odds
ratio) are not.
Many of the limitations of these choice models are being addressed in more complex
models made possible in recent times due to advances in computing power. These are
discussed in the following section.

FURTHER DEVELOPMENTS AND APPLICATIONS

The restrictive nature of the IIA property of the MNL model has led to efforts to
develop models that relax the distributional assumptions that give rise to it. The most

DWYER 9781781001288 Part I.indd 122 25/07/2012 11:07

 Discrete choice 123

straightforward of these is the nested multinomial logit (NMNL) model. In this model,
the choice is structured as a hierarchical, multi-stage process with similar alternatives
grouped together and a choice amongst groups preceding the choice within the chosen
group. The distributional assumptions and IIA property now only need to hold within
each group, not across choices in different groups. This is readily extended to more than
two stages (groups within groups). It does not require that the actual decision process
is multi-staged, just that it is modelled that way to allow for correlations within groups.
But the grouping of alternatives needs to be pre-determined by the analyst, and both
the grouping and the ordering of the nesting can affect the results. For example, in an
exercise modelling tourists destination choices, classing destinations into (say) Europe
or Asia, and then into adventure- or sun-based, will give different results to doing the
adventure- or sun-based classification first and then the Europe or Asia grouping, even
if the final sets of destinations are the same (Nevo, 2000). As the example suggests, this
is not always a satisfactory feature as a hierarchy of classifications is often not obvious.
Bresnahan et al. (1997) have put forward a technique that may alleviate this problem.
Further extensions of the MNL model allow alternatives to be, to various extents, in
more than one group of the NMNL model (called generalized nested multinomial logit
in Koppelman and Sethi, 2005). The heteroskedastic MNL model relaxes the identically
distributed errors assumption to allow for differences in the variances of the errors.
Koppelman and Sethi (2005) have proposed a generalization of the MNL model that
allows for heterogeneous errors, heterogeneous covariances also and flexibility in elas-
ticities of substitution between alternatives (i.e., relaxing the IIA restriction as in nested
logit models). This results in a version of the MNL (called the heterogeneous generalized
nested logit model: HGNL) that allows for the variancecovariance matrix of the errors
to be quite general (not restricted to independent or constant variances) but which is still
computationally feasible without resort to numerical integration.
Alternatively, the multinomial probit model allows for the heterogeneity of the
HGNL model in a conceptually simpler and more appealing manner (normally distrib-
uted errors with a general variancecovariance matrix), but at the price of computa-
tional difficulties. The probit model has generally been confined in practice to analysing
binary choices, due to the problems of computing multinomial probit models (without
the independence assumption, Equation (6.6) involves m 1 integrals, where m is the
number of alternatives). Recent simulation based methods can be used to approximate
the complex integrals involved and open up the possibility of such models being tractable
(Weeks, 1997). These techniques can relax the independence and homogeneity restric-
tions. Chaminuka et al.s (2011) study of ecotourism in the Kruger National Park is a
recent example with three alternatives (a status quo alternative of the current offering,
plus two ecotourism alternatives defined in terms of accommodation location, price and
availability, or not, of visits to craft markets and village tours). Larger MNP models are
still difficult to implement, but are becoming possible.
Variations across individuals in preferences are most readily addressed by including in
the models variables for individuals characteristics (such as income, age, gender, previ-
ous visits) as main effects and in interaction with alternatives measures variables. The
random coefficients (or mixed MNL) model allows the parameters (coefficients in b) to
vary across individuals, to allow for differences in tastes. Again, this has become recently
estimable in practice with simulation-based (numerical integration) methods (McFadden

DWYER 9781781001288 Part I.indd 123 25/07/2012 11:07

 124 Handbook of research methods in tourism

and Train, 2000; Nevo, 2000). Random coefficients models can be a good approximation
to any discrete choice random utility model (McFadden and Train, 2000), hence they are
becoming a popular applied technique. It has been noted that controlling for variation
across individuals generally increases the observed impact of the other variables (Baltas
and Doyle, 2001).
Rather than using the individual level data directly, an alternative approach is to
model the observed frequencies of the choices, which requires using Weighted Least
Squares (or, the arguably more efficient Feasible Generalised Least Squares see an
advanced econometrics text such as Greene, 2003). Morley (1994a) is a tourism applica-
tion of this approach, which analysed the impacts of price components on Malaysians
choice of international destination for a holiday. Nevo (2000) expounds on random coef-
ficients logit models estimated on frequencies (market share) data.
Further developments can be expected in the joint modelling of both discrete and con-
tinuous variables, such as the choice of destination and time spent there, or the modelling
of spending at a resort (which is zero if the resort is not chosen, but has a range of posi-
tive values if it is chosen). Alegre et al. (2011) in their study of the length of stay of tour-
ists in the Balearic Islands have made a start in this direction with a binary logit model
for the choice coupled with a Poisson distributed length of stay. The multinomial tobit
model can be applied and the simulation-based techniques of likely value in the MNP
case are just as likely to be of value here. The multinomial tobit has not yet been applied
in tourism, but there is room for it to be considered.
Later developments in modelling techniques have increased in complexity and the
estimation and use of these models results is more difficult. Such models may also be less
robust than the simple, tried and tested forms. These trade-offs are yet to be determined.
The conduct of the choice experiments can be advanced through the use of streaming
video and other information technology. The choices can be presented in more real-
istic, dynamic, visual and tailored ways than the static listing of attributes that has to
date been the norm (Verma and Plaschka, 2003) who report that they have extensively
used web-based technologies (with hyperlinked pictures or written illustrations, brand
logos, and audio and video files) to realistically illustrate choice scenarios (Verma and
Plaschka, 2003, p. 160) although this raises further potential issues of extraneous vari-
ables in the presentation affecting the results.
Overall the combination of experimental design, stated preference data and discrete
choice models form a powerful body of techniques for analysis of tourism demand.
Often general descriptive results, such as the significance of particular explanatory vari-
ables and differences between segments of tourists defined by demographic variables, are
apparent from simple data analysis using contingency tables and aggregate level logit
modelling. This is because a lot of analytic power comes from the experimental protocols
and design. The more sophisticated choice models derived from discrete choice theory
quantify the impact of variables for detailed policy analysis, strategic market planning
and economic analysis.

DWYER 9781781001288 Part I.indd 124 25/07/2012 11:07

 Discrete choice 125

REFERENCES

Albaladejo-Pina, I.P. and M.T. Daz-Delfa (2009), Tourist preferences for rural house stays: evidence from
discrete choice modelling in Spain, Tourism Management, 30, 805811.
Alegre, J., S. Mateo and L. Pou (2011), A latent class approach to tourists length of stay, Tourism
Management, 32, 555563.
Baltas, G. and P. Doyle (2001), Random utility models in marketing research: a survey, Journal of Business
Research, 51, 115125.
Brau, R. (2008), Demand-driven sustainable tourism? A choice modelling analysis, Tourism Economics, 14,
691708.
Bresnahan, T., S. Stern and M. Trajtenberg (1997), Market segmentation and the sources of rents from inno-
vation: personal computers in the late 1980s, RAND Journal of Economics, 28, S17S44.
Chaminuka, P., R.A. Groenveld, A.O. Sleomane and E.C van Ierland (2011), Tourist preferences for eco-
tourism in rural communities adjacent to Kruger National park: a choice experiment approach, Tourism
Management, doi:10.1016/j.tourman.2011.02.016.
Cox, D.R. (1958), Planning of Experiments, New York: Wiley.
Cramer, J.S. (2007), Robustness of logit analysis: unobserved heterogeneity and mis-specified disturbances,
Oxford Bulletin of Economics and Statistics, 69, 545555.
Eymann, A. and G. Ronning (1997), Microeconomic models of tourists destination choice, Regional Science
and Urban Economics, 27, 735761.
Greene, W.H. (2003), Econometric Analysis, 5th edition, New Jersey: Prentice Hall.
Hausman, J. and D. McFadden (1984), Specification tests for the multinomial logit model, Econometrica, 52,
12191240.
Hensher, D.A., P.O. Barnard and P.T. Truong (1988), The role of stated preference methods in studies of
travel choice, Journal of Transport Economics and Policy, 22, 4558.
Hensher, D.A. and L.W. Johnson (1981), Applied Discrete Choice Modelling, London and New York: Halstead
Press.
Hensher, D.A., P.R. Stopher and J.J. Louviere (2001), An exploratory analysis of the effect of numbers
of choice sets in designed choice experiments: an airline choice application, Journal of Air Transport
Management, 7, 373379.
Koppelman, F.S. and V. Sethi (2005), Incorporating variance and covariance heterogeneity in the Generalized
Nested Logit model: an application to modelling long distance travel choice behaviour, Transportation
Research Part B, 39, 825853.
Kroes, E.P. and R.J. Sheldon (1988), Stated preference methods: an introduction, Journal of Transport
Economics and Policy, 22, 1125.
Lindberg, K., B. Dellaert and C.R. Rassing (1999), Resident tradeoffs: a choice modelling approach, Annals
of Tourism Research, 26, 554569.
Louviere, J.J. (1988), Conjoint analysis modelling of stated preferences, Journal of Transport Economics and
Policy, 22, 93119.
Louviere, J.J. and H. Timmermans (1990), Stated preference and choice models applied to recreation research:
a review, Leisure Sciences, 20, 923.
Louviere, J.J. and G. Woodworth (1983), Design and analysis of simulated consumer choice or allocation
experiments: an approach based on aggregate data, Journal of Marketing Research, 20, 350367.
Manski, C.F. and D.L. McFadden (1981), Structural Analysis of Discrete Data with Econometric Applications,
Cambridge MA: MIT Press.
McFadden, D. and K. Train (2000), Mixed MNL models for discrete response, Journal of Applied
Econometrics, 15, 447470.
McFadden, D.L. (1984), Econometric analysis of qualitative response models, in Z. Griliches and M.D.
Intriligator (eds), Handbook of Econometrics, Volume II, Amsterdam: Elsevier Science Publishers, pp.
13951457.
Morley, C.L. (1992), A micro-economic theory of international tourism demand, Annals of Tourism Research,
19, 250267.
Morley, C.L. (1994a), Discrete choice analysis of the impact of tourism prices, Journal of Travel Research,
33 (2), 814.
Morley, C.L. (1994b), Experimental destination choice analysis, Annals of Tourism Research, 21 (4), 780791.
Nevo, A. (2000), A practitioners guide to estimation of random-coefficients logit models of demand, Journal
of Economics and Management Strategy, 9, 513548.
Nicolau, J.L. (2010), Differentiated price loss aversion in destination choice: the effect of tourists cultural
interest, Tourism Management, doi:10.1016/j.tourman.2010.11.002.

DWYER 9781781001288 Part I.indd 125 25/07/2012 11:07

 126 Handbook of research methods in tourism

Nicolau, J.L. and F.J. Ms (2006), The influence of distance and prices on the choice of tourist destinations:
the moderating role of motivations, Tourism Management, 27, 982996.
Sheldon, P.J. and J. Mak (1987), The demand for package tours: a mode choice model, Journal of Travel
Research, 25, 1340.
Tapsuwan, S. (2010), A multivariate probit analysis of willingness to pay for cave conservation: a case study
of Yanchep National Park, Western Australia, Tourism Economics, 16, 10191035.
Verma, R. and G. Plaschka (2003), The art and science of customer-choice modelling: reflections, advances
and managerial implications, Cornell Hotel and Restaurant Administration Quarterly, 44, 156165.
Weeks, M. (1997), The multinomial probit model revisited: a discussion of parameter estimability, identifica-
tion and specification testing, Journal of Economic Surveys, 11, 297320.
Witt, S.F. (1980), An abstract mode abstract (destination) node model of foreign holiday demand, Applied
Economics, 12, 163180.
Witt, S.F. (1983), A binary choice model of foreign holiday demand, Journal of Economic Studies, 10, 4659.

DWYER 9781781001288 Part I.indd 126 25/07/2012 11:07

 7 Panel data analysis
Neelu Seetaram and Sylvain Petit

INTRODUCTION

Panel data sets are also known as longitudinal data or cross-sectional time series data.
They have spatial (N) and temporal (T) dimensions. They constitute of a number
of observations over time on a number of cross-sectional units such as individu-
als, firms, and countries allowing researchers to analyse the dynamics of change in
short time series data. According to Baltes and Nesselroade (1979), longitudinal data
and techniques involve a variety of methods connected by the idea that the entity
under investigation is observed repeatedly as it exists and evolves over time. These
methods have been applied in different research disciplines. Frees (2004) posits that
they have developed because important databases have become available to empirical
researchers.
The term panel data was introduced by Lazarsfeld and Fiske (1938, in Frees 2004) in
their study of the effect of the relationship between radio advertising and product sales,
where they proposed to interview a panel of consumer over time. Toon (2000 in Frees
2004), acknowledges Engels 1857 budget surveys as the earliest application of longitu-
dinal data. In this survey Engel collected data on the expenditure pattern from the same
set of subjects over a period time. The aim was to study expenditure on food as a func-
tion of income. Panel data modelling and estimation techniques were developed in the
second half of the twentieth century. Early applications of this technique are those of
Kuh (1959), Johnson (1960), Mundlak (1961) and Hoch (1962) who used the fixed effect
models (explained later in the text) and Balestra and Nerlove (1966) and Wallace and
Hussain (1969) who used the random effect models (explained later in the text).
This chapter focuses on the application of panel data techniques in the tourism litera-
ture. The chapter is organized as follows. The next section explains panel data model-
ling technique emphasizing the difference between fixed and random effects. Dynamism
in modelling is introduced and the implication for model estimation is discussed. The
chapter then devotes a section to unit root and cointegration tests which is followed by
an illustration of the application of panel data analysis in tourism research, namely, in
tourism demand modelling and in explorations of the relationship between tourism and
economic growth. The conclusion spells out the limitations of this technique and direc-
tions for future research.

PANEL DATA MODELLING TECHNIQUES AND BENEFITS

Panel data analysis offers several advantages. The most obvious is that inferences are
performed using a larger sample and the lack of degrees of freedom is fairly unlikely to
occur. According to Baltagi (2005), more complex relationships can be modelled, for

127

DWYER 9781781001288 Part I.indd 127 25/07/2012 11:07

 128 Handbook of research methods in tourism

example temporal changes in cross-section can be analysed. One of the most important
advantages, however, is that panel data modelling allows for the control of heterogeneity
in the sample.
A standard approach to model the relationship between Y (dependent variable) and X,
a set of explanatory variables, is given below where eit is the stochastic error term which
takes into account the variation in the expected value of Y which cannot be explained
by the Xs.

yit 5 x9itbi 1 eit (7.1)

For example in a tourism demand model, Y can stand for the number of arrivals to
a particular destination while the Xs include factors affecting demand, such as income
in the home country, relative prices, marketing expenditure, transportation cost and
so on. The Xs or explanatory variables can be included in the model so long as they
are observable and measurable. There are, however, factors such as culture and other
unique characteristics of the individuals or groups under study which are not observable
or measurable but which influence the outcome of expected Y. These factors are referred
to as the heterogeneity and are not directly part of Equation (7.1). The effect is incorpo-
rated in the one of the bs, should they be correlated with the respective X or otherwise
included in eit. As a result, in the first case, the estimated b will not reveal the true effect of
the variation in X on expected Y. By modelling the relationship between X and Y, using
the panel data technique, the researcher is able to separate the effect of the heterogeneity
from that of b.
Suppose K number of subjects are observed over time. The response of subject Y1t
will tend to be similar to responses in previous years but different to the rest Y2t,Y3t ...
Yk 2 1,t. That is, there is uniqueness in the behaviour of this subject and this is assumed to
be constant over time. The uniqueness of Y1 can be attributed to factors such as culture,
and past experience (for a person) business practices, (for a firm), history, or system of
government (for a country). For example, the cultural and historical links that Australia
and the UK share may be expected to promote international tourism flows between the
two countries. In this respect, tourism flows to Australia from the UK are expected to
be influenced by factors which are unique to travellers from this source and not perti-
nent for travellers from other home countries. In a panel data model, the uniqueness of
each market in the sample is captured by mi which is the unobserved heterogeneity in the
sample as given in Equation (7.2) below.

yit 5 mi 1 x9itbi 1 eit (7.2)

The additional benefit of including mi in the model is that it offers a potential solution
to the problem of omitted variables and measurement errors in data. Lack of data is
the most common problem faced by researchers. In regression analysis, it often results
in the omission of relevant variables from the model. This may give rise to potential
model specification errors for example, due to difficulties in obtaining data on market-
ing expenditure, this variable is often omitted from tourism demand model, although a
priori, it is expected to be relevant in explaining demand.
Consider Equation (7.3), where Yit is determined by three variables.

DWYER 9781781001288 Part I.indd 128 25/07/2012 11:07

 Panel data analysis 129

yit 5 b1x1it 1 b2x2it 1 b3x3it 1 eit (7.3)

omitting x3 from the model will reduce Equation (7.3) to Equation (7.4), where the effect
of X3 is dampened by the eit. The actual model estimated is

yit 5 b1x1it 1 b2x2it 1 b3x3it 1 uit (7.4)

where uit 5 x3t 1 eit.

In this case the residuals include the effect of x3t and will display patterns leading to the
conclusion that the model may be suffering from serial correlation (Green, 1999). The
implications for the b^ s will depend on whether the Xits are correlated with X3t. If so, the
estimated coefficients will be biased. If on the other hand, the covariance between X3t and
the Xits is equal to zero, b^ s will be unbiased and consistent. According to Wooldridge
(2002), the panel data modelling technique offers an effective solution to this problem.
The inclusion of mi in Equation (7.4) will solve the problem as it will absorb the effect
of X3t. This solution is also applicable when measurement errors are present in the data.
Taking the example of tourism model, Seetaram (2010b) explains the complexities
that arise when faced with the computation of airfare elasticities. Airfare data are often
plagued with measurement errors which arise mainly because of the wide array of airfares
and travel class categories which are prevalent in the market. This makes the task of the
researcher complex as often no choice is left but to use an average airfare to represent the
transportation cost to the destination. Average airfare is not always a good representa-
tion of actual airfare. Suppose that airfare, x3t is measured with errors such that the actual
variable which is included in the model is x* 3t 5 x3t 1 vt. The model estimated is given by:

yt 5 b1x1 1 b2x2 1 b3x3t* 1 hit (7.5)

where hit 5 eit 1 vt.

In Equation (7.5), the stochastic error term is given by ht. The cov (hi, x3t*) 5 2bd2v
where d2v is the variance of the measurement error (Green, 1999). This violates the crucial
assumption of non-correlation between any explanatory variable and the residuals of the
equation (Green, 1999). As a consequence, all the Ordinary Least Squares (OLS) estima-
tors, b^ s will be biased and inconsistent and b^ 3 will be attenuated. If instead the relation-
ship between Y and the explanatory variables are modelled using the format of Equation
(7.2), the measurement errors, vt, will be absorbed by the unobserved heterogeneity mi
leaving eit free from its effect.

Fixed Effect and Random Effects

There are two ways of modeling mi, namely the fixed and random effect. The choice
between these two depends on whether mi is correlated to any of the other explanatory
variables of the model (Wooldridge, 2002). Equation (7.2) is formulated using the fixed
effect (FE) technique. This method assumes that the heterogeneity in the model is mi time
invariant and specific to the individual group. In Equation (7.2), the slopes are fixed but
the intercepts vary for each cross section. This is equivalent to adding a dummy specific
for each cross section which is why it is also referred to as the Least Square Dummy

DWYER 9781781001288 Part I.indd 129 25/07/2012 11:07

 130 Handbook of research methods in tourism

Variable (LSDV) method. The slopes are treated as constant across group and across
time. It is however, possible to allow the slope to vary across groups, across time or both
(Hsiao, 2003). The rationale behind this modelling approach is that since mi accounts for
time invariant characteristics of the group, it removes the pernicious effect of omitted
variables (Allison, 2005). Fixed effect is often chosen as a precaution against omitted
variable bias. The drawback is that mi cannot be used to assess the effect of characteristics
which change overtime.
The FE technique explores the relationship between explanatory and dependent varia-
bles within one individual group. For each group, the variations of the all variables from
their mean values are considered and the estimated coefficients are also known as the
within estimates. This can be a limitation of the FE method as in-between variations are
ignored. Furthermore, only the effect of variables with sufficient variability can be ana-
lysed. For example, in a longitudinal study of an individual tourists perception of the
quality of destination attributes, the effect of gender and ethnicity cannot be analysed.
The modeller is expected to make a trade-off between sample variability and omitted
variable bias (Allison, 2005). However, as explained before the effects of these time-
invariant factors are controlled in the FE model. In circumstances when the in-between
variation is not relevant, the FE model makes use of maximum information, yielding
error terms with smaller variations (Allison, 2005). FE models may additionally include
an error component which changes over time but not for each unit, tt. tt is treated as a
constant in the model.

yit 5 x9itb 1 mi 1tt 1 eit (7.6)

Taking the example of the tourist demand model, consider a sample which includes
arrivals from ten sources over a period of 10 years. The aim is the find the income and
price elasticities of demand. Each market is a group in the sample. muk will take into
account all characteristics of the UK other than price and income that will influence
arrivals to Australia. The UK as a market has certain characteristics which may or may
not influence income and relative prices in the country. For example, mi can stand for
system of government, democracy, which may or may not be related to income level in
the UK. However, in the instant that there exists such a relationship, then the inclusion
of mi controls for the effect of democracy and allows the estimation of the net effect of
income on number of arrivals from the UK. The FE technique also assumes that m is
unique to the UK and is not correlated to the characteristics of other countries in the
sample. Any correlations between muk and musa are ignored.
If, however, the mis are correlated to one another and to the error terms of other
groups, the resulting variance will be high making statistical inference dubious. A
better approach in this case will be to use the random effect (RE) technique. In the RE
approach, variation across entities is assumed to be random and uncorrelated with inde-
pendent variables in the model. The RE model is given as

yit 5 x9itb 1 mit 1 tt 1 eit (7.7)

mit is referred to as the between-group error. The advantage of RE is that since variation
across the sample are considered, it permits the study of time invariant factors such as

DWYER 9781781001288 Part I.indd 130 25/07/2012 11:07

 Panel data analysis 131

gender, ethnicity and race in the model. The RE method uses variations both within and
between individuals and typically has less sampling variability than fixed effects methods
(Allison, 2005). The problem, however, is that all relevant measurable variables need to
be included in the model and data on a few may not be available therefore leading to
omitted variable bias in the model.
The choice between FE and RE depends on whether mi is correlated to any of the other
explanatory variables of the model (Wooldridge, 2002). When such a correlation exists,
the fixed effect technique is superior. Otherwise, the random effect is more parsimonious
and gives more efficient estimates (Wooldridge, 2002). A formal test for assessing the
correlation between the unobserved heterogeneity and other explanatory variable is the
Hausman (1978) specification test. In the tourism literature, the FE method has been
more frequently applied since the groups under observations are often markets, or desti-
nations which have characteristics which influence the other explanatory variables of the
model. The rest of the chapter will focus on the FE modelling method.

Dynamic Panel Data Models

By nature, all panel data models are dynamic since they take into account the time
series dimension of the sample. However, functions which specifically model the effect
of lagged dependent variables are referred to as dynamic panel data models. A general
dynamic panel data model with FE effect is given as Equation (7.8) below.

yit 5 giyi,t 2 1 1 b0 1 x9iktbk 1 mi 1 eit (7.8)

It is assumed that:

1. mi ~ (N, sm2) and eit ~ (N, s2e) where sm2 0 and s2e > 0
2. The explanatory variables are strictly exogenous, that is they are not correlated with
the error terms. i.e., E(eit ejs) 5 0 for i j or t s
3. The unobserved heterogeneity, if it is present, is random. i.e., E(mi mj) 5 0 for i j
4. The unobserved heterogeneity is uncorrelated within the countries and with the
error i.e., E(mi ejs) 5 0 for 5 i, j, t, s
5. The explanatory variables are strictly exogenous, that is they are not correlated with
the error terms. i.e., E(xit ejs) 5 0 for 5 i, j, t
6. The unobserved heterogeneity are correlated with the predetermined variables i.e.,
E(xit mj) 5 0 for 5 i, j, t
7. gi0 is uncorrelated with the error term i.e., E(gi0 ejt) 5 0 for 5 i, j, t
8. gi0 can be correlated with the unobserved heterogeneity. i.e., E (gi0 mj) 5 unknown for
5 i, j

Dynamic panel data analysis is becoming increasingly popular in the tourism litera-
ture modeling. In tourism demand model, gi accounts for destination loyalty and repeat
visitations. It takes into account the extent to which current visits are dependent on the
number of past visits. It takes into account the effect of habit persistence in demand and
the extent to which consumer react to ex-post information available. gi is an indication
of the efficiency of information diffusion through word-of-mouth.

DWYER 9781781001288 Part I.indd 131 25/07/2012 11:07

 132 Handbook of research methods in tourism

Generally, the functional form utilized for Equation (7.8) is that of double logarithm
implying that the b^ s are the short term elasticities. The long term elasticities the long term
elasticities b^ * may be obtained by

b^
b^ * 5 (7.9)
12g

The most widely used estimation technique for dynamic panel data sets in the tourism
literature has been has been the Arellano Bond (1991). Examples of studies which have
applied this technique are Garn-Muos (2006), Khadaroo and Seetanah (2007) and
Eugenio-Martn et al. (2004).

Estimation Technique

Estimating b^ using standard LSDV method yields biased and inconsistent estimators.
b^ LSDV estimator, also referred to as the covariance estimator, is given by:
21
b^ LSDV 5 c a a (xit 2 xit) (xit 2 xit) r d c a a (xit 2 xit) (yit 2 yit) r d
N T N T
(7.10)
i50 t51 i50 t51

where yit 5 1/TSTt51yit and x 5 1/TSTt51xit, and b^ LSDV is the estimated true coefficient of
the exogenous variable xit. xit and yit are the mean of xit and yit respectively. b^ LSDV will be
biased and inconsistent unless T S ` (Anderson and Hsiao, 1981; Arellano and Bond,
1991; Judson and Owen, 1999; Kiviet, 1995; Nickell, 1981). This occurs because in
Equation (7.8), yit21 will be correlated with the mean of the stochastic error term models
eit by construction and will be correlated to eit21 which is contained in eit (Hsiao, 2003).
Anderson and Hsiao (AH) (1981) and Arellano and Bond (AB) (1991) show that the
bias may be reduced by first differencing Equation (7.8) and using the lagged level values
of the yit as instruments. Consider Equation (7.11) below which is similar to Equation
(7.8) but for simplicity, the vector of exogenous variables x is left out.

yit 5 gyit21 1 mi 1 eit (7.11)

i 5 1, 2 ... N, t 5 1, 2 ... T

mi is the fixed effect which is the cause of the bias in the estimation by LSDV. To eliminate
mi, Anderson and Hsiao (1981) suggest that first difference transformation be applied to
Equation (7.11). First differencing Equation (7.11) gives the following:

(yit 2 yit21) 5 g (yit21 2 yit22) 1 (eit 2 eit21) (7.12)

i 5 1, 2 ... N, t 5 1, 2 ... T

Equation (7.12) is a first difference autoregressive process of order one with no exog-
enous regressors. Dyit21 5 (yit21 2 yit22) , is correlated with the error (eit 2 eit21) . The
second lag yit22, and the first difference of this second lag, Dyit22 5 (yit22 2 yit23) , are

DWYER 9781781001288 Part I.indd 132 25/07/2012 11:07

 Panel data analysis 133

possible instruments, since they are both correlated with (yit21 2 yit22) but are uncorre-
lated with (eit 2 eit21) , as long as the eit themselves are not serially correlated (Anderson
and Hsiao,1981). Using the second lag and the first difference of this second lag as instru-
mental variables, two estimators g^ IV and g^IV * can be developed. These are given in (7.13)
and (7.14).
N T

a a (yit 2 yit21) (yit22 2 yit23)

i51 t53
g^ IV 5 N T
(7.13)
a a (yit21 2 yit22) (yit22 2 yit23)
i51 t53

N T

a a (yit 2 yit21) yit22

i51 t52
g^IV
*5 N T
(7.14)
a a (yit21 2 yit22) yit22
i51 t52

These estimators are consistent when N S ` or T S ` or both. AB argue that more

efficient estimators can be obtained by taking in additional instruments whose validity is
based on orthogonality between lagged values of the dependent variable yit and the errors
eit. These results are confirmed by Kiviet (1995) and Judson and Owen (1999). However
the bias persists in samples with small T (Kiviet, 1995; Judson and Owen, 1999). In fact
it increases with the value g and decreases with T (Kiviet, 1995). An estimator that relies
on lags as instruments under the assumption of white noise errors will lose its consistency
if the errors are serially correlated (Kiviet, 1995).
Using simulations to generate data, Judson and Owen (1999) performed an exercise
with a panel of varying size. The number of cross sections, N, takes the value of 20 or
100 and the number of time periods, T, is given the values of 5, 10, 20 or 30. Judson and
Owen (1999) show that although the value of the bias falls as T increases, it is neverthe-
less still considerable at T 5 30 and can be as high as 20 per cent of the true value of the
parameter. They prove that estimates resulting from the AB technique have higher vari-
ances than the LSDV estimators supporting earlier results of Kiviet (1995).
LSDV estimates are more efficient than any other class of estimates developed for
autoregressive panel data models (Judson and Owen, 1999; Kiviet, 1995). The removal
of the bias in LSDV estimates opens the possibility of obtaining more robust estimates
(Kiviet, 1995). Kiviet (1995) evaluated the bias in the true parameters based on a
Monte Carlo study and developed a method to correct for potential bias in the esti-
mated parameters when the true parameters are known. On a practical level, however,
true parameters are seldom known, in which case Kiviet (1995) suggests that estimates
be obtained using the techniques proposed by AH (1981) and AB (1991). These esti-
mates can then be corrected for the bias by applying the Kiviet (1995) corrected LSDV
(CLSDV) method. This method is only applicable to balanced samples samples which
contain the same number of observations for each cross section. In the tourism litera-
ture, authors such as Corts-Jimnez (2008) in her study of regional tourism in Spain
and Italy and Soukiazis and Proena (2008) examining regional tourism in Portugal have
employed this technique.

DWYER 9781781001288 Part I.indd 133 25/07/2012 11:07

 134 Handbook of research methods in tourism

Unit Root and Cointegration Tests

Classical statistical inferences rely on data being mean reverting. However, economic var-
iables which tend to evolve over time are not always stationary and failure to account for
this will result in spurious regression results. To circumvent such problems, unit roots are
carried out to ascertain that regression results are valid. However, while testing for unit
root is standard in the time series literature, it is quite recent in panel data (Baltagi, 2005).
In the tourism literature although not very common, the availability of samples with
fairly large time dimensions has resulted in more testing for unit root in the panel data set
up, for example, Lee and Chang (2008) in a study of tourism development and economic
growth and Seetaram (2010a, 2012) in the context of Australian outbound tourism.
In the panel data setup, panel unit roots tests have higher power than unit root tests
based on individual time series for each of the cross section, since the later performs
poorly when data periods are short (Baltagi, 2005; Banerjee, 1999; Banerjee et al., 2004;
Levin et al., 2002; Im et al., 2003 and Pedroni, 1999). Several tests for unit roots in
panel data have been developed. A few examples are Choi (2001), Breitung (2000) and
Maddala and Wu (1999). According to Baltagi (2005), however, the two most efficient
tests for stationarity in a panel data setting are Levin, Lin and Chu (hereafter LLC, 2002)
and Im et al. (hereafter IPS, 2003). The fundamental difference between these two tests
rests on the assumption made regarding the autoregressive process (Baltagi, 2005):

1. LLC assumes that the autoregressive process is common for all cross sections, that
is r 5 ri in Equation (7.15) give below.
2. IPS assumes that the persistence parameter, ri is allowed to vary across the cross sec-
tions.

Both tests are based on estimating the following equation:

zi
Dyit 5 riyit21 1 a fij Dyit2j 1 eit (7.15)
j51

yit is the dependent variable being tested for unit root. D denotes the first difference in
the dependent variable yit. zi is the number of lags to be included in the testing. eit are the
error terms. ri and fij are parameters.
LLC assumes that the error term eit is independent across the units of the sample and
have a fixed variance. LLC tests the hypothesis that each of the series in the panel con-
tains a unit root against the hypothesis that all individual series are stationary. This can
be written as:

H0: r 5 ri 5 0 for all i ~ All of the individuals have a unit root.

H1: r 5 ri < 0 for all i ~ All the individual series are stationary.

The IPS test assumes that the panel is balanced. It hypothesizes that each of the series
contains a unit root against the alternative hypothesis, that at least one of the series is
stationary. These can be represented as follows:

DWYER 9781781001288 Part I.indd 134 25/07/2012 11:07

 Panel data analysis 135

H0: ri 5 0 for all i. ~ All the series have unit roots.

ri 5 0 for i 5 1, 2, . . ., N1
H1: e ~ Some of the series have unit roots.
r , 0 for i 5 N 1 1, N 1 2, . . ., N
are the cross sections in the panel data set. This method involves regressing the individual
series for each of the cross sections. The critical value for the hypotheses is obtained by
taking the average of the student t-statistics for the ri from the individual regressions.
This is given by:

a a tiT (ri) b
N

i5t
tNT 5 (7.16)
N
Series containing unit roots arenon-stationary processes which have time-varying mean
and variance that increases as sample size grows (Baltagi, 2005).
When variables are individually integrated of the same order, such as the ones in
this study, a linear combination of these variables can still be stationary (Baltagi, 2005;
Banerjee, 1999; Banerjee et al., 2004; Pedroni, 2004). If the series are found to be cointe-
grated then there must be at least one cointegrating vector which renders the combina-
tion of variables stationary. Furthermore, it implies that there is a long-run relationship
among the variables.
The conventional cointegration tests suffer from low power when applied to samples
with a small time dimension, such as the one used in this chapter (Baltagi, 2005;
Banerjee, 1999; Banerjee et al., 2004 Pedroni, 2004). Panel cointegrating techniques
have been developed to allow researchers to pool information regarding common
long-run relationships from across the panel. In addition, such techniques allow the
associated short-run dynamic and fixed effects to be heterogeneous across the different
members of the panel (Baltagi, 2005; Banerjee, 1999; Banerjee et al., 2004; Pedroni,
1999, 2004). This chapter focuses on the Pedroni (1999) test as it is one of the most
widely used tests for cointegration in panel data. The first step in this test is to estimate
a cointegrating relationship with fixed effects and heterogeneous time trends for each
of cross section of the study individually. Then the cointegration tests are performed
based on the residuals obtained (Banerjee, 1999; Banerjee et al., 2004; Pedroni, 1999,
2004).
Pedroni (1999) proposes seven tests for cointegration in the panel data frame-
work. Four of these tests are referred to as the panel cointegrating statistics or the
within-dimension based statistics (Pedroni, 1999, p. 658). In these tests, he assumes
that there is a common cointegrating relationship among the variables. For these four
tests, the residuals are pooled across the time dimension of the panel. An autoregres-
sive function is regressed and the autoregressive coefficient is given by ui. This method
assumes that ui 5 u, that is, it is common for all countries. The alternate hypothesis
is that there is a cointegrating relationship for all the cross sections given as (Pedroni,
1999):

H0: : u 5 ui 5 1 ~ All of the individuals are not cointegrated

H0: : u 5 ui < 1 ~ All of the individuals are cointegrated

DWYER 9781781001288 Part I.indd 135 25/07/2012 11:07

 136 Handbook of research methods in tourism

By contrast, the remaining three tests are called the group mean cointegrating statistics
or the between-dimension. These tests statistics are based on pooling the residuals of the
regression along the cross sections of the panel (Pedroni, 1999). In these tests, estima-
tors average the individually estimated autoregressive coefficient for each cross section
(Pedroni, 1999). The hypotheses here are given by:

H0: : ui 5 1 ~ All of the individuals are not cointegrated

H0: : ui < 1 ~ All of the individuals are cointegrated

The group mean statistics can be considered as more accurate, as they allow for more
heterogeneity among the countries, and produce consistent estimates (Pedroni, 2001).
The higher value of the group mean statistics can be considered to be a more accurate
representation of the average long-run relationship (Pedroni, 2004).
It can be noted, however, that in general the unit root cointegration tests increase the
probability of determining if data are stationary or not and whether they are cointe-
grated (Banerjee, 1999; Banerjee et al., 2004). However, one limitation of these tests is
that they assume no cross-sectional correlation in the sample (Banerjee, 1999; Banerjee
et al., 2004). Banerjee et al. (2004) showed that the results of cointegration tests were
susceptible to dependence among the cross sections. This means that the power of the
tests is reduced in cases where the cross sections are not independent. In spite of this, in
panel data sets, the problem of spurious regression results is not expected to be as serious
as in pure time series. As demonstrated by Phillips and Moon (2000), noise in time series
regression is lessened by pooling cross-section and time series observations, implying
that the model may be estimated in level form without risking spurious results.

APPLICATIONS OF PANEL DATA TECHNIQUES IN TOURISM

DATA ANALYSIS

The empirical literature based on the panel data methods and applied to the tourism
sector, can be divided into two broad areas. The first concerns the determinants of
tourism demand. The topic of demand modelling is not recent and goes back to Rugg
(1972). The study of international tourism arrivals and receipts is a vibrant area of
research and it cannot be confined to gravity models due to factors such as seasonality,
cultural links or time constraints which are particular to the consumption of tourism
products. The second area consists mainly of analysis of the link between tourism and
economic growth. Since, several destinations, for example, Spain, Italy and Portugal,
have simultaneously experienced economic development and expansion of their tourism
sector, especially following the adoption of the Millennium Goals for Economic
Development, many authors have attempted to estimate the real impact of international
tourism on economic growth.
In tourism demand modelling, after a short period of application of standard and
classic panel data models (OLS with fixed effects or random effects), the literature is
characterized by a high volume of dynamic panel models. Additionally, from the second
half of 2000 onwards, several of the works employ cointegration techniques. The first

DWYER 9781781001288 Part I.indd 136 25/07/2012 11:07

 Panel data analysis 137

empirical papers, based on the classic panel data methods, were published in the begin-
ning of the 2000s. These studies focused generally on the prices elasticities or on the
impact of political risk and violence on international tourism arrivals. For example,
Ledesma- Rodrguez and Navarro-Ibez (2001) used annual data from 1979 to 1997
to study factors affecting arrivals to Tenerife from 13 markets. They found arrivals to
be elastic with respect to income and inelastic with respect to prices and transport cost
in the long run. Espinet et al. (2003) used panel data with random effects for a hedonic
evaluation based on 86 000 prices between 1991 and 1998. The data concerned hotels in
the southern Costa Brava region. Their results indicate a real and significant effect from
the quality to the price.
Eilat and Einav (2004) used multinomial logit estimations, based on bilateral data
during the period 198598, to study the leisure tourism determinants. This was the first
study based on a three-dimensional panel data (year, destination, origin). They showed
that political risk is a very important determinant of the tourist arrivals. The most recent
paper based on the traditional panel data method is that of Arita et al. (2011). They
analysed the impact of ADS (Approved Destination Status, beginning of the 1990s)
agreements on international tourism arrivals in China. They integrated, in the estima-
tion, fixed effects, to compare results before and after ADS and also between destinations
inside and outside the ADS.
As the relevant data available is often annual, there are many empirical works which
use dynamic panel data methods to test long-run relationships. The most commonly used
estimation technique in the literature is that of AB although authors such as Maloney
and Montes Rojas (2005) made used of General Method of Moment (GMM) suggested
by Blundell and Bond (1998). They measured the tourism price elasticity in Caribbean
countries, with bilateral data (tourists came from the United States, Canada, United
Kingdom, Germany, Netherlands, Italy and Spain) from 1990 to 2002. They estimated
a large price elasticity of 4.9.
As mentioned above, the AB technique is the most widely used in the tourism litera-
ture. For example, Neumayer (2003) estimated the effect of violence, risks, freedom and
human rights violations on annual international tourism arrivals during 19772000.
This study was based on a sample which contained more than 100 countries. Neumayer
employed two methods for this estimation: a traditional data panel model with fixed
effects and a dynamic panel data model. He found that in most cases, these explanatory
variables had a real and significant effect on tourism. Garn-Muos and Montero-Martin
(2007) studied yearly data from 1991 to 2003 to assess factors affecting the number of
arrivals to the Balearic Islands. They found a high level of consumer loyalty to tourism
in the Balearic Islands and recommended that suppliers of tourism products should raise
the quality of their products and should improve their brand image.
Khadaroo and Seetanah (2007) used data on arrivals to Mauritius during the period
1978 to 2003 to assess the relative importance of transport infrastructure as a demand
determinant. Transport infrastructure was approximated by the net investment in land,
air and sea infrastructure at the current market price while non-transport infrastructures
were measured by net investment at current market price on communication, energy,
wastewater and defence infrastructure. Transport infrastructure was found to be an
important determinant of demand for travellers from Asia, Europe and the United
States while the latter two groups are also influenced by the non-transport infrastructure.

DWYER 9781781001288 Part I.indd 137 25/07/2012 11:07

 138 Handbook of research methods in tourism

However, they use a poor proxy for prices in their model. They sought to capture the
price effect by using the real value of the Mauritian rupees in US dollars. The deprecia-
tion or appreciation of the Mauritian rupee against the US dollar cannot be expected to
reflect changes in the cost of holidays for visitors from Africa, Asia and Europe. It is not
surprising that they found prices to be insignificant in determining international arrivals
to Mauritius.
Naud and Saayman (2005) analysed annual data from 1996 to 2000 to estimate
tourism arrivals for 43 African countries. Their estimation yielded a negative coefficient
for the lagged dependent variable suggesting that tourism in South Africa is taking
a downward trend as currently level of arrivals was negatively related to past levels.
Political stability was a key determinant of arrivals. Their model included lagged values
of explanatory variables. It is not unlikely that the variables are highly collinear to their
lagged values and therefore the model suffers from multicollinearity. This may explain
why only a few of their estimated coefficients were statistically significant.
A dynamic data panel model has the fundamental characteristics to establish the long-
run relationship between two variables. As the question of difference between short-
and long-run estimation is important (the results can differ as a short-run estimation
can be amplified or cancelled in the long run), the dynamic model is often used along
with another method of estimation which has the objective to estimate the short-run
impact. For example, Kuo et al. (2008) and Kuo et al. (2009) tried to estimate impact of
infectious diseases (Avian Flu) and severe acute respiratory syndrome (SARS) on inter-
national tourist arrivals in selected Asian, African and European destinations, for the
sample period of 2001 to 2006 and 2004 to 2006 respectively. The long-run estimation
is based on the GMM procedure and the short-run estimation is provided by ARMAX
models for each country. They established different results between short and long-run
estimations: SARS had a significant impact on tourist arrivals in the long- and short-run
whereas infectious diseases had only an impact on the long run. Garn-Muos (2006)
used annual data from 1992 to 2002 to estimate factors affecting arrivals to the Canary
Islands from 15 of its markets. Demand was found to be inelastic in the short run but
income and price elasticity was greater than one in the long run. Demand was elastic with
respect to changes in transport cost in the short run and in the long run.
Another way of comparing short and long-run results is to employ cointegration
techniques and dynamic model. Seetaram (2010a) and Seetaram (2012) analysed tourist
arrivals and departures in Australia during the period 19912008. After testing for sta-
tionarity and cointegration (based on the suggestion of Pedroni, 1999 and 2004), she
employed GMM and Corrected Least Square Dummy Variable, CLSDV (provided by
Kiviet, 1995), to calculate short-run and long-run elasticities. The justification for using
this estimation technique was that the temporal dimension was relatively small. For the
tourist arrivals, she found that demand is inelastic in the short-run, with respect to its
determinants, and elastic in the long-run. Concerning outbound tourism, the results
indicated an elasticity of migration on outbound tourism of 0.2 per cent and amplified in
the long run with a value of 0.6 per cent. However, contrary to the inbound model, there
were no effects detected from the price index in the outbound model.
Finally, another way is to use an estimator directly for the cointegrated model: Fully
Modified Ordinary Least Squares (FMOLS) or Dynamic Ordinary Least Squares
(DOLS). Seetanah (2010) employed a gravity model to test price and income elastici-

DWYER 9781781001288 Part I.indd 138 25/07/2012 11:07

 Panel data analysis 139

ties. However, this paper suffers from a problem of methodology as he integrates many
exogeneous variables into the regression. For a set of n variables, there can be up to n1
independent cointegrating vectors (Harris, 1995). If explanatory variables were cointe-
grated among them, there would more than n1 cointegrating vectors. So it is not possi-
ble to put explanatory variables that are cointegrated simultaneously into the regression
(FMOLS or DOLS).
Regarding tourism and economic growth, the lack of theoretical foundation explain-
ing the mechanism through which tourism causes economic growth has prompted several
authors to apply cointegration techniques in order to explain the causality between the
two variables. This has resulted in a small but growing number of empirical studies using
dynamic panel data techniques. The first studies have analysed the effect of a growing
tourism sector on the economic growth of the destination (tourism growth S economic
growth). Eugenio-Martn et al. (2004) estimated the effect from growth of tourism per
capita to income per capita with a sample including 21 Latin American countries from
1985 to 1998. They used a dynamic panel data model with AB estimator and they cat-
egorized the countries into three groups based on their respective income per capita:
rich, medium and poor. The results established that tourism led to economic growth for
medium and poor countries while the reverse was true for rich countries.
Corts-Jimnez (2008) analysed the impact of international and domestic tourism
arrivals on economic growth (by using the GDP per capita) for 17 regions of Spain and
20 regions of Italy during the period 19902000. She used two estimators for the dynamic
panel model: GMM and CLSDV. She found that tourism impacted positively on eco-
nomic growth in coastal and Mediterranean regions. Her results are, however, surprising
for the inland regions in her sample. Here she found that while domestic tourism fostered
economic growth, international tourism had the reverse effect.
Soukiazis and Proena (2008) studied the effect of tourism activity on the economic
convergence of Portuguese regions between 1993 and 2001. Convergence was approxi-
mated by the difference of GDP per capita. Three estimators were applied: GMM,
CLSDV and GLS (with random effects). The authors used bed capacity as a proxy for
tourism activity.
According to their results, a tourism capacity expansion of 1 per cent results in a
supplement of 0.01 per cent economic growth. Their results also indicated that tourism
accelerates economic convergence across the regions. Without tourism, the conver-
gence was estimated to take 11 years whereas when the effect of tourism was taken into
account, this reduced to only 10 years.
Sequeira and Mas Nunes (2008) used a sample of 94 countries (poor countries or
small countries, with a population less than 5 million) for the period 19802002. Three
indicators of tourism activity were employed (tourism arrival in population proportion,
tourism receipts as a percentage of exports and tourism receipts as a percentage of GDP).
They found that tourism had a positive effect on economic growth but only for the poor
countries.
In the tourism literature, the panel data cointegration technique has also been
employed in order to test for the causality between tourism growth and economic devel-
opment. Narayan et al. (2010) used cointegration tests to estimate long-run elasticities
of tourism to Pacific Islands (Tonga, Fiji, Solomon, Papua New Guinea) for the period
extending from 1988 to 2004. They then analysed the causality from tourism exports to

DWYER 9781781001288 Part I.indd 139 25/07/2012 11:07

 140 Handbook of research methods in tourism

GDP with an ECM model (as its structure permits to check at the short-run and at the
long-run). The tourism literature is not clear on the direction of the causality between
economic growth and tourism. There are schools of thoughts which suggest that eco-
nomic growth and development can also impact on tourism demand (economic growth
S tourism).
A few studies have empirically tested this link by using cointegration methods for
the data panel model. For example, Lee and Chang (2008) used panel cointegration
methods to determine the relationship and the causality between economic growth and
tourism development in 55 countries (developed and developing countries) covering
the time span of 1990 to 2002. After performing the IPS unit root and Pedroni (1999)
cointegration tests, they applied FMOLS estimator to estimate the long-run relationship
between tourism and economic developments. They then performed causality tests, the
results of which indicated a unidirectional causality for developed countries (tourism S
economic growth), whereas in developing countries there were bidirectional causality
relationships.
It is interesting to note that tourism has not only benefitted economic growth but also
international trade. Santana-Gallego et al. (2011) employed FMOLS and DOLS, while
Keum (2011) applied multiple dynamic causal patch analysis to investigate the causal-
ity between tourism and trade. Cassette et al. (2009) and Petit (2010) developed panel
data cointegration models to analyse whether international trade of goods and services
had any effect on income inequalities. Their results indicate that international trade
in tourism services results in income inequalities disadvantaging low-skilled workers.
Whilst tourism activities seem to benefit economic growth, the low-skilled workers seem
to be losing from globalization of the tourism services.

LIMITATIONS OF PANEL DATA ANALYSIS

The first disadvantage of this technique is that it is fairly complex to estimate and data
requirements are high. Observing a number of individuals over a period of time usually
results in data collection that can be tedious and expensive (Baltagi, 2005). From a sta-
tistical perspective, panel survey designs have some inherent disadvantages as noted in
Kitamura (2000, p. 127).

1. Respondents may find it cumbersome to participate regularly in the same survey,

which results in increasing non-response.
2. Attrition or dropout rate from the sample can be high.
3. Over time, the accuracy of data collection may decline. This is known as panel
fatigue.
4. The response of individuals may be influence by their responses from previous par-
ticipations.

These disadvantages can, however, be addressed, although solutions do come at a cost.

Solution proposed, include refreshing the survey design and adding fresh participants
at later stages. For more in-depth analysis of attrition in panel data, see Alderman et al.
(2001), Fitzgerald et al. (1998) and Uhrig (2008).

DWYER 9781781001288 Part I.indd 140 25/07/2012 11:07

 Panel data analysis 141

CONCLUSION AND FUTURE DIRECTIONS

The use of panel data modelling techniques is becoming more common in the analysis
of tourism data. In less than a decade, the literature has progressed from the develop-
ment of standard static fixed effect and random effect models in the early 2000s to the
more sophisticated dynamic panel data cointegration models. So far, the application
of this technique has been restricted to tourism demand studies and others which have
explored the relationship between the expansion of tourism sectors and economic
growth or international trade. It can be extended to other topics of research within the
tourism context.
One interesting area is the investigation of the supply side of the tourism industry.
For example, a scrutiny of regional differences in the productivity of this industry will
benefit from panel data modelling techniques, as models developed may be used to
control for the heterogeneity of each region. Other types of studies where this method
can prove to be useful are examinations of labour markets and their outcomes across
different districts within a destination or in international comparisons of destinations.
Examples of problems that can be addressed are analysis of wage differentials and
gender bias, or human capital formation and return to education within the tourism
industry.
While panel data techniques are powerful, and generally yield reliable estimates,
they nevertheless suffer from some weaknesses. Estimations can be complex and data
requirements are fairly large especially for dynamic panel models. The main limitation,
however, is that the panel data survey design is inevitably subject to problems related to
attrition. This drawback, however, does not seem to have affected tourism research as
the existing studies are based on secondary data or macro data where this occurrence
is either easier to deal with or less frequent. It is expected, however, that future research
may be based on survey data when researchers seek to find answers to the behaviour of
tourists using microlevel data.
Regarding unit roots and cointegration tests, it is noted that the tests applied can
be restrictive in that they assume that there is no cross-sectional correlation in the
sample. In the instances that this assumption is overly binding and inappropriate,
researchers may circumvent it by using alternative tests such as those of Westerlund
(2007) who has designed tests which perform well on smaller samples and have higher
power. Additionally, Westerlund (2006) and Westerlund and Edgerton (2006) have
developed cointegration tests for panel data sets with structural breaks. These tests
will be useful for researchers who are attempting to examine whether the effect of
shocks on the tourism data are permanent or transitory. The exploitation of these tests
will be made easier when software such as STATA, which is the dominant application
for estimating panel data models, catches up with theoretical development in this field
and incorporates it in newer versions of the program. Finally, it is expected that in the
future, the literature will see more use of panel-ECM and that application of panel
data techniques for analysing tourism data will extend to non-linear models including
binary models.

DWYER 9781781001288 Part I.indd 141 25/07/2012 11:07

 142 Handbook of research methods in tourism

REFERENCES

Alderman, H., J. Behrman, H.P. Kohler, J.A. Maluccio and S. Watkins (2001), Attrition in longitudi-
nal household survey data: some tests from three developing countries, Demographic Research, 5 (4),
79124.
Allison, P. (2005), Fixed Effects Regression Methods for Longitudinal Data Using SAS, Cary, NC: SAS
Press.
Anderson, T.W. and C. Hsiao (1981), Estimation of dynamic models with error components, Journal of the
American Statistical Association, 76, 589606.
Arellano, M. and S. Bond (1991), Some tests of specification for panel data: Monte Carlo evidence and an
application to employment equations, Review of Economic Studies, 58, 277297.
Arita, S., C. Edmonds, S. La Croix and J. Mak (2011), Impact of approved destination status on Chinese
travel abroad: an econometric analysis, Tourism Economics, 17 (5), 983996.
Balestra, P. and M. Nerlove (1966), Pooling cross section and time series data in the estimation of a dynamic
model: the demand of natural gas, Econometrica, 34 (3), 585612.
Baltagi, B. (2005), Econometric Analysis of Panel Data, 5th Edition, New York: Wiley and Sons.
Baltes, P.B. and J.R. Nesselroade (1979), History and rationale of longitudinal research, in J.R. Nesselroade
and P.B. Baltes (eds), Longitudinal Research in the Study of Behavior and Development, New York: Academic
Press, pp. 139.
Banerjee, A. (1999), Panel data unit roots and cointegration: an overview, Oxford Bulletin of Economics and
Statistics, 61, 607629.
Banerjee, A., M. Marcellino and C. Osbat (2004), Some cautions on the use of panel methods for integrated
series of macroeconomic data, Econometrics Journal, 7, 322340.
Blundell, R.W. and S.R. Bond (1998), Initial conditions and moment restrictions in dynamic panel data
models, Journal of Econometrics, 87, 115143.
Breitung, J. (2000), The local power of some unit root tests for panel data, in B. Baltagi (ed.), Advances in
Econometrics, Vol. 15: Nonstationary Panels, Panel Cointegration, and Dynamic Panels, Amsterdam: JAI
Press, pp. 161178.
Cassette A., N. Fleury and S. Petit (2009), Income inequality and international trade: short and long-run evi-
dence and the specific case of tourism services, paper presented to the 2nd Conference of the International
Association for Tourism Economics, Chiang Mai, Thailand, December 2009.
Choi, I. (2001), Unit root tests for panel data, Journal of International Money and Finance, 20, 249272.
Corts-Jimnez, I. (2008), Which type of tourism matters to the regional economic growth? The cases of Spain
and Italy, International Journal of Tourism Research, 10, 127139.
Eilat, Y. and L. Einav (2004), Determinants of international tourism: a three dimensional panel data analysis,
Applied Economics, 36 (12), 13151327.
Espinet, J.M., M. Saez, C. Coenders and M. Fluvi (2003), Effect on prices of the attributes of holiday hotels:
a hedonic prices approach, Tourism Economics, 9 (2), 165177.
Eugenio-Martn, J.L., N.M. Morales and R. Scarpa (2004), Tourism and economic growth in Latin American
countries: a panel data approach, FEEM working paper No. 2004.26.
Fitzgerald, J., P. Gottschalk and R. Moffitt (1998), An analysis of sample attrition in panel data: the Michigan
panel study of income dynamics, The Journal of Human Resources, 33, 251299.
Frees, E.W. (2004), Longitudinal and Panel Data: Analysis and Applications in the Social Sciences, West Nyack,
NY: Cambridge University Press.
Garn-Muos, T. (2006), Inbound international tourism to Canary Island: a dynamic panel data model,
Tourism Management, 27 (1), 281291.
Garn-Muos, T. and L.F. Montero-Martn (2007), Tourism in the Balearic Island: a dynamic model for
international demand using panel data, Tourism Management, 27 (5), 12241235.
Green, W. (1999), Econometric Analysis, 4th Edition, Upper Saddle River, NJ: Prentice Hall.
Harris, R.I.D. (1995), Using Cointegration Analysis in Econometric Modelling, Upper Saddle River, NJ:
Prentice Hall.
Hausman, J.A. (1978), Specification tests in econometrics, Econometrica, 46 (6), 12511271.
Hoch, I. (1962), Estimation of production function parameters combining time series and cross section data,
Econometrica, 30 (1), 3453.
Hsiao, C. (2003), Analysis of Panel Data, 2nd edition, New York: Cambridge University Press.
Im, K.S., M.H. Pesaran and Y. Shin (2003), Testing for unit roots in heterogeneous panels, Journal of
Econometrics, 115 (1), 5374.
Johnson, J. (1960), Econometric Methods, New York: McGraw-Hill.
Judson, R. and A. Owen (1999), Estimating dynamic panel models: a guide for macroeconomist, Economics
Letters, 65, 5378.

DWYER 9781781001288 Part I.indd 142 25/07/2012 11:07

 Panel data analysis 143

Keum, K. (2011), International tourism and trade flows: a causality analysis using panel data, Tourism
Economics, 17 (5), 949962.
Khadaroo, J. and B. Seetanah (2007), Transport infrastructure and tourism development, Annals of Tourism
Research, 34 (4), 10211032.
Kitamura, R. (2000), Longitudinal methods, in D.A. Hensher and K.J Button (eds), Handbook of Transport
Modelling, Netherlands: Pergamon, pp.1329.
Kiviet, J.F. (1995), On bias, inconsistency and efficiency of various estimators in dynamic panel data models,
Journal of Econometrics, 68 (1), 5378.
Kuh, E. (1959), The validity of cross-sectionally estimated behavior equations in time series applications,
Econometrica, 27 (2), 197214.
Kuo, H.-I., C.-L. Chang, B.-W. Huang, C.-C. Chen and M. MacAleer (2009), Estimating the impact of avian
flu on international tourism demand using panel data, Tourism Economics, 15 (3), 501511.
Kuo, H.-I., C.-C. Chen, W.-C. Tseng, L.-F. Ju and B.-W. Huang (2008), Assessing impacts of SARS and avian
flu on international tourism demand to Asia, Tourism Management, 29, 917928.
Ledesma-Rodrguez, F.J. and M. Navarro- Ibez (2001), Panel data and tourism: a case Study of Tenerife,
Tourism Economics, 7 (1), 7588.
Lee, C.-C. and C.-P. Chang (2008), Tourism development and economic growth: a closer look at panels,
Tourism Management, 29, 180192.
Levin, A., C.-F. Lin and C.-S.J. Chu (2002), Unit root tests in panel data: asymptotic and finite-sample prop-
erties, Journal of Econometrics, 108 (1), 120143.
Maddala, G.S. and S. Wu (1999), A comparative study of unit root tests with panel data and a new simple
test, Oxford Bulletin of Economics and Statistics, 61, 631652.
Maloney, W.-F. and G.-V. Montes Rojas (2005), How elastic are sea, sand and sun? Dynamic panel estimates
of the demand for tourism, Applied Economic Letters, 12, 277280.
Mudlak, Y. (1961), Empirical production function free of management bias, Journal of Farm Economics, 43
(1), 4456.
Narayan, P.K., S. Narayan, A. Prasad and B.C. Prasad (2010), Tourism and economic growth: a panel data
analysis for Pacific Island countries, Tourism Economics, 16 (1), 169185.
Naud, W.A. and A. Saayman (2005), Determinants of tourist arrivals in Africa: a panel data regression
analysis, Tourism Economics, 11 (3), 365391.
Neumayer, E. (2003), The impact of political violence on tourismdynamic econometric estimation in a cross-
national panel, Journal of Conflict Resolution, 48 (2), 259281.
Nickell, S. (1981), Biases in dynamic models with fixed effects, Econometrica, 49 (6), 14171426.
Pedroni, P. (1999), Critical values for cointegration tests in heterogeneous panels with multiple regressors,
Oxford Bulletin of Economics and Statistics, 61 (November), 653670.
Pedroni, P. (2004), Panel cointegration; asymptotic and finite sample properties of pooled time series tests with
an application to the PPP hypothesis, Econometric Theory, 20 (3), 597625.
Petit, S. (2010), An Analysis of International Tourism: Production Fragmented, Two-Way Trade,
Redistributive Effects, PhD thesis, University of Lille 1, France.
Phillips, P.C.B. and H.R. Moon (2000), Nonstationary panel data analysis: an overview of some recent devel-
opments, Econometric Reviews, 19, 263286.
Rugg, D. (1972), The choice of journey destination: a theoretical and empirical analysis, Review of Economics
and Statistics, 55 (1), 6472.
Santana-Gallego, M., F. Ledesma-Rodrguez and J.V. Prez-Rodrguez (2011), Tourism and trade in OECD
countries: a dynamic heterogeneous panel data analysis, Empirical Economics, 41 (2), 533554.
Seetanah, B. (2010), Using the panel cointegration approach to analyze the determinants of tourism demand
in South Africa, Tourism Economics, 16 (3), 715729.
Seetaram, N. (2010a), Use of dynamic panel cointegration approach to model international arrivals to
Australia, Journal of Travel Research, 49 (4), 414422.
Seetaram, N. (2010b), Computing airfare elasticities or opening Pandoras Box, Research in Transportation
Economics, Special Issue on Tourism and Transport, 26, 2736.
Seetaram, N. (2012), Estimating demand elasticities for Australias international outbound tourism, Tourism
Economics, (in press).
Sequeira, T.N. and P.M. Nunes (2008), Does tourism influence economic growth? A dynamic panel data
approach, Applied Economics, 40, 24312441.
Soukiazis, E. and S. Proena (2008), Tourism as an alternative source of regional growth in Portugal: a panel
data analysis at NUTS II and III levels, Portuguese Economic Journal, 7, 4361.
Uhrig, C.N. (2008), The nature and causes of attrition in the British Household Panel Study, ISER Working
Paper 200805.
Wallace, T.D. and A. Hussain (1969), The use of error components models in combining time-series with
cross-section data, Econometrica, 37 (1), 5572.

DWYER 9781781001288 Part I.indd 143 25/07/2012 11:07

 144 Handbook of research methods in tourism

Westerlund, J. (2006), Testing for panel cointegration with multiple structural breaks, Oxford Bulletin of
Economics & Statistics, 68 (1), 101132.
Westerlund, J. (2007), Testing for error correction in panel data, Oxford Bulletin of Economics and Statistics,
69 (6), 709748.
Westerlund, J. and D. Edgerton (2006), A simple test for cointegration in dependent panels with structural
breaks, Oxford Bulletin of Economics and Statistics, 70 (5), 665704.
Wooldridge, J. (2002), Econometric Analysis of Cross Section and Panel Data, Cambridge, MA: MIT Press.

DWYER 9781781001288 Part I.indd 144 25/07/2012 11:07

 8 The almost ideal demand system
Sarath Divisekera

INTRODUCTION

The classic paper by Deaton and Muellbauer (1980a) one of the top 20 papers which
appeared in American Economic Review during the first 100 years of its existence is an
established standard for applied demand analysis (Arrow et al., 2011). The fundamental
demand model established by this paper Almost Ideal Demand System or AIDS has
found widespread application in consumer demand analysis. The attraction of AIDS is
that it gives an arbitrary first-order approximation to any demand system; it satisfies the
axioms of choice (almost) exactly; it aggregates perfectly without invoking the assump-
tion of parallel linear Engel curves; and it has a functional form which is consistent with
known household budget data. As within the general economics literature, the AIDS
model is the principal established demand system employed by researchers in the tourism
field. Beginning from the work of White (1982) the earliest attempt to model tourism
demand based on a system approach to demand modelling to the most recent study
of Wu et al. (2011), the AIDS model has been the basis for modelling and estimation of
tourism demands.1 The key contributions include: Coenen and Eekeren (2003), De Mello
and Fortuna (2005), De Mello et al. (2002), Divisekera (2007, 2008, 2009a, 2009b, 2010a,
2010b, 2010c), Divisekera and Deegan (2010), Fujii et al. (1985), Han et al. (2006), Li
et al. (2004), Lyssiotou (2000), Mangion et al. (2005), OHagan and Harrison (1984),
Papatheodorou (1999), Sinclair and Syriopoulos (1993), White (1982, 1985), and Wu et
al. (2011).
In addition to the original non-linear AIDS model these studies employ its variants
and extensions including the linear, quadratic and dynamic versions. This may reflect,
among other things, complexities arising from modelling tourism demands that require
extensions to the base model, issues arising from particular data available that need
modifications, and adaptation of recent advances in the econometric methodology of
applied researchers. An understanding of the various issues confronting an applied
researcher who relies on the AIDS model to analyse tourism demand is a must, as
modifications and extensions are often made to accommodate and to deal with the prac-
tical problems the researcher faces. A systematic review of these issues in the context
of tourism demand modelling is the focus of this chapter. In this review, we intend to
provide a broad overview of the AIDS model focussing on its specification, properties,
extensions, modifications and applications to tourism.
The remainder of this chapter is organized as follows. We begin with a brief review
of the neoclassical theory of consumer behaviour, which is the cornerstone of modern
consumer demand analysis and the theoretical foundation of the AIDS model. Then
the specification of the AIDS model is presented and its properties are discussed. In the
following section, extensions and modifications to the base AIDS model are reviewed in
the light of the practical issues relating to their utilization in tourism demand analysis.

145

DWYER 9781781001288 Part I.indd 145 25/07/2012 11:07

 146 Handbook of research methods in tourism

These issues include incorporation of seasonality and structural change, and the short-
and long-run dynamics of tourism demands. In the final section, contributions to the
tourism demand literature are reviewed in the light of the application of the AIDS model
and its variants. In the final section we summarize key observations, we discuss possible
implications for policy formulation, and we conclude with some suggestions for future
research directions.

THEORETICAL BACKGROUND AND EVOLUTION OF THE

AIDS MODEL

The empirical analysis of consumer behaviour holds a central position in economic

analysis, and the underlying theory of consumer behaviour provides the structure for
model formulation and data analysis (Deaton, 1986). The theory of consumer behav-
iour postulates that consumers choose goods and services so as to maximize utility
subject to their budget constraint; the utility-maximizing consumers demand for any
commodity depends on the prices of all commodities available to them and on their
total expenditure on these commodities.2 This formulation presumes the existence of a
utility function that measures the level of satisfaction an individual achieves by consum-
ing goods and services. The utility function (u), representing consumer preferences, is
defined by

u 5 u(q) (8.1)

where q 5 (qi) is an n-element vector of levels of commodities consumed at a given point

in time.
The utility function (8.1) is assumed well behaved, strictly increasing, strictly quasi-
concave and twice continuously differentiable. The rational consumer maximizes utility
subject to a budget constraint:

pq 5 m (8.2)

where p 5 (pi) is the n-element column vector of prices and m is the consumers total
expenditure (income). Prices (and incomes) are assumed to be given. Maximization
of (8.1) subject to (8.2) yields n equations of q 5 q(p, m) which express the quantity
demanded of one good as a function of income and all prices,

qi 5 qi(p1, ... pn, m) (8.3)

To obtain a specific functional form for a complete set of demand equations given by
(8.3), it is necessary that the underlying utility function (8.1) is strictly quasi-concave
(that is, consumer preferences are convex). This poses limits on the types of preferences
that can be allowed in empirical demand analysis (Deaton and Muellbauer, 1980a).
However, an alternative approach not requiring strict concavity properties of consumer
preferences follows from duality theory (Diewert, 1971, 1974; Shephard, 1953). Note
that the maintained hypothesis of consumer theory is that a consumer chooses a set of

DWYER 9781781001288 Part I.indd 146 25/07/2012 11:07

 Almost ideal systems 147

goods and services that maximizes utility. The solution to the maximization problem (the
primal problem) yields a set of quantities demanded and a maximum utility level of u*.
This problem can be formulated equivalently so as to minimize the expenditure necessary
to achieve a given level of utility u*.
Minimize:

m 5 p9q subject to u(q) 5 u* (8.4)

The solution to this problem (the dual) yields the same set of quantities demanded, and
the minimum expenditure m* is equal to the given m in the primal problem. In both cases
primary and dual the optimal value of q is being sought.3 Since the two solutions
coincide, we have:

qi 5 qi(p, m) 5 hi(p, u*) (8.5)

Each of these solutions can be substituted back into their respective problems to yield
first, the maximum utility achievable, given prices p and expenditure m. The correspond-
ing function v(p, m) is called the indirect utility function.

u 5 u[q(p, m)] 5 v(p, m) (8.6)

Second is the minimum cost of achieving a fixed level of utility u at prices p. This is called
the cost (or expenditure) function (which is the mathematical inverse of the indirect utility
function v(p, m)).

m 5 (u, p) 5 c(u, p) (8.7)

The duality approach to the consumer choice problem makes derivation of complete
demand systems straightforward. For example, the Marshallian demand functions can
be derived simply by differentiating the indirect utility function defined in (8.9) applying
Roys identity. The Hicksian demand functions can be derived differentiating the cost
function defined in (8.10) using Shephards lemma. The cost function has become a major
tool in recent applications of consumer demand theory due to its attractive properties.
First, the cost function is homogeneous of degree unity in prices. Second, it is increasing
in u, non-decreasing in p, and increasing in at least one price. Third, the cost function is
continuous in p and first- and second-order derivatives exist everywhere except possibly
at specific price vectors. Fourth, the cost function is concave in prices. Most importantly,
the concavity of the cost function is independent of whether consumer preferences are
assumed to be strictly convex or not (Deaton and Muellbauer, 1980a; Thomas, 1987).
Thus, any cost function that possesses the above properties can be regarded as repre-
sentative of some underlying consumer preference ordering. Consequently, unlike the
traditional approach which imposes limits on consumer preferences the cost function
approach offers a flexible alternative in deriving demand functions. This is the approach
used in specifying the AIDS model.

DWYER 9781781001288 Part I.indd 147 25/07/2012 11:07

 148 Handbook of research methods in tourism

Approaches to the Specification of Complete Demand Systems and the Specification of

the AIDS Model

Since the advent of the Linear Expenditure System (Stone, 1954), a number of
approaches to the specification of complete demand systems consistent with the basic
postulates of consumer choice theory homogeneity, symmetry, negativity and adding-up
or aggregation have been advanced.4 One approach is to derive demand functions using
a utility function which satisfies the general restrictions implied by consumer theory.
The well-known examples are the Linear Expenditure System (LES) and the Indirect
Addilog (for details, see Berndt et al., 1977; Frisch, 1959; Houthakker, 1960; Stone,
1954). The second approach ignores the underlying utility function and starts with a set
of demand functions that do not necessarily satisfy theoretical restrictions. These are
then interpreted and restricted in the light of consumer demand theory. Powells System
of Additive Preferences and the Rotterdam Demand System (RTDS) belong to this class
of direct demand functions (Powell, 1966; Theil, 1975). The third approach is to employ
flexible functional forms.5 This method may be considered as an attempt to combine
the two early approaches by approximating utility functions (direct or indirect) and
cost functions. Flexible functional forms belong to the Translog family. The Translog
Demand System of Christensen et al. (1975) and the Almost Ideal Demand System of
Deaton and Muellbauer (1980a) belong to this class.6

The Almost Ideal Demand System (AIDS)

Deaton and Muellbauer (1980a) start the specification of the AIDS model with a
class of consumer preferences known as the Price Independent Generalized Log-Linear
(PIGLOG) form. This class of preferences is known to represent market demands as if
they are the outcome of decisions by a rational representative consumer (Deaton and
Muellbauer; 1980a). The preferences are represented by a cost function, c(u, p) that
defines the minimum expenditure necessary to attain a specific level of utility, u, at given
prices, p, and is given by:

Log c(u, p) 5 (1 2 u) log [a(P)] 1 u log [b(P)] (8.8)

where u denotes utility and P is a vector of prices. Note that the utility index can be
scaled to correspond to cases of subsistence (U 5 0) and bliss (U 5 1), in which case, a(P)
and b(P) can be interpreted as representing the cost of subsistence and bliss. Deaton and
Muellbauer use Translog and CobbDouglas functions to approximate a(P) and b(P)
respectively. The resulting cost function takes the following form:
n
1 n n n
log c (u, p) 5 a0 1 a ai log pi 1 a a g*ij log pi log pj 1 ub0 q pbi i (8.9)
i51 2 i51 j51 i51

where pi is the vector of prices, u is the level of utility and a0, ai, b0, bi and g*ij are param-
eters (i, j 5 1, ..., n). By logarithmic differentiation of (8.9), compensated (Hicksian)
demand functions can be obtained and the resulting AIDS demand model expressing
expenditure shares (wi) as a function of prices and utility (u) takes the following form:

DWYER 9781781001288 Part I.indd 148 25/07/2012 11:07

 Almost ideal systems 149

wi 5 ai 1 a gij log pj 1 biubo q pbi i (8.10)

j i

where
1 piqi
gij 5 (g*ij 1 g*ji) , wi 5 .
2 a piqi
The behavioural demand system given in (8.10) with the unobservable u can be converted
into an observable market demand system using the indirect utility function associated
with (8.9):
n
1 n n
log X 2 a0 2 a ai log pi 1 a a g*ij log pi log pj
i51 2 i51 j51
u5 n
(8.11)
ub0 q pbi i
i51

where log X 5 log c(u, p). By substituting (8.11) for u in (8.10), uncompensated (or
Marshallian) market demand functions (in expenditure or budget share form) are
obtained. The resulting AI demand functions are:

wi 5 ai 1 a gij log pj 1 bi log (X/P) (8.12)

j

where wi (wi K piqi/X) is the share of expenditure of the ith commodity, pj is the price of jth
commodity, X is the total expenditure on all goods and services (X 5 Spiqi), and P is an
aggregate price index defined as:

1
log P 5 a0 1 a ai log pi 1 a a gij log pi log pj (8.13)
i 2 i j

Linear homogeneity of the cost function, symmetry of the second-order derivatives

andadding-up across the share equations implies the following set of (equality) restric-
tions.

n n n
a i51 ai 5 1, a i51 gij 5 0, a i51 bi 5 0; a j gij 5 0, gij 5 gij (8.14)

Given the parameter restrictions, Equations (8.12) and (8.13) represent a system of
demand functions which add up to total expenditure (Si wi 5 1), are homogeneous of
degree zero in prices and total expenditure, and satisfy Slutsky symmetry.
Demand and substitution elasticities can be computed based on the estimated model
parameters. The implied expenditure elasticities, hi, reflecting the sensitivity of demands
to changes in expenditure, can be calculated using the formula:

%Dqi 0qi x bi
hi 5 5 . 5 11 (8.15)
%Dx 0x iq wi

where x is the total expenditure.

DWYER 9781781001288 Part I.indd 149 25/07/2012 11:07

 150 Handbook of research methods in tourism

Price elasticities of demand are defined as:

%Dqi 0qi pj
eij 5 5 . (8.16)
%Dpj 0pj qi

where eij is the price elasticity of demand (cross-price elasticity for i j and own price
elasticity (for i 5 j), pi is the price on the ith good, and qi is the quantity demanded for
the ith good. Price elasticities can be derived from either the Marshallian demand equa-
tion or the Hicksian demand equation. Given (8.8), the Marshallian (or uncompensated)
price elasticity for good i with respect to good j is:

gij 2 bi (wj 2 bj ln (X/P))

ij 5
eM 2 dij (8.16.a)
wi

where dij is the Kronecker delta; dij 5 1 if i 5 j and 0 otherwise (dij 1). Marshallian elas-
ticities can be transformed into Hicksian (compensated) elasticities through the Slutsky
equation given by:

eHij 5 eM
ij 1 wjhj (8.17)

where hj is the expenditure elasticity. Expenditure elasticity shows how the quantity
purchased changes in response to a change in the consumers expenditure/budget.
Uncompensated price elasticities indicate how a change in the price of a commodity
affects its own demand (own price elasticity) and the demands for all other commodities
(cross-price elasticities) included in the group. Compensated elasticities measure these
effects, assuming that real expenditures are held constant. Cross-price elasticities allow
the classification of commodity pairs to be substitutes or complements; negative cross-
price elasticities indicate that the two commodities are complements; positive values
indicate substitutes.
The AIDS model given by (8.12) and (8.13) is non-linear in parameters, thus
complex non-linear estimation methods are needed to estimate model parameters.
Deaton and Muellbauer (1980) proposed an alternative which allows the model to be
estimated using linear regression methods. That is, the replacement of the non-linear
price index (8.13) with an alternative price index that can be defined outside the AIDS
system, thus leaving a linear system of share equations. The proposed alternative was
that of the Stones price index, the values of which can be calculated prior to the esti-
mation as:
n
log P 5 a wi log pi (8.18)
i51

This version of the AIDS model is known as the Linear Approximate AIDS model or
LA-AIDS. With the use of the Stone price index approximation, Marshallian price elas-
ticities can be calculated using the following formula (White, 1985; Pashardes, 1993):7

eij 5 (1/wi) (gij 2 biwj) 2 dij (8.19)

DWYER 9781781001288 Part I.indd 150 25/07/2012 11:07

 Almost ideal systems 151

Model Extensions and Modifications

Over the years, several extensions and (or) modifications to the two versions of the AIDS
model have been proposed. Key modifications include the inverse version (IAIDS) due
to Eales and Unnevehr (1993, 1994), and the quadratic version Q-AIDS proposed by
Banks et al. (1997). The refinements include dynamic versions of the base AIDS model
due to Anderson and Blundell (1983), and error-correction versions of the LA-AIDS
models due to Attfield (1997), Karagiannisa and Mergos (2002), Ng (1995) and others.

Inverse AIDS (IAIDS) Model

As opposed to the conventional direct demand functions, inverse demand systems

explain price variations as functions of quantity variations. The particular version of
the IAIDS model proposed by Eales and Unnevehr (1993, 1994) considered applications
where the quantities tended to be fixed in the short run, and thus prices adjusted to clear
the market. The development of the IAIDS proceeds with the dual of the cost function,
the distance function. The distance function measures the proportion by which quantities
must be scaled back to reach a particular indifference curve (Deaton 1979). It possesses
the same properties as the cost function; it is linear homogeneous, concave, and non-
decreasing in quantities (as opposed to prices), and it is decreasing in utility (as opposed
to increasing) (Diewert, 1982). The distance function is defined as:

U{q/D(u, q)} 5 u (8.20)

where u equals utility, q is quantity and D is distance. The distance function is defined in
a manner analogous to the AIDS expenditure function:

1
log D (u, q) 5 a0 1 a aj ln qj 1 a a g*ij ln qi ln qj 1 ub0 q q2b
j
i (8.21)
2 i j j

Differentiation of the distance function (8.21) with respect to a quantity of a particular

good yields the compensated inverse demand for that good, in the same way that dif-
ferentiation of the cost function with respect to a particular price yields a compensated
demand function. Substitution of the inverted distance function (which is equivalent to
the direct utility function) uncompensate the functions and yields a system of inverse
demand share equations of the form:

wi 5 ai 1 a gij ln qj 1 bi ln Q (8.22)
j

where
1
ln Q 5 a0 1 a ln qj 1 a a gij ln qi ln qj (8.23)
j 2 i j
The model may be estimated by replacing the non-linear quantity index with a linear one,
such as Stones quantity index. With inverse demand models, sensitivities (elasticities)
are typically measured by flexibilities. Eales and Unnevehr (1993) provide the relevant

DWYER 9781781001288 Part I.indd 151 25/07/2012 11:07

 152 Handbook of research methods in tourism

formulas for the flexibilities when estimating the linear version (i.e., using Stones quan-
tity index) of the inverse AIDS model.

fij 5 2dij 1 { gij 2 biwj } /wi (8.24)

fi 5 21 1 bi/wi (8.25)

Equation (8.24) is defined for all own and cross flexibilities, where dij is the Kronecker
delta that equals 1 if i 5 j. This is analogous to the uncompensated own and cross-
price elasticities in the LA-AIDS model. Own price flexibilities describe the percent-
age change in the price of the ith good when the quantity demanded of that good
increases by 1 per cent. The demand for the ith commodity is flexible (inflexible) if a 1
per cent increase in the consumption of that commodity leads to a greater (less) than 1
per cent decrease in the marginal consumption value (its normalized price). Similarly,
cross-price flexibility is defined as the percentage change in the price of a good, where
the quantity demanded of competing good increases by 1 per cent. Goods are gross
quantity-substitutes (q-substitutes) if their cross-price flexibility is negative, and are
gross quantity-complements (q-complements) if their cross-price flexibility is positive.
Equation (8.25) is the analogue of the expenditure elasticity of the base AIDS model. It
is interpreted as a scale flexibility that measures the percentage change in the price of a
good brought about by a proportional change in the aggregate quantity or scale of con-
sumption. Scale flexibilities that are greater (less) than 1 in absolute value are necessities
(luxuries) (Eales and Unnevehr, 1993). The IAIDS model provides a viable alternative to
the conventional direct demand functions as it can be used to model tourist numbers as
a measure of tourist demand (as opposed to tourist expenditure).

Quadratic AIDS (QUAIDS)

Another extension to the AIDS model was proposed by Banks et al. (1997) who intro-
duced a quadratic version of the standard AIDS model. This may be considered an
attempt to improve the flexibility of the base AIDS model, which assumes linear Engel
curves. The rationale for the proposed extension is that the relationship between budget
shares and income is not always linear (however, nor is it always non-linear). Banks et
al. (1997) develop their quadratic version to be flexible enough so that the non-linear
Engel curve specification does not have to be applied to each good. The specification
of the QAIDS model starts with adding a quadratic logarithmic income term and nests
the standard AIDS model specification. The resulting QUAIDS model is given by share
equations of the form:

li
wi 5 ai 1 bi log (x/a (p)) 1 b
(x/a (p)) 2 1 a gij log pj (8.26)
q kpkk j

for i 5 1, ..., n and where log a(p) can be approximated by the Stone price index (8.18),
the QUAIDS budget shares reduce to those of AIDS if li 5 0 for all i.
Coenen and Eekeren (2003) used QUAIDS to analyse the demand for tourism goods
and services by Swedish households. Among other modifications, there are several

DWYER 9781781001288 Part I.indd 152 25/07/2012 11:07

 Almost ideal systems 153

attempts to redress one limitation of the AIDS model, namely its failure to satisfy the
global regularity condition (that is, the AIDS cost function is not globally concave),
although it can be regular over a restricted range (that is, locally flexible) (Diewert and
Wales, 1987; Lau, 1986). Although this is trivial, as Deaton (1986) noted, it is possible to
improve the regularity of the AIDS model by either imposing restrictions prior to esti-
mation, or by modifying the underlying cost functions (for example, see Chalfant, 1987;
Cooper and McLaren, 1988; Cooper et al., 1991).8 Chalfant (1987) proposed a globally
flexible, almost ideal demand system by using the Fourier flexible form in place of the
Translog to approximate budget share equations. To improve the curvature property
within AIDS, Cooper and McLaren (1992) developed a modified version of the AIDS
model (MAIDS), while Moschini (1998) developed a semi-flexible AIDS specification.
While these contributions are important in terms of guaranteeing global curvature con-
ditions, as implied in the consumer theory of choice, as Lau (1986) has shown flexibility
and regularity are incompatible. Thus, if regularity is to be incorporated into a flexible
functional form, it is at the expense of flexibility. Further, such modifications appear to
have implications for the economic plausibility of some derived elasticity parameters as
well.9 Therefore, any attempt to improve regularity is likely to impose some restrictions
on the preference structure, and the resulting parameter estimates could be biased, as
they are conditional upon prior restrictions.
Discussed above are key modifications to the AIDS model proposed in the literature
which are aimed at improving its flexibility. In addition, there are several other proposed
refinements that are necessitated by, or are consequences of, accommodating specific
issues arising when implementing and (or) applying the model to study particular issues.
These include the way in which applied researchers treat theoretical restrictions during
the estimation, the nature and issues related to the types of data used for estimation, and
the aim of the study.
An important property of the AIDS (and flexible functional forms in general) is that
they afford the estimation of complete demand systems consistent with the consumer
theory of choice, that is, models can be estimated while imposing theoretical restrictions
a priori as maintained hypotheses. Likewise, models may be estimated unrestricted and
the validity of the theoretical restrictions implied in the consumer theory of demand can
be tested. The imposition of theoretical restrictions is usually made as restrictions on the
AIDS model parameters (for example, Equation (8.14)).

(1) The adding-up (aggregation) restriction implying that all budget shares sum to unity,
requires:
n n n
a i51 ai 5 1, a i51 gij 5 0, a i51 bi 5 0 (8.27)

(2) The homogeneity restriction implying the absence of money illusion requires:

a jgij 5 0 (8.28)

(3) The symmetry restriction guaranteeing the consistency of consumer choice requires:

gij 5 gij (8.29)

DWYER 9781781001288 Part I.indd 153 25/07/2012 11:07

 154 Handbook of research methods in tourism

(4) A final restriction following from the theory, negativity, is satisfied if the Slutsky
matrix, sij, of compensated price derivatives is negative semi-definite:

sij 5 gij 2 bibj log (X/P) 2 widij 1 wiwj (8.30)

where dij is the Kronecker delta (dij 5 1, if i 5 j; and dij 5 0, otherwise), and wi is the pre-
dicted value of the ith budget share.
Of the four theoretical restrictions, the adding-up condition is automatically satisfied
in a system of share equations by the way of construction. The negativity condition can
be tested for its validity, but it cannot be directly imposed as a constraint upon econo-
metric estimation of the AIDS model.10 The remaining two restrictions of homogeneity
and symmetry can be imposed prior to the estimation or tested for their validity. While
some researchers estimate empirical models imposing theoretical restrictions homoge-
neity and symmetry as maintained hypotheses, others estimate unrestricted versions
and then test for the validity of the restrictions based on estimated model parameters.
Although no study has been specifically designed to test for the validity of the basic pos-
tulates of demand theory, many such tests have failed to verify the empirical validity of
the two theoretical restrictions (for a detailed discussion, see Keuzenkamp and Barten,
1995). Among the two restrictions, homogeneity is the one more often rejected. Of
the various explanations, errors associated with dynamic misspecification of empirical
models are the ones which led researchers to incorporate dynamic consumption behav-
iour into the AIDS model.11
Anderson and Blundell (1982) introduced a dynamic version of the AIDS model by
incorporating habit persistence. They showed that restrictions implied by economic
theory are not rejected when tested on the long-run dynamic structure. Thus, dynamic
modelling is claimed to be a better alternative in modelling demand using time series data
compared to static modelling. The first attempt to model tourism demands using this
dynamic version was by Lyssiotou (2000) who investigated the UK demand for tourism
in the United States, Canada, and 16 European destinations. An alternative explanation
advanced by Ng (1995) proposes that time-series issues unit roots being present are
partly responsible for the rejection of homogeneity.12 Thus she proposes the applica-
tion of recently developed techniques for estimating cointegrating vectors to account
for dynamic misspecification (see also Attfield, 1997; Karagiannisa and Mergos, 2002;
Lewbell and Ng, 2005). This led to another empirical version of the LA-AIDS model,
often coined the error-correction LAAIDS model (ECLAAIDS). This version is the
most widely used in recent empirical studies of tourism demand, including the studies of
Durbarry and Sinclair (2003), Li et al. (2004), Mangion et al. (2005) and Corts-Jimnez
et al. (2009).
Depending on the specific issues being investigated and the particular nature of data
available, some minor modifications to the model structure are usually introduced by
altering and (or) adding model parameters. For example, issues relating to seasonality
and structural change may arise in applications using data collected over time. This is
particularly the case with regard to tourism, given the apparent seasonality associated
with tourist demand (see for example, Divisekera, 2009a, 2010b; Lyssiotou, 2000).
Further, demand for tourism may be affected by random events including political
disturbances, social unrest, natural disasters and health concerns. These could lead

DWYER 9781781001288 Part I.indd 154 25/07/2012 11:07

 Almost ideal systems 155

to exogenous shifts in tourism demand, and thus they need to be accounted for in the
modelling (see for example, Divisekera, 2003; Han et al., 2006; Lanza et al., 2003). In
addition, if data are extracted from household surveys and the focus is on household
demand for tourism, data aggregated across households may raise questions regarding
the aggregation properties of a particular demand model (see, for example, Coenen and
Eekeren, 2003).
The standard practice is to use dummy variables in the basic share equations to
capture the effects of structural change, seasonality, and (or) any other exogenous shifts
(Divisekera, 2003; Han et al., 2006; Lyssiotou, 2000). In relation to linear trends, the
common approach is to add a simple linear trend to the share equations, which is analo-
gous to allowing the intercept term of each share equation to trend (see for example,
Fujii et al.1985; Lyssiotou, 2000; White, 1985). The same intuition carries over to more
complex factors that allow the intercept to shift according to the quarter or season of
the year of the observation, or even other things such as structural breaks in the time
series, and changes in tastes and preferences (Holt and Goodwin, 2009). When adding
exogenous intercept shift variables, one needs be mindful that their addition does not
violate the adding-up condition. Usually, time trend, seasonality, and structural shifts
are incorporated by replacing the intercept term ai (in Equation (8.12)) with ait (that is,
their effect is measured as a deviation from the intercept):

ait 5 a0i 1 d0it 1 a sdisDs (8.31)

where t 5 1, ..., T; Ds are seasonal dummy variables, S 5 1, ..., 4 and Ss dis 5 0. The
parameters d0i are interpreted as reflecting exogenous shifts in preferences (Lyssiotou,
2000).
The use of deterministic dummy variables to capture seasonal shifts in demands,
however, could lead to biased parameter estimates when turning points of the seasonal
time series are changing (Fraser and Moosa, 2002; Carlos et al., 2004). The proposed
alternative is to incorporate trigonometric variables to account for seasonality. This
approach allows the seasonal cycle to be dictated by the data, rather than using the deter-
ministic dummy variables that define the season. This is particularly appealing when
there are clear structural shifts in seasons. The application of this method is carried out
by augmenting the share equations (Equation (8.12)) with seasonal trigonometric vari-
ables and a time trend (Divisekera, 2009b; 2010a):

2pt 2pt
wi 5 ai 1 a gij log pj 1 bi log (X/P) 1 aci cos 1 asi sin 1 atit (8.32)
j 4 4

where aci and asi represent parameters on the trigonometric variables, and ati is the param-
eter on the time trend variable. To ensure that the adding-up condition is not violated
these parameters should satisfy the following conditions:
n n n
a i51ai 5 0, a i51ai 5 0, a i51ai 5 0
c s t
(8.33)

Once the demand system is specified incorporating all of the relevant variables and exog-
enous shift variables where necessary, the next step involves estimation.

DWYER 9781781001288 Part I.indd 155 25/07/2012 11:07

 156 Handbook of research methods in tourism

The Stochastic Specification and Estimation Methods

The model estimation is usually preceded by adding a stochastic random disturbance

or error term uit to the share equations. The general stochastic version of the model in
vector form is:

wit 5 i (fit, b) 1 uit (8.34)

where wit is the ith budget share at time t, fit is the set of explanatory variables, b is
the parameters to be estimated, and uit is a random disturbance term. Let t9 denote
the n-vector of disturbances at time t and the contemporaneous covariance matrix W.
Following convention, the covariance matrix for the share disturbance is assumed to
be the same for all observations, E(tt9) 5 W and the disturbances are uncorrelated
across observations, E(ts) 5 0 for s t. Note that given the dependent variables are
shares (wi) and that they add to unity implies that Si uit 5 0 for each t. Thus, the errors
cannot be distributed independently across equations and the W is singular (Barten,
1969; Pollak and Wales, 1992). Due to the singularity of W, the density of t may be
expressed in terms of the density of any (n 2 1) equations by deleting any arbitrary
element. The resulting parameter estimates are independent of the deleted equation
(Barten, 1969).
The joint estimation of parameters may be carried out using a non-linear estimation
method such as the Maximum Likelihood (ML) Estimator, the only known method
which yields an estimator invariant with respect to the equation which is dropped
during the estimation (Barten, 1969).13 With the arbitrary deletion of the nth equation,
let W be the corresponding covariance matrix and ut9 be the vector of disturbances. Then,
under the assumption that these independent disturbances are normally distributed with
zero mean and covariance matrix W, the log of likelihood function for t observations on
the (n 2 1) independent equations is given by:

a2 (n 2 1) log (2p) 2 log 0 W 0 b 2 a u W21ut

T 1 1 T
log L 5 (8.35)
2 2 2 t51

where

ut 5 (wt 2 ft); wt 5 (w1t ... wn 2 1t);

f1t (pt, xt; a, b, g)

ft 5
.
.
fn 21, t (pt, xt; a, b, g)

pt 5 (p1t, ... pnt); Xt 5 (Si 51, pit xit); a 5 (a1 ... an); b 5 (b1 ... bn); g 5 (g11.. gn1, ...
gnn)

ML estimation consists of choosing values of unknown parameters a, b, g and W in

order to maximize the log-likelihood function L. The ML procedure has attractive prop-

DWYER 9781781001288 Part I.indd 156 25/07/2012 11:07

 Almost ideal systems 157

erties: estimators are consistent, asymptotically efficient, and asymptotically normally

distributed (Judge et al., 1980; Malinvaud, 1980).

APPLICATIONS OF THE AIDS MODEL AND ITS VARIANTS IN

TOURISM DEMAND MODELLING

As stated at the outset, the AIDS model has been the principal and most widely used
established demand system employed by researchers in the tourism field.14 The available
empirical studies may be broadly classified into two categories: (1) destination choice
or demand for tourist destinations, and (2) the studies that seek to analyse the demand
for tourism goods and services. The former group of studies attempts to evaluate the
effects of incomes and prices as determinants of tourism demand for a group of destina-
tions. This has been the central focus of most of the early empirical studies of tourism
demand. In terms of a modelling strategy, except for the inverse AIDS model, all other
versions and (or) modifications of the AIDS model have been employed. The latter
group of studies investigates the aspects of tourism demands that focus on the demand
for destination-specific tourism goods and services. More specifically, they attempt to
analyse tourists consumption behaviour or allocations of tourists spending among
various goods and services at the destinations they visited. The economic parameters
associated with ex-post tourist demand are important for destination management and
for developing destination-specific strategies to maximize gains from tourism.

Empirical Analysis of Demand for International Tourism (Destination Choice)

The first attempt to model international demand for tourism in the system context
undertaken by White (1982, 1985) was based on the LAAIDS version of the model. In
this pioneering study, White analysed the demand for travel and tourism in Western
Europe by US residents. This was followed by OHagan and Harrison (1984), White
(1985), Sinclair and Syriopoulos (1993), and Papatheodorou (1999) who employed the
LAIDS model to study the US, European and international demands for tourism in
Mediterranean destinations. De Mello et al. (2002) concentrated on UK demand for
tourism in France, Portugal and Spain. This was followed by Divisekera (2003) who
developed multi-country demand systems for the demand for tourism to Australia from
Japan, New Zealand, the United Kingdom and the United States. In his most recent
study, Divisekera (2009c) modelled travel and tourism demands simultaneously for the
same group of countries.
Of the studies based on the LAAIDS model, the pioneering study of White (1982,
1985) modelled US demand for travel to Western Europe. The model consisted of eight
equations and one representing demand for travel. The study provides a comprehensive
set of expenditure, own, and cross-price elasticities among the eight Western European
destinations included in the model. White used consumer price indices (CPI) as a proxy
for destination prices; a proxy for travel prices between the United States and Western
Europe was constructed using airlines revenue data. The results allow the classification
of regions/destinations of Western Europe as substitutes or as complements according to
the preferences of travellers. The price elasticities are relatively low for France, Belgium,

DWYER 9781781001288 Part I.indd 157 25/07/2012 11:07

 158 Handbook of research methods in tourism

the Netherlands and Luxembourg. France and the United Kingdom exhibited high price
substitution effects as did France and Germany. Travel to most other countries might be
classified as complementary with respect to travel to France, and substitutes with respect
to travel to the United Kingdom. The expenditure elasticities allow the classification
of goods as normal or luxury goods.15 The expenditure elasticity for travel to Spain
and Portugal is estimated to be 1.3 and the expenditure elasticity for Norway, Sweden
and Denmark is 1.2. This indicates that among destinations, these countries could be
considered luxury destinations. The other countries (as well as transportation) have
expenditure elasticities that are not significantly different from one. Therefore, the share
of travel expenditures in these groups is not expected to increase as total expenditures
increase.
All of the studies below follow a similar modelling strategy; they use CPIs as their
proxy for tourism prices and they ignore travel costs. The only exception is Divisekera
(2003) who estimated four demand systems focussing on Japanese, New Zealand, UK
and US demand for Australian tourism. Divisekera (2003) departed from the rest of the
available empirical studies in several respects. First, he specified the model in the context
of a preference-consistent utility maximization framework. Second, his study uses
tourism price indices along with observed average daily spending to construct tourism
prices, a significant departure from the commonly used consumer price indices. Third,
his study incorporates the cost of international travel, a key economic determinant of
international tourism demand that was largely ignored by other researchers. Estimated
models are in conformity with the basic postulates of consumer theory, homogeneity
and symmetry. Derived elasticities reveal substantial cross-demand effects, reflecting the
diversity of preferences of tourists from different origins.
All of the forgoing studies were based on the LAAIDS version; the first to use an
alternative version was Lyssiotou (2000) who employed a dynamic version of the AIDS
model of Anderson and Blundell (1983). The focus of the study was British demand
for tourism to the United States, Canada and 16 European destinations. The key issue
examined was the extent to which current tastes are formed through past consumption
behaviour. The results show that price and expenditure elasticities of the various holiday
destinations differ between the short run and the long run. For example, UK tourism
demand for France is price inelastic in the short run but elastic in the long run. In the case
of GreeceItaly the demand is found to be price inelastic both in the short run and the
long run. Also, USCanada appears to be a luxury holiday destination in the long run,
although is not so according to the short-run expenditure elasticity. Overall, the long-run
expenditure elasticities appear to be close to 1 everywhere suggesting that in the long run
all the destinations included in the analysis benefit equally from an increase in British
tourism expenditure. These findings are consistent with Divisekera (1995) who examined
short- and long-run dynamics in tourism demand.
Durbarry and Sinclair (2003) investigated the demand for tourism by French nation-
als in Italy, Spain and the United Kingdom. They employ a linear version of the
dynamic error-correction AIDS model of Anderson and Blundell (1983) (henceforth
EC-LAAIDS). They estimate short-run and long-run demand elasticities. However,
due to poor statistical fit, only the long-run elasticities are reported and evaluated.
The results indicate that it is changes in relative effective own price competitiveness,
rather than in the tourists budget, that are the main drivers of changes in the destina-

DWYER 9781781001288 Part I.indd 158 25/07/2012 11:07

 Almost ideal systems 159

tion countries shares of the French tourism market. Li et al. (2004) used this approach
to model UK demand for tourism in neighbouring Western European countries. The
estimated expenditure elasticities show that travelling to most major destinations in
Western Europe is a luxury for UK tourists in the long run. The demand for travel to
these destinations is also found to be relatively more price elastic in the long run than
it is in the short run. The calculated cross-price elasticities suggest that the substitution
and (or) complementarity effects vary from destination to destination. The UK demand
for tourism in the Mediterranean destinations of Malta, Cyprus and Spain was the
focus of the study by Mangion et al. (2005). They applied the EC-LAAIDS model and
their particular emphasis was on price competitiveness among these three destinations
which are perceived as strong competitors. The results showed that in terms of price
competitiveness Malta is the most price-sensitive destination, followed by Cyprus and
Spain. The results failed to reveal any significant short-run demand effects. Han et al.
(2006) model US tourism demand for four European destinations France, Italy, the
UK and Spain. The results show that price competitiveness is important for US demand
for France, Italy and Spain but it is relatively unimportant in US tourism demand for the
UK. France and Italy are regarded as substitutes by US tourists, as are Spain and Italy.
As US expenditure rises, the market shares of Spain and the UK decline, while France
and Italy benefit. Han et al. (2006) show further that that the static LAIDS model gives
empirically better results than does the EC-LAIDS model. Somewhat contrary evidence
is found by Corts-Jimnez et al. (2009) who studied Italian demand for tourism in
four European destinations France, Germany, the UK and Spain. This study, based
on monthly data, incorporates seasonality using dummy variables and accounts for
preference changes using a linear time trend. While the study estimates both long- and
short-run dynamic versions of the model, demand elasticities are reported only for the
short run. This is somewhat puzzling because it is the long-run model that appears to
be theoretically consistent (in terms of the acceptance of homogeneity and symmetry
restrictions). The short-run elasticities revealed unusually high price-elastic demands
ranging from 28.7 for France, to 28.6 for Germany and 25.8 for Spain. The empirical
plausibility of these estimates needs to be questioned, as one would expect tourists to
adjust to price changes slowly in the short run. Further, the statistical fit of the restricted
EC model used for deriving elasticities is poor (only four of the ten own and cross-price
coefficient (cijs) effects are statistically significant and none of the expenditure coefficients
(bis) are statistically significant.

Empirical Analysis of Demand for Tourism Goods and Services (Ex-post Demands)

The second group of demand studies concern the demand for destination-specific
tourism goods and services. More specifically, they attempt to analyse tourists con-
sumption behaviour or allocations of tourist spending among various goods and services
at the destinations they visited. The economic parameters associated with ex-post tourist
demands are important for destination management and for developing destination-
specific strategies to maximize gains from tourism. Of the available research, one group
of studies focus on demands by landed foreign tourists and the others focus on demands
by domestic tourists. The studies that focus on ex-post demands include Divisekera
(2007, 2009b, 2010a), Divisekera and Deegan (2010), Fujii et al. (1985), Sakai (1988) and

DWYER 9781781001288 Part I.indd 159 25/07/2012 11:07

 160 Handbook of research methods in tourism

Wu et al. (2011). The studies investigating domestic demand for tourism goods and serv-
ices include Coenen and Eekeren (2003), Divisekera (2007, 2009a, 2009b, 2010a, 2010b)
and Pyo et al. (1991).
The first attempt to model ex-post tourism demand was Fujii et al. (1985). Based on
the LAIDS model, they estimated demand for six categories of goods and services pur-
chased by visitors to Hawaii. These categories were food and drink, lodging, recreation
and entertainment, local transport, clothing and other. Their estimated expenditure
elasticities are all positive, indicating that all goods purchased by visitors are normal
goods. All the own price elasticities are negative and significantly different from 0. Based
on the non-linear AIDS model, Divisekera (2009b) investigated the economic param-
eters underlying the ex-post demand for Australian tourism goods and services from
ten source markets in Asia, Europe and North America.16 These are respectively: China,
Korea, Hong Kong, Malaysia, Taiwan, Indonesia, Singapore, Thailand, Germany and
Canada. Five broad commodity aggregates representing a tourists consumption bundle
were used: accommodation, food, transport, shopping and entertainment. The necessary
data were obtained from quarterly International Visitor Surveys conducted by Tourism
Australia which provides itemized consumption expenditure data on a quarterly basis.
Price indices were constructed for each broad commodity aggregate using detailed price
data obtained from the Australian Bureau of Statistics. The study revealed substantial
differences in consumption patterns by tourists from different source countries in Asia,
North America and Europe. For example, expenditure elasticities of demand for the
food and shopping aggregate are broadly similar across the source markets, with elastic-
ity coefficients around unity in magnitude. However, elasticities with respect to accom-
modation varied across the markets. In general, demand for accommodation is the
most expenditure-elastic across all of the commodity aggregates, with the exception of
Chinese demand. Significant differences in the elasticities of demand for transportation
are also evident, in the majority of cases demands are unitary elastic; the exceptions are
Taiwanese, Thai, Korean and Chinese demands. Expenditure elasticities of demand for
entertainment are around unity across the markets, with the exceptions of Singapore and
China with elasticities greater than unity. In general, demands by tourists from all regions
were found to be price inelastic, implying that all goods are necessities from a tourists
point of view. This result is consistent with the empirical reality that once tourists arrive
at a destination, they are bound to consume the available goods and services. The cross-
price elasticities reveal a gross complementarity of demands. This indicates that tourists
tend to purchase a bundle of goods and services, and that all of them are necessary for
the maximization of their utility from the visit. Divisekera (2010b) employed the same
methodology to examine the consumption behaviour of foreign tourists from the four
major source markets to Australia New Zealand, Japan, the UK and the United States.
The results from this study are consistent with the general findings of Divisekera (2009b)
in that tourist demands for the bundle of goods and services are price inelastic, while
expenditure elasticities varied across the commodity groups and among source markets.
Divisekera and Deegan (2010) provided further evidence on the consumption behav-
iour of foreign tourists in their analysis of the demand for Irish tourism goods and serv-
ices. Using the non-linear AIDS model, they estimated five demand systems, with four
representing the major source markets of tourists to Ireland. These were Great Britain,
North America, mainland Europe and the rest of the world, and an aggregate model

DWYER 9781781001288 Part I.indd 160 25/07/2012 11:07

 Almost ideal systems 161

based on a pooled sample. Each system included six commodity aggregates including
food, lodging (accommodation), transportation, shopping, sightseeing and (or) enter-
tainment and a miscellaneous category. The data on various commodities consumed
by tourists were obtained from the Irish Tourist Board (Failte Ireland) based on sample
surveys. One finding of this study that is broadly consistent with the two previous studies
is the apparent insensitivity of foreigners demand in response to commodity prices.
While there are notable variations in the consumption patterns of tourists from differ-
ent source markets reflecting differences in consumer preferences and consumption
habits in general, tourists demand for the various Irish tourism goods and services is
found to be price inelastic. Similarly, the cross-price elasticity values for all commodities
across the source markets indicate gross complementarity. In relation to expenditure
elasticities, substantial variations across the source markets are found. For example, the
elasticity of demand for shopping is the highest among all commodities for the British
tourists, whereas in the case of North American tourists, demand for accommodation
is, relatively, the most expenditure elastic. In general, the North American demands for
most Irish tourism goods and services are expenditure elastic, except for the demand for
transportation, which is a necessity for North American tourists.
Wu et al. (2011), the most recent study on ex-post demands, concentrate on foreign
demand for tourism goods and services in Hong Kong. Based on the EC-LAAIDS model,
they estimate eight demand systems for Hong Kongs inbound tourism representing the
source markets of Australia, the UK, the United States, Japan, South Korea, Singapore,
Taiwan and mainland China. They classify tourist spending into four categories: shop-
ping, hotel accommodation, meals outside hotels and other. Evaluation of demand
parameters is carried out under three groups: long-haul destinations Australia, the UK
and the United States and the short-haul markets Japan, South Korea, Singapore,
Taiwan and mainland China. The empirical results reveal different types of consump-
tion behaviour among the eight source markets. For example, expenditure elasticities
reveal an elastic demand for shopping by the tourists from the long- and short-haul haul
markets implying that the tourists from these markets perceive shopping in Hong Kong
as a luxury. In contrast, mainland Chinese tourists regard shopping in Hong Kong as a
necessity, but view hotel accommodation as a luxury. In relation to price elasticities, the
analysis is based on compensated elasticities and reveals relatively price-inelastic demand
across the commodity groups. All the statistically significant cross-price elasticities are
positive implying that the three commodity groups are substitutes. Overall, the empiri-
cal findings in of this study in relation to the own price and expenditure elasticities are
consistent with those of Divisekera (2009b, 2010a) and Divisekera and Deegan (2010),
however, the cross-price effects reveal contrary evidence.17

Analysis of Domestic Tourism Demands

The second group of studies addressing the consumption behaviour of tourists focus on
domestic demands. The related studies based on the AIDS model include Coenen and
Eekeren (2003) who studied Swedish demand and Divisekera (2008, 2009a, 2010c).18
The first attempt to model domestic demand for tourism goods and services using the
AIDS model was by Coenen and Eekeren (2003). They used a quadratic version of
the AIDS model to analyse the demand for domestic tourism by Swedish households.

DWYER 9781781001288 Part I.indd 161 25/07/2012 11:07

 162 Handbook of research methods in tourism

Using microeconomic data drawn from the Swedish tourism and travel database, they
estimated ten individual demand systems, one aggregate demand and six additional
systems distinguished by demographic (five income groups) and household composi-
tional factors (four household groups). Each system included five commodity aggre-
gates: groceries, restaurant meals, accommodation, transportation and shopping. The
model for total expenditure (their aggregate model) shows that income elasticities for
domestic tourism demand for the different household categories were all around 1,
while the price elasticities reveal highly elastic demand. Price elasticities of demand for
accommodation were the highest among the five commodity aggregates and the lowest
is for restaurants.
Based on the non-linear AIDS model using expenditure data collected from
Australian National Visitor Surveys Divisekera estimated a number of demand models
focussing on different aspects of domestic demand for tourism by Australian households.
These include the regional demand for domestic tourism (Divisekera, 2008, 2010c),
aggregate demand for domestic tourism (Divisekera, 2009a) and leisure versus non-
leisure tourism demands (Divisekera, 2010b). Each variant contained five commodity
aggregates; these are food, accommodation, local transport, shopping and entertain-
ment. In relation to regional demand overall, the estimated own price elasticities across
the eight states and territories reveal relatively price inelastic demand for the five com-
modity aggregates, while the demands were found to be relatively expenditure elastic.
The degree of sensitivity of demands to expenditure and prices, however, varied signifi-
cantly across the eight states and territories. These variations may reflect the diversity of
the preferences of the tourists from different origins.
In relation to leisure and non-leisure tourist demand, the demands by the leisure tour-
ists are found to be relatively more sensitive to prices than are those of the non-leisure
tourists. The cross-price elasticities derived from both models reveal gross complementa-
rities of demands, implying that tourists overall utility depends on their (joint) consump-
tion of a bundle of goods and services. The observed price inelastic demands, coupled
with the apparent complementarities of demands, may reflect the possibility of a latent
price sensitivity associated with tourist demand.

CHALLENGES AND CONCLUSIONS

The AIDS model devised by Deaton and Muellbauer (1980) has come to be the industry
standard in modern empirical demand analysis. Since its publication in 1980, several
modifications and extensions have been proposed. The extensions include dynamic,
inverse, quadratic and error correction versions. All of these variants of the AIDS model
except the inverse AIDS have been employed by researchers of tourism demand. The
AIDS model dominates as the single most widely used complete demand system by
researchers in the tourism field. In addition to its attractive theoretical properties and
flexibility, the popularity of this model among the applied researchers who rely on estab-
lished demand systems to analyse specific issues of interest is due largely to the ease of
estimation afforded by its linear version. This version, known as the LAAIDS, avoids
the need for using complex non-linear estimation methods compared with a similar class
of models such as the Translog demand systems of Christensen et al. (1975).19 It is this

DWYER 9781781001288 Part I.indd 162 25/07/2012 11:07

 Almost ideal systems 163

version of the AIDS model (and its variants) that has been the principal base model used
by researchers in tourism. The exceptions are Lyssiotou (2000) and studies by Divisekera
(2009a, 2010a) that used the non-linear AIDS model.
The array of empirical studies using various versions of the AIDS model has explored
different aspects of tourism demands, including international, domestic, and regional
demands. Of the available empirical studies based on the AIDS model, the first group
of studies attempt to evaluate the effects of incomes and prices as determinants of
tourism demand for a group of destinations. Thus, international or destination-choice
studies have been the central focus of most early empirical studies. The second group
of studies focus on the demand for destination-specific tourism goods and services or
ex-post tourism demands. They attempt to analyse tourists consumption behaviour or
allocations of tourist spending among various goods and services at the destinations
they visited. The economic parameters associated with the ex-post tourist demands are
important for destination management and for developing destination-specific strategies
to maximize the gains from tourism. The second group of studies addressing the con-
sumption behaviour of tourists focus on domestic demands.
These latter studies have generated a large number of economic parameters enriching
our knowledge and understanding about the dynamics of this important activity that
generates multiple socio-economic benefits. Regardless of the particular version of the
model employed, the data used for estimation and their focus, all studies provide similar
if not identical sets of economic parameters associated with tourism demands. They are,
respectively, expenditure and (or) income elasticities and price elasticities. Expenditure
elasticities highlight the effects of total tourist spending and (or) incomes on demand,
and how sensitive their choices are to changes in total spending. Likewise, price elastici-
ties indicate how the prices of tourist products affect the demand for a given product
(own price elasticities) and the cross-price effects (cross-price elasticities) or interactions
among and between alternative products.
The AIDS methodology allows evaluations of price effects in a consistent manner
from two theoretical perspectives: Marshallian or uncompensated, and Hicksian or
compensated price effects. Marshallian elasticities indicate how consumers (tourists)
react to changes in the market price of a product, and the Hicksian price elasticities
reveal the same effects under the proposition that consumers (tourists) have been com-
pensated for price changes. The latter implies that a consumers real income remains
unchanged (following a price change), thus the Hicksian price elasticities reveal pure
substitution effects of a price change, whereas the Marshallian elasticities reflect both
income and substitution effects associated with price changes. The relevance of the two
elasticity concepts as a guide to understanding market behaviour, or to develop appro-
priate policy measures in response to them, and (or) to exploit them for own advantage
is different. For example, if the focus is to develop pricing and other strategies to exploit
market conditions, the appropriate guide is provided by the Marshallian price elastici-
ties. Alternatively, if the focus is on the welfare of consumers and the consideration is to
ensure that price changes leave consumers unaffected in terms of their level of consump-
tion (after the price change), then the concept of Hicksian price elasticities is the relevant
concept that should be used as a guide to develop appropriate compensation measures.
The empirical studies of tourism demands reviewed in the preceding sections of this
chapter report both types (in majority of cases), or at least one type of elasticity. Most

DWYER 9781781001288 Part I.indd 163 25/07/2012 11:07

 164 Handbook of research methods in tourism

of these studies are specific in the sense that the focus is demand from a given origin to a
group of destinations or countries. Most early studies concentrated on US demand for in-
ternational travel to European and Mediterranean destinations (Han et al., 2006; OHagan
and Harrison, 1984; Papatheodorou, 1999; Sinclair and Syriopoulos, 1993; White, 1982,
1985). De Mello et al. (2002) and Li et al. (2004) concentrated on UK demand for tourism
in Europe. All of these studies are based on the AIDS model and use similar types of price
measures; often the real exchange rate is used as a proxy for destination prices.
Despite the fact that almost all of the studies use the same methodology and similar
types of explanatory variables, there is hardly any agreement in terms of the observed
price and expenditure effects. For example, in the case of UK demand, the expenditure
elasticity for Portugal was estimated at 0.04 by Papatheodorou (1999) compared to
1.58 by Syriopoulos and Sinclair (1993). Li et al. (2004) report a very high elasticity
for Greece (-2.75), whereas Papatheodorou (1999) found a relatively low value (-0.93)
for the same country. The same applies to many other available international demand
studies. The variations in elasticities between studies are so plain that no generalization
can be made about the nature and magnitudes of elasticities of demands from a given
origin to chosen destinations. Thus, for specific purposes such as developing policy
measures to exploit or correct prevailing market conditions, demand parameters need
to be estimated by taking into account both prevailing market conditions and they must
use the most current data.
Most attention among researchers in recent times appears to be on keeping up
with the changing econometric methodology, rather than on improving the modelling
process. For example, none of the studies except for White (1985) and Divisekera (2003,
2009c) attempted to incorporate one crucial cost factor affecting international tourism
demands, namely the cost of transport. Similarly, all the studies continue to use the CPI
(adjusted for exchange rates) as a proxy for tourism prices. This is a major drawback, as
they grossly overestimate or underestimate the real cost of tourism goods and services.
Thus, there is a need to develop appropriate tourism price indices as inputs for demand
model estimation. Future studies need to concentrate on improving databases and devel-
oping reasonable price measures that capture both the costs of travel and tourism goods
which are the key determinants of tourism demands.

NOTES

1. Of the other existing demand systems, two of the studies (Bakkal, 1991; Bakkal and Scaperlanda, 1991)
were based on the Translog demand system and three on the Linear Expenditure System (Pyo et al., 1991;
Sakai, 1988; Smeral, 1988).
2. The neoclassical view of consumer demand has several important underpinnings. A central concept is the
neoclassical view of the consumer as a rational choosing agent who is motivated to seek the highest level
of satisfaction or utility. A consumers wants depend on the level of utility that a particular purchase will
generate. This implies that the value of a particular good is dictated not solely by its price, but by a con-
sumers subjective feelings toward it. The neoclassical theory of consumer behaviour is well documented
in the work of Hicks (1956) and Samuelson (1947). An extensive treatment of consumer theory leading
to systems of demand functions can be found in Barten (1977), Deaton (1986), Deaton and Muellbauer
(1980a), Goldberger (1967), Phlips (1974), Powell (1974) and Theil (1975, 1976).
3. Note, however, that the resulting demand functions are different. In the primal case, the solution is a set
of Marshallian or uncompensated demand functions, and in the dual case they are Hicksian, or compen-
sated, demand functions.

DWYER 9781781001288 Part I.indd 164 25/07/2012 11:07

 Almost ideal systems 165

4. The consumers maximization problem implies four constraints on the demand equations. The first is
demand homogeneity which states that an equiproportional change in prices has no effect on the quanti-
ties demanded when real income is held constant. The second constraint is symmetry of the substitution
effects, or Slutsky symmetry. This states that when real income is held constant, the effect of a $1 rise in
the price of commodity x on the consumption of commodity y is exactly equal to the effect on x consump-
tion of a $1 rise in the price of y. The third is the negativity condition (the law of demand, namely that
demand curves slope down when real income remains constant). A fourth is the aggregation or adding up
restriction. The adding up condition implies that expenditures on individual goods must add up to total
expenditure (Spiqi 5 m).
5. A functional form for a complete system of demand functions is said to be flexible if at any given set of
non-negative prices, incomes, and parameters, demand functions can be chosen so that the own, cross-
price and income elasticities are capable of assuming arbitrary values, subject only to the requirement of
theoretical consistency (Pollak and Wales, 1992).
6. Other available flexible functional forms include the Generalized Leontief (Diewert, 1971, 1974) and the
Generalized CobbDouglas (Gallant, 1984; Lewbel, 1987). These are not widely used in the empirical
literature. For an excellent review of alternative flexible functional forms see Thompson (1988).
7. Despite its popularity and wide applications, the relative merits of the LA-AIDS have been debated and
one particular issue concerns the calculation of price elasticities (for details, see Green and Alston, 1990;
1991; Pashardes, 1993; and Buse, 1994). This linear version is the most widely used AIDS variant in the
literature. Despite its popularity and wide application, the relative merits of LA-AIDS have been debated
on several grounds. One issue concerns the specification and estimation of price elasticities (for details,
see Buse, 1994; Pashardes, 1993; Green and Alston, 1990, 1991). Another concerns the overall properties
of the LA-AIDS model (for details see Buse, 1998; Eales and Unnevehr, 1988; LaFrance, 2004; Moschini,
1995). In any event, as Holt and Goodwin (2009) pointed out, all issues pertaining to the specification,
estimation, and interpretation of the LA-AIDS model are rendered moot if instead the Translog price
index a(p) in (8.12) is simply used in estimation.
8. According to Lau (1986), among the five criteria the only area where a compromise can be made is in the
domain of applicability (regularity). This is because most practical applications can be accommodated,
even if the functional form is not globally theoretically consistent, as long as it is theoretically consistent
within a sufficiently large subset of the space of independent variables. For another detailed review of
flexible functional forms and selection criteria see Thompson (1988).
9. A comparison of expenditure elasticities in relation to necessities derived from a version of a regular
flexible functional form (McFadden and CobbDouglas) with the LES and AIDS revealed some
interesting evidence. The movements in expenditure elasticities in relation to food (also clothing and
tobacco) derived from the CooperMcLaren model and the LES follow the same trends. The expendi-
ture elasticity of food, for example, rose from 0.33 to 0.70 from 195354 to 199091 as in the case of
the elasticities derived from the LES (in the LES case, the elasticities increased from 0.42 to 0.64). By
contrast, the elasticities derived from the AIDS took the opposite trend, declining from 0.64 to 0.39. The
movements in the expenditure elasticities derived from the AIDS are closer to the observed facts (that
is, when real incomes grow all goods, and in particular necessities, become less luxurious (Flood et al.,
1984). The Chalfant (1987) results, drawn from his globally regular version of the AIDS, also indicate
irregularities.
10. An implication of the negativity condition is that Hicksian or compensated demand functions will be
non-increasing in own price (that is, the Slutsky matrix will be negative semi-definite). Thus, a simple rule
of thumb to test empirically whether the negativity condition is met is to verify that the compensated own
price elasticities are negative. An alternative suggested in the literature is to re-parameterize the Slutsky
matrix so that the negativity condition may be directly imposed at a point during estimation (for details,
see Diewert and Wales, 1988a; 1988b; Moschini, 1998; Ryan and Wales, 1998).
11. The literature offers several explanations for the over-rejection of the homogeneity condition. These
include particular parameterization of demand systems (see, for example, Byron, 1970a, b), aggrega-
tion biases in commodity groups (Parikh, 1988), misspecifications caused by omitted variables (such as
changes in taste), systematic errors associated with price and quantity data, and errors associated with
dynamic misspecification (Anderson and Blundell (1982).
12. Using techniques developed for estimating cointegrating vectors in the presence of deterministic trends,
Ng (1995) finds that homogeneity holds in many cases.
13. Given the additivity of regression disturbances, one can use the Zellner (1962) method of Seemingly
Unrelated Regressions (SUR) to obtain parameter estimates. However, under this procedure estimates
depend on the equation dropped during estimation (Pollak and Wales, 1992).
14. A few authors have used two alternative models, the Linear Expenditure System (LES) (Pyo et al., 1991;
Sakai, 1988; Smeral, 1988) and the Translog Demand System (Bakkal, 1991; Bakkal and Scaperlanda,
1990). All other studies are based on the Almost Ideal Demand System (AIDS) and its variants.

DWYER 9781781001288 Part I.indd 165 25/07/2012 11:07

 166 Handbook of research methods in tourism

15. Expenditure elasticity shows how the quantity purchased changes (how sensitive it is) in response to a
change in the consumers expenditure/budget (a proxy for income). In the AIDS context, where demand
is measured in terms of expenditure/budget shares, expenditure elasticity refers to changes in expenditure/
budget share of a commodity (say share of expenditure on a particular destination) arising from a change
in the travel budget of a traveller. This is represented by the ratio between percentage change in quantity
demanded and percentage change in expenditure. If the percentage change in the quantity demanded is
greater than the percentage change in consumer expenditure, demand is said to be expenditure elastic or
responsive to changes in consumer expenditure. If the percentage change in the quantity demanded is
less than the percentage change in consumer expenditure, the demand is said to be expenditure inelastic
or not responsive to changes in consumer expenditure. The higher the expenditure elasticity, the more
sensitive consumer demand is to expenditure changes. Although, the measure of expenditure elasticity is
different from that of income elasticity (which is calculated using changes in the income of the traveller),
it is customary to interpret expenditure elasticities as analogous with income elasticities. If the income
(expenditure) elasticity of demand is positive, the good is considered a normal good, and if the income
(expenditure) elasticity of demand is negative, the good is classified as an inferior good. Similarly, if the
percentage change in the quantity demanded is greater than the percentage change in consumer income
(expenditure), the demand is said to be income (expenditure) elastic, or responsive to changes in consumer
income (expenditure). If the percentage change in the quantity demanded is less than the percentage
change in consumer income (expenditure), the demand is said to be income (expenditure) inelastic, or not
responsive to changes in consumer income (expenditure). Those commodities with income (expenditure)
elasticities greater than unity are classified as luxuries and the commodities with income (expenditure)
elasticities less than one (positive in value) are classified as necessities.
16. Chronologically, the next to follow was the study by Sakai (1988) who estimated two demand systems
for business and pleasure travellers based on the Linear Expenditure System for Hawaii. Sakai used
microexpenditure data from individual tourist parties to Hawaii drawn from the Hawaii Visitors Bureau
expenditure surveys. Broad commodity groups included food, lodging, recreation, local transportation,
clothing and a miscellaneous category. Sakai found that for business travellers, lodging is a luxury good.
For pleasure travellers, lodging and transportation are luxury goods, while food and recreation are neces-
sities. Further, he found that business travellers demands are less price sensitive than pleasure travellers
demands are.
17. It should be noted that no strict comparison is warranted here, given that the present study concerns only
a comparatively limited number of commodity groups. Further, the analysis is based on compensated
demand elasticities (cross-price) which measure the pure substitution effect of a price change. This means
either that the real expenditure is held constant, or the consumer has been compensated for the loss of
income associated with the change in the price of the commodity in question.
18. Pyo et al. (1991) were the first to examine the demand for tourism goods and services by domestic
tourists. Based on the Linear Expenditure System they estimated US domestic tourist demand for five
commodity groups: transportation, lodging, food, entertainment and recreation and other goods and
services. Transportation is found to be the most price sensitive of all, and they argued further that the
widely held perception that tourism products are luxuries may result from the income effect on transpor-
tation.
19. Each of these demand systems has identical properties: both have budget share Engel curves that are
linear in the log of total expenditures, and similar aggregation properties; and they have indirect utility
functions that are built up of polynomials in log prices. Moreover, as Lewbel (1989) has shown, the AIDS
and Translog demand systems are almost equal in terms of explanatory power and goodness of fit.

REFERENCES

Anderson, G. and R. Blundell (1983), Testing restrictions in a flexible dynamic demand system: an application
to consumers expenditure in Canada, The Review of Economic Studies, 50, 397410.
Anderson, G.J. and R.W. Blundell (1982), Estimation and hypothesis testing in dynamic singular equation
systems, Econometrica: Journal of the Econometric Society, 15591571.
Arrow, K.J., B.D. Bernheim, M.S. Feldstein, D.L. McFadden, J.M. Poterba and R.M. Solow (2011), 100
years of the American Economic Review: the top 20 articles, The American Economic Review, 101 (1), 18.
Attfield, C.L.F. (1997), Estimating a cointegrating demand system, European Economic Review, 41 (1), 6173.
Bakkal, I. (1991), Characteristics of West German demand for international tourism in the northern
Mediterranean region, Applied Economics, 23 (2), 295304.

DWYER 9781781001288 Part I.indd 166 25/07/2012 11:07

 Almost ideal systems 167

Bakkal, I. and A. Scaperlanda (1991), Characteristics of US demand for European Tourism: a translog
approach, Weltwirtschaftliches Archiv, 127, 119137.
Banks, J., R. Blundell and A. Lewbel (1997), Quadratic Engel curves and consumer demand, Review of
Economics and Statistics, 79 (4), 527539.
Barten, A.P. (1969), Maximum likelihood estimation of a complete system of demand equations, European
Economic Review, 1, 773.
Barten, A.P. (1977), The systems of consumer demand functions approach: a review, Econometrica, 45,
2351.
Berndt, E.R., M.N. Darrough and W.E. Diewert (1977), Flexible functional forms and expenditure distribu-
tions: an application to Canadian consumer demand functions, International Economic Review, 18 (3),
651675.
Buse, A. (1994), Evaluating the linearized almost ideal demand system, American Journal of Agricultural
Economics, 781793.
Buse, A. (1998), Testing homogeneity in the linearized almost ideal demand system, American Journal of
Agricultural Economics, 80, 208220.
Byron, R.P. (1970a), A simple method for estimating demand systems under seperable utility assumptions,
Review of Economic Studies, 37, 271274.
Byron, R.P. (1970b), The restricted aitken estimation of sets of demand functions, Econometrica, 38, 816830.
Carlos, A., P. Daniel and M. Gehlhar (2004), Locating seasonal cycles in demand models, Applied Economics
Letters, 11, 533535.
Chalfant, J.A. (1987), A globally flexible, almost ideal demand system, Journal of Business & Economic
Statistics, 5 (2), 233242.
Christensen, L.R., D.W. Jorgenson and L.J. Lau (1975), Transcendental logarithmic utility functions, The
American Economic Review, 65 (3), 367383.
Coenen, M. and L. Van Eekeren (2003), A study of the demand for domestic tourism by Swedish households
using a two-staged budgeting model, Scandinavian Journal of Hospitality and Tourism, 3 (2), 114133.
Cooper, R.J. and K.R. McLaren (1988), Regular alternatives to the almost ideal demand system, Department
of Econometrics Working Paper No. 12/88, Monash University.
Cooper, R.J. and K.R. McLaren (1992), An empirically oriented demand system with improved regularity
properties, Canadian Journal of Economics, 25 (3), 652668.
Cooper, R.J., K.R. McLaren and P. Parameshwaran, (1991), A system of demand equations satisfying effec-
tively global curvature conditions, Department of Econometrics Working Paper No. 7/91, (Revised Version
1992), Monash University.
Corts-Jimnez, I., R. Durbarry and M. Pulina (2009), Estimation of out bound Italian toursim demand: a
monthly dynamic EC-LAID model, Toursim Economics, 15 (3), 547565.
De Mello, M.M. and N. Fortuna (2005), Testing alternative dynamic systems for modelling tourism demand,
Tourism Economics, 11, 517537.
De Mello, M.M., A. Pack and M.T. Sinclair (2002), A system of equations model of UK tourism demand in
neighbouring countries, Applied Economics, 34, 509521.
Deaton, A. (1979), The distance function in consumer behavior with applications to index numbers and
optimal taxation, Review of Economic Studies, 46 (July 1979), 391405.
Deaton, A. (1986), Demand Analysis, in Z. Griliches and M.D. Intrilligator (eds), Handbook of Econometrics,
Vol III, Amsterdam: North-Holland, pp. 17671839.
Deaton, A. and J. Muellbauer (1980a), An almost ideal demand system, American Economic Review, 70 (3),
312326.
Deaton, A. and J. Muellbauer (1980b), Economics and Consumer Behavior, Cambridge: Cambridge University
Press.
Diewert, W.E. (1971), An application of the Shephard duality theorem: a generalized Leontief production
function, Journal of Political Economy, 79, 481507.
Diewert, W.E. (1974), Applications of duality theory, in M.D. Intriligator and D.A. Kendrick (eds), Frontiers
of Quantitative Economics, Vol. II, Amsterdam: North-Holland, Chapter 3.
Diewert, W.E. (1982), Duality approaches to microeconomic theory, in K.J. Arrow and M.D. Intriligator
(eds), Handbook of Mathematical Economics, Vol II, Amsterdam: North Holland.
Diewert, W.E. and T.J. Wales (1987), Flexible functional forms and global curvature conditions, Econometrica,
55, 4368.
Diewert, W.E. and T.J. Wales (1988a), Normalized quadratic systems of consumer demand functions, Journal
of Business & Economic Statistics, 6 (3), 303312.
Diewert, W.E. and T.J. Wales (1988b), A normalized quadratic semi flexible functional form, Journal of
Econometrics, 37, 327342.
Divisekera, S. (1995), An econometric model of international visitors flows to Australia, Australian Economic
Papers, 34 (65), 291308.

DWYER 9781781001288 Part I.indd 167 25/07/2012 11:07

 168 Handbook of research methods in tourism

Divisekera, S. (2003), A model of demand for international tourism, Annals of Tourism Research, 30, 3149.
Divisekera, S. (2007), Modelling And Estimation of Tourism Demand Elasticities: A Study of Tourist Expenditure
Allocation in Australia, Gold Coast, Australia: CRC Sustainable Tourism.
Divisekera, S. (2008), Modelling regional demand for tourism services: an analysis of regional demand for
domestic tourism in Australia, Global Business & Economics Anthology, 2, 318326.
Divisekera, S. (2009a), Economics of domestic tourism: a study of Australian demand for tourism goods and
services, Tourism Analysis, 14 (3), 279282.
Divisekera, S. (2009b), Ex-post demands for Australian tourism goods and services, Tourism Economics, 15
(1), 153180.
Divisekera, S. (2009c), An analysis of demand for international air transportation and tourism: interac-
tions and dependencies, presented at the second conference of the International Association for Tourism
Economics (IATE), Chiang Mai, Thailand.
Divisekera, S. (2010a), Economics of tourists consumption behaviour: some evidence from Australia,
Tourism Management, 31 (5), 629636.
Divisekera, S. (2010b), Economics of leisure and non-leisure tourist demand: a study of domestic demand for
Australian tourism, Tourism Economics, 16 (1), 117136.
Divisekera, S. (2010c), Regional variations in tourist consumption patterns: a model of tourist expenditure
allocation, Global Business & Economics Review, 2, 101111.
Divisekera, S. and J. Deegan (2010), An analysis of consumption behaviour of foreign tourists in Ireland,
Applied Economics, 42, 16811697.
Durbarry, R. and Sinclair, M.T. (2003), Market shares analysis: the case of French tourism demand, Annals
of Tourism Research, 30 (4), 927941.
Eales, J.S. and L.J. Unnevehr (1988), Demand for beef and chicken products: separability and structural
change, American Journal of Agricultural Economics, 70 (3), 521532.
Eales, J.S. and L.J. Unnevehr (1993), Structural change in U.S. meat demand, American Journal of
Agricultural Economics, 75, 259268.
Eales, J.S. and L.J. Unnevehr (1994), The inverse almost ideal demand system, European Economic Review,
38 (1), 101115.
Flood, L.R., R. Finke and M.C. Rosalsky (1984), How trustworthy are the standard errors of translog coef-
ficients?, Economics Letters, 16, 5962.
Fraser, I. and I.A. Moosa (2002), Demand estimation in the presence of stochastic trend and seasonal-
ity: the case of meat demand in the United Kingdom, American Journal of Agricultural Economics, 84,
8389.
Frisch, R. (1959), A complete scheme for computing all direct and cross demand elasticities in a model with
many sectors, Econometrica, 27, 177196.
Fujii, E., M. Khaled and J. Mark (1985), An almost ideal demand system for visitor expenditures, Journal of
Transport Economics and Policy, 19, 161171.
Gallant, A.R. (1984), The Fourier flexible form, American Journal of Agricultural Economics, 66, 204208.
Goldberger, A.S. (1987), Functional Form and Utility. A Review of Consumer Demand Theory, Boulder, CO:
West View Press.
Green, R. and J. Alston (1990), Elasticities in AIDS models, American Journal of Agricultural Economics, 72,
442445.
Green, R. and J. Alston (1991), Elasticities in AIDS models: a clarification and extension, American Journal
of Agricultural Economics, 73, 874875.
Han, Z., R. Durbarry and M.T. Sinclair (2006), Modelling US tourism demand for European destinations,
Tourism Management, 27 (1), 110.
Hicks, J.R. (1956), A Revision of Demand Theory, Oxford: Oxford University Press.
Holt, M.T. and B.K. Goodwin (2009), The almost ideal and translog demand systems, in D.J. Slottje (ed.),
Quantifying Consumer Preferences (Contributions to Economic Analysis, Volume 288), Bradford, UK:
Emerald Group Publishing Limited, pp. 3759.
Houthakker, H.S. (1960), Additive preferences, Econometrica, 28, 244257.
Judge, G.G., W.E. Griffiths, C.R. Hill and T.-C. Lee (1980), The Theory and Practice of Econometrics, New
York: John Wiley.
Karagiannisa, G. and G.J. Mergos (2002), Estimating theoretically consistent demand systems using cointe-
gration techniques with application to Greek food data, Economics Letters, 74, 137143.
Keuzenkamp, H.A. and A.P. Barten, (1995), Rejection without falsification, on the history of testing the
homogeneity condition in the theory of consumer demand, Journal of Econometrics, 67 (1), 103127.
LaFrance, J.T. (2004), Integrability of the linear approximate almost ideal demand system, Economic Letters,
84, 297303.
Lanza, A., P. Temple and G. Urga (2003), The implications of tourism specialization in the long run: an
econometric analysis for 13 OECD economies, Tourism Management, 24, 315321.

DWYER 9781781001288 Part I.indd 168 25/07/2012 11:07

 Almost ideal systems 169

Lau, L. (1986), Functional forms in econometric model building, in Z. Griliches and M.D. Intriligator (eds),
Handbook of Econometrics, Amsterdam: North-Holland, pp. 15161566.
Lewbel, A. (1987), Fractional demand systems, Journal of Econometrics, 36, 331337.
Lewbel, A. (1989), Nesting the AIDS and translog demand systems, International Economic Review, 30,
349356.
Lewbell, A. and S. Ng (2005), Demand systems with non-stationary prices, Review of Economics and
Statistics, 87, 479494.
Li, G., H. Song and S.F. Witt (2004), Modelling tourism demand: a dynamic linear AID approach, Journal
of Travel Research, 43 (22), 141150.
Lyssiotou, P. (2000), Dynamic analysis of British demand for tourism abroad, Empirical Economics, 25,
421436.
Malinvaud, E. (1980), Statistical Methods of Econometrics, Amsterdam: North-Holland.
Mangion, M.L., R. Durbarry and M.T. Sinclair (2005), Tourism competitiveness: price and quality, Tourism
Economics, 11 (1), 4568.
Moschini, G. (1995), Units of measurement and the stone index in demand system estimation, American
Journal of Agricultural Economics, 77, 6368.
Moschini, G. (1998), The semi flexible almost ideal demand system, European Economic Review, 42, 349364.
Ng, S. (1995), Testing for homogeneity in demand systems when the regressors are non-stationary, Journal of
Applied Econometrics, 10, 147163.
OHagan, J.W. and M.J. Harrison (1984), Market shares of US tourist expenditure in Europe: an econometric
analysis, Applied Economics, 16, 919931.
Papatheodorou, A. (1999), The demand for international tourism in the Mediterranean region, Applied
Economics, 31 (5), 619630.
Parikh, A. (1988), An econometric study on estimation of trade shares using the almost ideal demand system
in the world link, Applied Economics, 20, 10171039.
Pashardes, P. (1993), Bias in the estimating the almost ideal demand system with the Stone index approxima-
tion, The Economic Journal, 103, 908915.
Phlips, L. (1974), Applied Consumption Analysis, 2nd edition, Amsterdam: North-Holland.
Pollak, R.A. and T.J. Wales (1992), Demand System Specification and Estimation, Oxford: Oxford University
Press.
Powell, A.A. (1966), A complete system of demand equations for the Australian economy fitted by a model of
additive preferences, Econometrica, 34, 661675.
Powell, A.A. (1974), Empirical Analytics of Demand Systems, Lexington, Washington, DC: D.C. Heath.
Pyo, S.S., M. Uysal and R.W. McLellan (1991), A linear expenditure model for tourism demand, Annals of
Tourism Research, 18, 443454.
Ryan, D.L. and T.J. Wales (1998), A simple method for imposing local curvature in some flexible consumer
demand systems, Journal of Business and Economic Statistics, 16, 331338.
Sakai, M.Y. (1988), A micro-analysis of business travel demand, Applied Economics, 20, 14811495.
Samuelson, P. (1947), Foundations of Economic Analysis, Cambridge, MA: Harvard University Press.
Shephard, R. (1953), Cost and Production Functions, Princeton, NJ: Princeton University Press.
Sinclair, M.T. and T.C. Syriopoulos (1993), An econometric study tourism demand: the AID model of U.S.
and European tourism in Mediterranean countries, Applied Economics, 25 (12), 15411552.
Smeral, E. (1988), Tourism demand, economic theory and econometrics: an integrated approach, Journal of
Travel Research, 26, 3842.
Stone, R. (1954), Linear expenditure systems and demand analysis: an application to the pattern of British
demand, The Economic Journal, 64, 511527.
Theil, H. (1975), Theory and Measurement of Consumer Demand, Vol. I, Amsterdam: North-Holland.
Theil, H. (1976), Theory and Measurement of Consumer Demand, Vol. II, Amsterdam: North Holland.
Thomas, R.L. (1987), Applied Demand Analysis, London and New York: Longman.
Thompson, G.D. (1988), Choice of flexible functional forms: review and appraisal, Western Journal of
Agricultural Economics, 13 (2), 169183.
White, K.J. (1982), The demand for international travel: a system-wide analysis for US travel to Western
Europe, Discussion Paper No. 82/28, University of British Columbia, Department of Economics.
White, K.J. (1985), An international travel demand model, US Travel to Western Europe, Annals of Tourism
Research, 12 (4), 529545.
Wu, D.C., G. Li and H. Song (2011), Analyzing tourist consumption: a dynamic system-of-equations
approach, Journal of Travel Research, 50 (1), 4656.
Zellner, A. (1962), An efficient method of estimating seemingly unrelated regressions and tests for aggregation
bias, Journal of the American Statistical Association, 57, 348368.

DWYER 9781781001288 Part I.indd 169 25/07/2012 11:07

 9 Hedonic price analysis
Andreas Papatheodorou, Zheng Lei and
Alexandros Apostolakis

INTRODUCTION

Pricing is a strategic choice for all firms. It not only generates revenue for a company to
survive but can also be used as a communicator, as a bargaining tool and a competitive
weapon. The consumer can use price as a means of comparing products, judging relative
value for money or product quality (Brassington and Pettitt, 2007). In tourism, pricing
decisions are often complex. For example, the price of a hotel room is normally affected
by seasonality, type of the room, facilities provided, or even attributes of the external
environment such as noise, pollution, distance from a specific landmark, or outside
views.
From a managerial perspective, it is critically important to understand consumer
perceptions of each of the attributes associated with the price: characteristics that a
customer is willing to make an extra payment for and those which are irrelevant in
the determination of consumer choices and preferences (Chen and Rothschild, 2010).
Hedonic price analysis (HPA) makes it easier to discern which characteristics are valued
by consumers and to what extent (Falk, 2008). In addition, HPA is also able to put a
price estimate onto nonmarket product or service characteristics. For example, Mahan
(1997) found that proximity to river streams had a significant influence on willingness
to pay (i.e., respondents valued the proximity to river streams equal to $13.81 per foot).
Hence, this technique is particularly useful for managerial decision making and evaluat-
ing individual preferences. In this chapter, we will explain the theory of HPA, followed
by an illustration of its application in tourism research, and concluded by a discussion of
managerial implications.

NATURE OF HEDONIC PRICE ANALYSIS AND ITS

EVOLUTION
HPA is an economic valuation technique based on revealed preferences. Economic valu-
ation has received considerable attention in the tourism literature recently as a result of
various developments in the sector. On the one hand, increased levels of competition in
the industry (emergence of new destinations, expansion of the boundaries of the tourism
economy) imply that suppliers of tourism products and services have to improve their
competitiveness in relation to close rivals and incumbents. Practically, this means that
suppliers need to identify the aspect of their business that generates a competitive advan-
tage and capitalize on it. To do so, tourism suppliers have to evaluate individual prefer-
ences for particular attributes of the product or service they bring to the market. The

170

DWYER 9781781001288 Part I.indd 170 25/07/2012 11:07

 Hedonic price analysis 171

above rationale is equally applicable to both destinations and specific tourist attractions
or resources.
On the other hand, the recent emergence of the evidence based principle in tourism
policy and decision making (Banks, 2009; The Countryside Agency, 2006) implies the
need for further justification of public investment or private expenditure in tourism.
In other words, the need for accountability regarding expenditure levels in tourism has
driven stakeholders to adopt economic valuation techniques that reveal individual will-
ingness to pay for a given level of product characteristic. A positive willingness to pay for
a particular product attribute or characteristic is taken to signify a positive contribution
to individual utility levels. Thus, the positive contribution of a given change in the prod-
ucts or services configuration merits or justifies the allocation of additional financial
resources on the basis of improving individual welfare levels.
On these grounds, HPA is used to estimate economic values for attributes or product
characteristics that directly affect prices for tourism products and services in the mar-
ketplace. Revealed preference methods use information from tourists expressed prefer-
ences on existing markets to make statements regarding the demand for tourism goods
and services. As such, revealed preference methods are quite useful in making inferences
regarding the effect of a number of variables or explanatory factors on tourist expendi-
ture patterns (Apostolakis and Jaffry 2009).

BACKGROUND AND TYPE OF PROBLEMS THAT HEDONIC

PRICE ANALYSIS IS DESIGNED TO HANDLE

The HPA has its origin in Lancasters (1971) theory of consumer demand. The theory
holds that consumer products are essentially bundles of separable characteristics
such as comfort, convenience and reliability that can be assessed accurately by value-
conscious consumer (Schwieterman, 1995, p. 292). The theory of HPA was formally
formulated by Rosen (1974) in a seminal paper where he uses a conventional utility-
maximizing approach to derive implicit attribute prices for multi-attribute goods under
conditions of perfect competition (Andersson et al., 2010). The value that consumers
attach to the characteristics is reflected in the price of the differentiated product.
HPA thus aims to disentangle the impact of the various attributes on implicit prices
where positive attributes are expected to boost the overall price and have a positive
effect on individual utility levels. Correspondingly, negative attributes exert downward
pressure on the price and have a negative effect on utility levels (Thrane, 2005). Tourism
resources are characterized as multi-attribute and multi-value resources (Oh, 2009;
Riganti and Nijkamp, 2008), so the application of HPA in this setting offers considerably
better and richer results as compared to other economic valuation techniques.
Another problem addressed by HPA is when researchers have to evaluate non-
market elements of the tourism offering. Increasingly these days the interaction of
tourism and the environment has become more complex than ever before. This is
mainly due to changing consumer preferences and lifestyle and greater environmen-
tal awareness. The increasing convergence between tourism and the environment
implies that a rising number of tourism product attributes cannot be evaluated using
conventional market-based methods, simply because there is no established market

DWYER 9781781001288 Part I.indd 171 25/07/2012 11:07

 172 Handbook of research methods in tourism

for these attributes or resources. Hence, traditional market-based approaches cannot

really capture the effect of an unspoilt ecosystem, or the existence of an area of natural
outstanding beauty in the tourist destination. At the same time, we know that tourists
derive pleasure by visiting or vacationing in or near pristine and unspoilt environments
(Maddison and Bigano, 2003). HPA offers a solution to this conundrum by establishing
surrogate markets to place a premium to the effect of non-market-based attributes on
the tourism product or service on offer (Lise and Tol 2002, Maddison 2001). Regression
techniques make it easier to estimate the implicit price for each attribute (Andersson et
al., 2010) with the parameter of the hedonic price function revealing the marginal value
consumers place on each of the individual attributes (Rigall-I-Torrent and Fluvia,
2011). While semi-logarithmic, log-linear, and linear BoxCox functional forms of the
regression model are all compatible with the theory of hedonic price analysis, most of
the previous research seem to have followed Rosens (1974) advice using the semi-log
form (Andersson et al., 2010). The latter implies that the dependent variable (price of
the evaluated product or service) is of a logarithmic form, whereas all the explana-
tory variables enter the equation in a linear format. The relationship between a tourist
product or service attributes and its associated price may not be a linear one simply
because the market may not be in long run equilibrium. In tourism, this may arise due
to supply side disequilibrium conditions, or temporal variations in supply and product/
service quality.
A basic hedonic price model for the price of a tourism product may be specified as a
function of a number of attributes:

ln Pi 5 a 1 bXij 1 ei (9.1)

where ln Pi is the natural logarithm of the price of a particular tourism product i; Xij is
a vector of attributes j associated with the tourism product; a is the intercept; and ei is a
random error term which is independent and identically distributed with zero expecta-
tion and constant variance. Xij may be measured in logs or levels. The partial derivative
of Pi with respect to each of the characteristics j, Pi/Xij, refers to the marginal implicit
price, which represents the consumers valuation of (or willingness to pay for) the par-
ticular attribute of the tourism product. The predicted price can also be compared with
the observed price to reveal whether the tourism product is significantly over- or under-
priced (Falk, 2008).
A unique feature of hedonic price analysis is the extensive use of dummy variables to
measure the qualitative attributes of a product. To provide precise explanation of each
coefficient associated with the dummy variables, it is necessary to transform the estimated
coefficient by e b 21, where b is the coefficient and e is the base of the natural logarithm.
This transformation gives the estimated effect of the dummy in percentage terms, while
the monetary values can be obtained by multiplying e b 21 by the average level value of
the dependent variable in the sample (Monty and Skidmore, 2003). Moreover, and while
the traditional HPA precludes the introduction of non-attribute variables into the econo-
metric function, the relaxation of this constraint can offer useful insight from a competi-
tiveness and/or marketing perspective. For example, in a HPA study of Mediterranean
tourism, Papatheodorou (2002) used as explanatory variables not only characteristics of
holiday packages but also dummy variables to represent specific tour operators and des-

DWYER 9781781001288 Part I.indd 172 25/07/2012 11:07

 Hedonic price analysis 173

tinations. In this case, a negative coefficient must imply either a bargain or a poor brand
image; conversely, a positive coefficient is consistent with both bad-value-for-money and
a strong brand name. To resolve the conundrum further information on the prevailing
demand conditions is required.
HPA requires a considerable degree of data homogeneity that would allow relevant
comparisons possible. In a study to measure how the overall price of a sun-and-beach
package holiday was determined by the choice of tour operator, choice of destination,
and a number of different attributes associated with the package tour, Thrane (2005)
adopted the following measures to homogenize the data. First, only tours to the Canary
Islands from Oslo airport in Norway were considered. This is because the airfare is a
major determinant of the price of a package holiday. With approximately equal travel
distance from Oslo to the four Canary Islands examined, the effect of distance on the
airfare has been minimized. Second, recognizing that seasonality has a significant effect
on package price, Thrane selected the same week, i.e. the first week of November 2003
as a departure date to control for the effect of seasonality. Finally, as all four Canary
Islands are Spanish, the tourism-production costs at the destinations are assumed at least
partially constant.

APPLICATIONS OF HEDONIC PRICE ANALYSIS IN TOURISM

The hedonic price valuation approach has seen wide applications in the environmental
economics literature (Boyle et al., 1999; Earnhart, 2001; Gibbons et al., 2011; McConnell
and Walls, 2005). In fact, the majority of the applications of the technique are drawn
from the wider environmental economics and management literature (Hunt et al. 2005).
This is mainly due to the multi-attribute nature of the evaluated resources. HPA has
also been extensively used for housing, computer equipment, and agricultural products
(Hamilton, 2007). Recently, there has been a growing interest in applying this method in
tourism research.
Espinet et al. (2003) examined how holiday hotel prices are affected by different hotel
characteristics in the sun-and-beach segment. The effect on price is estimated under the
hedonic function perspective by random effect models. Table 9.1 shows a comprehensive
list of attributes considered by these authors.
The study reveals that compared to the reference 4-star hotels, prices were 64 percent
lower for 1-star, 61 percent lower for 2-star, and 50 percent lower for 3-star hotels. Hotels
located in front of the beach were more expensive, while distance to the town center
had no significant effect. With regards to hotel services, availability of parking space
increased prices by 8.5 percent, but recent renovation, special room equipment, avail-
ability of a garden, swimming pool, or sporting facilities had no significant effect on price.
In a study of 73 hotels in Taipei, Chen and Rothschild (2010) separated hotel rates
into weekdays and weekends. The authors found that internet access and the presence
of a fitness center had a significant effect on weekday rates, while room size had a sig-
nificant effect on weekend rates. In contrast to the Espinet et al. (2003) study, Chen and
Rothschild (2010) found that in Taipei there was a negative relationship between prox-
imity to city center and room rates, on both weekdays and weekends: this is probably
due to the fact that hotels outside the city often offer a wider range of amenities such

DWYER 9781781001288 Part I.indd 173 25/07/2012 11:07

 174 Handbook of research methods in tourism

Table 9.1 List of attributes considered in Espinet et al. (2003)

Town in which hotel is located Parking place

Star category Hairdresser
Number of rooms Outdoor swimming pool
Accepts credit cards Indoor swimming pool
Special access for disabled Music or radio in the room
Admits pets Conference halls
Air-conditioning Sauna
Satellite TV Hairdryer in the room
Lift 24-hour service
Cafeteria Medical care
Shops Squash
Safe Telephone in the room
Heating TV set in the room
Money change Tennis court
Close to town centre Simultaneous translation
Discotheque Video in the room
Located in a historical building Games and recreational activities
Located in a picturesque building Casino
Car repair Recently renovated
Gymnasium Water and wine included in full board price
Golf Number of operators including it in their catalogue
Kindergarten Childrens swimming pool
Suites with living room Restaurant
Garden or terrace Close to the beach
Bicycle rental Balcony in the room
Mini golf View to the sea
Mini-bar in the room Year of first opening
Park for children

Source: Espinet et al. (2003).

as hot springs and sports facilities, hence a positive impact on the price, while the more
competitive environment in the city center exerts a downward pressure on room rates.
In another study, Hamilton (2007) specifically focused on the effect of the coastal
landscape on hotel prices. She found that an increase in the length of open coast in
Schleswit-Holstein, Germany resulted in a rise in the average price of accommodation,
while an increase in the length of dikes would lead to a reduction in the average price of
accommodation. Complementing the above studies, Monty and Skidmore (2003) exam-
ined the willingness to pay for bed and breakfast amenities in southeast Wisconsin in the
United States. The findings showed willingness-to-pay for specific characteristics such as
a hot tub, a private bath, and a larger room. Location and spatial characteristics (such as
day of the week, and period of the year) were also found to be important. However, fire-
places, architectural themes, scenic views, and room service were not statistically signifi-
cant determinants of price. In another study, Garcia-Pozo et al. (2011) applied a hedonic
model to the analysis of campsites in Spain. The results revealed that the variables with

DWYER 9781781001288 Part I.indd 174 25/07/2012 11:07

 Hedonic price analysis 175

the greatest influence on prices were coastal location, the quality of the facilities, quality
certification, and managerial compliance with ecological procedures.
As tourism products are often consumed and produced simultaneously, the physi-
cal environment may have a substantial effect on prices (Rigall-I-Torrent and Fluvia,
2011). Consider for example, two hotels of the same quality and standards, where
one is located in a run-down environment, while the other is located in the same city
but in a more pleasant environment with the presence of attractive cultural facilities
and good provision of restaurants and pubs. The latter is more likely to command a
price premium to reflect its location advantage. To measure the effects of the so-called
public attributes on hotel prices, Rigall-I-Torrent and Fluvia (2011) used a sample
of 279 coastal hotels in Catalonia, Spain. A number of public attributes were consid-
ered, including exclusivity (population), complementary products and services (cultural
and sports facilities, restaurants, marinas), crowdedness, natural environment (caves),
and public safety. They found that most of these public attributes have significant
impact on hotel prices.
HPA has also been applied to study competitiveness levels among holiday resorts.
Using brochure information, Papatheodorou (2002) examines destination competitive-
ness in the Mediterranean region. The price of holiday packages is regressed on a number
of package characteristics as well as operator and location factors for core and periph-
ery destinations, defined by their popularity as featured in the selected tour operators
brochures. Those characteristics that are statistically significant are subsequently used
in a correlation analysis, providing useful insights into the observed price differentials
among core and peripheral holiday destinations. The evidence from the empirical results
suggests that fewer than half of core tourist destinations command a (statistically sig-
nificant) price premium as compared to the benchmark tourist destination (Es Cana,
Spain). In particular, the two Cypriot destinations (Paphos and Protaras) are associated
with the higher willingness to pay, equal to 9.10 percent and 10.3 percent, respectively.
On the other hand, Hammamet in Tunisia and Bugibba in Malta were sold at the highest
discount prices (11.6 percent and 20.9 percent less, respectively). Moreover, the effect
of the hotel chain affiliation is positive on peripheral but negative on core destinations
(positive effect). This difference may be justified in terms of the brand name effect on less
familiar and standardized destinations. An alternative but conceptually similar explana-
tion could be attributed to the effect of information asymmetries between known and less
well-known destinations.
Compared to the growing number of HPA studies in the hotel and tour operating
sectors, research into other areas of tourism is somewhat limited. In a study using a
sample of 10 000 published airfares, Schwieterman (1995) explored the relationship
between carrier reputation for service quality and the price of air travel in the US
market. The study demonstrated that airlines established prices on the basis of 23 service
attributes ranging from ticketing restrictions to flight frequency and travel requirements.
Based on a database of 84 ski resorts, Falk (2008) investigated the relationship between
lift ticket prices and the ski resorts characteristics. The results indicated that ski runs,
transport capacity, share of modern high-speed chairlifts and gondolas, measures of
snow conditions, and access to neighboring ski areas covered by the same lift pass all had
a positive and significant effect on the price of a 1-day lift ticket and a 6-day ski pass. A
ranking of the ski resorts according to their quality characteristics was also conducted.

DWYER 9781781001288 Part I.indd 175 25/07/2012 11:07

 176 Handbook of research methods in tourism

ADVANTAGES AND LIMITATIONS OF HEDONIC

PRICE ANALYSIS CONCEPTUALLY AND FOR POLICY
FORMULATION

There are a number of factors explaining the popularity of HPA in tourism research.
First, tourism competitiveness is, inter alia, a function of the various destination char-
acteristics (Dwyer et al., 2000). In this respect, regressing tourist expenditure on a des-
tinations characteristics could reveal useful information to policy makers on how to
improve competitiveness. Taking advantage of the product characteristics background,
the HPA is more capable of focusing on non-price competitive practices (Rosiers and
Theriault, 1996; Vanslebrouck et al., 2005). This is because consumers (tourists) explic-
itly provide an account of how important a particular product characteristic is to their
(positive) appreciation of a destination. The evidence generated from HPA applications
can assist tourism policy makers in identifying those characteristics enhancing financial
returns from tourist activity. Tourism practitioners can then use this information for
destination management purposes. Hence, HPA is particularly adept in dealing with the
high degree of competitiveness in the contemporary tourism sector (Andropoulos 2011).
Second, HPA is considered a better alternative for the economic evaluation of tourism
resources than other techniques as tourism products and services are dependent upon
natural (and thus non-market) characteristics. For example, the attractiveness of a
Mediterranean destination or resort is largely dependent upon the climate and the nice
weather. In addition, the probability of a tourist visiting an archaeological site (e.g.,
Pompeii in Italy, Knossos Palace in Greece, or Stonehenge in Britain) is also dependent
upon natural and weather conditions. The fact that HPA is not solely dependent upon
the price attribute means that it can offer useful insights to managers and practitioners
regarding the effect of non-price attributes of a products or a services marketing mix
on tourist choice and preference patterns. For example, Aguil et al. (2001) have found
that certain hotel attributes were significant predictors of the overall price that tourists
were willing to pay.
Third, the technique is more flexible and accommodating compared to other more
traditional approaches to tourism demand. This is because HPA has the ability to incor-
porate both demand and supply-side market characteristics (Apostolakis and Jaffry,
2009; Chen and Rothschild, 2010; Monty and Skidmore, 2003). In tourism, as in any
service-related industry, demographic characteristics (e.g., size of family, existence of
dependents) are as important as the quality of product characteristics. By involving and
interacting demand- and supply-side influences, HPA provides more realistic estimates.
Finally, the HPA tourism literature review proves that this is particularly versatile
and adapted to consider several possible interactions in terms of incorporating indi-
vidual characteristics and interactions between market goods and environmental quality
(Champ et al., 2003; Cheshire and Sheppard, 1998; Gibbons et al., 2011).
As for the limitations of HPA one issue arises out of the ex-post (or retrospective)
nature of the technique compared to other economic evaluation methodologies. In
particular, HPA is limited to evaluating the current state of the world. Although useful,
this information offers little opportunity to managers, practitioners, and policy makers
regarding the exploitation of potential competitive advantage a destination or a resort
may have. Thus, HPA offers little guidance regarding future policy initiatives and how

DWYER 9781781001288 Part I.indd 176 25/07/2012 11:07

 Hedonic price analysis 177

these may affect tourist preferences and choice patterns. The ability to offer an insight on
the effect of future policy and managerial decision making on prices and willingness to
pay would be particularly appreciated in a highly competitive tourism industry.
Another issue is related to price changes. In practice, HPA assumes that the price
dependent variable adjusts automatically to changes in the nature of the product/service
attributes. In reality, a time lag may exist between price changes occurring as a result
of attribute changes. This problem may also appear when dealing with new or evolving
tourist destinations, as the number of visitors is still small and there is little information
on product attributes and set prices. In other words, there is heavy dependence of the
technique on data availability and the strong assumption of perfect information from the
consumer perspective to evaluate the effect of changes in product or service attributes on
price levels in an accurate way.
From an econometric perspective, multicollinearity is another problem that may arise.
It may well be the case that superior quality hotels and resorts are only found in green
and environmentally unspoilt locations and environments, whereas lower quality small
establishments with rundown stock of facilities tend to be located in underdeveloped or
poor environments. In such situations, it is difficult for the HPA researcher to disentan-
gle environmental conditions from the nature of the hotel and its facilities in an accurate
and convincing way. The dependence of the HP valuation methodology on spatial effects
(Paez et al. 2001) and the existence of spatial autocorrelation and heterogeneity preclude
the straightforward use of the technique (Long et al. 2007).

FURTHER DEVELOPMENTS AND APPLICATIONS OF

HEDONIC PRICE ANALYSIS IN TOURISM RESEARCH

The previous discussion on the hedonic price analysis may be of particular interest to
policy makers and practitioners focusing on the marketing of specific tourism products,
services and destinations: knowing whether and why these are sold at a discount or a
premium can provide useful insight for the strategy to be followed in terms of market
positioning, pricing, and branding. Interestingly, the hedonic price framework may also
be used in the context of competition analysis to examine market power abuse as a result
of dominance and/or to appraise the pricing effects of potential mergers especially of hor-
izontal nature. The critical and ultimate question to be answered by every competition
inquiry is whether the business entities under investigation can (or have already) profit-
ably abuse(d) their market power by raising their prices to the detriment of consumer
welfare (Stabler et al., 2010). In the presence of horizontal (i.e., variety) and vertical (i.e.,
quality) disparities in the market, purged prices should be first obtained before focusing
on the actual competitive conduct. On these grounds, the hedonic price framework has
been used to quantitatively evaluate the fundamentals behind the observed car price dif-
ferentials in the European Union (LECG, 1999). So far, however, and to the best of this
authors knowledge, the framework has not been used in the context of tourism-related
industries in spite of the existence of powerful oligopolies and oligopsonies especially in
the transport for tourism, hotel and tour operations sectors (Papatheodorou, 2006).
In the following, the application of the technique in the context of a potential merger
appraisal between two airlines (A and B) of the same country is highlighted. For

DWYER 9781781001288 Part I.indd 177 25/07/2012 11:07

 178 Handbook of research methods in tourism

simplicity, we will assume that both airlines share airport hub C as the main base of their
fleet and operations. Airport C is the most important hub in the country; moreover, the
joint market share of A and B in airport C is highly dominant. A and B announce their
intention to hypothetically merge on 1 June 2011 and the Competition Commission
in charge decides to undertake the hedonic price analysis 6 months later, i.e., in early
December 2011 (this time gap is regarded as normal in competition inquiries), to find
out whether the proposed merger would result in a substantial lessening of competition
and higher prices thus harming passengers and tourism in general. The sample should
refer to monthly data for at least the last 36 months and the estimation method should
best rely on panel data analysis comparing the results of Pooled Ordinary Least Squares
(POLS) with models incorporating the existence of fixed and/or random effects (Lei and
Papatheodorou, 2010).
The dependent variable can be average fare on a monthly basis per operating airline
for all domestic and international city pairs using airport hub C as a point of origin or
destination. From a commercial perspective, this is equivalent to the average passenger
revenue per route on a monthly basis: for example, what was the average fare of airline
A for the itinerary CD in March 2011; or what was the average fare of airline B for the
itinerary EC in November 2011? The implementation of yield management techniques
may render the calculation of average fares a difficult task; however, knowing the load
factor and the revenue per leg can prove of great help. Ideally, the average fare should
be provided on a single ticket (i.e., per leg) basis, i.e., the CD fare should be quoted
separately from the DC. The analysis should be based on net fares but taxes and other
charges should also be considered separately. Moreover, all city pairs to/from C where
A and B operate should be taken into account. The focus on C stems from the concerns
regarding the potential abuse of market power by the two airlines in their hub airport.
The Competition Commission in charge does have the legal power to ask A and B to
provide the suitable data. Alternatively, these may be purchased by Global Distribution
Systems operators (such as Amadeus, Galileo, etc.) who store such data on Marketing
Information Data Transfer (MIDT) tapes.
The first set of independent variables should consider cost factors. Ideally, the
Competition Commission in charge should be provided with explicit cost data by the
airlines under investigation; nonetheless, the reliability of such data is not always easy
to validate. As an alternative, four major variables should be included: (1) average jet
fuel price on a monthly basis, (2) representative type of operating aircraft per leg on a
monthly basis, (3) average load factor per leg on a monthly basis, and (4) block time
(i.e., flight time plus taxiing) involved in each city pair. Fuel cost is undoubtedly of major
importance in the aviation industry. Moreover, airline productivity is largely affected
by the type of aircraft whereas the load factor can provide an indication of the average
cost. Finally, block time may provide a proxy for labor, maintenance, and other costs.
These independent variables are expected to be positively associated with the dependent
variable.
The second group of independent variables should consider the general macro-
economic environment. Three major variables may be included, i.e., inflation (expected
to have a positive association with the dependent variable); unemployment (expected to
have a negative association), and GDP growth rate (expected to have a positive associa-
tion) all expressed on a monthly basis. Given the existing concerns about the level of

DWYER 9781781001288 Part I.indd 178 25/07/2012 11:07

 Hedonic price analysis 179

sovereign debt in various countries, the interest rate spread between the government
bond and a risk-free benchmark may also be used as a macroeconomic indicator.
The third set of independent variables should consider mode and carrier substitutabil-
ity. In particular, a dummy variable can be included for mode substitutability. It can take
the value of zero when the distance of a city pair is less than 500 km and surface transport
is available and/or when travelling by sea is an acceptable alternative time-wise; it can
take the value of one in all other cases. This dummy variable is expected to be positively
related with the dependent variable, i.e., a lower degree of mode substitutability should
be associated with higher prices. Regarding carrier substitutability, a variable showing
the average number of airlines operating directly in each city pair on a monthly basis
should be included; as a better alternative, the average monthly value of the Herfindahl
Hirschman concentration index (i.e., which is equal to the sum of the squared market
shares in terms of capacity) should be used. The lower the number of operating carriers
(and the higher the value of the HH index) is, the higher the average fare is expected to
be. Only direct flights should be considered given the inherent inconvenience involved in
indirect flights.
The fourth group of independent variables should consist of dummy variables taking
the value of one when the flight is operated by a particular carrier and zero in all other
cases. In line with the discussion on hedonic price analysis in previous sections of this
chapter, a positive association between the dummy variables and the dependent variables
is consistent with the conclusion that certain airlines can command a price premium pos-
sibly due to their strong brand name; conversely, a negative association may reveal that
certain airlines sell their product at a discount possibly due to a poor brand image. It
would also be advisable to add a dummy variable taking the value of zero when the flight
is related to the domestic network and one when the flight is international to highlight
any potential differences in the pricing policies implemented by the various airlines.
The final set of independent variables should consist of dummy variables related to
time. The first dummy variable should be associated with seasonality, i.e., taking the
value of one for months considered to be of peak period for the city pairs involving
airport C. Understandably, this dummy variable should be positively associated with
the dependent variable as a result of the use of price discrimination techniques by the
airlines. Most importantly, however, another dummy variable should be included to
take account of the merger announcement. This can take the value of zero for the whole
sample period including 31 May 2011 and one thereafter (i.e., from 1 June 2011 when
airlines A and B announced their intention to merge until the date that the hedonic price
analysis is undertaken sometime in early December 2011).
In fact, the very essence of the econometric analysis lies on the coefficient of the last
dummy variable. In particular:

A statistically significant positive coefficient means that the average air fare after
the merger announcement has been higher than before as a result of both unilateral
(i.e., related to the fact that the merged entity can feel confident enough to plan
ahead its pricing strategy without considering the interdependence with other car-
riers) and coordinated (i.e., related to the increased possibility of collusion among
the incumbent airlines as a result of the reduction in the number of participating
carriers in the market due to the merger) effects. A fortiori, if airlines A and B

DWYER 9781781001288 Part I.indd 179 25/07/2012 11:07

 180 Handbook of research methods in tourism

found it profitable to increase their average fare after the announcement but before
the actual approval of the merger it is almost certain that they would continue
doing so after the approval of the merger. In other words, a positive coefficient
means that the competitive pressure has been reduced allowing market partici-
pants to increase prices. As a result and using backward induction logic, it will be
possible to consistently argue that airlines A and B are neighboring competitors.
In conjunction with the coefficients of the cost variables, substitutability and vari-
ables related to company brand name and nature of network, it will be possible to
derive concrete conclusions on whether there are differences between domestic and
international flights and whether the proposed merger will result in cost savings for
the consumers or not.
A statistically significant negative coefficient means that the average air fare after
the merger announcement has been lower than before. This is consistent with
a benevolent stance of the airlines under investigation as a result of increased
efficiency or alternatively a manipulative behavior to secure merger approval. In
this case and to avoid witch-hunting, it would be appropriate to conduct sup-
plementary qualitative analysis possibly based on the implementation of a Delphi
technique with focus groups of experts expressing their opinions on the possible
impacts of the merger. Supplementary analysis should also be undertaken, if the
coefficient proves to be statistically insignificant.

Having the above in mind, the hedonic price analytical framework can be applied as a
valid technique in competition analysis to appraise among others the effect of a poten-
tial merger on prices. Given the existence of large oligopolies (and oligopsonies) in the
tourism sector as well as the complementarity which characterizes the various compo-
nents of the aggregate tourism product, HPA may prove useful to tourism competition
economists in the future.

CONCLUSIONS

This chapter presented the basic foundations of hedonic price analysis applied in tourism
research. HPA is an economic valuation approach that, conceptually speaking, belongs
to the revealed preferences family. It is based on Lancasters characteristics approach
and postulates that price differences among competing tourism destinations or resources
reflect differences in the benefits or value to the consumer derived from the consumption
of various characteristics in each destination or resource. The methodological reliance
of the technique on Lancasters characteristics approach, make hedonic price analysis
particularly adept to the study of a multi-attribute and multi-value sector such as the
tourism industry (Papatheodorou, 2001). The effect of each individual attribute or char-
acteristic could be measured through the use of regression analysis. In particular, coef-
ficients with a positive sign are taken to imply a positive effect on prices (and consumers
valuation of the tourist resource), whereas the opposite is the case for coefficients with a
negative sign.
HPA has received considerable attention in tourism recently due to two main reasons.
First, increased levels of competition both within the industry and as a result of the expan-

DWYER 9781781001288 Part I.indd 180 25/07/2012 11:07

 Hedonic price analysis 181

sion of the boundaries of the tourism economy make it imperative to estimate properly
the effect of each component of the tourism product on offer. Second, the expansion of
the evidence based requirement in terms of justification of tourism-related policy making
makes the technique particularly appealing to managers, practitioners and policy makers
(Rigall-I-Torrent and Fluvia, 2011). In addition, the recent convergence of tourism and
non-market environmental resources (illustrated through the increasing popularity
of the sustainability concept in tourism operations, environmentally friendly tourism
practices, and ecotourism resources) necessitates the need for an evaluation approach
that would allow to estimate the associated value placed by consumers/tourists on these
non-market based characteristics of the tourism product or service on offer. Again, HPA
is well suited to accommodate this need. The technique is based on the construction of
surrogate markets to evaluate the effect of each product or service attribute on implicit
prices consumers/tourists are willing to pay. In this respect, and despite a number of
caveats discussed earlier, HPA offers a valid alternative method of examining demand
patterns compared to other (market-based) methods.

REFERENCES

Aguil, P., J. Alegre and A. Riera (2001), Determinants of the price of German tourist packages on the island
of Mallorca, Tourism Economics, 7, 5974.
Andersson, D.E., O.F. Shyr and J. Fu (2010), Does high-speed rail accessibility influence residential property
prices? Hedonic estimates from southern Taiwan, Journal of Transport Geography, 18, 166174.
Andropoulos, C. (2011), Latest developments at EU level concerning tourism competitiveness, The 10th
European Tourism Forum, Krakow Poland, The European Commission.
Apostolakis, A. and S. Jaffry (2009), Examining expenditure patterns of British tourists to Greece,
International Journal of Tourism Policy, 2 (3), 187205.
Banks, G. (2009), Evidence based policy making: what is it? How do we get it?, Australian Government
Productivity Commission, Canberra.
Boyle, K., J. Poor and L. Taylor (1999), Estimating the demand for protecting freshwater lakes from eutrophi-
cation, American Journal of Agricultural Economics, 81 (5), 11181122.
Brassington, F. and S. Pettitt (2007), Essentials of Marketing 2nd edition, Harlow: Pearson Education.
Champ, P.A., K.J. Boyle and T.C. Brown (2003), A Primer on Nonmarket Valuation, Boston: Kluwer
Academic Press.
Chen, C. and R. Rothschild (2010), An application of hedonic pricing analysis to the case of hotel rooms in
Taipei, Tourism Economics, 16 (3), 685694.
Cheshire, P. and S. Sheppard (1998), Estimating the demand for housing, land and neighbourhood character-
istics, Oxford Bulletin of Economics and Statistics, 60 (3), 357382.
Dwyer, L., P. Forsyth and P. Rao (2000), The price competitiveness of travel and tourism: a comparison of 19
destinations, Tourism Management, 21, 922.
Earnhart, D. (2001), Combining revealed and stated preference methods to value environmental amenities at
residential locations, Land Economics, 77 (1), 1229.
Espinet, J.M., M. Saez, G. Coenders and M. Fluvia (2003), Effect on prices of the attributes of holiday hotels:
a hedonic prices approach, Tourism Economics, 9 (2), 165177.
Falk, M. (2008), A hedonic price model for ski lift tickets, Tourism Management, 29, 11721184.
Garcia-Pozo, A., J.L. Sanchez-Ollero and D.M. Marchante-Lara (2011), Applying a hedonic model to the
analysis of campsite pricing in Spain, International Journal of Environmental Research, 5 (1), 1122.
Gibbons, S., S. Mourato and R. Guilherme (2011), The amenity value of English Nature: a hedonic price
approach, Spatial Economics Research Centre (SERC) Discussion Paper DP0074, London School of
Economics and Political Science.
Hamilton, J.M. (2007), Coastal landscape and the hedonic price of accommodation, Ecological Economics,
62, 594602.
Hunt, L., P. Boxal, P. Englin and W. Haider (2005), Remote tourism and forest management: a spatial
hedonic analysis, Ecological Economics, 53, 101113.

DWYER 9781781001288 Part I.indd 181 25/07/2012 11:07

 182 Handbook of research methods in tourism

Lancaster, K.J. (1971), Consumer Demand: A New Approach, New York: Columbia University Press.
LECG (1999), Quantitative Techniques in Competition Analysis, Hayes, UK: Office of Fair Trading.
Lei, Z. and A. Papatheodorou (2010), Measuring the effect of low-cost carriers on regional airports commer-
cial revenue, Research in Transportation Economics, 26, 3743.
Lise, W. and R. Tol (2002), Impact of climate on tourism demand, Climatic Change, 55 (4), 429449.
Long, F., A. Paez and S. Farber (2007), Spatial effects in hedonic price estimation: a case study in the city of
Toronto, Centre for Spatial Analysis Working Paper Series N.020, McMaster University.
McConnell, V. and M. Walls (2005), The Value of Open Space: Evidence from Studies of Nonmarket Behaviour,
Washington DC: Resources for the Future.
Maddison, B. and A. Bigano (2003), The amenity value of the Italian climate, Journal of Environmental
Economics and Management, 45 (2), 319332.
Maddison, D. (2001), In search of warmer climates? The impact of climate change on flows of British tourists,
Climate Change, 49 (2), 103208.
Mahan, B. (1997), Valuing wetlands: a property pricing approach, in Evaluation of Environmental
Investments Research Group, US Army Corps of Engineers, Portland, USA.
Monty, B. and M. Skidmore (2003), Hedonic pricing and willingness to pay for bed and breakfast amenities
in Southeast Wisconsin, Journal of Travel Research, 42, November, 195199.
Oh, C. (2009), Assessing tourists multi-attribute preferences for public beach access, Coastal Management,
37 (2), 119135.
Paez, A., T. Uchida and K. Miyamoto (2001), Spatial association and heterogeneity issues in land price
models, Urban Studies, 38 (9), 14931508.
Papatheodorou, A. (2001), Why people travel to different places?, Annals of Tourism Research, 28 (1),
164179.
Papatheodorou, A. (2002), Exploring competitiveness in Mediterranean resorts, Tourism Economics, 8 (2),
133150.
Papatheodorou, A. (2006), Corporate rivalry, market power and competition issues in tourism: an introduc-
tion, in A. Papatheodorou (ed.) Corporate Rivalry and Market Power: Competition Issues in the Tourism
Industry, London: IB Tauris, pp. 119.
Rigall-I-Torrent, R. and M. Fluvia (2011), Managing tourism products and destinations embedding public
good components: a hedonic approach, Tourism Management, 32, 244255.
Riganti, P. and P. Nijkamp (2008), Congestion in popular tourist areas: a multi-attribute experimental choice
analysis of willingness to wait in Amsterdam, Tourism Economics, 14 (1), 2544.
Rosen, S. (1974), Hedonic prices and implicit markets: product differentiation in pure competition, Journal of
Political Economy, 82 (1974), 3455.
Rosiers, F. and M. Theriault (1996), Rental amenities and the stability of hedonic prices: a comparative analy-
sis of five market segments, The Journal of Real Estate Research, 12 (1), 1736.
Schwieterman, J. (1995), A hedonic price assessment of airline service quality in the US, Transport Reviews,
15 (3), 291302.
Stabler, M., A. Papatheodorou and T. Sinclair (2010), The Economics of Tourism, 2nd edition, London:
Routledge.
The Countryside Agency (2006), Planning for sustainable settlements in the High Weald AONB; evidence
based policy making, High Weald Area of Outstanding Natural Beauty Authority.
Thrane, C. (2005), Hedonic price models and sun-and-beach package tours: the Norwegian case, Journal of
Travel Research, 43 (2005), 302308.
Vanslebrouck, I., G.V. Huylenbroeck and J.V. Meensel (2005), Impact of agriculture on rural tourism: a
hedonic pricing approach, Journal of Agricultural Economics, 56 (1), 1730.

DWYER 9781781001288 Part I.indd 182 25/07/2012 11:07

 10 Factor analysis
Lindsay W. Turner and Chau Vu

INTRODUCTION

Factor analysis as a generic term that is increasingly used to encompass all forms of
analysis intended to uncover latent structure within a set of data. The very intent of
such an analysis is fundamental to a wide variety of problems encountered in tourism
because, as a social science, there is significant requirement to measure unstructured con-
cepts, as opposed to structured items with readily measurable attributes such as height,
weight, length, and width. In tourism the concepts requiring measurement relate to social
systems involving values, attitudes, motivations, risks, satisfaction, and beliefs. Tourism
is a combination of study fields in the social sciences where a great many problems
contain such unstructured elements.
Consequently, there is a huge body of published research in tourism that uses factor
analytic methods. The objective of this chapter is to discuss the background of the
methods involved, explain how the methods work, highlight some examples of their
applications and discuss their advantages and limitations.

EVOLUTION OF FACTOR ANALYSIS

Spearman (1904) is given the recognition for developing the factor analysis technique
but like all complex methods there is a history of considerable research development
reaching through to current use. Spearmans model focused upon deriving just one
common function to derive intellectual activity. The work continued through to 1930
with the major contributors Cyril Burt, Karl Holzinger, Truman Kelly, Karl Pearson,
and Godfrey Thomson. It was Pearson in 1901 who was arguably the first to develop a
practical method for factor analysis when he proposed the method of principal axis fac-
toring. By 1930 the idea of a single function was understood to be too limiting in describ-
ing latent structure. In 1937 Holzinger and Swineford developed bi-factoral theory
whereby there remained one single overall factor for all variables, but also several other
factors describing subsets of variables in the total set of variables. The most significant
development of modern factor analysis was made by Thurstone (1931, 1938, 1947) when
he developed the centroid calculation method (1931) commonly used before the develop-
ment of computers, and the concept of simple structure (1947).
The description of simple structure is stated technically by Harman (1976, p. 87):

The truly remarkable contribution of Thurstone was the generalization of Spearmans tetrad-
difference criterion to the Rank of the correlation matrix as the basis for determining the
number of common factors. He saw that a zero tetra-difference corresponded to the vanishing
of a second-order determinant, and extended this notion to the vanishing of higher order deter-
minants as the condition for more than a single factor.

183

DWYER 9781781001288 Part I.indd 183 25/07/2012 11:07

 184 Handbook of research methods in tourism

It is thought likely here that this statement explains nothing to most tourism readers.
Therefore the approach taken in the further discussion tends to be less technical.
Thurstone set conditions for simple structure and these conditions mean that, ideally,
only one factor should contribute substantially to a given variable, and the other factors
should contribute not at all. The definition of contribution set the scene for much debate.
The value of contribution set as 1 is considered overly restrictive and unlikely, but the
condition should still be high, and is commonly interpreted to be between 0.6 and 0.8. So
a factor should load highly for one variable (0.8) and hardly at all on the other factors,
preferably zero or close to zero. Hence, with several variables related to a single factor
and all variables relating to a much smaller set of factors the factors have the capacity to
be defined (named) as constructs and derive simple structure.
This early development occurred before computers, and so the calculations were
theoretical, for example at that time a regression line of best fit would be selected by eye
from a graph. The development of computers allowed for the application of the centroid
method, and was aided by Hotelling (1933) who developed the mathematics for princi-
pal components. So from the late 1950s just about every unstructured set of data was
analysed and the methods moved beyond principal components to image factor analysis
(Guttman, 1953), canonical factor analysis (Harris, 1962; Rao, 1955), alpha factor analy-
sis (Kaiser and Caffrey, 1965) and minres (Harman and Jones, 1966). The development
away from principal components is possibly best defined by Jreskog (1966, 1967, 1969,
1970) who used earlier work by Bock and Bargmann (1966) and Lawley (1940) to test
hypotheses, and developed confirmatory factor analysis.
A more detailed discussion of this history is given in an excellent text by Lindeman
(1980).
Hence, several different factor analytic methods developed under the general heading
factor analysis with different intent, and these differences have caused considerable
confusion ever since. Many tourism research articles do not clearly specify which tech-
nique is used, although in more modern times this is not so much the case. Moreover,
the development of hypotheses is often necessarily vague in factor analysis and leads to
further confusion. Finally, it is not adequate to replace the specific method title such as
principal components analysis (PCA) or confirmatory factor analysis (CFA) with the
generic title factor analysis, because this leaves the reader uncertain, and it is not nec-
essary to do so.

TYPES OF METHOD

Principal Component Analysis (PCA) is the most common use of factor analysis because
it is descriptive, but provides significant insight into the latent structure of data that
can often be used for further analysis, for example in structural equation modeling
(SEM). The SEM software can also be used specifically to do factor analytic methods.
Consequently, having described the latent structure in one set of data, hypotheses could
be developed to use CFA analysis on a different set of data, or if the sample is large
enough, on a subset of the original sample. In PCA, the factors are used to account for all
of the variance for each variable, including variance shared with other variables (covari-
ance). Consequently, there must be the same number of factors as variables in order to

DWYER 9781781001288 Part I.indd 184 25/07/2012 11:07

 Factor analysis 185

explain all the variance both unique to a variable and shared between them because, at
least theoretically, each variable will have some variance unique to itself. However, the
objective is to reduce the number of original variables to a smaller set of components that
describes hidden structure within the data set, so that only the components with several
variables with high simple structure loadings are the components that are named.
In more technical terms the analysis uses the complete correlation matrix of the
original data with one on the diagonals. A covariance matrix could be used if the volume
of the data has special relevance such as the volume of tourist arrivals. Principal com-
ponents analysis calculates the component accounting for the most variance first and
stepwise successively less variance each time, so the last components extracted can be
expected to have only one variable loaded to them which must represent just the unique
variance for that variable.

Mathematically, PCA is clean and avoids the question of estimating communalities

that form the diagonal of the reduced correlation matrix in CFA. PCA also allows
component scores to be directly calculated. The wide range of topics analysed in
tourism are given by example in Table 10.6. Confirmatory Factor Analysis (CFA) is used
to determine whether a particular set of hypotheses defining the number and type of
factors applies to particular data. The hypotheses require specific theoretical definition,
presumably justified by previous research, or theoretical construction such as a concep-
tual framework or definition from PCA. At the very least, the factors must be named
in advance on the basis of theory, and possibly the variables defining these names at
least suggested. As such, CFA is potentially more powerful in deriving conclusions to
research problems. The method analyses only the variance in the correlation matrix
that is shared (common) with other (not necessarily all) variables. Consequently, the
number of factors extracted may be less than the number of variables, which in research
terms is more parsimonious in defining the original data set as founded on underly-
ing structure. The problem is determining the shared variance (covariance) in the first
place.
The most common method is to calculate the squared multiple correlation coefficient
of each variable against all the other variables to create the diagonal in the matrix (com-
munality), so the sum of these diagonal elements becomes the total common variance.
This is a reasonable guess of the communality but is not theoretically pure. The conse-
quence of this process is to weight the importance of each variable by the strength of
its correlation with all other variables. When the factors are extracted they are pulled
closer to those variables with the higher communalities. zgener and raz (2006) studied
customer relationship management in small and medium sized tourism enterprises using
CFA and Susskind et al. (2003) examined consumer apprehensiveness toward internet
use.

Image Factor Analysis (IFA) is an alternative method of calculating the common vari-
ance and as such is an alternative to CFA. It uses multiple regression to predict the values
of each variable using the other variables as independent variables. The part of the vari-
able that can be predicted becomes the image and the rest the anti-image. The R2 values
become the communalities in the diagonal of the matrix and the off diagonal elements
are adjusted so no eigenvalues (see below) are negative (referred to as gramian). There

DWYER 9781781001288 Part I.indd 185 25/07/2012 11:07

 186 Handbook of research methods in tourism

are few examples of the method and it can be argued to have the same diagonals as CFA
as a starting point so there is no particular advantage in using the method.

Alpha Factor Analysis assumes the variables are randomly sampled from a large set of
variables to test for reliability in the factors generated. It uses Cronbachs alpha which
is a measure of the intercorrelations between the variables as a measure of reliability,
and only factors obtaining positive alphas are extracted in the order highest to lowest.
Generally, few factors are extracted and this makes an analysis structure difficult to
interpret.

Canonical Factor Analysis, termed Raos canonical factoring, involves re-scaling the
correlation matrix by the unique variance. The higher the uniqueness of a variable
the less weight it has in the re-scaling. In this way the variables that have the highest
common variance have more weight in extracting the factors. The factors that are
derived and their scores are invariant of scale and this is the main reason for selecting
this technique, because it removes the decision as to whether a correlation or covariance
matrix is used. However, this aspect is also evident in both image and alpha factoring
and is not unique to this method. In the methods above the fundamental differences are
as follows.
PCA uses the correlation matrix with one in the diagonal; CFA uses the correlation
matrix with communalities in the diagonal; IFA uses the covariance matrix with squared
multiple correlations in the diagonal, and anti-image factor analysis uses the covariance
matrix with one minus the squared multiple correlations in the diagonal; and canonical
factor analysis uses a matrix of correlations divided by the product of the standard devia-
tions of the uniqueness. As such there is a strong mathematical relationship between
these techniques as shown by Harris (1962).
Another interesting aspect is that, in analysis conditions, the results from PCA and
CFA are often similar, and indeed as the level of unique variance declines the closer the
results become. This adds yet another potential confusion to factor analysis. However,
this issue cannot be used to justify indiscriminant use of the methods as they are dis-
tinctly different.

Factor Regression Analysis is a combination of a factor analysis and regression. The

main reason for doing this is to overcome multicollinearity between independent vari-
ables. PCA can reduce a number of correlated variables to a smaller number of uncor-
related latent variables provided the analysis remains orthogonal, refer to Massy (1965).

ANALYSIS

In order to explain clearly the analytic procedure an example using a simple random
data set is used. It is more mathematically direct to use an example of PCA as a start-
ing point because it avoids the need to calculate communalities, it allows factor scores
to be directly calculated rather than estimated, there is an empirically based method for
determining the number of components and most of the discussion equally relates to all
forms of factor analysis.

DWYER 9781781001288 Part I.indd 186 25/07/2012 11:07

 Factor analysis 187

Correlations

The discussion above has assumed the reader understands correlation, and it is necessary
to take the discussion to some further detail and this is done here in graphic form, or geo-
metrically, to avoid the more technical approach of matrix algebra.
Assuming we have two variables X1 and X2 and the correlation between these two vari-
ables is 0.75 then r12 5 r21 5 0.75. Assuming the data for the two variables are normally
distributed then the mean is zero and the standard deviation (S) is one.
Hence we can show the correlation between the two variables as a straight line:
_______r12 __________ ______r21 __________
X1 :__________________:_________ X2 :__________________:_________
S1 S2

Figure 10.1 Correlation between two variables

so that the correlation r 5 0.75 is three quarters of the length of the standard deviation.
Of course if the correlation was perfect then the value of r would be 1 and both lines
would be the same length.
For the two lines to be compared on the same diagram they must have the same origin
point, but this requires knowledge of the angle between the two lines. The geometric rule
is that a perpendicular line from the three-quarters point (in this case) on each variable
must meet the end of the line representing the standard deviation. It is common to think
of this as one line casting a shadow on the other, in this case, with a length of 0.75, see
Figure 10.2.

A D

0.75

41 41
O 0.75 C B

Figure 10.2 Comparison of lines

Basic trigonometry can be used to calculate the angle, where the cosine of the angle
between the two lines represents the correlation between the two variables. Cosines like
correlations range between 21 and 11. A cosine of zero is equal to an angle of 90o or an
r 5 zero. A line at 90o or a right angle is termed orthogonal and represents an orthogonal
relationship. So cos21 0.75 5 41o 419.
The lines representing the correlations in a correlation matrix will almost certainly
create a diagram in three-dimensional space. So assuming the values given in Figure
10.3.

DWYER 9781781001288 Part I.indd 187 25/07/2012 11:07

 188 Handbook of research methods in tourism

Correlation Matrix Angles

X1 X2 X3 X1 X2 X3
X1 1.00 0.69 0.92 X1 0 46 23
X2 0.69 1.00 0.20 X2 46 0 79
X3 0.92 0.20 1.00 X3 23 79 0

Figure 10.3 Correlation matrix and angles

It is not possible to draw three lines in two-dimensional space, whereby X3 is both 23o
from X1 and 79o from X2, but it is possible in three-dimensional space. The number of
dimensions is equal to the number of variables.

Extracting Components

The objective in deriving the components is to derive new variables that represent groups
of existing variables. Each grouping variable (component) theoretically represents what-
ever is described by the group of variables as loading on the component; or stated oth-
erwise, the underlying hidden (latent) structure described by that group of variables that
is not described until the group is identified to exist. This process is similar to calculating
the mean of a set of data.
Each component is extracted in turn with the first component as the line that is as
close as possible to all the lines representing all the variables at the same time using the
centroid method. Given our correlation matrix, Figure 10.4.

X1 X2 X3
X1 1.00 0.69 0.92
X2 0.69 1.00 0.20
X3 0.92 0.20 1.00
SUM 2.61 1.89 2.11 6.6 = Total Sum
1.01 0.74 0.82 = SUM/!ww
Total Sum
0 42 35 = Angles

Figure 10.4 Correlation matrix

The ratio of the sum of the correlations for each variable to "TS is the correlation of
each variable with the component or how close it is to the average. The angles and other
values are rounded, and given the component is also the same as an existing variable, it
has a standard deviation of one and hence the same line length.
The matrix above shows that X1 has the highest intercorrelation at 2.61 with all the
other variables including itself. The highest possible value for the total sum (TS) is the
number of variables squared (n2) assuming the highest correlation possible (r 5 1), or 33
5 9. So the maximum sum for each variable is "TS 5 "9 5 3. Therefore "TS is the
new variable (principal component).

DWYER 9781781001288 Part I.indd 188 25/07/2012 11:07

 Factor analysis 189

X1

X2

X3
X5
C1
X6

X7 X4

C2
X8

Figure 10.5 The centroid method

This centroid method is not fully accurate, so it is not used given the capacity to
conduct fast matrix algebra calculations by computer. However, the result is the same in
terms of locating the new variable in multi-dimensional space. In order to demonstrate
the procedure of locating each component a matrix greater than order three (as used in
the simple example above) is required, because we need to locate more than one principal
component and to do this we need at least six variables. For each of the eight variables
in the following example three separate values are calculated:

(1) The correlation (cosine of the angle)

(2) The angle between the first component and each variable.
(3) The correlation squared (coefficient of determination r2) which is the proportion of
the variance associated with the component for each variable.

The correlations are termed component loadings and interpreted the same way as
Pearson product moment coefficients, so that the square of the component loading (L)
shows the proportion of the variance for each variable associated with the component
(C1). In the example above the highest contributor to the component one is variable X4
with 97 percent of the variance in C1 related to X4. The sum of the squared component
loadings is the eigenvalue, usually referred to as lambda and here lambda one (l1) for the
first component.
Consequently the formula for an eigenvalue is:

DWYER 9781781001288 Part I.indd 189 25/07/2012 11:07

 190 Handbook of research methods in tourism
n
li 5 a L2ij (10.1)
j51

Table 10.1 Correlation and angle values

Variable Correlation (r) Angle r2

X1 0.1736 80 0.0301
X2 0.8910 27 0.7939
X3 0.9511 18 0.9046
X4 0.9659 15 0.9330
X5 0.7986 143 0.6378
X6 0.6428 130 0.4132
X7 0.4695 118 0.2204
X8 0.4848 61 0.2350
S 5 4.1680 5 l1

Where: Lij is the loading for variable j on the component i; and li is the eigenvalue for
component i.
The eigenvalue must be compared to the total variance in the correlation matrix to be
interpreted. As already stated the correlation matrix is made up of a number of variables
(n) each with a standardized variance of one, so the total variance is n 1.02 5 n. Also
note that the variance is the square of the standard deviation. Therefore the eigenvalue
can be expressed as a percentage of n, and for the first component becomes:

100/(l1/n) 5 100(4.1680/8) 5 52.10%

Therefore, component 1 accounts for 52 percent of the total unique and common
variance in the analysis. Component 1 is the first average variable and the component
loadings (equivalent to correlation coefficients) state how much of the variance for
each variable is accounted for by component 1. In this case the variables that load
highly are X4, X3, X2, and X5, but what of X6? This question becomes an issue of debate
discussed later, the question being how high does a variable need to load on a compo-
nent to be a significant contributor to account for the variance of a given component?
There is no test of significance for factor analysis and this is an issue of criticism. If a
variable loads above 0.5 then it is not possible for any other loading on any other com-
ponent to be higher, because there is only 0.49 or so residual variance. However, com-
ponents can theoretically be equally explained by equal proportions of one variable.
To move on and derive the second component, which seems a likely prospect, because
100 52.10 5 47.90% of the total variance is residual and remains unaccounted for, we
assume the correlation r12 5 0.6743 and take L1 5 0.1736 (the first loading for X1 on the
first component) and L2 5 0.8910; with L12 5 0.0301 as part of the correlation r11 associ-
ated with component one, L22 5 0.7939 as part of r22 associated with component one, and
L1 L2 as part of r12 associated with the component then:

0.1736 3 0.8910 5 0.1547

DWYER 9781781001288 Part I.indd 190 25/07/2012 11:07

 Factor analysis 191

Table 10.2 Possible values for the second component

Variable Correlation (r) Angle r2

X1 0.9848 170 0.9698
X2 0.6691 132 0.4477
X3 0.3420 110 0.1170
X4 0.1737 80 0.0302
X5 0.4226 65 0.1786
X6 0.7660 40 0.5868
X7 0.9063 25 0.8214
X8 0.7193 44 0.5174
S 5 3.6689 5 l2

and

r12 2 (L1)(L2) 5 0.6743 2 0.1547 5 0.5196

Thus, the correlation between X1 and X2 with the correlation of the first component
removed is 0.5196. This correlation is termed a partial correlation. The influence of the
first component is removed in the above manner to produce a new correlation matrix
made up of partial correlations (rij 2 (Li)(Lj), where i and j are variables).
In the calculation of the residual variance the partial correlations are low for variables
that have already correlated highly with component 1, and potentially higher for succes-
sive new components, where the variable has not already correlated highly with compo-
nent 1. Also the successive components must be orthogonal to the previous component,
because the residuals are not correlated with the previous component(s) (otherwise they
would not be residual). So on Figure10.5 the second component is depicted at right
angles to the first component.
The second component from a correlation matrix is thus the average pattern among
the residuals of the first component, and the second eigenvalue is less than the first,
because the amount of variance available for the second component is less than the first,
and so each successive component will have lower eigenvalues. Given the sum of the first
two eigenvalues:

4.1680 1 3.6689 5 7.8369

Nearly all the total variance (n 5 8) is explained by the two components.

Communality

A loadings matrix results from the analysis and is commonly presented in the form of
Table 10.3.
The communality is the sum of the squared loadings for each variable (h18). It is the
proportion of the variance for each variable accounted for by all the components. The
communalities in Table 10.3 are close to one because the two components account for
97.96 percent of the variance. This is not usual because in most studies there are a larger

DWYER 9781781001288 Part I.indd 191 25/07/2012 11:07

 192 Handbook of research methods in tourism

Table 10.3 Loadings matrix

Loadings
C1 C2 C3 ... n h2
Variable X1 0.1736 0.9848 ... 1.0
X2 0.8910 0.6691 ...
X3 0.9511 0.3420 ...
X4 0.9659 0.1737 ...
X5 0.7986 0.4226 ...
X6 0.6428 0.7660 ...
X7 0.4695 0.9063 ...
X8 0.4848 0.7193 ...
Eigenvalue 4.1680 3.6689 ...
% 52.10 45.86
Cumulative % 52.10 97.96

number of variables in the study and unless the number of components extracted equals
the number of variables the communality will be less than one. If a communality value
is greater than one this indicates there is a technical problem with the analysis, such as a
sample size that is too small.

Component and Factor Scores

Scores give the strength of the relationship between each case in the sample and each
principal component. It is common practice to standardize the scores as Z scores with a
mean of zero and standard deviation of one, so that they are comparable with each other.
The interpretation for isolating outlier cases is similar to analysing the residuals from
regression. The larger the score the further away the case is from the line of best fit or
the mean line which is the component. It is possible to determine if there are any patterns
in the sample data (characteristics of particular cases) that cause particular components
to load in a particular way. There will be as many scores as there are cases times the
number of components K.
Since the total explained variance of a component is given as the eigenvalue the stand-
ard deviation of all the non-standardized scores on a component is equal to the eigen-
value in principal components. Therefore, the simplest way to standardize component
scores is to divide the raw score by the eigenvalue. Drey and Sarma (2010) used standard
scores in the information search for motivation based tourism, and Vu and Turner (2009)
used component scores to list 150 countries comprising different development structures,
based on different components, and to assess the economic structure of world tourism.
Factor scores cannot be directly calculated because CFA factors contain only
common variance, whereas the components have all variance both common and unique.
Factor scores must be derived from regressing the original data matrix onto the loadings
matrix. This assumes that, on average, the proportions of common and unique variance
are the same for each observation and this is likely not the case. The only way around
this problem is to use image factor analysis, which uses only the common variance after

DWYER 9781781001288 Part I.indd 192 25/07/2012 11:07

 Factor analysis 193

X1

X2

X3

X5

X6
X4
X7

C1
X8

Figure 10.6 Alternative location for original specification

the unique variance has been removed, so the factor scores can be calculated directly.
However, because only the unique variance is used in image factor analysis the origi-
nal data has been transformed and the loadings matrix is different to CFA. Correia et
al. (2008) used CFA factor scores in a second order analysis to assess satisfaction with
quality and price in hospitality examining satisfaction with the gastronomic experience.

Rotation

The initial solution may not be optimal because the mean location for the first compo-
nent or factor may miss obvious variable groupings. For example, what if the original
specification had placed the first location as in Figure 10.6.
This is a reasonable mean location for all the variables and misses the potential for
the two groups identified previously. However, the calculation of the loadings may be as
given below with an eigenvalue of 2.8, which is much lower than the previous eigenvalues
calculated originally for components 1 and 2.
The obvious solution to the problem is to rotate the components, probably clock-
wise, but at the same time maintaining their orthoganality to achieve a better solution.
Originally this was done by eye but obviously this creates the problem that different
analysts could determine different rotations for the same data.
The first issue to determine is the objective of the rotation. The objective is given by
Thurstone (1947) is the description of simple structure, whereby variables would load

DWYER 9781781001288 Part I.indd 193 25/07/2012 11:07

 194 Handbook of research methods in tourism

Table 10.4 Correlation and angle values

Variable Correlation (r) Angle r2

X1 0.0872 95 0.0052
X2 0.5736 55 0.3290
X3 0.6691 48 0.4477
X4 0.9272 22 0.8597
X5 0.4540 117 0.2061
X6 0.3420 110 0.1170
X7 0.2588 105 0.0670
X8 0.8746 29 0.7649
S 5 2.7966 5 l1

Table 10.5 Ideal simple structure

Factor
Variable 1 2 3
1 X 0 0
2 X 0 0
3 X 0 0
4 0 X 0
5 0 X 0
6 0 X 0
7 0 0 X
8 0 0 X

highly on only one component and close to zero on all others, so that an ideal simple
structure would be as shown in Table 10.5.
Where there are three components accounting for the eight original variables in
descending order of importance from 1 to 3. Alternatively, the objective could simply be
to define more distinct groupings.
The rotation of principal components strictly speaking makes little sense because the
centroid method should have already located optimally, and the concept of simple struc-
ture as a target for rotation is not relevant to principal components analysis, because it
relates only to common variance. The issue is a difficult one, and confusing because what
is rotated within a principal components analysis is the smaller set of components that
have highly loaded variables on them, with the other components ignored. When this
smaller set of variables is rotated this is akin to factor analysis and this is why there is
often little difference between the results of either method on the same data.
Rotation of components will likely improve the fit of the components to the vari-
ables and allows principal components to be the best method for group identification.
However, if the objective is to achieve simple structure (and this should already be
hypothesized to exist) then CFA is the better method.
The significant issue with rotation is to find an objective procedure for the rotation,
and there are several choices, but only two general groupings orthogonal and oblique

DWYER 9781781001288 Part I.indd 194 25/07/2012 11:07

 Factor analysis 195

rotation. Orthogonal rotation keeps the components or factors at right angles as they
rotate and oblique rotation allows the angle to be other than 90o.

Orthogonal Rotation

The original orthogonal rotation (Carroll, 1953; Ferguson, 1954; Neuhaus and Wrigley,
1954; Saunders, 1953) was termed Quartimax. It works on the row loadings on the basis
that the optimal result occurs when the factor (component) passes directly through
the variable. So if b1 and b2 are the co-ordinate locations of the variable, if a factor
(component) passes directly through the point its product will be minimized. The term
Quartimax refers to maximizing the sum of the fourth powers of the rotated factor coef-
ficients, and this can be shown to be equivalent to maximizing the sum of variances of the
squared factor loadings in the rows of the loading matrix.
An alternative is Varimax rotation (Kaiser, 1956, 1958), which is superior in terms of
achieving simple structure. Varimax works on the column loadings as opposed to the
row loadings. Varimax maximizes the sum, over the m factors, of the column variances
of the squared coefficients of the loading matrix. If all the coefficients in the column are
either 0 or 1, then one aspect of simple structure is achieved of maximized variance. Most
importantly, Kaiser showed that the solution is factorally invariant. That is if successive
samples are drawn from the same population the factor loadings remain invariant using
Varimax. This gives the method strength or power, because it allows for inferences about
the total population to be made from a sample set of measures. Examples of tourism
research using varimax rotation are widespread and include Chu and Chois (2000)
examination of factors determining perceived hotel performance and satisfaction, Lee et
al.s (2006) examination of the motivations of gamblers, and Yuan et al.s (2004) exami-
nation of the motivations of wine festival attendees.
There are other less used orthogonal rotation methods including Transvarimax, which
contains Equamax and Ratiomax (Saunders, 1962), and Parsimax (Crawford, 1967).

Oblique Rotation

Oblique rotation may result in obtaining a closer fit to simple structure and as such is
more relevant to CFA. By allowing the angle between the factors to change from 90o
in a general sense there is a higher likelihood of identifying groups according to simple
structure, but the difference between these groups may not be high (orthogonal) and so
there is a built in assumption that the resulting factor structure is yielding truly independ-
ent latent measures, which may not always be the case. Ultimately, it follows that if all
the factors are rotated, that groups originally identified will be subdivided by subsequent
rotations and ultimately all the variables will each have their own factor.
There are two main oblique methods, Oblimin (Carroll, 1960) and direct Oblimin
(Jennrich and Sampson, 1966). Oblimin is the most commonly used method and it is
a trial and error procedure to determine if there is a better fit to simple structure. The
method was used by Galloway (2002) for the psychographic segmentation of park visi-
tors, and by Lim (2010) to analyse the variables affecting e-marketing adoption by UK
independent hotels. In oblique rotation two factor loadings are obtained, the structure
loading and pattern loading. The structure loading is the same as the orthogonal loading

DWYER 9781781001288 Part I.indd 195 25/07/2012 11:07

 196 Handbook of research methods in tourism

matrix in regard to interpretation, where it describes the proportion of the variance asso-
ciated by a given variable to a given factor. The pattern loading identifies the correlation
between the variable and the factor independent of other factors, and as such is a partial
correlation. As such the pattern loading explains how much of the proportion of the
structure matrix is unique variance.
The communalities, eigenvalues and sums of squared loadings are not interpretable
from an oblique rotation.
Other methods of oblique rotation include Little Jiffy (Kaiser, 1970), Oblimax
(Saunders, 1961), Promax (Hendrikson and White, 1964), and Maxplane (Cattell and
Muerle, 1960). There are few research articles in tourism using these rotations although
some published studies have used Promax including Snyman and Saayman (2009) who
attempt to identify the key factors influencing foreign direct investment in the South
African tourism sector, and Yoo and Chon (2008) who assess the factors affecting con-
vention participation decision making.

CONFIRMATORY FACTOR ANALYSIS USING LISREL AND

AMOS

The alternative to using a standard statistical package for estimating a CFA is to use
the LISREL or Amos software developed from modeling by Karl Jreskog. Amos
is the more commonly used package (although it is less sophisticated than LISREL)
because it is interfaced with SPSS. The terminology used by Amos differs to the stand-
ard factor analytic terms, whereby loadings are referred to as weights and variables are
sometimes referred to as common factors. The method used is not principal factoring
as described above, but the Maximum Liklihood method (ML). The main drawback
is that the method assumes the data is normally distributed. If the data is strongly dis-
tributed away from normal this method is questionable (Curran et al., 1996; Hu et al.,
1992). These methods derive a CFA based upon oblique rotation and that assumes the
resulting factors are correlated and not orthogonal. The degree of correlation can be
measured, and this aspect, along with the capacity to define the specific hypothesized
relationship to be tested (which variables account for which factor) and to define this
relationship by a path diagram (graphically) provides added strength to undertaking
factor analysis, provided that the factors do not need to be orthogonal, and provided
that the theory tested is quite strongly founded. Examples of the many studies include
Chu (2008) who used Lisrel 8.3 to study the work values of hospitality students, and
Reisinger and Turner (1999a) who studied the cultural dimensions of Japanese tourism.
The analysis result from graphics defined theory based relationships can also be dis-
played graphically in standardized form, and this is also a significant advantage. For a
fictional analysis of motivations the outcome might be two latent dimensions (factors)
of Risk and Escape defined by three explanatory variables each. This relationship is
displayed graphically in Figure 10.7 where the value on the curved line gives the squared
multiple correlation weight between the two factors as 0.55. The loadings (weights) are
given as squared multiple correlations so that 87 percent of the variance in Freedom is
accounted for by the factor Escape. The remaining 13 percent is not accounted for and
comprises error.

DWYER 9781781001288 Part I.indd 196 25/07/2012 11:07

 Factor analysis 197

0.37
Excitement e1
0.61
0.33
Experience e2
0.57
Risk
0.72
Adventure e3
0.85

0.55 0.87
Freedom e4
0.93
Escape 0.66 0.44
0.72 Nature e5
0.52
My Achievement e6

Figure 10.7 Analysis of motivations

The model allows for additional statistical measures of the reliability of the resulting
model. There are several measures and the most commonly reported measures are Chi-
square (CMIN) and its associated p-value, GFI (General Fit Index), NFI (Normed Fit
Index), CFI (Comparative Fit Index), RFI (Relative Fit Index), (PNFI) (Parsimonious
Normed Fit Index) and RMSEA (Root Mean Squared Error of Approximation).
Chi-square is a measure of the overall fit of the model but is very sensitive to sample
size. In particular, if the sample size exceeds 300 the likelihood of a small chi-square is
defined as unlikely (Hakstian et al., 1982; Harris and Harris, 1971, MacCallum, 1990).
Additionally, if the data is not normally distributed then chi-square can inflate. So in
most applications chisquare is not appropriate (Jreskog and Srbom, 1989). The
GFI measures the relative amount of the variance and covariance accounted for by the
model and varies between 0 and 1 where the marginal acceptance level is 0.9. It is better
adjusted for degrees of freedom as the AGFI. The NFI, CFI, and RFI are all relative
values comparing the model against a baseline model (independence model) which is a
particularly bad model fit. The values tend to range between zero and one, or are forced
to, so values close to one indicate the better fit relative to the baseline model. There are
several options for the baseline model and it can be argued the independence model is
not the best choice. It also needs to be understood that these measures are not absolute
measures of fit. Therefore, a successful measure close to one may only indicate that the
model is relatively better than a very poor option, but not that the model is intrinsically a
good fit. The general argument is that the value of NFI should be 0.9 or higher, and this
tends to imply the same for the other measures. The PNFI is the NFI adjusted by degrees
of freedom and this is done as a parsimony adjustment of the degrees of freedom for
the model, divided by the degrees of freedom for the baseline model times the NFI. The
RMSEA is a useful statistic to allow for data size and a large number of variables and

DWYER 9781781001288 Part I.indd 197 25/07/2012 11:07

 198 Handbook of research methods in tourism

factors (that is a complex model) because F0 (a reliability measure which incorporates

no penalty for model complexity) is adjusted as the RMSEA by degrees of freedom, and
gives an approximation of the error in the path diagram. The RMSEA should be below
0.08 and certainly not above 0.1.
This model is very simple and can be made more complex with second order con-
structs, and the identification of correlation between the variables. As such, it is very easy
to re-run the model with adjustments to derive the best fit. This issue becomes debatable
and is not to be recommended, because the loss of theory that this adjustment implies
creates a data-mining solution that might have little relationship to the original theory.
Refer to Reisinger and Turner (1999b) for a detailed discussion on the use of the meth-
odology in tourism.
Because SEM allows the modeling of factors as predictors there is no need to calcu-
late factor scores, and this is another advantage of this technique, as it overcomes the
problem of estimating factor scores from CFA.

APPLICATIONS IN TOURISM

There are many examples of factor analysis used in tourism research and too many to
examine all publications. Taking a sample from the Journal of Travel Research (JTR)
from 2000 (Volume 39) to July 2011 (Volume 50) there are 61 articles published using
factor analysis of the 587 articles published in total, or 10 percent of all articles used factor
analysis. JTR provides a good example of a cross section of tourism research, as it is a
quality general tourism research journal publishing across a wide range of study fields.
Table 10.6 displays a summary of the articles published in JTR, including the area
of study. These papers exclude studies that use factor analysis as a preliminary step
to structural equation modeling (SEM). This is somewhat ambiguous, and it could be
argued SEM is a form of regression extension to factor analysis, but since the objective is
different in the conceptual frame, these studies are excluded. Studies that use regression
as a further or preliminary step in the analysis either of the factor (component) scores, or
the original data, are noted.
Table 10.7 provides a summary of the papers in JTR that use the Jreskog (ML regres-
sion) approach to CFA.
Table 10.6 provides several insights into the use of factor analysis in tourism research.

(1) Although the articles tend not to discuss the issue specifically, there is a large ratio
between the number of variables and cases. On average 19 cases are used for each
original variable.
(2) The vast majority of papers use PCA, and the vast majority of papers are clear in
stating what type of factor analysis is used.
(3) Most papers use a Varimax rotation and are intending to maximize the difference
between the derived components.
(4) Most papers present the loadings, eigenvalues, and levels of explained variance and
where they do not it is often possible to calculate the statistics from the information
provided.
(5) Cronbachs alpha is the most common measure of reliability.

DWYER 9781781001288 Part I.indd 198 25/07/2012 11:07

 Table 10.6 Summary of factor analysis 20002011 in Journal of Travel Research

Authors Vol. No. Yr Sample Vars. Type Factors Rotation Eigen- % var. Loadings Relia- Corr. Criteria Regres- Study
(N) values bility Cut sion Area
Jang et al. 38 (3) 2000 523 13 PCA Oblique Yes Yes Yes Cronbach 0.298 1 One Scores Gambling
KMO Scree

DWYER 9781781001288 Part I.indd 199

Kaiser
Henthorne 38 (3) 2000 1500 19 PCA Varimax No No Yes No 0.587 NS No Marketing
Phillips and 38 (4) 2000 100 18 CFI 6 Varimax Yes Yes Yes Cronbach 0.61 1 One Scores Planning
Moutinho 60% var
Scree
Heung and 38 (4) 2000 220 15 PCA 4 Varimax Yes Yes Yes Cronbach 0.4 1 One Scores Retail
Cheng Perception
Nicholson 39 (4) 2001 1300 20 PCA 4 Varimax Yes Yes Yes Cronbach 0.408 1 One No Events
and Pearce 20 PCA 5 Motivations
20 PCA 5
20 PCA 6
Burnett and 40 (1) 2001 NS 30 PCA 4 Oblique Yes Yes Yes NS 0.59 1 One No Disability

199
Baker Attitudes
Beliefs
Baloglu and 40 (1) 2001 234 14 PCA 4 NS Yes Yes Yes NS 0.4 1 One Yes Senior Travellers
Shoemaker 13 PCA 7 Yes Yes Yes Motivations
Preferences
Pennington- 40 (1) 2001 485 27 PCA 9 Varimax Yes Yes Yes Cronbach 0.465 1 One No Women
Gray and Travellers
Kerstetter Benefits
Wong and 40 (1) 2001 200 13 PCA 4 Varimax Yes Yes Yes KMO 0.6 1 One No Culture
Lau Bartlett Values
Cronbach
Juric et al. 40 (3) 2002 636 15 PCA 1 Varimax No No No Cronbach 0.573 NS Scores Ecotourism
Reisinger 40 (3) 2002 618 73 PCA 5 Oblique R No Yes Yes Cronbach 0.61 1 One No Culture
et al. Rules
Values
Interaction
Perceptions
Satisfaction

25/07/2012 11:07
 Table 10.6 (continued)

Authors Vol. No. Yr Sample Vars. Type Factors Rotation Eigen- % var. Loadings Relia- Corr. Criteria Regres- Study
(N) values bility Cut sion Area

DWYER 9781781001288 Part I.indd 200

Johns and 40 (3) 2002 1099 12 PCA 4 Varimax No Yes Yes Cronbach 0.527 Scree No Marketing
Gyimthy 19 PCA 4 Varimax No Yes Yes Cronbach 0.543 Scree
30 PCA 4 Varimax No Yes Yes Cronbach 0.491 Scree
Hsu and Lee 40 (4) 2002 817 55 PCA 6 Varimax Yes Yes Yes Cronbach 0.4 1 One No Senior
Travellers
Attributes
Perdue 41 (1) 2002 6342 3 PCA 1 Varimax Yes No No Cronbach, 0.6 1 One No Skiing
4 PCA 1 Varimax KMO 0.5 Crowding
Bartlett Satisfaction
Sirakaya 41 (1) 2002 464 82 PCA 7 Varimax Yes Yes Yes Cronbach 0.4 1 One Scores Development
et al. 464 24 PCA 2 Varimax Attitudes
Perceptions
Kim and 41 (2) 2002 2688 6 PCA 2 Varimax Yes Yes Yes Cronbach 0.5 1 One Yes Marketing

200
Crompton 2373 5 PCA 1 Varimax Yes Yes Yes Cronbach 0.57 Behaviour
Loyalty
Vincent and 41 (2) 2002 969 27 PCA 5 Varimax Yes Yes Yes Bartlett 0.3 1 One No Ecotourism
Thompson Cronbach Scree Perceptions
Sonmez and 41 (2) 2002 552 56 PCA 6 Varimax Yes Yes Yes Cronbach 0.45 1 One No Destination
Sirakaya 26 PCA 4 Varimax Yes Yes Yes Cronbach 0.51 KMO Image
17 PCA 2 Varimax Yes Yes Yes Cronbach 0.49
Susskind et 41 (3) 2003 518 25 PCA 2 NS Yes Yes Yes Cronbach 0.4 Scree Yes Internet
al. 243 21 CFA 2 NS Yes Yes Yes Cronbach 0.4 Scree Apprehension
329 14 CFA 2 NS Yes Yes Yes Cronbach 0.4 Scree
Morais et al. 41 (4) 2003 279 40 PCA 1 Promax Yes Yes Yes Cronbach 0.416 1 One No Resource
279 29 PCA 2 Promax Yes Yes Yes Cronbach Theory
Perceptions
Bencken- 42 (1) 2003 407 10 PCA 3 Varimax Yes Yes Yes Cronbach 0.401 1 One No Attractions
dorff and Perceptions
Pearce
Revilla and 42 (1) 2003 139 25 PCA 5 Varimax Yes Yes Yes Cronbach 0.39 1 one No Authenticity
Dodd Scree Perceptions

25/07/2012 11:07
 Poria et al. 43 (1) 2004 304 NS PCA 3 Oblimin No Yes Yes Cronbach 0.4 NS No Heritage
136 NS PCA 3 Oblimin No Yes Yes Cronbach 0.4 NS Motivations
Reisinger 43 (3) 2005 582 NS PCA 3 Varimax Yes Yes Yes Cronbach 0.6 1 One Yes Travel Risk
and
Movondo
Pearce and 43 (3) 2005 1012 74 PCA 14 Varimax Yes Yes Yes KMO 0.4 No Motivations to

DWYER 9781781001288 Part I.indd 201

1 One
Lee Cronbach travel
Choi and 43 (4) 2005 308 54 PCA 7 Varimax Yes Yes Yes KMO 0.442 1 One No Sustainable
Sirakaya 48 PCA 7 Varimax Yes Yes Yes Bartlett 0.540 Tourism
Cronbach Attitudes
Kwan and 44 (2) 2005 519 30 PCA 8 Varimax Yes Yes Yes Cronbach 0.43 1 One No Gaming
McCartney Scree Perceptions
Poria et al. 44 (3) 2006 282 17 PCA 5 Oblique Yes Yes Yes Cronbach 0.505 1 One No Heritage
Motivations
Formica and 44 (4) 2006 135 20 NS 4 NS Yes Yes Yes Cronbach 0.427 1 One Yes Destination
Uysal Attractions
Hsu et al. 44 (4) 2006 464 16 NS 3 Varimax Yes Yes Yes Cronbach 0.429 1 One No Marketing
KMO Benefits

201
Bartlett Sought
Hudson and 44 (4) 2006 140 31 PCA 4 NS No Yes Yes Cronbach 0.7 1 One Yes Film Tourism
Ritchie Marketing
Initiatives
Ekinci and 45 (2) 2006 250 27 PCA 3 Varimax No Yes Yes KMO 0.45 1 One No Branding
Hosany Bartlett Perceptions
Cronbach
Ollenburg 45(4)2007 250 15 PCA 5 Varimax No Yes Yes Bertlett 0.5 1 One No Farm Tourism
and Oblimin KMO Motivations
Buckley
Murphy et 46 (1) 2007 124 20 NS 4 NS Yes Yes Yes NS 0.455 1 One No Branding
al. 126 20 NS 3 NS Yes Yes Yes 0.430 Personality
Walters et al. 46 (1) 2007 180 NS NS 2 NS NS Yes Yes Cronbach 0.691 NS No Marketing
Reichel et al. 46 (2) 2007 579 32 PCA 8 Varimax NS Yes Yes Cronbach 0.55 NS No Perceived Risk
Luo and 46 (4) 2008 335 15 PCA 3 Varimax Yes Yes Yes Cronbach 0.54 1 One No Environment
Deng 14 PCA 4 Varimax 0.48 Beliefs
Motivations

25/07/2012 11:07
 Table 10.6 (continued)

Authors Vol. No. Yr Sample Vars. Type Factors Rotation Eigen- % var. Loadings Relia- Corr. Criteria Regres- Study
(N) values bility Cut sion Area

DWYER 9781781001288 Part I.indd 202

Rittichai- 46 (4) 2008 251 19 PCA 4 Promax Yes Yes Yes Cronbach 0.51 1 One No Motivations
nuwat
Wang and 47 (1) 2008 130 20 PCA 2 Varimax Yes Yes Yes KMO 0.43 1 One No Attitudes
Pfister Cronbach
Yoo and 47 (1) 2008 279 42 PCA 6 Promax Yes Yes Yes Cronbach 0.47 1 One No Conventions
Chon Scree Decision-
making
Brey et al. 47 (2) 2008 2716 18 PCA 3 Varimax Yes Yes Yes Cronbach 0.451 1 One No Hospitality
Ratings
Vong 47 (3) 2009 495 21 PCA 5 Varimax Yes Yes Yes Cronbach 0.47 1 One No Gambling
Attitudes
Stepchen- 47 (4) 2009 540 100 PCA 15 Direct No Yes Yes Cronbach 0.416 NS No Destination
kova et al. 540 100 PCA 15 Oblimin No Yes Yes Image
Gil et al. 47 (4) 2009 252 16 PCA 5 Varimax No Yes Yes Cronbach NS NS Yes Museum

202
7 PCA 1 Varimax No Yes Yes KMO 0.524 Image
13 PCA 4 Varimax No Yes Yes Bartlett NS Motivations
3 PCA 1 Varimax No Yes Yes 0.919 Satisfaction
Zhou and 48 (1) 2009 1165 26 PCA 4 Varimax Yes Yes Yes KMO 0.427 1 One No Olympics
Ap Cronbach Perceptions
Poria et al. 48 (1) 2009 166 NS PCA 2 Promax Yes Yes Yes Cronbach 0.553 1 One No Heritage
Bartlett Scree Preferences
Woosnam 49 (3) 2010 69 71 PCA 4 Varimax Yes Yes Yes KMO 0.481 1 One No Beliefs
et al. 69 71 PCA 4 Varimax Yes Yes Yes Bartlett 0.439 Scree Behavior
455 71 CFA 4 NS NS NS Yes Validity 0.513 Interaction
Solidarity
Hosany et al. 49 (4) 2010 200 44 PCA 3 Varimax Yes Yes Yes NS 0.55 1 One No Destination
Emotions
Bronner and 50 (1) 2011 419 20 PCA 5 Varimax No Yes Yes Cronbach 0.53 1 One No Internet
de Hoog Motivations
Wilkins 50 (3) 2011 300 7 PCA 2 Oblique No No Yes Cronbach 0.508 NS No Souvenirs
Attitudes

25/07/2012 11:07
 Table 10.7 Summary of CFA using LISREL (Amos) 20002011 in Journal of Travel Research

Authors Vol. No. Yr. Sample No. Vars. Type Number of Measures of Reliability Study Area
(N) Analysis Factors
Petrick and 41 (1) 2002 325 7 SAS 2 CMIN, CFI, GFI, NFI Golf Tourism

DWYER 9781781001288 Part I.indd 203

Backman Perceived Value
Morais et al. 41 (4) 2003 279 14 SAS 4 CMIN, CFI, NNFI Resource
Theory
Perceptions
Petrick 43 (1) 2004 792 25 EQS 5 CFI, NFI, RMSEA Perceived Value
Al-Sabbahy et al. 42 (3) 2004 116 12 LISREL VIII 2 GFI, AGFI, NFI, CFI, Hospitality
120 12 LISREL VIII 2 RMSEA, Cronbach Perceived Value
Snepenger et al. 45 (2) 2006 353 12 AMOS 5.0 1 to 2 CMIN, GFI, CFI, NFI, Motivations
RFI, PNFI, RMSEA
Nyaupane and 46 (4) 2008 332 10 LISREL 8.2 3 CMIN, GFI, CFI, RMSEA, Travel
Andereck Cronbach Constraints

203
Sirakaya-Turk et al. 46 (4) 2008 1817 7 LISREL 8.72 7 CMIN, GFI, AGFI, CFI, Sustainable
NFI, RMSEA Cronbach Tourism
Attitudes
Yoo and Chon 47 (1) 2008 279 25 LISREL 8.3 5 CMIN, GFI, CFI, NNFI, Conventions
RMSEA Decision
Making
Woosnam and 49 (3) 2010 455 37 EQS 6.1 4 CFI, RMSEA, Cronbach Beliefs,
Norman Behavior,
Interaction,
Solidarity
Hosany et al. 49 (4) 2010 200 23 LISREL 8 5 CMIN, GFI, AGFI, Destination
NNFI, CFI, RMSEA, Emotions
SRMR
Wilkins 50 (3) 2011 300 7 NS 7 CMIN, AGFI, CFI, Souvenirs
RMSEA, SRMR, Attitudes
Cronbach

25/07/2012 11:07
 204 Handbook of research methods in tourism

(6) The cut-off for significant components is mostly determined by an eigenvalue of one.
(7) The level of variable allocation to a component is in most cases very low at 0.4.
(8) There is a wide range of study areas examined using the technique.

Limitations

These points raise several questions about the use of factor analysis in tourism research.

(1) What sample size is required?

(2) Is the correct type of analysis applied to the problem?
(3) Is the correct validation given for the choice of rotation?
(4) What statistics should be reported? What reliability is required?
(5) What cut-off determines the number of useable components?
(6) What level of loading is required for a variable on a component?

It is not necessarily correct to say that current practice, as sampled by the articles in JTR,
are the required standards.
The sample size required depends upon several factors including the size of the original
population of data. In tourism the population sizes are usually very large because they
relate to total numbers of tourists. In this case the accuracy of the sampling frame can be
more important than the number of cases and a carefully selected representative sample
might not be large. Consequently, the manner in which the sample is selected can be just
as important as the ratio of cases to variables. If the selection of the cases is random then it
is generally argued that the ratio is excellent at 20:1 or above, good at 10:1, and acceptable
only at 5:1. Despite the high overall average of cases to variables in the JTR sample there
are many examples where the sample size is very small relative to the number of variables
analysed. However, just because a recommendation on ratios can be found in a statistics
text does not make it correct; for example Gorsuch (1983) recommends a sample size of
5:1 and Nunnally (1978) and Everitt (1975) recommend 10:1. These ratios are assuming
that a multivariate analysis technique is to be used. Another approach is suggested by
MacCallum et al. (1996) of calculating the sample size required by determining the power
level required from the analysis, given a population size estimate. Another strategy is
to compare the sample characteristics to the known population equivalents, to ensure
sample representation of the population which in turn may allow for a smaller sample.
In the sample of JTR papers most of the factor analysis types are PCA and there is a
clear understanding that the analysis is exploratory or to reduce the number of variables.
The PCA is sometimes followed by a CFA analysis on a second sample (if the sample size
is very large the sample might be split in half to conduct the exploratory and then con-
firmatory analyses), and these papers are more common in Table 10.7. There appears to
be no confusion as to the type of factor analysis to use. However, in presenting research
this point still needs to be explained. There is confusion in the general literature arguably
resulting from a conflict in opinion about which is better PCA or CFA, when really there
is no better or worse option. PCA is superior for data reduction and exploratory work
and CFA to test hypotheses. Both are capable of identifying latent constructs (despite
arguments to the contrary in the general literature) within these contexts, and so both are
a form of factor analysis.

DWYER 9781781001288 Part I.indd 204 25/07/2012 11:07

 Factor analysis 205

The type of rotation used in the tourism sample with PCA is mostly Varimax and the
intention is to describe orthogonal components. Varimax is recognized as the best form
of orthogonal rotation. Where an oblique rotation is used there should be more explana-
tion as to why it is used. Also there is nothing to stop an oblique rotation from deriving
an orthogonal solution, and the ML method described above provides the advantage of
measuring the correlation between the factors in a CFA. Where the ML method shows
there is correlation between factors, this might suggest a higher order of constructs
exists, although this point is often missed by researchers. Promax oblique is actually a
little more common than Oblimin in the JTR sample, but this may not be representative
of all tourism factor analysis publications. Promax derives its name from Procrustean
rotation. It uses Varimax as the target rotation and uses a least square fit to obtain the
oblique rotation, and as such is computationally more direct.
The question of what statistics should be reported is reasonably straightforward for
most types of analysis and includes eigenvalues and explained variance along with factor
loadings.
The question of reliability is less certain. Cronbachs alpha is the most common reli-
ability measure followed by KMO and Bartlett. Alpha measures the degree to which a
set of variables explain a latent construct, and a measure above 0.7 in the range 0 to 1 is
considered sufficient while a measure of 0.8 is considered good, and as such it is akin to
a measure of correlation. The main problem with the measure is that it is affected by the
number of items in the scale, and so it does vary in accuracy becoming less accurate as
the number of variables increases. The KaiserMeyerOklin (KMO) statistic measures
the proportion of variance among the variables that might be common variance. The
measure ranges between 0 and 1 and 0.5 and above is considered adequate to proceed
with a factor analysis, although Kaiser has recommended a much higher value of 0.7 or
above. KMO compares the magnitude of the correlation coefficients against the magni-
tude of the partial correlation coefficients.
None of the papers use the MSA which measures the common variance for each vari-
able, and this is quite reasonable as the MSA is not easily interpreted as a measure of
reliability.
Bartletts test of sphericity tests whether the correlation matrix is an identity matrix
(made up of ones in the diagonal and zeros off the diagonal). If all the correlations are
one on the diagonal the variables are perfectly correlated with themselves and if zero
off diagonal, they are totally uncorrelated with any other variable and there can be no
common variance. It is a chi-square approximate test and the objective is to reject the
null hypothesis of no correlation. This test suffers from the problem of sample size, and
becomes less valid with smaller sample sizes. The Bartlett test is not highly discriminating
and usually always rejects.
On balance the Cronbachs alpha value and KMO are the better reliability measures.
When an ML analysis is used with CFA, the test statistics are larger in number and the
suggested minimum number of statistics to report is given in the section discussing ML
analysis.
The question of what cut-off should be used to determine the number of components
to be rotated is reasonably straightforward. Components with eigenvalues greater than
one should be selected. In the case of CFA the number of factors to be rotated should be
selected from the correlation matrix with one on the diagonals not the reduced matrix

DWYER 9781781001288 Part I.indd 205 25/07/2012 11:07

 206 Handbook of research methods in tourism

with communalities. Obviously if an eigenvalue is very close to one, it is valid to round

the eigenvalue up from 0.5 (0.95 5 1) but doing this needs to be noted and is not ideal.
The logic for this stated criterion is sound because the sum of all eigenvalues is equal to
the number of variables, so a value of one is average, and for another component to be
selected it should exceed the average of one.
The alternative of using the scree plot by Cattell (1966) should be avoided, as it is
subjective and often not conclusive. There is a suspicion that the scree plot is refer-
enced in research to justify the use of components below an eigenvalue of one, in order
to strengthen research discussion and in this context the use would be invalid. Ideally
the scree plot should be presented if used, regardless of the journal editors desire to
keep an article short, so that the cutoff point is clearly obvious and justified. The main
advantage of the scree method is with CFA analysis, because it applies to the reduced
matrix with communalities and not the original matrix and so reflects directly on the
CFA factors.
There is a third method called parallel analysis (Horn, 1965; Humphreys and Ilgen,
1969; Montanelli and Humphreys 1976) but it is not available in current programs, and
is difficult to apply. The method has not been used in published tourism research, and it
is recommended that further reference is made to Lidesma and Valero-Mara (2007) for a
detailed discussion on how to apply the methodology.
The process of selecting the number of relevant factors in ML analysis using SEM is
more subject to issues of data mining and must be done more carefully. The process is
one of trial and error running the analysis several times with different numbers of factors,
and attempting to maximize the fit statistics. The process is accepted generally, but raises
the question of tailoring the results to the hypotheses being tested. So justification of the
process needs careful explanation in theory. Quite often the adjustments yield only a
small improvement and raise the question of whether they are necessary at all.
The question of the required loading of each variable for it to be allocated as signifi-
cantly loaded on a component is the most vexing question. The original answer that is
accepted in an unquestioning manner by tourism researchers was given by Stevens (1986)
and is usually referenced in Hair (various editions with 1992 most commonly cited in
the tourism literature, although in later editions this rises to 0.5) that a loading must at
least be 0.4, because it is suggested that would indicate at least 30 percent of the varia-
tion in a variable loaded on a single factor. This was then followed by Comrey (1988)
who stated that this might only apply if the loading of 0.4 was unique to only one factor.
Consequently, the rule is often applied of accepting a value as low as 0.4 provided there
is no other factor loading close to 0.4 for the same variable.
Given the loading in a PCA is a correlation statistic, it is difficult to see that 0.4 is an
acceptable level of loading and in the literature on the evolution of factor analysis this
low value was not accepted. In a regression analysis or correlation analysis a correlation
of 0.4 would never be accepted, yet it is universally accepted in factor analysis. Although
many tourism articles use the 0.4 criterion they usually state at least 0.4 and often have
a lowest loading above 0.4 up to 0.55. There is an inconsistency in quoting one rule and
then using another, simply because that was the computational result. It is also clear that
a loading of 0.4 is far lower than a loading of 0.8, yet in the interpretation of the meaning
of a component the variables loaded on the component receive an equal weighting for
importance, carrying the same weight in the naming of the component. An equal weight-

DWYER 9781781001288 Part I.indd 206 25/07/2012 11:07

 Factor analysis 207

ing of importance might be acceptable if the cutoff value is 0.6 or higher, but not 0.4 or
0.5. A cut off of 0.4 or lower may also be acceptable if the PCA is a preliminary analysis
for a CFA because the role of the PCA is simply to reduce the number of variables and
give direction to the research. Results from the analysis need to be careful in defining the
meaning of a component with careful interpretation of the meaning of the variables that
load onto it, taking into account the relative weight of each loading variable from the
strength of the loading. There is not much evidence of this happening in tourism research
from the sample of articles examined.

CONCLUSION

Factor analysis is a complex methodology that provides an objective way of determining

latent (hidden) concepts within social data and as such it is an important tool for analys-
ing tourism data, and potentially solving tourism-related problems. There is a need for
researchers to understand the technique, how it is calculated at least to some degree, and
what the main features are of the procedure. A researcher needs to be aware of the types
of applications suited to factor analytic analysis and the limitations to the methodology.
In the current published literature there are now so many factor analysed data sets
that it is becoming possible to compare results on similar problems across different data
surveys. More needs to be done developing hypothetical structures from previous studies
and confirming constant relationships in future research. In order to aid this process
there needs to be agreement and understanding of the basic elements from analysis that
are to be reported, what statistics and survey elements should be clearly evident, and
what standards of interpretation are required.
The major limitation in the method is interpretation of the meaning of the latent
structures and more effort is needed in explaining the meaning of the survey items and
the meaning and relevance of the latent constructs that are derived. Too often a meaning
to a latent construct is stated with little or no justification of why the variables loading
on the construct can be interpreted to have the given meaning. Alternatively, a discus-
sion approach could be taken that gives more than one potential meaning to a latent
construct, and debate follow as to the likely better interpretation.
Factor analysis will continue as a major tourism research tool but there is room for
improvement in the use of the technique to raise the standard of research analysis and
consequent findings.

REFERENCES
Al-Sabbahy, H.Z., Y. Ekinci and M. Riley (2004), An investigation of perceived value dimensions: implica-
tions for hospitality research, Journal of Travel Research, 42 (3), 226234.
Baloglu, S. and S. Shoemaker (2001), Prediction of senior travelers motorcoach use from demographic, psy-
chological, and psychographic characteristics, Journal of Travel Research, 40 (1), 1218.
Benckendorff, P.J. and P.L. Pearce (2003), Australian tourist attractions: the links between organizational
characteristics and planning, Journal of Travel Research, 42 (1), 2435.
Bock, R.D. and R.E. Bargmann (1966), Analysis of covariance structures, Psychometrika, 31, 507534.
Brey, E.T., D.B. Klenosky and A.M. Morrison (2008), Standard hospitality elements at resorts, Journal of
Travel Research, 47 (2), 247258.

DWYER 9781781001288 Part I.indd 207 25/07/2012 11:07

 208 Handbook of research methods in tourism

Bronner, F. and R. de Hoog (2011), Vacationers and eWOM: who posts, and why, where, and what?, Journal
of Travel Research, 50 (1),1526.
Burnett, J.J. and H.B. Baker (2001), Assessing the travel-related behaviors of the mobility-disabled consumer,
Journal of Travel Research, 40 (1), 411.
Carroll, J.B. (1953), An analytical solution for approximating simple structure in factor analysis,
Psychometrika, 18, 2328.
Carroll, J.B. (1960), IBM 704 program for generalized analytic rotation solution in factor analysis, unpub-
lished manuscript, Harvard University.
Cattell, R.B. (1966), The scree test for the number of factors, Multivariate Behavioral Research, 1, 245276.
Cattell, R.B. and J.L. Muerle (1960), The maxplane program for factor rotation to oblique simple struc-
ture, Educational and Psychological Measurement, 20, 569590.
Choi, H.C. and E. Sirakaya (2005), Measuring residents attitude toward sustainable tourism: development of
sustainable tourism attitude scale, Journal of Travel Research, 43 (4), 380394.
Chu, K.H. (2008), A factorial evaluation of work value structure: second-order confirmatory factor analysis
and its implications, Tourism Management, 29, 320330.
Chu, R.K.S. and T. Choi (2000), An importance-performance analysis of hotel selection factors in the
Hong Kong hotel industry: a comparison of business and leisure travelers, Tourism Management, 21,
363377.
Comrey, A.L. (1988), Factor-analytic methods of scale development in personality and clinical psychology,
Journal of Consulting and Clinical Psychology, 56, 754761.
Correia, A., M. Moital, C. Ferreira da Costa and R. Peres (2008), The determinants of gastronomic tourists
satisfaction: a second-order factor analysis, Journal of Foodservice, 19, 164176.
Crawford, C. (1967), A general method of rotation for factor analysis, paper presented at the meeting of the
Psychometric Society, April.
Curran, P.J., S.G. West and J.F. Finch (1966), The robustness of test statistics to nonnormality and specifica-
tion error in confirmatory factor analysis, Psychological Methods, 1, 1629.
Drey, B. and M.K. Sarma (2010), Information source usage among motive-based segments of travelers to
newly emerging tourist destinations, Tourism Management, 31, 341344.
Ekinci, Y. and S. Hosany (2006), Destination personality: an application of brand personality to tourism
destinations, Journal of Travel Research, 45 (2), 127139.
Everitt, B.S. (1975), Multivariate analysis: the need for data, and other problems, British Journal of
Psychiatry, 126, 237240.
Ferguson, G. (1954), The concept of parsimony in factor analysis, Psychometrika, 19, 281290.
Formica, S. and M. Uysal (2006), Destination attractiveness based on supply and demand evaluations: an
analytical framework, Journal of Travel Research, 44 (4), 418430.
Galloway, G. (2002), Psychographic segmentation of park visitor markets: evidence for the utility of sensation
seeking, Tourism Management, 23 (6), 581596.
Gil, S.M. and J.R. Brent Ritchie (2009), Understanding the museum image formation process: a comparison
of residents and tourists, Journal of Travel Research, 47 (4), 480493.
Gorsuch, R.L. (1983), Factor Analysis, 2nd edition, Hillside, NJ: Erlbaum.
Guttman, L. (1953), Image theory for the structure of quantitative variates, Psychometrika, 18, 277296.
Hair, J.F. (1992), Multivariate data analysis, 3rd edition, New York: Macmillan.
Hakstian, A.R., W.T. Rogers and R.B. Cattell (1982), The behavior of number-of-factors rules with simulated
data, Multivariate Behavioral Research, 17, 193219.
Harman, H.H. (1976), Modern Factor Analysis, 3rd edition, Chicago, IL: University of Chicago Press.
Harman, H.H. and W.H. Jones (1966), Factor analysis by minimizing residuals (minres), Psychometrika, 31,
351368.
Harris, C.W. (1962), Some Rao-Guttman relationships, Psychometrika, 27, 247263.
Harris, M.L. and C.W. Harris (1971), A factor analytic interpretation strategy, Educational and Psychological
Measurement, 31, 589606.
Hendrikson, A.E. and P.D. White (1964), PROMAX: a quick method of rotation to oblique simple structure,
British Journal of Mathematical and Statistical Psychology, 17, 6570.
Henthorne, T.L. (2000), An analysis of expenditures by cruise ship passengers in Jamaica, Journal of Travel
Research, 38 (3), 246250.
Heung, V.C.S. and E. Cheng (2000), Assessing tourists satisfaction with shopping in the Hong Kong Special
Administrative Region of China, Journal of Travel Research, 38 (4), 396404.
Holzinger, K. and F. Swineford (1937), The bi-factor method, Psychometrika, 2, 4154.
Horn, J.L. (1965), A rationale and technique for estimating the number of factors in factor analysis,
Psychometrika, 30, 179185.
Hosany, S. and D. Gilbert (2010), Measuring tourists emotional experiences toward hedonic holiday destina-
tions, Journal of Travel Research, 49 (4), 513526.

DWYER 9781781001288 Part I.indd 208 25/07/2012 11:07

 Factor analysis 209

Hotelling, H. (1933), Analysis of complex statistical variables into principal components, Journal of
Educational Psychology, 24, 417441; 498520.
Hsu, C.H.C. and E. Lee (2002), Segmentation of senior motorcoach travelers, Journal of Travel Research, 40
(4), 364373.
Hsu, C.H.C., S.K. Kang and T. Lam (2006), Reference group influences among Chinese travelers, Journal of
Travel Research, 44 (4), 474484.
Hu, L., P.M. Bentler and Y. Kano (1992), Can test statistics in covariance structure analysis be trusted?,
Psychological Bulletin, 112, 351362.
Hudson, S. and J.R.B. Ritchie (2006), Promoting destinations via film tourism: an empirical identification of
supporting marketing initiatives, Journal of Travel Research, 44 (4), 387396.
Humphreys, L.G. and D.R. Ilgen (1969), Note on a criterion for the number of common factors, Educational
and Psychological Measurements, 29, 571578.
Jang, H., B. Lee, M. Park and P.A. Stokowski (2000), Measuring underlying meanings of gambling from the
perspective of enduring involvement, Journal of Travel Research, 38 (3), 230238.
Jennrich, R.I. and P.F. Sampson (1966), Rotation for simple loadings, Psychometrika, 31, 313323.
Johns, N. and S. Gyimthy (2002), Market segmentation and the prediction of tourist behavior: the case of
Bornholm, Denmark, Journal of Travel Research, 40 (3), 316327.
Jreskog, K.G. (1966), Testing a simple structure hypothesis in factor analysis, Psychometrika, 31, 165178.
Jreskog, K.G. (1967), Some contributions to maximum likelihood factor analysis, Psychometrika, 32,
443482.
Jreskog, K.G. (1969), A general approach to confirmatory maximum likelihood factor analysis,
Psychometrika, 34, 183202.
Jreskog, K.G. (1970), A general method for analysis of covariance structures, Biometrika, 57, 239251.
Jreskog, K.G. and D. Srbom (1989), LISREL-7 Users Guide, Mooresville, IN: Scientific Software.
Juric, B., T.B. Cornwell and D. Mather (2002), Exploring the usefulness of an ecotourism interest scale,
Journal of Travel Research, 40 (3), 259269.
Kaiser, H. (1956), The Varimax method of factor analysis, unpublished PhD dissertation, University of
California, Berkley.
Kaiser, H. (1958), The Varimax criterion for analytic rotation in factor analysis, Psychometrika, 23, 187200.
Kaiser, H. (1970), A second generation little-jiffy, Psychometrika, 35, 401415.
Kaiser, H. and J. Caffrey (1965), Alpha factor analysis, Psychometrika, 30, 114.
Kim, S. and J.L. Crompton (2002), The influence of selected behavioral and economic variables on percep-
tions of admission price levels, Journal of Travel Research, 41 (2), 144152.
Kozak, M. and M. Rimmington (2000), Tourist satisfaction with Mallorca, Spain, as an off-season holiday
destination, Journal of Travel Research, 38 (3), 260269.
Kwan, F.V.C. and G. McCartney (2005), Mapping resident perceptions of gaming impact, Journal of Travel
Research, 44 (2), 177187.
Lawley, D.N. (1940), The estimation of factor loadings by the method of maximum likelihood, Proceedings
of the Royal Statistical Society of Edinburgh, 60, 6482.
Lee, C., Y. Lee, B.J. Bernhard and Y. Yoon (2006), Segmenting casino gamblers by motivation: a cluster
analysis of Korean gamblers, Tourism Management, 27, 856866.
Lidesma, R.D. and P. Valero-Mara (2007), Determining the number of factors to retain in an EFA: an easy
to use computer program for carrying out parallel analysis, Practical Assessment, Research and Evaluation,
12 (2), 111.
Lim, W.M. (2010), Factor analysis of variables affecting e-marketing adoption by UK independent hotels, in
U. Gretzel, R. Law and M. Fuchs (eds), Information and Communication Technologies in Tourism 2010, New
York, Springer, pp. 3950.
Lindeman, R.H. (1980), Introduction to Bivariate and Multivariate Analysis, London: Scott, Foresman and
Company.
Luo, Y. and J. Deng (2008), The new environmental paradigm and nature-based tourism motivation, Journal
of Travel Research, 46 (4), 392402.
MacCallum, R.C. (1990), The need for alternative measures of fit in covariance structure modeling,
Multivariate Behavioral Research, 24, 5969.
MacCullum, R.C., M.W. Browne and H.M. Sugawara (1996), Power analysis and determination of sample
size for covariance structure modelling, Psychological Methods, 1, 130149.
Massy, W.F. (1965), Principal components regression in exploratory statistical research, Journal of the
American Statistical Society, 60, 234256.
Montanelli, R.G. Jnr. and L.G. Humphreys (1976), Latent roots of random data correlation matrices with
squared multiple correlations on the diagonal: a monte carlo study, Psychometrika, 41, 341348.
Morais, D.B., S.J. Backman and M.J. Dorch (2003), Toward the operationalization of resource investments
made between customers and providers of a tourism service, Journal of Travel Research, 41 (4), 362374.

DWYER 9781781001288 Part I.indd 209 25/07/2012 11:07

 210 Handbook of research methods in tourism

Murphy, L., G. Moscardo and P. Bekerndorff (2007), Using brand personality to differentiate regional
tourism destinations, Journal of Travel Research, 46 (1), 514.
Neuhaus, J.O. and C. Wrigley (1954), The Quartimax method: an analytical approach to orthogonal simple
structure, British Journal of Mathematical and Statistical Psychology, 7, 8191.
Nicholson, R.E. and D.G. Pearce (2001), Why do people attend events: a comparative analysis of visitor moti-
vations at four South Island events, Journal of Travel Research, 39 (4), 449460.
Nunnally, J.C. (1978), Psychometric Theory, 2nd edition, New York: Mc Graw-Hill.
Nyaupane, G.P. and K.L. Andereck (2008), Understanding travel constraints: application and extension of a
leisure constraints model, Journal of Travel Research, 46 (4), 433439.
Ollenburg, C. and R. Buckley (2007), Stated economic and social motivations of farm tourism operators,
Journal of Travel Research, 45 (4), 444452.
zgener, . and R. raz (2006), Customer relationship management in small-medium enterprises: the case of
Turkish tourism industry, Tourism Management, 27, 13561363.
Pearce, P.L. and U. Lee (2005), Developing the travel carer approach to tourist motivation, Journal of Travel
Research, 43 (3), 226237.
Pearson, K. (1901), On the lines and planes of closest fit to systems of points in space, Philosophical Magazine,
6, 559572.
Pennington-Gray, L.A. and D.L. Kerstetter (2001), What do university-educated women want from their
pleasure travel experiences?, Journal of Travel Research, 40 (1), 4956.
Perdue, R.R. (2002), Perishability, yield management, and cross-product elasticity: a case study of deep dis-
count season passes in the Colorado ski industry, Journal of Travel Research, 41 (1), 1522.
Petrick, J.F. (2004), First timers and repeaters perceived value, Journal of Travel Research, 43 (1),
2938.
Petrick J.F. and S.J. Backman (2002), An examination of the construct of perceived value for the prediction of
golf travelers intentions to revisit, Journal of Travel Research, 41 (1), 3845.
Phillips, P.A. and L. Moutinho (2000), The strategic planning index: a tool for measuring strategic planning
effectiveness, Journal of Travel Research, 38 (4), 369379.
Poria, Y., A. Biran and A. Reichel (2009), Visitors preferences for interpretation at heritage sites, Journal of
Travel Research, 48 (1), 92105.
Poria, Y., R. Butler and D. Airey (2004), Links between tourists, heritage, and reasons for visiting heritage
sites, Journal of Travel Research, 43 (1), 1928.
Poria, Y., A. Reichel and A. Biran (2006), Heritage site perceptions and motivations to visit, Journal of Travel
Research, 44 (3), 318326.
Rao, C.R. (1955), Estimation and tests of significance in factor analysis, Psychometrika, 20, 93111.
Reichel, A., G. Fuchs and N. Uriely (2007), Perceived risk and the non-institutionalized tourist role: the case
of Israel student backpackers, Journal of Travel Research, 46 (2), 217226.
Reisinger, Y. and F. Movondo (2005), Travel anxiety and intentions to travel internationally: implications of
travel risk perception, Journal of Travel Research, 43 (3), 212225.
Reisinger, Y. and L.W. Turner (1999a), A cultural analysis of Japanese tourists: challenges for tourism mar-
keters, European Journal of Marketing, 33 (12), 12031227.
Reisinger, Y. and L.W. Turner (1999b), Structural equation modeling with LISREL: application in tourism,
Tourism Management, 20 (1), 7188.
Reisinger, Y. and L.W. Turner (2002), Cultural differences between Asian tourist markets and Australian
hosts, Part 1, Journal of Travel Research, 40 (3), 295315.
Reisinger, Y. and L.W. Turner (2003), Cross-Cultural Behaviour in Tourism, Concepts and Analysis, Oxford:
Butterworth Heineman.
Revilla, G. and T.H. Dodd (2003), Authenticity perceptions of Talavera pottery, Journal of Travel Research,
42 (1), 9499.
Rittichainuwat, B.N. (2008), Responding to disaster: Thai and Scandinavian tourists motivation to visit
Phuket, Thailand, Journal of Travel Research, 46 (4), 422432.
Saunders, D.R. (1953), An analytical method for rotation to orthogonal simple structure, Educational Testing
Service Research Bulletin RB-53-10.
Saunders, D.R. (1961), The rationale for an Oblimax method of transformation in factor analysis,
Psychometrika, 26, 317324.
Saunders, D.R. (1962), Trans-varimax, American Psychologist, 17, 395.
Sirakaya, E., V. Teye and S. Sonmez (2002), Understanding residents support for tourism development in the
Central Region of Ghana, Journal of Travel Research, 41 (1), 5767.
Sirakaya, E., M. Uysal and C.F. Yoshioka (2003), Segmenting the Japanese tour market to Turkey, Journal
of Travel Research, 41 (3), 292304.
Sirakaya-Turk, E., Y. Ekinci and A.G. Kaya (2008), An examination of the validity of SUS-TAS in cross
cultures, Journal of Travel Research, 46 (4), 414-421.

DWYER 9781781001288 Part I.indd 210 25/07/2012 11:07

 Factor analysis 211

Snepenger, D., J. King, E. Marshall and M. Uysal (2006), Modeling Iso-Aholas motivation theory in the
tourism context, Journal of Travel Research, 45 (2), 140149.
Sonmez, S. and E. Sirakaya (2002), A distorted destination image? The case of Turkey, Journal of Travel
Research, 41 (2), 185196.
Spearman, C. (1904), General intelligence, objectively determined and measured, American Journal of
Psychology, 15, 201293.
Stepchenkova, S., A.P. Kirilenko and A. Morrison (2009), Facilitating content analysis in tourism research,
Journal of Travel Research, 47 (4), 454469.
Stevens, J. (1986), Applied Multivariate Statistics for the Social Sciences, Hillsdale, NJ: Erlbaum.
Snyman, J.A. and M. Saayman (2009), Key factors influencing foreign direct investment in the tourism indus-
try in South Africa, Tourism Review, 64 (3), 4958.
Susskind, A.M., M.A. Bonn and C.S. Dev (2003), To look or book: an examination of consumers apprehen-
siveness toward internet use, Journal of Travel Research, 41 (3), 256264.
Thurstone, L.L. (1931), Multiple factor analysis, Psychological Review, 38, 406427.
Thurstone, L.L. (1938), Primary mental abilities, Psychometric Monographs, 1, Chicago, IL: University of
Chicago Press.
Thurstone, L.L. (1947), Multiple Factor Analysis, Chicago, IL: University of Chicago Press.
Vincent, V.C. and W. Thompson (2002), Assessing community support and sustainability for ecotourism
development, Journal of Travel Research, 153160.
Vong, F. (2009), Changes in residents gambling attitudes and perceived impacts at the fifth anniversary of
Macaos gaming deregulation, Journal of Travel Research, 47 (3), 388397.
Vu, C.J. and L.W. Turner (2009), The economic structure of world tourism, Tourism Economics, 15 (1), 521.
Walters, G., B. Sparks and C. Herington (2007), The effectiveness of print advertising stimuli in evoking elabo-
rate consumption visions for potential travelers, Journal of Travel Research, 46 (1), 2434.
Wang, Y. and R.E. Pfister (2008), Residents attitudes toward tourism and perceived personal benefits in a
rural community, Journal of Travel Research, 47 (1), 8493.
Wilkins, H. (2011), Souvenirs: what and why we buy, Journal of Travel Research, 50 (3),239247.
Woosnam, K.M. and W.C. Norman (2010), Measuring residents emotional solidarity with tourists: scale
development of Durkheims theoretical constructs, Journal of Travel Research, 49 (3), 365380.
Wong, S. and E. Lau (2001), Understanding the behavior of Hong Kong Chinese tourists on group tour pack-
ages, Journal of Travel Research, 40 (1), 5767.
Yoo, J.J. and K. Chon (2008), Factors affecting convention participation decision-making: developing a meas-
urement scale, Journal of Travel Research, 47 (1), 113122.
Yuan, J., L.A. Cai, A.M. Morrison and S. Linton (2004), An analysis of wine festival attendees motivations:
a synergy of wine, travel and special events?, Journal of Vacation Marketing, 11 (1), 4158.
Zhou, Y. and J. Ap (2009), Residents perceptions towards the impacts of the Beijing 2008 Olympic Games,
Journal of Travel Research, 48 (1), 7891.

DWYER 9781781001288 Part I.indd 211 25/07/2012 11:07

 11 Cluster analysis
Liz Fredline

INTRODUCTION

The purpose of this chapter is to describe and illustrate the analytical technique known
as cluster analysis and to outline its application in tourism research. This technique can
be extremely useful for certain research questions, but a review of literature shows that
its application has been quite infrequent. The chapter begins with a brief summary of the
nature of the technique followed by a summary of some of the tourism literature which
has employed cluster analysis. Following on from this is a summary of a worked example
(using IBM SPSS 19.0) of a two-stage cluster analysis procedure. Finally, the chapter
concludes with a discussion of the advantages and limitations of the technique as well as
the possible future advances in tourism research which may be possible with the applica-
tion of cluster analysis.
Cluster analysis is a family of multivariate techniques useful for analysing cases based
on their scores on a range of measured variables. Essentially the technique identifies
cases with a comparable pattern of responses that can be regarded, for the purposes of
the analysis, as similar. In so doing a large number of individual cases are summarized
into a smaller, more manageable number of clusters. The outcome of a successful cluster
analysis would be a small number of highly homogeneous clusters that are substantially
different to each other (Hair et al., 1998). Cluster membership can then be used as a vari-
able for further analysis aimed at understanding the clusters and the bases on which they
were formed.
In the tourism context there are many possible applications for cluster analysis. It is
useful in any situation where the researcher wants to identify groups of cases that are
similar to each other but different from other groups, that is, segmentation of the popula-
tion. This segmentation can be based on any variables that are regarded as important for
discriminating between groups and could include simple demographic characteristics or
psychographic characteristics such as attitudes, perceptions, and motivations. The popu-
lations being studied may be tourism or hospitality customers (see for example, Arimond
and Elfessi, 2001; Jurowski and Reich, 2000; Ryan and Huyton, 2000) or residents of a
region that is visited by tourists (Davis et al., 1988; Fredline and Faulkner, 2000, 2002;
Madrigal, 1995; Perez and Nadal, 2005). This latter application is quite common in the
tourism literature aimed at assessing residents perceptions of the impact of tourism.

NATURE OF CLUSTER ANALYSIS AND ITS EVOLUTION

Cluster analysis has its origins in the biological sciences where it was first used in taxo-
nomical studies to classify flora and fauna into phyla, classes and orders. Gilmour (1951)
describes its development from the mid-1800s. However, it was not until the second half

212

DWYER 9781781001288 Part I.indd 212 25/07/2012 11:07

 Cluster analysis 213

of the twentieth century that advances in the use of computers for statistical analysis
allowed the technique to be updated for modern application. Authors such as Sneath
and Sokal (1973) and Anderberg (1973) were instrumental in this process (cited in
Jurowski and Reich, 2000).
Cluster analysis is an interdependence rather than a dependence technique. This
means that it does not aim to assess the relationship between independent and depend-
ant variables; rather it is focused on the underlying structure of variables and the useful-
ness of this structure in categorizing cases (Hair et al., 1998). Thus, it does not have a
specific test of significance indicating the outcome of the analysis. For this reason many
researchers inexperienced in its use are sometimes uncomfortable with it as they are used
to having a definitive indication of the solution. Instead interpretation of the results of
cluster analysis must be based on researcher knowledge and experience, as well as the
logic of the solution and the extent to which clusters demonstrate high internal homoge-
neity and high external heterogeneity. As Jurowski and Reich (2000, p. 69) suggest the
technique can be considered an art as well as a science.
Like any analysis there are a number of options and therefore several decisions the
researcher must make. First, the data must be assessed for their suitability for analysis.
Cluster analysis does not hinge on assumptions about the distribution of the data so
normal considerations about the normality, linearity and homoscedasticity of the data
are not relevant (Hair et al., 1998). However, there are three issues which do require
some consideration; outliers, multi-collinearity and the extent to which the data are rep-
resentative of the population being studied.
Outliers can substantially affect the results so it is important to try to identify multi-
variate outliers (those which represent an unusual combination of scores on the linear
combination of variables) and remove these. Mahalanobis distance can be used for this
purpose and can be accessed through the regression procedure in SPSS. Multicollinearity
is a problem to the extent that if two or more variables included in the analysis are highly
correlated, while other variables are not, then those variables have more weight in deter-
mining the outcome. Any high correlations need to be examined and if two variables are
determined to be effectively the same, then one should be removed or the two variables
could be collapsed together into a composite. Like any other analytical technique, the
results will only be generalizable beyond the specific cases included in the analysis if the
sample was representative of the population of interest (Aldenderfer and Blashfield,
1984). Consideration of representativeness should reflect on the sampling procedures
used as well as the response rate. If response rates were low, then some check should be
made of non-response bias.
If the variables to be analysed are not measured on the same scale then consideration
should be given to standardizing them prior to analysis. This is because variables with a
much larger range will have a greater influence on the solution than those with a smaller
range. For example, if one variable is measured on a scale ranging from 1100 while
another ranges from 110, then the former could potentially carry far more weight in
determining the similarity or differences between cases. Standardization will ensure that
all variables are having a similar impact on the outcome (Hair et al., 1998).
The next decision that researchers need to consider is how the variables which are to
form the basis of segmentation should be combined into a composite score known as
a cluster variate. There are three main choices here: correlational measures, distance

DWYER 9781781001288 Part I.indd 213 25/07/2012 11:07

 214 Handbook of research methods in tourism

measures, and association measures. The choice depends on the form of the variables
and the purpose of the research. If the data are non-metric, then association measures are
most appropriate. Metric data can be combined using either correlation measures, if the
aim is to identify patterns in the data set, or distance measures, if the aim is to identify
cases which are closest together (Hair et al., 1998).
The next decision, and possibly the most fundamental one, is between hierarchical
and non-hierarchical clustering, as these represent quite different approaches to segmen-
tation. The former begins with each case in its own cluster and in the first step it joins
together the two closest cases based on their scores on the cluster variate. In each step
there is one less cluster and the process continues until all cases are joined together in
a single cluster. It is quite a computationally intensive technique, as the centroids (the
multivariate means of the scores on the variate) must be recalculated as each new cluster
is formed. Thus a large number of cases and/or variables will require substantial compu-
tational power.
Once the analysis is complete it is then necessary to select which solution, that is, which
number of clusters, is the best. SPSS produces two outputs that can help in this regard:
the agglomeration schedule and the dendrogram. The agglomeration schedule shows the
clustering coefficient at each step. A small increase in the coefficient indicates that the
clusters being combined are relatively homogeneous, while a large increase indicates that
they were fairly heterogeneous, so the basic principle involves stopping the analysis just
before a large increase. However, Hair et al. (1998) suggest that this may tend to indicate
too few clusters. The dendrogram shows the distances between clusters being combined
at each stage. Similarly, the idea would be to select a solution just before a large distance
is observed.
The researcher can then examine the solution at each point to determine where the
explanation is most appropriate. It is typical to start by examining the two-cluster solu-
tion and work backwards. The researcher should look at the means of the two clusters
on the original variables to see how these differ. Then the three cluster solution should
be examined. This solution will contain one cluster that was present in the two cluster
solution, while to other two will have resulted from the splitting of another so it is useful
to try and understand the basis for this split. The researcher must then ask themselves
whether this split results in a more interpretable and useful segmentation than the two
cluster solution. If they feel this to be the case then they should proceed to the four
cluster solution and so on until it is felt that the outcome provides good discrimination
into manageable and meaningful groups. This process, in combination with the inter-
pretation of the agglomeration schedule and/or dendrogram, should help researchers to
decide on the most appropriate solution.
One weakness associated with the method is that cases can become trapped in a
cluster. Once a case is allocated to a cluster it must stay there even if, by the process of
recalculating centroids at each step, a new cluster later emerges that it would be closer
to. As Arimond and Elfessi (2001, p. 391) suggest, hierarchical methods only provide
a starting point and do not consistently produce clusters that are clearly homogeneous
and well balanced.
The alternative method, non-hierarchical clustering (also known as quick cluster or
K-means clustering), takes a quite different approach. In this case, the researcher must
pre-specify the number of clusters to be derived and a starting variate score (or seed) for

DWYER 9781781001288 Part I.indd 214 25/07/2012 11:07

 Cluster analysis 215

Table 11.1 Advantages and disadvantages of hierarchical and non-hierarchical techniques

Hierarchical Non-hierarchical
Advantages Can be used to determine the best Allows cases to be reassigned
number of clusters Provides additional diagnostic tables
Disadvantages Computationally intensive Most suitable when the researcher has
Cases can become trapped some knowledge about the number of
clusters and their starting points

Table 11.2 Clustering algorithms

Hierarchical Non-hierarchical
Single Linkage joins clusters based on Sequential Threshold Method selects seeds one
shortest distance to any case within the cluster. at a time and includes all cases within a specified
Also known as nearest neighbour. distance. Then moves onto the next seed.
Complete Linkage joins clusters based on Parallel Threshold Method processes all cluster
longest distance between any two cases. seeds simultaneously and assigns cases to the
Average Linkage joins clusters based on the closest seed.
average distance between all cases. Less affected Optimizing Procedure similar to the parallel
by outliers. method but allows reassignment of cases.
Centroid Method joins clusters based on the
distance between centroids.
Wards Method joins clusters based on the
sums of squares within the clusters. Tends to
produce clusters of a similar size.

each cluster. For this reason non-hierarchical clustering is far more successful if there is
a substantial body of previous literature to inform these decisions. It is possible to ran-
domly select seeds but these will substantially affect the outcome. It is likely that different
randomly selected seeds will produce somewhat different results.
A key benefit of non-hierarchical clustering is that it allows cases to be reassigned if
they become closer to a different cluster centroid. Also the non-hierarchical clustering
method in IBM SPSS 19.0 provides some diagnostic tables which facilitate better under-
standing of the adequacy of the solution.
Thus, each method has its advantages and disadvantages as summarized in Table 11.1.
For this reason, a common approach is to use a combination of the two; hierarchical
clustering followed by non-hierarchical. This allows the researcher to select the most
appropriate solution in terms of number of clusters, and then ensure the best possible
allocation of cases to clusters.
The next decision for the researcher is which clustering algorithm to use. There are
many choices here. In hierarchical clustering the options include single linkage, com-
plete linkage, average linkage, Wards method and the centroid method. The methods
differ in how they decide which clusters should be joined together at each iteration of the
analysis. In non-heirarchical clustering the options include sequential threshold, parallel
threshold and optimization. Table 11.2 provides a brief explanation of these but a fuller

DWYER 9781781001288 Part I.indd 215 25/07/2012 11:07

 216 Handbook of research methods in tourism

description can be found in many multivariate statistics textbooks such as Hair et al.
(1998) and Jobson (1991).

APPLICATIONS OF CLUSTER ANALYSIS IN TOURISM

As mentioned previously, the two most common applications of cluster analysis in

tourism research are market segmentation and segmentation of host community atti-
tudes or perceptions toward tourism. Below is a brief summary of some of the work in
these two fields.

Market Segmentation

A common application of cluster analysis in tourism is market segmentation as the tech-

nique has been adopted from more mainstream marketing research where it has been
widely used. In this form, the analysis allows tourism researchers and operators to better
understand the various groups that exist within their market and therefore to modify
their product to specifically target selected markets or alternatively to attempt to meet
the differing needs of all segments. It allows more efficient allocation of resources which
should then lead to the generation of additional income.
An excellent example of this type of study is Jurowski and Reich (2000) who classified
potential hotel customers based on 24 items relating to their interest in entertainment
options which might be offered in the future by a casino in Arizona. They used hierar-
chical clustering (with between groups linkage as the method and squared Euclidean
distances as the measure) to identify three clusters which they described as diners,
dancers and drinkers. The results of the study provided valuable information to
the property developers on the interests of different market segments and how these
segments could be best targeted. The authors identified that the cluster most likely to
increase patronage in response to the development of new facilities was the dancers and
therefore, if this market segment were targeted then the potential for additional revenue
could be maximized.
Jurowski and Reich (2000) also provide a very good discussion of the application of
cluster analysis with detailed discussion on the selection of methods and interpretation
of output from SPSS.
Cha et al. (1995) used cluster analysis to examine the travel motivations of Japanese
outbound tourists. They initially used factor analysis to reduce 30 motivational items
to 6 motivational factors, and then used the factor scores in an hierarchical analysis
of 5 per cent of the cases. The means observed were then used in a non-hierarchical
analysis of all 1199 cases. Although the authors do not explicitly mention a reason
for this approach, it is likely that the large number of cases and items was beyond the
computational power of their desktop computers as this research was conducted more
than 20 years ago. The use of factor analysis prior to clustering was very common in
early research most likely for this reason. However, Rolph (1970, cited in Hair et al.,
1998) says there is evidence to suggest the variables which truly discriminate between
underlying groups are not well represented in many factor solutions. In recent years
most clustering has been conducted on original variables. Following on from the cluster

DWYER 9781781001288 Part I.indd 216 25/07/2012 11:07

 Cluster analysis 217

analysis, Cha et al. (1995) used multiple discriminate analysis to explain the differences
between the three cluster they identified; sport seekers, novelty seekers and family/
relaxation seekers.
Another application of cluster analysis to market segmentation is found in Ryan and
Huyton (2000). These researchers used 28 items relating to desired tourism experiences in
the Northern Territory, Australia, to identify seven market segments and their differing
levels of interest in aboriginal tourism products.
Mueller and Lanz Kaufmann (2000) conducted an interesting study on the health
tourism market. They used a two-stage process (hierarchical followed by non- hierarchi-
cal) to identify four clusters based on the importance ratings of various wellness com-
ponents offered by hotels in Switzerland. These included tangible components such as
swimming pools, saunas and medical centres, as well as intangibles such as information,
courses and medical advice. The identified clusters included demanding health guests,
independent infrastructure users, care intensive cure guests and undemanding recrea-
tion guests.
Arimond and Elfessi (2001) used an interesting combination of multiple correspond-
ence analysis followed by a non-heirarchical clustering procedure. This was because of
the format of their data, which was non-metric categorical data. The seven variables
all related to desired services and benefits that customers seek when staying at a bed
and breakfast accommodation property, and for each variable respondents were asked
to select only the most desirable option, rather than rate each option on a Likert type
scale. The results from the multiple correspondence analysis were then input into the
non-hierarchical clustering procedure. Because the researchers were unsure about the
appropriate number of clusters they undertook a number of analyses and the solutions
were compared. The researchers settled on a four cluster solution which produced the
clearest distinguishable segments.
Najimi et al. (2010) used cluster analysis to segment the market of tourists to
Iran based on two variables: profitability and cultural compatibility. Using the
average linkage method and Euclidean distances as the measure, they identified three
clusters: semi-compatibles, compatibles and dissatisfied. They then undertook a
further segmentation of the compatibles cluster into Iranian-compatibles and Arab-
compatibles based on a common sense approach, and the fact that this delineation
would be more useful in targeting appropriate tourists. This demonstrates the applica-
tion of art in cluster analysis rather than just science as described by Jurowski and
Reich (2000).
The above studies demonstrate the usefulness of cluster analysis in market segmenta-
tion as it helps to identify important sub-groups of the total market which have different
needs and wants. However, as each application is in a very different tourism market, the
literature does not really form a body of knowledge beyond development of the methods
used. The next example application, however, resident perceptions of tourism, does
form a body of knowledge, because although the case studies are in different locations,
people are similar all over the world and there are commonalities in their reactions to
tourism.

DWYER 9781781001288 Part I.indd 217 25/07/2012 11:07

 218 Handbook of research methods in tourism

Cluster Analysis in Host Community Attitude/Perception Research

The assessment of how residents of a host community feel about tourism has seen numer-
ous applications of cluster analysis with the aim of understanding the impacts of tourism
on host communities.
An early example of this application was Davis et al. (1988) who used cluster analysis
to segment Florida residents in terms of their attitudes toward tourism development.
Using a hierarchical procedure they identified five clusters labelled as haters, lovers,
cautious romantics, in betweeners and love em for a reason. The first two groups
obviously represented extreme opposite opinions, while cautious romantics appeared
to be in favour of tourism but not pro-growth. The in betweeners had middle of the
road opinions, while members of love em for a reason had similar responses to lovers
but not as strong.
Similarly, Madrigal (1995) examined community reactions to tourism in two different
types of cities in the United States and United Kingdom, finding three clusters, haters,
lovers and realists. Haters agreed primarily with the negative impacts of tourism,
lovers primarily with the positive impacts, whereas realists agreed with both positive
and negative impacts.
Fredline and Faulkner (2000) used a similar approach to segment the host community
in terms of their perceptions of the impact of a motorsport event. They employed a two-
stage cluster analysis (hierarchical followed by non-hierarchical) and, using 36 event
impact statements measured on a Likert-type scale, identified five clusters labelled as
ambivalent supporters, haters, realists, lovers and concerned for a reason.
In a follow up study, Fredline and Faulkner (2002) conducted a similar analysis, this
time comparing the host communitys reactions to different motorsport events in dif-
ferent cities. Like Madrigals (1995) comparison they found that similar clusters can be
observed in different locations indicating that perceptions of the impacts of tourism (and
events) have some common elements regardless of the context. Fredline and Faulkner
(2002) then used logistic regression to explain cluster membership.
Other studies which have used cluster analysis in a similar way include Andriotis and
Vaughan (2003), Perez and Nadal (2005), Ryan and Montgomery (1994) and Williams
and Lawson (2001). In all these cases between three and five clusters were identified
representing different perceptions held by various segments of the community.

AN EXAMPLE OF CLUSTER ANALYSIS IN TOURISM

RESEARCH

The following is a worked example of an application of cluster analysis in tourism

market segmentation. The data were collected from competitors in a sport tourism event,
the Asia Pacific Masters Games. This is a multi-sport event held biennially on the Gold
Coast, Australia. The surveys were included in the participants registration packages
along with a reply paid envelope. There was a total of 1936 responses and 1905 of these
contained all data necessary for inclusion in the analysis.
The variables used to segment the market were the responses to nine statements
enquiring about competitors motivations for attending the event. The list of motivations

DWYER 9781781001288 Part I.indd 218 25/07/2012 11:07

 Cluster analysis 219

Table 11.3 Motivation items used in cluster analysis with means and standard deviations

Mean Std. Deviation

A I wanted to participate in my sport 5.45 1.797
B I wanted to participate in the social activities of the event 2.63 2.691
C I wanted to see the Gold Coast 0.96 1.883
D I wanted to spend time with my family or friends 2.48 2.852
E I wanted to meet new friends 2.88 2.673
F I wanted to relax and escape from everyday life 2.93 2.918
G I wanted to do something different 2.27 2.713
H I wanted to have fun 4.97 2.460
I I wanted to learn new things 1.83 2.584

is presented in Table 11.3. Each item was measured on a seven point Likert-type scale
ranging from not very important to extremely important.
The raw responses to the motivation items exhibited low variance as shown in Table
11.3, so the scores were standardized in order to magnify the differences. Thus the scores
represent deviations from the mean rather than a specific point on the scale.
A two-stage cluster procedure was used utilizing a hierarchical procedure followed
by non-hierarchical. The hierarchical procedure used Wards method and squared
Euclidean distances as the measure. A distance measure was selected because the aim
was to identify cases closest together in terms of their motivations. Wards method was
selected because of its tendency to produce more evenly sized clusters. Most other meas-
ures have a tendency to produce one large and numerous much smaller clusters which is
less useful for market segmentation (Hair et al., 1998).
As mentioned previously the hierarchical process produces n 2 1 solutions but for the
sake of parsimony it makes sense for the researcher to begin with the two cluster solution
and work backwards to select the smallest possible solution that appears to adequately
explain differences between the groups based on the means for each group on the original
variables. The agglomeration schedule suggested a three cluster solution as the percent-
age change in coefficient at this step was 2.5 per cent compared with 0.7 per cent at the
previous stage. However, interpretation of the means showed a meaningful split into
four clusters, and given Hair et al.s (1998) suggestion that agglomeration coefficients
can underestimate the number of clusters, a decision was made to proceed with the four
cluster solution.
In the second stage, the means of each of the four clusters identified in the first stage
were used as starting points or seeds for a non-hierarchical (quick cluster) procedure.
This allowed for the reclassification of cases so that they could move to the closest
cluster which is not possible in the hierarchical procedure. In this analysis, just over 16
per cent of cases were reassigned, most of these moving from cluster one to the other
three groups.
In SPSS the quick cluster procedure produces a number of output tables which can
provide other useful information. One of these is a table showing one-way ANOVA tests
of difference between the clusters on the original variables as shown in Table 11.4. While
it is inappropriate to use these to confirm the adequacy of the solution (it is a tautological
argument given that the clusters were derived by trying to maximize the differences on

DWYER 9781781001288 Part I.indd 219 25/07/2012 11:07

 220 Handbook of research methods in tourism

Table 11.4 Mean scores for each cluster and ANOVA results

I wanted to ... ANOVA results

and Novelty Seekers

and Friend Seekers

Relaxation Seekers
Cluster Two Fun
Cluster One Fun

Cluster Three

Cluster Four
Sport Focused
A. Participate in my sport 0.61 1.10 0.68 2.29 F (3, 1905) 5 530.58, p<0.001
B. Participate in the social activities of 20.51 0.14 0.01 20.24 F (3, 1905) 5 77.365, p<0.001
the event
C. See the Gold Coast 21.21 20.68 20.84 20.32 F (3, 1905) 5 134.032, p<0.001
D. Spend time with my family or friends 20.54 20.15 0.20 20.23 F (3, 1905) 5 80.604, p<0.001
E. Meet new friends 20.16 0.18 20.06 20.25 F (3, 1905) 5 32.046, p<0.001
F. Relax and escape from everyday life 0.23 20.77 0.80 20.32 F (3, 1905) 5 934.003, p<0.001
G. Do something different 0.63 20.60 20.64 20.31 F (3, 1905) 5 487.809, p<0.001
H. Have fun 0.80 1.26 0.87 20.35 F (3, 1905) 5 744.516, p<0.001
I. Learn new things 0.15 20.47 21.02 20.27 F (3, 1905) 5 314.130, p<0.001

these variables), it does provide some insight into which of the variables best explain the
differences between the groups.
In this study the greatest differences were observed with regard to items F (Escape), H
(Have Fun) and A (Participate in Sport). This can also be observed in a line chart of the
means for each cluster as presented in Figure 11.1.
Another interesting output table produced in SPSS is the distance between final
cluster centres. Theses distances have been reproduced in Table 11.5 along with the
mean distances of each case from its cluster centroid on the diagonal. This facilitates
comparison of the within and between group variability of the solution. As can be
seen, Cluster Four is the most homogeneous as its members have a mean distance from
their cluster centroid of 1.14. Conversely, the least homogeneous is Cluster One. A clear
discrimination between the groups cannot be claimed as the average distance of cluster
one members from the centroid is the same as the distance between centroids one and
three. However, this is to be expected given the highly specific nature of the event and
the low variance observed in the original variables. It does not create a problem for
interpretation given that the purpose of the study is to better understand market seg-
ments rather than to definitively classify specific cases into groups for some further
action.
The next step in interpretation of the analysis is to consider what is now known about
the clusters identified. In this case the mean scores on the original variables is the logical
starting point as these describe the motivations of the different segments. Figure 11.1
displays these means, making it clear what the main motivating factors for each group
were. Following on from this a more useful interpretation of the clusters can be explored
if additional data exist for the cases. In this study, a number of demographic and sport-
related variables were collected and these can be examined for differences between and
relationships with cluster membership to facilitate a fuller understanding of the clusters.
A brief discussion of these results is shown below as an example.

DWYER 9781781001288 Part I.indd 220 25/07/2012 11:07

 2.5
Cluster 1 Cluster 2 Cluster 3 Cluster 4

2.0

DWYER 9781781001288 Part I.indd 221

1.5

1.0

0.5

221
0.0

0.5

1.0

1.5
A. B. Social C. See D. Friends E. New F. Escape G. Do H. Have I. Learn
Participate activities Gold Coast & family friends something fun new things
in sport different

Figure 11.1 Means of clusters

25/07/2012 11:07
 222 Handbook of research methods in tourism

Table 11.5 Comparison of the within and between group variability

Cluster 1 2 3 4
1 2.07
2 2.08 1.81
3 2.07 1.82 1.93
4 2.54 2.19 2.561 1.14

Cluster One: Fun and Novelty Seekers

The first group has been labelled as Fun and Novelty Seekers, as these are the
motivations they rated as being relatively more important (not including sport par-
ticipation which was rated highly by all groups). They rated social activities, spend-
ing time with family and friends and seeing the Gold Coast as unimportant factors.
Relationships between cluster membership and other variables were examined in
contingency tables evaluated using the chi-square statistic. These results are shown in
Table 11.6.
Table 11.6 indicates that people in the fun and novelty group were more likely to be
younger and female, with dependent children at home. A larger than expected propor-
tion had part time jobs (based on the standardized residuals in the chi-square contin-
gency table analysis). They were most likely to be attending their first Masters Games
event, and a large proportion had been participating in their sport for less than 5 years.
They were more likely than other groups to rate their abilities as below average, and less
likely to be aiming to win their events.

Cluster Two: Fun and Friend Seekers

The next group rated fun and sport participation as the most important motivators for
attending the event, but the scores that differentiate them from other groups are for
social activities and meeting new friends. While most groups rated these as unimportant
factors, cluster two (on average) saw them as moderately important factors. Table 11.6
shows that they are somewhat more likely to be locals (Gold Coast residents) who are
older and retired.

Cluster Three: Relaxation Seekers

Like the other groups, the third cluster was highly motivated by sport and fun, but they
were also highly motivated by a desire for relaxation and escape from everyday life. This
group was the most likely to be working full time, and to have dependent children, so
perhaps they see the event as an opportunity to get away from lifes pressures. This group
was more likely to include repeat attendees at this type of event, with nearly half having
attended two or more Masters Games in the past. They were more likely to have com-
menced played sport at a young age (childhood or 20s) and therefore more likely to have
been playing for many years.

DWYER 9781781001288 Part I.indd 222 25/07/2012 11:07

 Cluster analysis 223

Table 11.6 Cross tabulations of cluster membership with demographic and event related
variables

Cluster Cluster Cluster Cluster Total

One Two Three Four N 5
Fun and Fun and Relax- Sport 1905
Novelty Friend ation Focused
Seekers Seekers Seekers n 5 285
n 5 515 n 5 573 n 5 532
Gender Male 45.5 57.9 53.8 68.8 1048
c2(3) 5 42.8, p<0.05 Female 54.5 42.1 46.2 31.2 856

Age Up to 40 years 28.9 20.9 23.9 15.2 435

c2(6) 5 91.2, p<0.05 4060 years 62.7 61.6 67.4 55.8 1182
Over 60 years 8.4 17.5 8.7 29.0 270

Dependant children? Yes 59.2 47.1 60.5 37.6 996

c2(3) 5 54.6, p<0.05 No 40.8 52.9 39.5 62.4 894

Employment status Full time employed 57.8 55.9 63.7 50.9 1090
c2(21) 5 104.0, p<0.05 Part time employed 23.0 15.2 18.4 13.8 339
Unemployed seeking work 1.0 0.9 0.6 2.1 19
Retired 6.3 15.3 6.7 24.4 223
Home duties 6.1 6.3 6.1 3.5 109
Student 1.4 0.7 0.2 0.4 13
Volunteer worker 0.8 1.4 1.5 0.7 22
Other 3.7 4.2 2.9 4.2 70

Place of origin Local 19.6 32.6 12.6 26.0 429

c2(3) 5 68.1, p<0.05 Non-local 80.4 67.4 87.4 74.0 1476

Previous event First event 54.6 41.4 29.7 35.1 776

attendance Second event 22.1 24.8 21.8 22.8 435
c2(12) 5 96.8, p<0.05 25 previous events 15.7 22.7 30.5 26.7 449
610 previous events 6.6 8.7 15.0 12.6 200
More than 10 previous 1.0 2.4 3.0 3.5 45
events

Type of event Team events 57.3 58.4 70.3 45.8 1129

c2(6) 5 60.6, p<0.05 Individual events 28.7 26.3 21.6 41.9 530
Both 14.0 15.3 8.1 12.3 237

Age commenced in sport Childhood/20s 39.3 41.7 48.8 37.5 784

c2(6) 5 33.0, p<0.05 30s/40s 50.8 44.2 43.2 43.7 843
Over 40 9.9 14.1 8.1 18.8 221

Years of participation Less than 5 years 34.8 30.6 24.0 22.8 529
in sport 510 years 15.6 10.5 13.7 12.7 241
c2(9) 5 30.8, p<0.05 1020 years 19.4 21.6 21.9 25.4 399
More than 20 years 30.2 37.3 40.4 39.1 673

Perception of ability 1 Very Low 1.6 1.6 1.7 0.7 28

c2(18) 5 42.2, p<0.05 2 3.4 2.1 2.3 2.5 48
3 8.1 6.6 3.4 4.3 108
4 Average 28.8 24.7 25.0 19.4 470
5 30.8 27.0 31.3 29.5 554

DWYER 9781781001288 Part I.indd 223 25/07/2012 11:07

 224 Handbook of research methods in tourism

Table 11.6 (continued)

Cluster Cluster Cluster Cluster Total

One Two Three Four N 5
Fun and Fun and Relax- Sport 1905
Novelty Friend ation Focused
Seekers Seekers Seekers n 5 285
n 5 515 n 5 573 n 5 532
6 19.6 26.0 25.4 27.0 454
7 Very High 7.7 11.9 10.8 16.5 209

Achievement aspirations I wanted... to win 17.3 24.3 18.5 32.3 417

c2(15) 5 57.4, p<0.05 ...to get a place in my 26.5 20.5 23.3 24.5 445
(our) event
... to perform better than 26.8 22.2 20.3 23.4 439
ever before
... to perform as well as 16.9 19.6 22.2 14.2 357
usual
I didnt care for myself 7.2 7.0 9.8 3.5 139
but did not want to let
the team down
Not at all concerned 5.3 6.5 6.0 2.1 102

Notes: and denote standardized residuals greater than or less than expected

Cluster Four: Sport Focused

The final group is clearly the most focused on sport and has rated all other motives as
relatively unimportant. As shown in Table 11.4, they rate fun as being much less impor-
tant than any of the other groups, and rate participating in sport as being much more
important. This group is predominantly male, and more likely to be older with 29 per
cent over 60 years of age. They are therefore more likely to be retired and less likely to
have dependent children. This group are more often involved in individual sports rather
than team sports and a relatively large proportion of them commenced in their sport at
over 40 years of age. Over half the group aspired to winning or placing in their event,
and they tended to rate their ability highly with a significantly higher proportion rating
their ability as very high.

Implications

As the above results show, this kind of analysis would be very valuable to the organizers
of this event in determine the future direction. They could elect to concentrate on one
or two of these clusters or alternatively to attempt to cater to the needs and wants of all
these groups. For example, they may wish to offer more associated social functions to
facilitate meeting new friends for the fun and friend seekers. Alternatively they could
offer more sport focused activities such as training and nutrition seminars if they feel
this is the target market they would prefer to pursue.

DWYER 9781781001288 Part I.indd 224 25/07/2012 11:07

 Cluster analysis 225

ADVANTAGES AND LIMITATIONS OF CLUSTER ANALYSIS

The main advantage of cluster analysis is that it provides an effective and meaningful
reduction of data. All tourism customers have slightly different needs and wants but it is
impossible to treat them all as individuals. For this reason it has been common practice
to either assume they all have the same needs or to segment them based on geographic
or demographic variables. Cluster analysis provides a better segmentation approach that
allows groups to be defined by their needs and therefore identifies groups who can be
targeted in a similar way because they are genuinely similar.
Similarly, in resident perception research, the tendency has been to consider the
majority opinion so that if more than 50 per cent of the population appear to support
a tourism issue then no problem is perceived. However, the cluster analysis approach
provides a much deeper understanding of gradations in perceptions of tourism and the
specific issues that appear to be of concern. This may encourage policy makers to con-
sider attempting to better cater to all segments of the community rather than just the
majority.
One of the major disadvantages of the technique in terms of its application is that the
solution is not definitive. That is, there is no single correct solution, all possible cluster
solutions are correct, it is just that some do not explain the data in a manner that is as
easy to interpret and use. Researchers who are looking for a black and white solution
can become frustrated with this lack of certainty.
Therefore it can be quite a labour-intensive process for the researcher to identify the
best solution, that is, the one that summarizes and simplifies the data set in a way that
makes it most manageable and provides insights that are useful.

FURTHER DEVELOPMENTS AND APPLICATIONS OF

CLUSTER ANALYSIS

In earlier days, the computational requirements of hierarchical analysis often meant

that researchers with data sets comprising many variables and cases would have to com-
promise by analysing only a subset of these. Obviously as desktop computers continue
to advance in terms of their capacity to process information such compromises will be
reduced.
Given the rapid pace of development of online surveys and handheld mobile devices
which can be used for field surveys there is potential for researchers to embed cluster
analysis into their data collection phase. Respondents could answer a range of questions
and it may be possible for the data to be processed during the survey so that respond-
ents could be assigned to a cluster immediately (using a variation of a non-hierarchical
approach). They could then be funnelled to a specific bank of questions applicable to
their cluster. This could either reduce the length of the questionnaire because the ques-
tions are only asked of those they are applicable to, or alternatively could lead to much
richer data being collected.

DWYER 9781781001288 Part I.indd 225 25/07/2012 11:07

 226 Handbook of research methods in tourism

CONCLUSION

This chapter has attempted to provide an overview of the usefulness of cluster analysis
in tourism research. A full discussion of the technicalities of the technique is beyond the
scope of this chapter but researchers who are interested in learning more are directed
to the many good texts on multivariate statistical techniques. Hair et al. (1998) is a par-
ticularly good example for non-mathematicians as its explanations of various issues are
simple and clear.
The chapter has also summarized some of the work in two areas which represent the
most common applications of cluster analysis in tourism research: market segmentation
and resident perceptions of tourism. It is hoped that researchers in these fields will con-
tinue to advance knowledge through the application and further development of cluster
analysis.

REFERENCES

Aldenderfer, M. and R. Blashfield (1984), Cluster Analysis, Newbury Park, CA: Sage Publications.
Andriotis, K. and R. Vaughan (2003), Urban residents attitudes toward tourism development: the case of
Crete, Journal of Travel Research, 42, 172185.
Arimond, G. and A. Elfessi (2001), A clustering method for categorical data in tourism market segmentation
research, Journal of Travel Research, 39, 391397.
Cha, S., K. McCleary and M. Uysal (1995), Travel motivations of Japanese overseas travellers: a factor
cluster segmentation approach, Journal of Travel Research, 34, 3339.
Davis, D., J. Allen and R. Cosenza (1988), Segmenting local residents by their attitudes, interest and opinions
toward tourism, Journal of Travel Research, 27 (2), 28.
Fredline, E. and B. Faulkner (2000), Host community perceptions: a cluster analysis, Annals of Tourism
Research, 27 (4), 763784.
Fredline, E. and B. Faulkner (2002), Residents reactions to the staging of major motorsport events within
their communities: a cluster analysis, Event Management, 7, 103114.
Gilmour, G. (1951), The development of taxonomic theory since 1851, Nature, 168, 400402.
Hair, J., W. Black, B. Babin, R. Anderson and R. Tatham (1998), Multivariate Data Analysis, Upper Saddle
River, NJ: Prentice-Hall.
Jobson, J. (1991), Applied Multivariate Data Analysis, Volume II: Categorical and Multivariate Methods, New
York: Springer-Verlag.
Jurowski, C. and A. Reich (2000), An explanation and illustration of cluster analysis for identifying hospitality
market segments, Journal of Hospitality and Tourism Research, 24, 6791.
Madrigal, R. (1995), Residents perceptions and the role of government, Annals of Tourism Research, 22,
86102.
Mueller, H. and E. Lanz Kaufmann (2001), Wellness tourism: market analysis of a special health tourism
segment and implications for the hotel industry, Journal of Vacation Marketing, 7 (1), 517.
Najimi, M., A. Sharbatoghlie and A. Jafarieh (2010), Tourism market segmentation in Iran, International
Journal of Tourism Research, 12, 497509.
Perez, E.A. and J.R. Nadal (2005), Host community perceptions: a cluster analysis, Annals of Tourism
Research, 32 (4), 925941.
Ryan, C. and J. Huyton (2000), Who is interested in Aboriginal tourism in the Northern Territory? A cluster
analysis, Journal of Sustainable Tourism, 8 (1), 5388.
Ryan, C. and D. Montgomery (1994), The attitudes of Bakewell residents to tourism and numbers in com-
munity responsive tourism, Tourism Management, 15, 358369.
Williams, J. and R. Lawson (2001), Community issues and resident opinions of tourism, Annals of Tourism
Research, 28, 269290.

DWYER 9781781001288 Part I.indd 226 25/07/2012 11:07

 12 Inputoutput and SAM models
Clemente Polo and Elisabeth Valle

NATURE OF THE TECHNIQUE AND ITS EVOLUTION

Inputoutput (IO) and social accounting matrix (SAM) models are tools that economists
have used for many decades to measure sectoral interdependencies, compare the eco-
nomic structure of economies, quantify production impacts, measure structural and pro-
ductivity changes, study the effects of redistribution policies, calculate the energy content
of commodities, estimate CO2 emissions, etc. The IO model is a simple linear model
first proposed by Leontief (1937) to determine production quantities and prices in a set
up where commodities are produced with commodities. It was as Leontief himself put
it an attempt to apply the economic theory of general equilibrium or rather, general
interdependence in an empirical study of the interrelationships between different parts
of the economy .... In Leontief s model quantities of different commodities and prices
were treated as variables and the technical and natural conditions of production and the
tastes of consumers as data.
The applied nature of his endeavour and the lack of detailed technological inves-
tigations to cover the entire field of agricultural, mineral and industrial production
led Leontief to specify the technical set up of each industry by as many homogeneous
linear equations as there are separate cost factors involved.1 Following Walras, Leontief
termed coefficients of production the parameters included in those linear equations and
assigned them numerical values employing a statistical table constructed for the United
States that included detailed information on quantitative input and output relations for
44 industries in 1919 (Leontief, 1936). The first tableau economique based on observed
data, or an IO table, of a national economy and the first numerical general equilibrium
model based on observed data had been born.
The first contributions of Leontief spurred the construction of IO tables to analyse
a wide range of issues with IO models.2 Stone (1961) pioneered the integration of IO
tables in the National Accounts framework taking the disaggregation of the production
account as a starting point. Stones report also included a self-contained account of the
algebra of the so-called Leontief s open and closed quantity and price models, extensions
of the basic model in various directions (restricted, dynamic and interregional models)
and a review of some applications to real economies. Other statements of IO analysis
can be found in Bulmer-Thomas (1982), Leontief (1966), Miernyk (1965) and Miller and
Blair (1985, 2009).
The extension of the IO table and Leontief s open model into a SAM and SAM model,
respectively, was a quite natural development. In 1969, the new system of national
accounts elaborated by the Statistical Office of the United Nations (SNA, 1969) first
included a square matrix representation of national accounts. Each account is fully
described by a row where revenues from all accounts are recorded and a corresponding
column where expenditures to all accounts are recorded. The matrix representation is

227

DWYER 9781781001288 Part I.indd 227 25/07/2012 11:07

 228 Handbook of research methods in tourism

not only a more transparent and efficient way to present national accounts but also a
very flexible tool since accounts can be easily subdivided.
It was only a matter of time until disaggregated national accounts were presented in a
matrix form to integrate the institutional accounts with the IO table and that they were
employed to specify numerically economic models. Pyatt et al. (1972) and Thorbecke
and Sengupta (1972) were the first to use it to analyse redistribution policies and growth.
Keuning and Ruijter (1988) signal Pyatt and Thorebeckes (1976) study of Bangladesh
as the first comprehensive presentation of a SAM and its uses. Other classical references
are Cohen (1989), Pyatt (1988), Pyatt and Round (1977, 1979), Pyatt et al. (1977) and
Thorbecke (2000).3
SAM models have two main advantages over IO models. First, the household(s)
and capital (savingsinvestment) accounts are balanced and the induced effects
caused by income generation (consumption and savingsinvestment) can be analysed
in a more consistent way than in extended IO models. Second, SAM models can
analyse the distributive consequences of injections, a subject that has no room in the
IO models. There are disadvantages too. On the one hand, SAM models require a
SAM and only a few Statistical Offices that routinely elaborate IO tables for national
or regional economies also compile SAMs. On the other hand, the extension of fixed
coefficients beyond the production sphere is even more questionable than in the IO
model.
The chapter is divided in four sections. The next section presents the basic algebra
employed in IO and SAM applications. Then how the models can be applied to tourism
and numerous IO and SAM applications to tourism since the 1970s are discussed. The
strengths and weaknesses of the models are then considered. The final section includes
a quest for improving the data sets and the way research results are presented. It also
discusses how tourism satellite accounts and some environmental approaches can be
combined with IO models, two topics that have received considerable attention in the
last decade.

BACKGROUND AND TYPES OF PROBLEMS THAT THE

TECHNIQUE IS DESIGNED TO HANDLE

Since the 1970s, IO models have been employed to asses tourism impacts. More recently
SAM models have also been used. In this section, we briefly present the algebra of IO and
SAM models and discuss their numerical specification.

A Basic IO Model

There are 1, 2, ..., M sectors in the economy. Sector i produces product i employing
intermediate products (Xji) j51,2,. . .,M and labour Li and capital services Ki. The ith sectors
output is determined by a Leontief linear homogeneous production function

Yi 5 min a , , b
X1i X2i XMi Li Ki
, , . . ., (12.1)
a1i a2i aM1 li ki

DWYER 9781781001288 Part I.indd 228 25/07/2012 11:07

 Inputoutput SAM models 229

where (aji) j51, 2,. . ., M , li and ki are sectors i technical coefficients (Leontief s production
coefficients) or unitary requirements of all products, labour and capital, respectively. It
follows that the amount of inputs that minimize the cost of production

Xji 5 ajiYi, Li 5 liYi, Ki 5 kiYi, j, i 5 1, 2, . . ., M (12.2)

are independent of production and factor prices.

The production level Yi that satisfy a given set of net (final) demands
Di $ 0, i 5 1, 2, . . ., M is therefore
N
Yi 5 a aijYj 1 Di, i, j 5 1, 2, . . ., M. (12.3)
j

The equilibrium conditions (12.3) can be expressed in matrix notation

y 5 Ay 1 d (12.4)

where A is the M 3 M matrix of technical or unit requirement coefficients and y and

d the column vectors of production and exogenous final demand, respectively. The
elements in the ith column of matrix A are the unit requirements of sector i.Thesolution
(provided A is invertible) of the system of Equation (12.4)

y 5 (I 2 A) 21d 5 Md (12.5)

is a linear transformation of the vector of final demands. M is known as Leontief s mul-

tiplier matrix and it can be easily checked that
` `
M 5 I 1 A 1 A2 1 . . . 5 a Ai 5 I 1 a Ai (12.6)
i50 i51

Substituting (12.6) in (12.5)

`
y 5 d 1 Ad 1 A2d 1 . . . 5 d 1 a Aid
i51

total output is broken down into two terms: output required to satisfy final demand and
output used by other sectors (indirect or intermediate demand). Since Equation (12.5) is
valid for any nonnegative d, the elements of the ith column of M, m # i 5 (mji) j51, 2, . . ., M ,
can be interpreted as sector j production needed to produce one net unit of i. This prop-
erty of matrix M led to define interdependency indexes (Beynon et al., 2009; Chenery and
Watanabe, 1958; Rasmussen, 1956) known as backward and forward linkages employ-
ing the column and row totals of M, respectively.
Labour and capital inputs required to produce d are also linear transformations of
output and therefore of final demand

DWYER 9781781001288 Part I.indd 229 25/07/2012 11:07

 230 Handbook of research methods in tourism

M `
L 5 a liYi 5 lTy 5 lTMd 5 lTd 1 a lTAid
i51 i51
M `
(12.7)
K 5 a kiYi 5 kTy 5 kTMd 5 kTd 1 a kTAid
i51 i51

As in the case of output, labour and capital requirements to produce d can be split into
direct and indirect labour and capital requirements.
A more detailed account of output, labour and capital requirements can be obtained
substituting vector d by a diagonal matrix D where the ith diagonal element dii 5 di

Y 5 MD, l 5 lTMD, k 5 kTMD (12.8)

where Yji indicates the amount of product j required to produce di units of i, and l i and
ki the total amount of labour and capital, respectively, required by all sectors to produce
di units of i.4 Since (12.5), (12.7) and (12.8) are linear transformations, the impact of dis-
crete changes in d are

Dy 5 MDd, DL 5 lTMDd, DK 5 kTMDd (12.9)

(12.3), (12.7) and (12.9) can be used to calculate the amounts of output, labour and
capital required to produce d or Dd, provided A, l and k remain unchanged.5
The basic open model just presented can be partially closed. Leontief (1937) already
treated households as any other industry. The household industry can be endog-
enized assuming it produces a composite consumption good that is required by all other
industries. Thus, Equation (12.4) becomes

y 5 A*y 1 d (12.10)

where A* is a (M 1 1) 3 (M 1 1) coefficients matrix and d is the final demand vector

net of households consumption. The M 1 1 column of A* includes the technical coef-
ficients of the household industry employed to produce the composite consumption
commodity and the M 1 1 row the composite consumption requirements by all indus-
tries.6 As households are included as an industry the new (M 1 1) 3 (M 1 1) multiplier
matrix

M* 5 (I 2 A*) 21

captures not only the direct and indirect effects but also those induced by the new indus-
try.
The basic model presented has intentionally ignored that products can be distin-
guished by origin. Following the national accounts identity, exports of different products
are included in d while product imports are added to domestic production to insure
equality of supply and demand

y 5 Ay 1 d 2 m (12.11)

DWYER 9781781001288 Part I.indd 230 25/07/2012 11:07

 Inputoutput SAM models 231

where m is the vector of equivalent imports. We shall come back to this point later.

Numerical Specification of the IO Table

IO tables present the identity of the production account for each sector: the value of
domestic production plus imports equals the value of intermediate consumption plus
final demand. Leontief (1936) presented the first IO table for the US economy in 1919
and his work spurred the construction of similar tables for other countries, states and
|
regions. A standard IO table includes an intermediate transactions matrix (Xji) , a factor
matrix F that shows sectoral payments to labour, capital, government and foreign
sectors and a final demand matrix D that provides the breakdown of private and public
consumption, investment and exports (excluding tourism exports). From the informa-
tion in the IO table, expenditure coefficients can be obtained for each sector
| | |
Xji pjXji | Li wLi | Ki rKi
|
a ji 5 | 5 , li 5 | 5 , ki 5 | 5 (12.12)
Yi piYi Yi piYi Yi piYi

If units for products, labour and capital are chosen in such a way that their prices are
equal to 1 (pi 5 w 5 r 5 1) , then the technical coefficients are equal to the expenditure
coefficients. Therefore, if Dd is measured in the same units, (12.9) can be interpreted as
changes in output, labour and capital quantities. Notice that column totals of M add up
quantities of different products; however, since prices are equal to 1 they can be inter-
preted as the total value of output. For the same reason,

DYj piDYj
mji 5 5
Ddi pjDdi

can be interpreted as the increase in the value of production of sector j brought about by
one additional unit spent in product i.

SAM and SAM Models

Let X be a SAM, i.e., a square N 3 N matrix that records transactions arising in the
circular flow of income (production, income generation, income distribution and income
expenditure) of an economy. The elements in the ith row indicate the sources of income
accruing to the ith account, and those in the ith column its expenditures. Thus, row
(column) sums indicate the total income accruing (spent or saved) by accounts. In a
SAM, row totals (revenues) equal column totals (outlays):
N
| | N
|
a X ij ; Y i ; a Xji (12.13)
j51 j51

A SAM may include accounts for industries, commodities, institutional sectors (house-
holds, non-profit institutions, corporate sector, public administrations and foreign
sectors), capital (savingsinvestment), and as many auxiliary accounts as needed.
Equation (12.13) implies that the value of sales equals total costs in any industry; the

DWYER 9781781001288 Part I.indd 231 25/07/2012 11:07

 232 Handbook of research methods in tourism

value of demands equals the value of supply for commodities; and total revenues equal
total expenditures and savings for institutions.
Expenditure coefficients can be defined as
|
Xij
|
a ij ; | (12.14)
Yj

and (12.13) can be expressed in terms of them

| N
|
Yi ; a |
a ijYj (12.15)
j51

The right hand term in Equation (12.15) can be split in two terms: one for the income
accruing from the accounts the first M accounts, considered endogenous, and the other
for remaining accounts taken as exogenous:
M N
Yi ; a |
a ijYj 1 a |
a ijYj i 5 1, 2, . . ., M (12.16)
j51 j5M 11

Identities (12.16) become a system of equations when the exogenous incomes

and the expenditures coefficients are assumed to be fixed. Assuming that units of
products, commodities and factors are chosen so that their prices are equal to 1,
some fixed expenditure coefficients (interindustry, industrycommodity, primary
factors) can be interpreted as technical coefficients and the results obtained with
SAM models compared with those of IO models.7 For unitary prices, balance con-
ditions (12.16) can be interpreted as zero profit conditions for industries, demand
and supply equilibrium conditions for commodities and budget constraints for
institutions. Having made these assumptions we drop the tilde from the coefficient
expenditures.
Once the partition of accounts is done, matrix A can be partitioned accordingly

A5a b
Amm Amn
Anm Ann

and (12.16) be written as

ym 5 Ammym 1 Amnyn (12.17)

where ym and yn are the vectors of endogenous and exogenous income, respectively.
From Equation (12.17) one can derive the vector of endogenous income

ym 5 (I 2 Amm) 21Amnyn 5 MmAmnyn 5 Mmdm (12.18)

being dm the vector of exogenous income Amnyn. As in the IO model, Mm is the multiplier
matrix and mji can be interpreted as the increase in income of account j brought about
by a unit injection into account i.

DWYER 9781781001288 Part I.indd 232 25/07/2012 11:07

 Inputoutput SAM models 233

Income Redistribution

The effects of exogenous injections on relative incomes can be easily calculated. First, the
vector of relative incomes is written in matrix form

ym Mmd
y5 M
5 (12.19)
eTym
a Yj
j51

being eT a row vector of ones. Then variation of the vector of relative incomes resulting
from exogenous injections is given by the redistribution matrix

0y1 0y1 0y1

...
0d1 0d2 0dm
0y2 0y2 0y2
...
MmeTym 2 ymeTMm
5 T m aMm 2
ymeTMm
b
0y 0d1 0d2 0dm 1
R5 5 5
0d ... (e y )
T m 2
e y eTym
0ym 0ym 0ym (12.20)
...
0d1 0d2 0d2m

whose rji is
M
Yj a mki
mji 2
0yj 1 k 51
5 (12.21)
0di eTym M

a Yk
k 51

The sign is determined by the parentheses and is positive if and only if

mji Yj
M
$ M
a mki a Yk
k 51 k 51

i.e., if the marginal relative effect on j (the term on the left-hand side of the equality) is
greater than the relative income of j (the right-hand term of the inequality). It can easily
be checked that the sum of the elements of each column of the redistribution matrix is
zero

5 T m aeTMm 2
eTymeTMm
b50
0y 1
eT
0d e y eTym

APPLICATIONS OF THE TECHNIQUE TO TOURISM

Economists interest in tourism came to age when the rapid increase in per capita
income in the western economies and the reduction of transportation costs after World
War II gave rise to mass international and domestic tourism. By the 1970s, servicing

DWYER 9781781001288 Part I.indd 233 25/07/2012 11:07

 234 Handbook of research methods in tourism

international tourism had become a key export industry that accounted for large shares
of output, income, labour and tax revenues in the recipient countries (Archer, 1982).
At the same time, domestic tourism had also become an important source of revenues
for regions, counties, cities and recreational areas (Archer, 1978). Not surprisingly, IO
models (or approaches inspired by it) were used since the 1960s to quantify the impact
of international tourism in large, medium and small national economies as well as in
regions, counties, cities and recreational areas all over the world. In the last two decades,
SAM and CGE models have also been applied to that end.

How to Apply the Models

Leaving aside for the moment concerns about the underlying assumptions made in IO
and SAM models, the fact is that they have been routinely employed to measure the
weight of tourism in the economy and quantify tourism impacts on output, income and
employment. These models can be applied if the appropriate information is available: an
IO table (or a SAM) for the economy at hand and a vector of tourists expenditures by
sector d valued at the base year prices. Of course, one may subdivide the tourists vector
into as many vectors (d 5 d1 1 d2 1 . . . 1 dK) as types of tourists data allows to differ-
entiate and calculate their individual contributions and impacts using Equations (12.5),
(12.7), (12.9), (12.19) and (12.21).8
If an IO table is available, the only difficulty to apply the model is to cast tourists
expenditures into a vector d congruent with the classification employed and the way
transactions are valued in the IO table. To begin with, when the IO table is dated at year
t and the impact occurs at t 1 s, vector et1s has to be valued at time t prices.9 Moreover,
tourists expenditures include distribution and transportation margins and product
taxes that need to be eliminated to make them congruent with IO tables. Reported
expenditures categories are often too broad to be assigned to specific IO sectors (Khan
et al., 1990). Moreover, as Cooper and Pigram (1984) point out, they have be valued as
transactions are in the table, i.e., excluding margins and taxes. In the case of domestic
tourism, adjustments have also to be made to deduct expenditures that would have been
made anyway and add pre-trip expenditures not included in reported tourists expen-
ditures. The development of Tourism Satellite Accounts (TSA) can provide adequate
estimates of tourists demand to be used jointly with IO tables to analyse its impact on
the economy. We shall return to this point.
A final point refers to the treatment of import leakages since not all tourists expenditures
are demand for local products and not all inputs employed by industries are of domestic
origin. As indicated earlier, an accepted solution to that problem (Lee and Taylor, 2005) is
to deduct imports from final demand as in (12.11). However, a more accurate way to deal
with imports is replace Equation (12.5) and (12.7) by their domestic counterparts

yI 5 (I 2 AI) 21dI 5 MIdI (12.5)

M `
L 5 a lI # iYI # i 5 lTIyI 5 lTMIdI 5 lTdI 1 a lTAiIdI
i51 i51
M `
(12.7)
K 5 a kI # iYI # i 5 k y 5 k MIdI 5 k dI 1 a k A d
T
I I
T T T i
I I
i51 i51

DWYER 9781781001288 Part I.indd 234 25/07/2012 11:07

 Inputoutput SAM models 235

yM 5 AMyI 1 dM 5 AMMIdI 1 dM

where the subscript I (M) denotes the matrix of technical coefficients and the net demand
vector of domestic (imported) origin. Unfortunately, even recent IO studies (Kweka
et al., 2001; Mazumder et al., 2011; Steenge and Van de Steeg, 2010; and Oosterhaven
and Fan, 2006) overlook import specification or apply a domestic vector dI to the total
coefficient matrix A. It should be emphasized that the distinction of products by origin
can also be incorporated into a SAM model, although the distinction has been generally
ignored in tourism studies.10

The Multipliers Fever

IO tables are often unavailable for small national, regional and local economies. The
rapid growth of tourism in the 1970s and early 1980s gave rise to a multipliers fever and
many studies were completed to calculate (personal) income and employment tourism
multipliers and provide marketing clues to policy makers. Archer and Owen (1971)
proposed a method that Henderson and Cousins (1975) perfected to calculate income
and employment multipliers. Following Archer (1976) and Vanhove (1981), total income
(employment) generated was the sum of direct, indirect and induced income (employ-
ment). Direct and indirect household income were approximated by
K M

a a qkiIi (12.22)
k 51 i51

where qki is the proportion of the expenditure by tourist k directed to business i over total
tourists expenditures and Ii the direct and indirect income generated by 1 unit spent in
business i. Total income was obtained multiplying (12.22) by a Keynesian multiplier
defined by
1
M
1 2 c a qriIi
i51

where c is the average propensity to consume over disposable income and qrithe ratio of
residents consumption of locally produced commodity i over total consumption.11 Ii was
in turn estimated from net of tax factor receipts and purchases to other businesses. Liu
and Var (1982) for Victoria BC and Liu et al. (1984) for Turkey employed the multiplier
matrix M to estimate Ii.
The typical study included the completion of two surveys: one directed to local
businesses to determine their cost structure and the other to incoming tourists to esti-
mate their expenditures. The results obtained (Archer, 1976, 1982; 1984; Milne, 1987;
Vanhove, 1981; Var and Quayson, 1985) indicate that the size of multipliers depends
(as expected) on three factors: the importance of import leakages, the value added coef-
ficients of tourists oriented sectors and the strength of linkages between tourist-oriented
sectors and other sectors. Thus, income multipliers for states (Hawaii and Missouri) and
small islands with a developed economic base (Dominican Republic and Hong Kong)
or high value added coefficients in tourists sectors (Bermuda and Indian Ocean) are

DWYER 9781781001288 Part I.indd 235 25/07/2012 11:07

 236 Handbook of research methods in tourism

large (0.91.30) while for small islands with a poor economic base (Cayman and British
Virgin) and most counties in the United States and UK are small (0.30.6). Liu and Vars
results for Turkey give a direct and indirect income multiplier of 0.79 and a total multi-
plier (household endogenous) of 1.87 for all visitors. Differences among various types of
visitors were negligible but Turkish tourists from abroad had the largest impact.

National and Regional IO Impact Studies

The ad hoc tourism multipliers approach (Fletcher, 1989) was gradually abandoned in
favour of standard IO modeling.12 IO tables (and to a less extent SAMs) are elaborated
regularly along with national accounts in many developed countries since the 1980s and
IO tables have also been compiled occasionally for some African and Asian countries,
regions and small islands. However, many of these databases have been constructed for
other purposes and do not provide a vector of tourists expenditures to be fed into IO
and SAM models. This is so even in countries where international tourism is a major
export industry and domestic tourism accounts for a sizable share of total residents
expenditures.13
The following indicative list includes some IO studies that have analysed the structure
of tourism or estimated its impact on the reference economy. At a national level, one can
mention the studies of Cooper and Pigram (1984) for Australia; Henry and Deane (1997)
for Ireland; IET (1997) for Spain;14 Kweka et al. (2001) for Tanzania; Liu et al. (1984)
for Turkey; Mazumder et al. (2011) for Tanzania; and Oosterhaven and Fan (2006) for
China. Also from a national perspective, there are studies that focus on specific tourist
activities such as those of Felsenstein and Freeman (1998) of gambling in Israel; Kim et
al. (2003) of conventions in South Korea; and Lee and Taylor (2005) of a mega sport
event (FIFA World Cup).15 The IO model has also been applied to analyse tourism
in regions and islands: Archer (1995) in the Bermudas; Archer and Fletcher (1996) in
Seychelles; Archer and Wanhill (1980) and Archer (1985) in Mauritius; Baster (1980) in
Scotland; Heng and Low (1990) and Khan et al. (1990) in Singapore; Jones and Munday
(2010) in Wales; Lin and Sung (1984) in Hong Kong; Polo and Valle (2008a and 2011)
in the Balearic Islands; Pratt (2011) in Hawaii; Ruz (1985) in Puerto Rico; Steenge and
Van de Steeg (2010) in Aruba; and West and Gamage (2001) in Victoria.
The studies for Australia, Turkey, Spain and Ireland differentiate international
and domestic tourism. Baster (1980) for Scotland and Polo and Valle (2008a) for the
Balearic Islands employ just one vector for non-residents demand, while West and
Gamage (2001) distinguish four tourists types: international, interstate, intrastate and
day-trippers. In contrast, the China study and most studies of small islands and develop-
ing countries only deal with international tourism, although Archer (1995), Archer and
Fletcher (1996), Heng and Low (1990) and Mazumder et al. (2011) differentiate them by
nationality.
As to the integration of tourists expenditures into the IO framework, generally
national and regional IO tables present a fine sectoral breakdown that facilitates tourism
impact studies. An exception is Oosterhaven and Fan that use a Chinese table with only
17 sectors. In those cases where IO tables were specifically compiled to carry out tourism
studies (Archer, 1985, 1995; Archer and Fletcher, 1996) all important tourism-oriented
sectors (hotels, other tourist accommodation, restaurants, travel agencies, car renting,

DWYER 9781781001288 Part I.indd 236 25/07/2012 11:07

 Inputoutput SAM models 237

etc.) are included in the tables and the impact of tourists expenditures directed to those
sectors is calculated. As to the estimation of the tourists vector, sometimes it is described
in great detail (Cooper and Pigram, 1984) or dispatched in a cursory way (Oosterhaven
and Fan, 2006).
It seems fair to say that less attention than desirable has been paid to model import
leakages. For instance, Oosterhaven and Fan (2006) adjust consumption coefficients but
say nothing on the tourists demand vector and the coefficients matrix. What for a large
developing country such as China might be understood, it can greatly distort the results
obtained for regions and small islands. Archer and Fletcher (1996) indicate that a large
quantity of additional data relating to imports by establishment had to be obtained in
order to obtain a detailed breakdown of intermediate consumption into its local and
imported components. Polo and Valle (2002, 2008a) employed only domestic technical
coefficients and domestic non-residents demand. Steenge and Van de Steeg (2010) also
indicate that imports and the taxes less subsidies on products were removed from
intermediate consumption, final consumption expenditure, gross capital formation and
exports. But in other studies, it is hard to guess what the authors actually did and how
they did it.
The most complete reports provide output multipliers for the basic and extended
IO models, backward and forward linkages, multipliers for value added, income and
employment economy, tax revenues and imports. However, some studies include only
weighted multipliers; others only output multipliers for the basic or the extended model;
in one case (Henry and Deane, 1997) the Government is also endogenous; and others
report only absolute impact figures. It is hard to compare results coming from such
heterogeneous studies.16 Table 12.1 presents some (average) output, income/value added
and employment multipliers for the basic model (unless otherwise indicated). They are
either reported in the studies or calculated by us.
As indicated, many of these studies provide other useful pieces of information on the
economies studied of great interest for policy makers: backward linkages of tourism
sectors; the relative importance of tourism respect to other exports; government revenues
derived from tourism; imports required to satisfy tourists demands and multipliers for
different tourist services (accommodation types, non-real estate renting, food, entertain-
ment, car renting, etc.) and different tourists types (residents and non-residents classified
by origin or trip motivation).
Some of them also calculate the production, value added and employment shares
accounted for tourists demand and measure the weight of tourism under alternative
assumptions on final demand. Tourists contributions to total income and employ-
ment (measured by GVA, GDP or GNP) go from very low levels in China (1.64 and
1.01 per cent in 1997) and Australia (2.92.8 and 3.32.6 per cent in 197475 and
198081), to medium range values for Ireland (4.54.9 and 5.66.8 per cent in 1990
95) and high values for Spain (9.33 per cent of total value added in 1992). Archer
(1995) and Archer and Fletcher (1996) reported even larger values for Bermuda
(45 per cent of income) and the Seychelles (23.5 per cent of GDP) with residents
consumption endogenous.
Polo and Valle (2007, 2008a) estimated with the IO model that non-residents demand
accounts for 36.2 per cent of value added and 30.12 per cent of employment in the
Balearic Islands in 1997, figures that increase up to 72.7 and 65.11 per cent, respec-

DWYER 9781781001288 Part I.indd 237 25/07/2012 11:07

 238 Handbook of research methods in tourism

Table 12.1 Output, income and employment multipliers: direct 1 indirect

Authors, country/ island or region and base year Output Income/ Employment c
VA b
Liu, Var and Timur (1984): Turkey,1973a 0.79
Cooper and Pigram (1984): Australia (197475 and 198081)
IET (1997): Spain, 1992 a 1.62 0.96
Henry and Deane (1997): Ireland 1990 and 1995 0.73 45.35
0.74 39.69
Felsenstein and Freeman (1998): Israel, 1982a, d, f 3.59 2.87 25.12
Tohamy and Swinscoe (2000): Egypt, 1991/92 d 2.64 / 755.18
Kim, Chon and Chung (2003): South Korea, 1998e 1.67 0.36/ 105.9
0.88
Kweka, Morrisey and Blake (2001): Tanzania, 1992e 3.37 / 7.78
Lee and Taylor (2005): South Korea, 1998d 1.92 0.59/ 60.07
1.37
Oosterhaven and Fan (2006): China, 1997d 2.80 1.80 74.57
Baster (1980): Scotland, 1973a 0.32 24.18
Lin and Sung (1984): Hong Kong, 1973 0.98 0.59 26.7
Ruz (1985): Puerto Rico, 1972 d 2.08 1.24 142.10
Archer (1985): Mauritius, 1982d 1.03 5.32
Heng and Low (1990): Singapore, 1983 1.46 0.75 26.66
Khan, Seng and Cheong (1990): Singapore, 1983 1.48 0.69 24.84
Archer (1995): Bermuda, 1985c n.a. 0.91 29.39
Archer and Fletcher (1996): Seychelles, 1991d n.a. 0.88 17.09
West and Gamage (2001): Victoria, 199394a 1.47 1.11/ 8.62
1.31
Polo and Valle (2008a): Balearic Islands, 1997a 1.28 /0.78 19.06
Steenge and Van de Steeg (2010): Aruba, 1999 0.68 n.a.
Mazumder et al. (2011): Malaysia, 2000 1.42 0.34 17.36
054
Polo and Valle (2011): Balearic Islands, 2004a 1.73 /0.71 15.31
Pratt (2011): Hawaii, 2005 /0.70

Notes:
Reported values and own calculations for foreign tourists unless otherwise indicated.
a
Domestic and international tourists distinguished.
b
Income when just one figure is reported.
c
Employment multipliers indicate the number of jobs created by unit of tourists expenditures. Ireland:
per million IR pound; Israel: per million $; Egypt: per million $; South Korea: per million $; Tanzania:
per million TShs; China: per million Yuan; Scotland: per 1000 pound; Hong Kong: per million HK$;
Puerto Rico: per million $; Mauritius: per 100 000 Rs; Singapore: per million $; Bermuda: per million $;
Seychelles: per 100 000 SEYRs; Balearic Islands: per million ; and Malaysia per 100 million Ringitt.
d
Includes induced effects.
e
For the hotels and restaurants sector (tourism) output and employment multipliers are 1.84 and 4.85.
f
International convention industry.
g
Casino gambling.
h
2002 FIFA World Cup.

DWYER 9781781001288 Part I.indd 238 25/07/2012 11:07

 Inputoutput SAM models 239

tively, when consumption and depreciation investment are endogenous. Polo and Valle
(2011) calculated that value added and employment shares accounted for non-residents
demand were 36.59 and 32.65 per cent, respectively, in the Balearic Islands in 2004.17
They also obtained that the value added and employment shares accounted for non-
residents demand exceeds 70 per cent of value added in renting (non-real estate) and
holiday home lodging and 90 per cent in lodging services and passenger supporting
services. Pratt (2011) shows how value added multipliers have evolved in Hawaii since
1967 until 2005.

Small Areas, Single Resource and Event Tourism Impact IO Studies

Hundreds of studies have been done to quantify the impact of tourism in small areas
and some of them have relied on the IO model (or models close to it). No matter the
visitors purpose (relax, enjoy nature, practice sports, visit cities historical landmarks
and museums, gamble in casinos, assist in events, etc.), their expenditures have economic
effects on the communities where the attraction is located or the event takes place, as well
as in the surrounding areas. Archer (1976) already reported all purpose tourism income
multipliers for a considerable number of counties in the UK and United States. Since
then, the emphasis has shifted towards the assessment of a single tourist resource or
tourist activity (Johnson and Moore, 1993; Mescon and Vozikis, 1985).18 Consequently,
the number of studies has grown exponentially19 and several authors have presented
guides to avoid methodological pitfalls (Burgan and Mules, 1992; Crompton et al.,
2001; Gelan, 2003; Getz, 1989; Matheson, 2002; Mules and Faulkner, 1996; Stynes,
1999; Tyrrell and Johnston, 2001).
Indeed, the assessment of tourism impacts whether all purpose, single resource or
event related in small economies present some peculiarities regarding both the esti-
mation of the injection and the model employed to simulate the effect. First, recorded
tourists expenditures may not reflect the amounts really spent in the area, or those
spent solely due to the resource (Johnson and Moore, 1993) or the event object of study
(Burgan and Mules, 1992; Lee and Taylor, 2005; Tyrrell and Johnston, 2001).20 Second,
since small areas are generally very open economies and leakages (product imports and
factor payments to nonresidents) and feedbacks (product exports and factor payments
to residents) may be quite substantial, considerable attention must be paid to the deline-
ation of the impact region (Stevens and Rose, 1985). Third, only local expenditures that
would have been spent out of the area in the absence of the resource or the event should
be added to visitors expenditures (Johnson and Moore, 1993; Mules, 1999). Therefore,
surveys conducted to estimate the expenditures in the destination area or due to the
resource or event should take into account all these specificities.
A final problem is that IO tables are not usually available for small areas. In some
cases, the estimated tourists injection is simulated either with a regional IO model
(Leistritz, et al., 1982; Mescon and Vozikis, 1985) or with a local inputoutput technical
coefficient matrix (Var and Quayson, 1985).21 But most applications rely on ready-made
models. IMPLAN, funded by the USDA Forest Services (Alward and Palmer, 1983)
provides in its last version a complete set of 2008 county level SAMs with 440 sectors and
9 households, output income and employment effects, and simple and total (households
endogenous) multipliers. BEA-RIMS II is a model constructed and supported by the

DWYER 9781781001288 Part I.indd 239 25/07/2012 11:07

 240 Handbook of research methods in tourism

Bureau of Economic Analysis that calculates county level simple and total multipliers
as well as output, income and employment effects. Both models regionalize the national
IO model employing regional location earning quotients (BEA-RIMS II) or regional
purchase coefficients and value added-output ratios (IMPLAN). The REMI model,
first implemented by Treyz et al. (1980) and Treyz (1981), includes also an IO block that
forms the core of the model (Treyz et al., 1991), but differs from REMI and RIMS II
in an important respect: prices are included in the model and interact with quantities.
Bushnell and Hyle (1985) compared earlier versions of the three models and Rickman
and Schwer (1995) compared their relative performance.
These models have been applied to estimate all sorts of tourist effects in regional
and local economies. Here is an illustrative sample: state parks on state economies
(Bergstrom et al. 1990), a single resource in a local economy (Johnson and Moore, 1993),
tourism spending in Washington DC (Frechtling and Horvath, 1999) or Michigan state
(Stynes, 2000), the Atlanta Olympic Games in Georgia (Humphreys and Plummer,
1995), a festival in Ocean City (Crompton et al. (2001), a youth softball tournament in
(Daniels, 2004) in fairs on Washington state (WSDA 2007), etc. Daniels (2004) proposes
an interesting extension of IO, occupation-based modelling, a method to assign sector
labour incomes across different occupations employing the Bureau of Labor Statistics
wage-occupational data bank. In the absence of an IO table, the use of ready-made
models may be the only cheap alternative to evaluate minor tourism events. However,
the output that comes out of these models generates probably more value added in the
consultancy business than in the search of economic truth.

SAM Models

SAM models have also been employed to study tourism impacts in national, regional
and small economies during the last two decades. Since expenditure coefficients are ratios
of SAM flows to column totals, the only requirements to estimate tourism impacts with
a SAM model are a SAM of the economy studied and a vector of tourists impacts. The
fact that many national statistical offices in developed economies publish IO tables along
with national accounts but only a few of them elaborate SAMs has hampered its use in
tourism studies. For many developing countries, SAMs have been assembled to explore
the links between growth, inequality and employment, and ... how the extent of poverty
and changes in it are related to familiar issues of savings and investment, balance of pay-
ments, production and distribution (Pyatt and Thorbecke, 1976). SAMs constructed for
those purposes can nevertheless be employed to quantify the role tourism in the economy
and its impact on all endogenous accounts including households.

National, Regional or Local SAM-based Studies

An early application is Wests (1993) analysis of tourism in Queensland (Australia)

which combines a regional SAM with econometric time series analysis. The Queensland
integrated model includes, along with the SAM, other blocks where labour equations,
income relationships and household expenditure by commodity and sector are calculated
to reduce the static limitations of the model and, to some extent, the linearity relations.
The SAM included the following accounts: ten sectors, two factors (labour and capital),

DWYER 9781781001288 Part I.indd 240 25/07/2012 11:07

 Inputoutput SAM models 241

three institutions (households, enterprises and Government), savingsinvestment and

the foreign sector. The basic information for activities was drawn from the 198586
Queensland IO table while visitors expenditure, subdivided in intrastate, interstate (the
largest) and overseas, was estimated with the information provided by the Queensland
Tourist and Travel Corporation. Tourism impacts for the three tourists types were cal-
culated by comparing the values of various economic indicators derived by the model
with and without tourism expenditure, assuming that activities, factors and institutions
are all endogenous. The following average multipliers: 0.70 for GDP; 0.53 for household
income and 26.48 jobs per million AU$.
Wagner (1997) developed a SAM for Guaraqueaba to study the impact of tourism
and its effects on households. The SAM comprises four types of accounts: eight activi-
ties (rudimentary rural farms, rural entrepreneurs, construction, manufacturing, com-
merce, service, transportation and government); three factors (labour:households,
labour:salaried, and capital rent) and indirect taxes; nine institutions (subsistence, low
income, medium income and high income households; enterprises, the capital account
and three accounts for local, state and federal governments); and the importexport
account. A peculiarity of the Guaraqueaba SAM is that the base year is 198994 instead
of a single year, a decision Wagner justified by the singularity of the economy and data
availability.
The basic source for activities transactions patterns were the 1980 IO table and the
1975 SAM for Brazil. Sector totals were calculated from a 1989 Government report on
value added by sector for municipalities. The distribution of capital rents and consump-
tion expenditures relied on the 1992 budget research study conducted by the Brazilian
Government. Consumption expenditures were then valued at production prices employ-
ing distribution and transportation margins from the 1980 IO table. As for labour,
official statistics only provided information on workers holding formal contracts but no
information was available for labourers in the informal economy that hold two or three
different jobs in a year. Number of labourers with and without formal contracts and
their annual salaries were obtained from structured interviews. Population data were
obtained from official sources and it was estimated that 95 per cent of the working-age
population worked without a formal contract. Data on the local Government receipts
and expenditures came from municipal sources. Notice that although national sources
were employed in many instances to obtain patterns, an effort was made to include in the
SAM the local information drawn from structured interviews with experts. The SAM
was balanced by adjusting the importexport entries, but taking into account that activi-
ties and final demand are mainly supplied with imports, and exports come exclusively
from the rural entrepreneurs and manufacturing sectors.
Multipliers for the endogenous accounts were obtained in three scenarios: first, only
productive accounts are endogenous; second, productive, labour accounts and house-
holds are endogenous; and finally, productive accounts, labour, capital payments,
households and enterprises are endogenous. In the first case only intergroup effects are
captured while in the other cases all intragroup (own), intergroup (closed loop) and
extragroup (open loop) effects are captured. As expected, activity multipliers increase
as more accounts are endogenous. The largest multiplier of private activities is that
of rudimentary rural farms in all scenarios; rural: entrepreneurs and commerce and
manufacturing are also large when factors income, households and the enterprise are

DWYER 9781781001288 Part I.indd 241 25/07/2012 11:07

 242 Handbook of research methods in tourism

endogenous. Labour:households and household:subsistence multipliers are the largest in

the factor income and institutions blocks, respectively.
Employing survey data on the number of visitors and average expenditure, Wagner
performed the following experiment: what will be the effect of the arrival of 100 tourists
who spent on average $15.15 per day? Output impacts are 1563, 2097 and 3261 dollars
and employment impacts are 0.28, 0.31 and 0.43 (full time equivalent) jobs in the three
scenarios. Assuming 7.500 tourist average arrivals every day to Guaraqueaba, the
total impact on output and jobs is obtained multiplying the previous figures by 75 is
$244.575 worth of goods and services, $19 425 salaries and 32 full time equivalent jobs.
Sensitivity analysis indicated that the results were reasonably robust to measuring errors.
Wagner concluded that although tourism impact was non-negligible, it will do nothing
to break the cycle of deficit spending by local households and advised strengthen the
linkages between rudimentary rural farmers and subsystem households and the rest of
the economy.
Polo and Valle (2007 and 2008a) constructed a 72 72 square SAM for the Balearic
Islands. It included 54 sectors in the IO table, two factors of production, several auxil-
iary accounts for taxes, subsidies and transfers, a resident household and a non-resident
household, a corporation sector, the local and general governments, the savingsinvest-
ment account and one foreign sector that includes the rest of Spain and the rest of
the world. Fourteen sectors that produce tourist services are included in the SAM.
Unfortunately, the SAM only includes one resident and one non-resident household and
does not distinguish commodities by origin.
Polo and Valle (2007) compared the effects of exogenous injections in three scenarios:
one, only the 54 production accounts are endogenous (SAM: 54); two, factors accounts
and the resident household account are also endogenous (SAM: 54 1 F 1 H); and
finally, the savingsinvestment account is also endogenous (SAM: 54 1 F 1 H 1 I).
They also compared the effects of a 10 per cent fall in non-residents consumption in the
three models. In the first case, weighted average production falls 3.21 per cent, value
added 3.61 per cent and employment 3.02 per cent. The falls when the factors, resident
household and savingsinvestment accounts are also endogenous are much larger: 5.52,
6.0 and 5.5 per cent, respectively. Polo and Valle (2008a and 2009) compared these
results with those obtained with IO models specified with domestic coefficients.
Polo et al. (2006 and 2008) also analysed the impact on employment and added value
of a hypothetical change in the expenditure distribution of the non-resident consumer
using both the IO and SAM models. In particular, they calculate the increase in the
demand of 45 star hotels services needed to offset a decline in the demand in the 123
star hotels category so that neither employment nor the added value would be affected.
Their results indicate that in order to keep value added constant a reduction of 1000 beds
in 123 stars hotel segment requires 505 (499) extra beds in the 45 stars category using
the IO (SAM) model, and including the complementary offer, these figures increase to
568 (562) extra beds in the 45 stars category. This information may be of interest for
those who view in a replacement policy a way to alleviate congestion and environment
degradation in a mass tourism destination.
Jones (2010) estimates the impact of tourism in Mozambique using a tourism-
focussed SAM. The paper only includes a highly aggregated standard and tourism-
focussed SAMs but more details can be found in Jones (2007). The standard SAM

DWYER 9781781001288 Part I.indd 242 25/07/2012 11:07

 Inputoutput SAM models 243

includes 17 production activities, 20 commodities including five characteristic products

(hotels, restaurants, land and air transport and travel agencies), four factors (no list pro-
vided) and 15 institutions including five taxes (although only four appear in the aggre-
gated sum). The institutions apparently include: firms, 5 households classified by income
quintiles, the government, investment and the rest of world.22 The tourism-focused
SAM includes some auxiliary accounts for domestic tourists (household, firms and
government, and investment) and foreign tourists (business, self-drive and other leisure
type) that appear twice in the activities and the commodity blocks. Foreign tourists in the
commodity block receive income from the rest of the world and give it to foreign tourists
in the activity block that use it to finance the purchase of commodities. In turn, domestic
tourists in the commodity block receive income from activities, the household and the
investment accounts and pass it to domestic tourists in the commodity block that use it
to buy commodities. Notice that the extraction of intermediate tourism from intermedi-
ate consumption distorts the production coefficients.
In the reported results, all accounts except investment and the rest of the world are
considered endogenous. As indicated earlier, the element mji in the multiplier indicates
the income accruing to the jth account after a unitary exogenous injection directed to
the ith account and the column sum the aggregate income caused by the injection. The
normalized column sums (column sums divided by the average value of all column sums)
shows that the injections directed to all types of foreign visitors have above average
impact whether one looks to all accounts, production accounts, household income
or value added. Domestic tourism impacts are slightly over the average but for two
accounts, firms and government and investment tourism, are below the average. Only
government revenue multipliers are larger than average and larger than foreign tourists
multipliers. Although these results indicate that tourism is a sector with comparatively
strong backward linkages across production, household income and value added, Jones
calls the attention to two weak points: the fragility of backward linkages of domestic
tourism and the shortage of scarce production factors.23

Small Areas Ready-made SAM-based Studies

Ready-made IMPLAN-based SAMs have also been applied to estimate tourism impacts
in small areas of the United States. Croes and Severt (2007) evaluate the effects of
tourism in Osceola County (Florida) and find the following average multipliers with
endogenous households: 1.21 for output, 0.66 for income, 0.38 for labour income, and
17.58 jobs per million dollars spent. Mansury and Hara (2007) propose the introduction
of organic food agritourism at Liberty in Sullivan County (New York) and propose
the convenience of implementing a successful campaign to promote organic agricul-
ture that has stronger linkages with the local economy. To simulate its effects, they
increase (reduce) the local (imported) coefficients of tourists sectors and conclude that
the change not only is expected to deliver higher production output, but also generates
a more egalitarian distribution of income. Hara and Naipaul (2008) analyse the effects
of a 50 million dollar injection in tourism expenditure in four counties in the vicinity
of Orlando (Florida). The estimated injection assumes that 1 per cent of all visitors to
central Florida would spend two days in the area enjoying agritourism activities and
spent two-thirds of their average expenditure. They calculate an output multiplier with

DWYER 9781781001288 Part I.indd 243 25/07/2012 11:07

 244 Handbook of research methods in tourism

endogenous households of 1.68 and 25.76 jobs created per million dollars; nothing is
said, however, on the investment required to develop agritourism activities. Daniels et al.
(2004) studied the Cooper River Bridge Run a sport event (a single day annual road race)
in Charleston. They compared SAM results with those obtained when labour income is
distributed employing the BLS occupation-based model for three levels of disaggrega-
tion. Their results indicate that the effects on households are sensitive to the specification
of the model and that using SAMs to estimate personal income effects across different
households may be inappropriate.

ADVANTAGES AND LIMITATIONS OF THE IO AND

SAM TECHNIQUES CONCEPTUALLY AND FOR POLICY
FORMULATION

It seems fair to say that the advantages and limitations of IO and SAM models are well
known by now. In this section, we first examine the relative merits and shortcomings of
IO models versus SAM models. Then, we address an issue that has attracted consider-
able attention in the last 15 years: the criticism of IO and SAM models spurred by advo-
cates of computable general equilibrium (CGE) models. They are in part responsible
for having given way to more sophisticated nonlinear models known as CGE or AGE
models. First, we compare the relative merits of IO and SAM models. Then, we address
an issue that has received considerable attention since IO models were applied to assess
economic impacts. This section ends with some general considerations on the interest of
this type of analysis.

IO Versus SAM Models

The main advantage of IO models is that all parameters involved (A, l, k) can be specified
numerically with the sole aid of an IO table, and IO tables are often provided by many
national (or even regional) statistical offices along with national (regional) accounts. The
expenditure coefficients (Amm and Amn) in a SAM model can also be specified having a
SAM, but SAMs require additional information to distribute sectors gross value added
among institutions (households, corporations, government and foreign sectors) and
determine their outlays (consumption and savings decisions). Unfortunately, only a few
national statistical offices that compile IO tables also assemble SAMs.
On the other hand, SAM models are conceptually more satisfactory than IO models
since the households and capital accounts are balanced. Although IO models can be
expanded to include consumption and investment industries in the model, the produc-
tion coefficients required by the other sectors of the added industries are rather arbi-
trary. It is rather more natural to make endogenous those activities when their accounts
are balanced. Therefore, one may say that all things equal SAM models are preferable
to IO models.
The all things equal condition is not satisfied when important details are lost in the
process of obtaining a SAM. The IO framework usually distinguishes commodities by
origin in both supply and use tables, a distinction crucial to assess economic impacts,
but few SAMs incorporate this feature. In that case, the results of both models are

DWYER 9781781001288 Part I.indd 244 25/07/2012 11:07

 Inputoutput SAM models 245

not comparable (even when the endogenous accounts in the SAM models are just
the traditional IO sectors) and IO simulations done with domestic coefficients may
give more accurate results than SAM simulations performed with total expenditure
coefficients. The comparison in this respect is not always favourable to IO models,
however. In particular, value-added estimates obtained with sectoral value added coef-
ficients calculated from an IO table may be too optimistic since wages and especially
capital rents may go to foreigners, leakages that are taken into account in the SAM
framework.

LIMITATIONS OF IO AND SAM MODELS

Since Leontief proposed his system of linear demand equations to analyse the interde-
pendencies in production, the fixed coefficients assumption has been considered one
of the greatest limitations of IO modelling. The proportionality between inputs and
outputs is a very stringent assumption in the short-run and even more so in the long-run.
Production coefficients do vary as new production processes brought about by changes
in relative prices and the introduction of technological innovations. It is unlikely that
production coefficients of industries will remain constant as energy prices soar, real
wages increase and new vintages of capital goods are installed. The proportionality
assumption is even less tenable when applied to factors, households, savingsinvestment
and government industries in extended IO and SAM models. Moreover, it is also ques-
tionable the use of tax coefficients, defined as ratios of tax revenues over total produc-
tion or outlays, to calculate tax revenues in these models.
Little can be said in favour of an assumption that is obviously restrictive beyond
pointing out that changes in the production coefficients of an entire industry are proba-
bly piecemeal. Thus, one can take the observed technical coefficients A (pt, t) , l (wt, rt, t)
and k (wt, rt, t) derived from an IO table at the base year t, as proxies of the true technical
coefficients A (pt1s, t 1 s) , l (wt1s, rt1s, t 1 s) and k (wt1s, rt1s, t 1 s) s years later. Thus,
the results of a simulation done with the observed coefficients to evaluate the effects of a
change in exogenous income, Dd, should be read as the impact that would have occurred
at time t had the prices remained unchanged after the impact. In other words, the rel-
evance of the simulation results increases the closer the present time is to the base year
and the smaller are the price changes caused by the impact itself. From this perspective,
the results of IO and SAM models can be viewed as a particular case of more general
models and it is not a coincidence that the SAM framework is also known as fix-price
multiplier analysis.24 At any rate, simulation results from IO and SAM models should
always be interpreted with caution.
Disregard for capacity restrictions and resource limitations have also been con-
sidered a serious limitation of IO and SAM models.25 In response to positive exog-
enous shocks, IO and SAM models assume that supplies of labour and capital can be
increased at will to satisfy their requirements as industries expand their outputs to clear
product markets.26 Where do they come from the extra supplies needed? People already
employed may work more hours and unemployed, inactive persons and immigrants
may join the labour force. Additional capital services can be obtained by increasing
utilization rates or installing new capital. Probably both things take time and the results

DWYER 9781781001288 Part I.indd 245 25/07/2012 11:07

 246 Handbook of research methods in tourism

obtained with IO and SAM models probably overstate the effects of positive injections
in the short-run. But as more workers and capital are added to the initial endowments,
IO and SAM results may not be so far off the target. Indeed, tourism has become a
key industry inmany regions and nations since 1960 and factor restrictions have not
deterred its growth. 27
What about the effects of negative exogenous shocks? In this case, the multiplier works
in the opposite direction: labour and capital demands fall with output, unemployment
rises and capacity utilization falls. It is true that in more general models, market clear-
ing conditions for labour and capital services ensure labour and capital endowments are
reallocated among sectors at the new equilibrium prices. But how long does it take to
reach the new equilibrium? We guess, a great deal. A fall in tourist demand will probably
result in unemployment and capacity underutilization for quite a while, as IO and SAM
models suggest, and it is not obvious how capital embodied in hotels and apartments to
service tourists might be cast into others types of capital useful in other industries. Polo
and Valle (2008b) show that IO and SAM models results are close to those obtained
with a general equilibrium model with a Keynesian closure. In our opinion, they do a
better job than standard neoclassical models in predicting short-run outcomes after a
negative demand shock.28
Going beyond impact analysis, Burgan and Mules (1992) pointed out that IO and
SAM models are not evaluative tools. A tourism impact simulation only quantifies the
effects on value added or employment of the injection, not the costs caused by it, and a
balanced approach to tourism requires weighting benefits and costs (Burgan and Mules,
2001; Dwyer and Forsyth, 1993; and Dwyer et al., 2011). Tourism may increase conges-
tion, raise land and real estate prices, damage the environment in different ways, increase
the (cost) of providing government services, alter the income distribution, etc., that have
been ignored in most studies. In our opinion, there is no objection to do so as long as the
results obtained with the models are interpreted as what they are: gross estimates of the
effects of tourism.

FURTHER DEVELOPMENTS AND APPLICATIONS OF THE

TECHNIQUE IN TOURISM RESEARCH

As indicated in the previous section, IO and SAM modelling is a mature applied field and
most users are well aware of the limitations of linear models. Some of those are intrinsic
to them and little can be done about it. But yet a great deal can be gained if only an effort
is made to improve the quality and timing of the databases employed in these studies and
the way the results are presented in academic journals.

Quest for Better Databases and Reporting of Results

The implementation of IO and SAM models require the existence of IO tables and SAMs
and national statistical offices that elaborate IO tables should take into consideration
analytical needs and shorten compilation periods. On this respect, what Stone wrote in
the 1961 report that played a crucial role in the gestation of the 1968 SNA has not lost
its relevance:

DWYER 9781781001288 Part I.indd 246 25/07/2012 11:07

 Inputoutput SAM models 247

These reports are intended to provide guidance on various conceptual issues and also to set a
system of definitions and classifications of general applicability, to indicate detail that is gener-
ally desirable for analytical purposes and to provide a framework for assembling the data of
various countries on a comparable basis. (Our emphasis.)

In our opinion, the link between the design of social accounting systems and its poten-
tial applications has become more tenuous with the passage of time. Offices in charge
of national accounts seem to be more concerned in developing IO tables and SAMs to
present the definitive National Accounts estimates than in providing the detail generally
desirable for analytical purposes. To put it shortly: IO tables and SAMs are not ends in
themselves but instruments to model the economy.
The numerical specification of a model is a crucial stage and should be done with the
best possible database. Archer (1978) recalled many years ago that the old adage of
rubbish in, rubbish out applies to multiplier work as much as other forms of analysis.
IO and SAM modelling require counting with good IO tables and SAMs. Although great
progress has been made in the last 50 years and statistical offices are aware that it is a
suitable instrument to ... guide the use of information in economic analysis, they seem
to value more its role as a framework to improve the basic statistics and evaluate the
validity of the first estimates of the production account (INE, 1985, pp. 9798). Perhaps,
for this reason, the information published quite often does not fit modellers needs. In
particular, those countries and regions where tourism is an important industry should
include all characteristic sectors and products in their IO framework.
Since IO tables have considerable delay, IO models (and all models based on IO
tables like SAM and CGE models) are condemned to be specified with outdated data,
as critics often point out.29 An 8 or 10 years lag is not infrequent as inspection of Table
12.1 confirms. Production coefficients do change and it is crucial to count with reliable
and updated databases to increase the confidence in the simulation results. Publication
lags should and can be shortened if technology and human resources are set at the task.
It may be a bit expensive but it is probably more costly to keep dozens and dozens of
economists working with outdated and non-reliable databases. Better data will not solve
all problems, but it will eliminate one source of uncertainty.30
In developed countries, statistical offices responsible for the elaboration of National
Accounts should also engage in the elaboration of SAMs. Individual scholars often have
to build SAMs making heroic assumptions to distribute factors income among differ-
ent agents. Statistical offices count with a great deal information (often undisclosed)
and could do a much better job.31 The European initiative to develop a SAM handbook
(Eurostat, 2003) deserves applause and those countries that decided not to participate in
the endeavour should reconsider their position. It would be nave to think that an official
SAM developed within the national accounting system will satisfy the needs of all poten-
tial users, but it would be a much better starting point, especially if modellers needs are
taken into account.
The IO model is well known but in applied work details matter and should be made
explicit by the authors. The main characteristics of the IO table or SAM, the data
on tourism expenditures and the adjustments made to obtain the vectors of tour-
ists expenditures should be clearly indicated. Academic articles should also include a
precise specification of the model to avoid unnecessary confusion and facilitate results

DWYER 9781781001288 Part I.indd 247 25/07/2012 11:07

 248 Handbook of research methods in tourism

comparisons.32 The implementation of the model deserves much more than a few cursory
remarks and the equations employed in the simulations should be presented. In this
review of the literature, we have found too many papers that are equation free (or almost
free) that included lots of irrelevant statistics and details about the country and tourism
trends. It is a bad practice that should be abandoned. Lack of space or the need to reach
a larger audience should no longer be an excuse in academic journals.

TSAs: The Promised Land?

The development of a tourism satellite account (TSA) has received a great deal of atten-
tion in the last decade (Libreros, et al., 2006). The publication of the United Nations
(2008) monograph Tourism Satellite Account: Recommended Methodological Framework
is the latest attempt to set up homogenous criteria to measure the contribution of tourism
activities.33 In essence, the TSA aims at providing a comprehensive and coherent account
of tourism from the demand side (visitors expenditure and consumption) and the supply
side (production of tourists services and employment). It also includes other accounts
to measure the contribution of tourism to capital formation, collective consumption
and nonmonetary indicators. Although the construction of the ten interrelated tables
that comprise the TSA has an interest of its own as an extension of the SNA framework
(Frechtling, 2010), the relevant issue from the viewpoint of IO, SAM and even CGE
modeling is whether or not the information actually provided facilitates the inclusion of
tourists households in the models.34
For Van de Steeg and Steenge (2008), tourism is implicitly already included in the
supply and use tables and ... the TSA is developed to make tourism explicit. For them,
a most important contribution of the TSA framework is the construction of a final
expenditure vector specific of tourism. Blake et al. (2001) recognize that TSAs represent
a major step forward in the measurement of the economic size of tourism and provide
detailed data on tourism activities that are not otherwise available in national accounts,
but also point out that TSA do not assess the total impact of tourism. Smeral (2006)
insists that it is necessary to adjust the TSA results for indirect effects and intermedi-
ate consumption. Jones and Munday (2008) also recognize that the TSA framework
offers complementary opportunities to improve the modeling of tourism activity and
policy, but they argue that it requires restructuring to become more useful for mod-
eling purposes. In other words, modelling remains essential to assess the importance of
tourism and TSAs may not provide the most suitable information to analyse tourism
with models.
Let us take the case of Spain a major international tourist destination to illustrate
some of the difficulties we encounter to combine the TSA and the IO frameworks. The
symmetric table provides a consumption vector by product in the territory and the
use table an estimate of total non-residents consumption in the territory (NRCT). To
simulate the impact of NRTC all is needed is to allocate that total among the different
products. Does the Spanish TSA do the job? The answer is ... definitely not. Although
TSAs are supposed to be a satellite of the SNA and the IO framework there are sub-
stantial differences between the numbers reported by both.35 Thus, the availability of
a TSA does not solve the allocation problem of NRCT.36 Would it not be less costly
and more interesting to focus the attention in improving the measurement of NRCT

DWYER 9781781001288 Part I.indd 248 25/07/2012 11:07

 Inputoutput SAM models 249

(distinguishing commodities by origin) and disaggregating it into several vectors (corre-

sponding to distinct tourists types) that would be displayed in the final demand matrix
of the IO framework? It seems that the elaboration of the TSA has become (at least in
some countries) an end in itself and overlooks the detail that is generally desirable for
analytical purpose.37

Energy and Environmental Impacts of Tourism

The notion that services energy direct requirements are generally smaller than in other
sectors (agriculture, manufacturing and construction) is probably behind the lack of
interest in estimating energy intensity and consumption and CO2 emissions associated
with tourism. However, the picture changes considerably when indirect energy require-
ments are taken into account (see Alcntara and Padilla, 2007; Cardenete et al., 2011).
In the case of tourism, the very concept of visitors involves the use energy-consuming
services such as transportation and accommodation.
Tabatchnaia-Tamirisa et al.s (1997) study of Hawaii used and IO model to estimate
direct and indirect energy use by domestic and foreign tourists in Hawaii. Energy inten-
sities of any tourism activity are given by the rows of the matrix Mr (T 3E), defined as the
transposed of the submatrix of M obtained by the intersection of the E energy rows and
the T tourism columns of the multiplier matrix M. The intermediate energy demand
vector associated with the production of the final tourist demand vector dT is given by

AEMdT

where AE is the E 3 M submatrix of direct energy coefficients. Finally, the total vector
of energy demand yTE is obtained by adding to the intermediate the final energy demand
vector dTE

yTE 5 AEMdT 1 dTE .

Tabatchnaia-Tamirisa et al. reported estimates of direct and indirect energy use by

domestic and foreign tourists. Their calculations indicate that indirect energy use is very
important both in absolute terms and as a ratio of visitors expenditure, especially for
foreign tourists.38
Briassoulis (1991) already signalled the assessment of the intangible social and envi-
ronmental impacts of tourism and linking them with the inputoutput model as an
important research topic. Among those intangible impacts, CO2 and more generally
greenhouse emissions have attracted much attention in the last decade. Having estimated
the energy consumption associated with tourism, the IO and SAM models can provide
estimates of, let us say, CO2 emissions by multiplying the energy vector yTE by a suitable
cr emissions vector.39 Lenzen (1998) applied the IO model to estimate energy intensities
and emissions of greenhouse gases (GHG) caused by final demand in Australia. Lenzen
et al. (2003) showed how the approach can also be used to assess other environmental
impacts (land disturbance, water use or emission of NOx and SO2) involved in the
construction of a second Sydney airport and their results indicate that total impacts
are considerably higher than the on-site impacts. For this reason, they suggest that

DWYER 9781781001288 Part I.indd 249 25/07/2012 11:07

 250 Handbook of research methods in tourism

inputoutput techniques could be incorporated as mandatory in environmental impact

assessment standards.40
Dwyer et al. (2007) and Lundie et al. (2009) have also employed IO analysis to provide
measures of environmental indicators for different tourist market segments of inbound
tourism in Australia. The direct onsite effects are estimated by the audit approach while
indirect effects are estimated with an IO model of Australia specified with a 199495 IO
table. The study distinguishes 29 fuels, 12 types of tourists, four environmental aspects
(primary energy consumption, water use, GHG emissions and land disturbance) and
calculates various measures of environmental yield (per visit, per visitor night and per
dollar spent). Their results indicate that markets associated with high daily expenditure
in Australia generate higher impacts to the environment compared with tourists who
spend less. UK repeaters and backpackers obtained the best scores in all environmental
accounts using per visitor night indicators.
Konan and Chan (2010) applied the IO model to estimate the petroleum use and GHG
emissions of tourism, Hawaiis key export industry. The model was specified with a 1997
IO table that included 131 sectors and seven fossil fuels.41 Out of 323 trillion BTU total of
primary energy consumption, direct requirements are very high in air transport (102.51),
electricity generation (99.89) and residents consumption (41.07). Visitors direct gaso-
line purchases are comparatively small (2.74) that apparently avails the view that visitors
GHG emissions are insignificant. However, visitors demand of transportation, accom-
modation services, restaurants, etc., imply a much larger indirect demand of fuels and
a greater responsibility for GHG emissions. Indeed, the IO analysis unveils that visi-
tors are responsible for 22 per cent of total emissions and emission rates per visitor are
between three and five times larger than those of residents depending on the type of fuel.
Collins et al. (2009) examined the relative merits of two approaches, environmental IO
modelling (EIO) and ecological footprint (EFP) analysis,42 to assess the environmental
impacts of major sport events and presented the results obtained for the 2004 Wales
Rally of Great Britain. Among the advantages of the EIO approach the authors men-
tioned the following: transparency and low cost, availability of IO tables and detailed
environmental satellite accounts, and the possibility of comparing economic benefits and
environmental costs resulting from the event. Despite the well-known limitations of IO
models and the ambiguity that surrounds EFP calculations, the two methods provide
two quantitative estimates that may help to compare event impacts across space and
time.43
Kytzia et al. (2011) employed an augmented IO model to simulate the effects of
changes in tourism patterns on land use efficiency indicators and employment at Davos
(Switzerland). The IO table was assembled combining the information drawn from
surveys that cover most sectors with an existing the IO table for an Austrian region.
The situation at the status quo (2002) was that accommodation in high and medium
class hotels and group accommodation are more efficient (in the sense they require
less ground floor area per overnight) than low class hotels, vacation rentals and second
homes. Then, they looked at the income generated by the different types and concluded
that tourists lodged in high class hotels and group accommodation establishments gener-
ate the largest income per unit of ground floor used per night. Finally, the authors simu-
lated three scenarios: reducing seasonality;44 increasing the number of high class tourists
and occupancy rates in high class hotels; and expanding accommodation capacity. The

DWYER 9781781001288 Part I.indd 250 25/07/2012 11:07

 Inputoutput SAM models 251

conclusions are not too surprising when they claim that an increase in the number of
tourists staying in high class hotels would increase land use efficiency. A serious draw-
back of this approach is that the efficiency indicator chosen (value added over ground
floor area per overnight) does not take into account land values. It is unclear to us why
value added per unit of ground floor should be preferred to the rate of return earned by
different types of accommodation services as an efficiency measure.
Berners-Lee et al. (2011) used the environmental IO model supplemented with direct
emissions estimates to estimate total emissions caused by small and medium size busi-
nesses in the tourism sector and evaluate the potential costs or savings from switching
activities. Their model that combines the top-down approach characteristic of IO
analysis with the process specific bottom-up Life-Cycle Analysis (LCA) is applied to the
characteristic tourism sectors: hotels and travel agencies. As the authors point out, LCA
suffers from subjectivity in the definition of the boundaries of supply chains that may
result in severe truncation errors (Lenzen, 2001).45 In turn, IO analysis lacks specificity
since industrys production coefficients are averages of different production processes.
Although the authors call it a hybrid model, the first 10 steps relate to the IO model and
provide a total emission matrix where each column indicates the emissions resulting from
producing one net unit of each product.46 Steps 11 and 12 are concerned with improving
the accuracy by incorporating elements of LCA. Since the nature of those adjustments
is not precisely stated, it is hard to evaluate the final result.47
All energy and environmental studies of tourism impact just reviewed employ the
basic IO model. An obvious extension would be to estimate those effects employing
extended IO and SAM models to capture induced effects. Moreover, more comprehen-
sive approaches like EFP and LCA suggest that the concept of intermediate consump-
tions should be expanded to include somehow the services of produced capital goods
and natural resource capital required to produce tourism services.48 Including invest-
ment coefficients along with intermediate coefficients may not be a satisfactory solution
but there is a need for a better integration of fixed capital (land, construction, natural
resources, etc.) in the model. Sustainability of tourism (especially in mature destinations)
will continue to be an important issue in the following decades and we expect IO and
SAM analysis will continue playing an important role side by side with more general
models and other complementary tools.

GENERAL APPRAISAL

Is it worthwhile to continue using IO and SAM models in tourism research?

Notwithstanding the limitations discussed, our answer is definitely yes. The models are
easy to specify numerically and provide quantitative estimates of the weight of tourism
in the economy, useful information on the interrelations between tourism characteristic
products and sectors with the rest of the economy, informed predictions of the magni-
tude of economic permanent impacts and an assessment of some environmental conse-
quences of tourism activities. What really matters is to count with a data basis (IO tables
and SAMs) assembled from basic statistical information on production, distribution
and expenditure activities. Ideally, it should be designed to be the basis of a wide range
of applications displaying a fine sectoral and commodity disaggregation. At least in

DWYER 9781781001288 Part I.indd 251 25/07/2012 11:07

 252 Handbook of research methods in tourism

those countries where tourism plays an important role, the final demand should include
vectors for representative domestic and foreign tourists. Moreover, tourism related
investment should also be singled out (Holtz-Eakin, 2001). A lot can be done in all those
areas to improve our models and their numerical specification.
It is true that linear models rule out substitution but the solution is by no means
to assume (as it often happens in nonlinear models) elasticity figures whose empirical
relevance for the economy studied has never been tested.49 It is probably true that IO
and SAM models overestimate the effects of positive impacts in the short run, but they
give good clues of what may happen in the medium run even for large economies after
labour and capacity constraints are adjusted; and their estimates in the short run can be
even more accurate than those obtained with more general nonlinear models when the
economy faces negative shocks and factor prices adjust slowly. It goes without saying
thattheir results should be taken as informed estimates and compared with those pro-
vided by more general models.50 Indeed, comparing the simulation results of IO-SAM
and more general models for different types of shocks and closure rules are issues that
deserve more attention in the future. What it seems undeniable is that IO and SAM
models have made a great contribution to understand the multiple ways tourism impacts
the economy and that they will probably continue playing an important role in the
future.

ACKNOWLEDGEMENTS

The authors gratefully acknowledged financial support from the Ministry of Education
of Spain, SEJ2007-61046/ECON. All errors are the responsibility of the authors.

NOTES

1. It should be noted that the industry covered not only agricultural, mineral and industrial sectors but also
services, household and the foreign sector.
2. Chenery and Watanabe (1958) indicated that at least 15 countries have now done serious inputoutput
work although not all of the results have been published.
3. The textbook by Hara (2008) covers the IO and SAM models.
4. Oosterhaven and Fan (2006) obtain l (k) multiplying the diagonal matrices L and K of factors coeffi-
cients, by the vector Md.
5. Observe that labour impacts are measured in the labour units chosen. To translate those effects into
number of jobs those figures need to be multiplied by the ratio of employees to labour units in the base
year.
6. Although is not as common, the investment industry can also be endogenized (Wolff, 1985). Henry and
Deane (1997) considered endogenous household consumption and the government. For an alternative
interpretation of M 1 1th row and column see Miller and Blair (2009).
7. There are expenditure coefficients that cannot be interpreted as technical coefficients at all. A conspicuous
case is the expenditure coefficient obtained when tax revenues are divided by total income.
8. They can be classified by origin (nonresidents and resident), nature of the trip (vacation, business, assist-
ance to a singular event, etc.), lodging characteristics (hotels, furnished houses, caravans, day-trippers,
etc.) and other socioeconomic considerations.
9. Dd 5 dt 1sPtt1s 2 dt where Ptt1s is a diagonal matrix whose ith element is pti/pt1si. An alternative is to
inflate the A matrix (Henry and Deane, 1997; Var and Quayson, 1985).
10. Prez-Viitez and Polo (2007) analysis of energy intensities in Catalonia is one of the few studies where

DWYER 9781781001288 Part I.indd 252 25/07/2012 11:07

 Inputoutput SAM models 253

an IO and SAM analysis is done with domestic and total coefficients and the conclusion reached is that
it matters.
11. Slightly different formulas were used to calculate the employment multiplier (Archer, 1976).
12. Vaughan has kept faithful to this approach (see Vaughan et al., 2000).
13. In Spain, for instance, the IO framework only provides the total consumption expenditures of non-
residents (residents) in (out of) the territory. The situation has improved since the National Statistical
Institute started publishing the Tourist Satellite Account in 2002 (INE, 2002).
14. The Institute of Tourist Studies (IET) is an organism dependent of the Ministry of Commerce and
Tourism.
15. There are many more articles that the interested reader can find in the references of those included here
that provide a fair representation of IO studies at different levels (nation, regions and small islands), geo-
graphical areas (Australia, Europe, Asia and Africa) and topics (general tourism and tourism events).
16. They differ in the aggregation of the IO table and the base year; the way the tourists vector was estimated
and the parameters specified; the income concept employed; (non-government income, value added,
GDP, GNP, etc.) and the characteristics of the tourism industry in each area.
17. Further results can be found in Polo and Valle (2007, 2008a, 2009) for 1997 and Polo and Valle (2011) for
1983, 1997 and 2004.
18. Since tourist resources and tourists activities attract visitors who may otherwise have gone elsewhere,
city officials strive to develop such resources to reinforce the economic base of their communities. In
the case of singular mega-events such as the Olympic Games, FIFA World Cup, World Exhibitions,
etc., they engage in long and costly contests to host them, although the benefits of mega-events exceeds
tourists expenditures due to the event and may have significant effects on the entire state where they are
held.
19. It is hard to find a festival, fair, race, game, mega-event, etc. whose impact has not been evaluated.
Crompton (2006) claims that most economic impact studies are commissioned to legitimize a political
position rather than to search for economic truth.
20. Guidelines have helped to clarify what should be included in the expenditure vector. However, not all is
clear cut. Tyrrell and Johnston (2001) indicate that if a visitor spends $15 on a meal and the reason to be
there is 97 per cent due to the activity only $14.75 should be included in meals entrance of the expenditure
vector. It is doubtful that those percentages can be taken seriously.
21. Saayman and Saayman (2006) estimate the impact of the Krueger National Park in Mpumalanga
province of South Africa with a national 1996 IO table and assume that provincial indirect effects are a
proportion sector dependent on the national indirect effects.
22. It is not totally clear how the SAM was assembled. Apparently, it is an update to 2003 of the SAM elabo-
rated by McCool et al. (2009) for 2001.
23. Jones (2010) defines in Appendix B (p. 696) two normalized adjusted value added multipliers defined as
the ratio of the average multipliers in a column by the scarce factor multiplier. Capital and skilled labour
are the two scarce factors considered. The normalized capital adjusted value added multipliers are slightly
over the average for foreign visitors but the normalized skilled labour adjusted multipliers well below the
average.
24. Notice, however, that this criticism also applies to CGE models that often assume Leontief production
functions for intermediate inputs.
25. For a recent statement of this view, see Dwyer et al. (2004). Dwyer et al. (2005) recognize, however, that
the inputoutput assumption of freely available resources is closer to the truth in the local case, because
labour and capital can flow to the area from other areas. Today, labour and capital mobility is high not
only in small areas.
26. In the tourism impact literature, impact restrictions were first introduced by Wanhill (1988).
27. This is by no means a novel view. McGregor et al. (1996) claimed that the IO system replicates the long-
run equilibria of a wide range of regional models, many of which do not operate as IO systems in the
short-run.
28. lvarez and Polo (2011) arrive at similar conclusions with a general equilibrium model of Spain.
29. Of course, IO tables and SAMs can be updated. But this poses another question: what confidence can we
have in updated tables if key sectoral information is not available when they are updated?
30. Another indication of the lack of attention to analytical needs of some national statistical offices is that
published IO tables do not include auxiliary price and tax tables required to analyse productivity changes
and evaluate the effects of fiscal policies.
31. It is unfortunate that some national statistical offices behave as if the statistical information collected with
citizens taxes was a private good.
32. One or two equations are sometimes included in appendixes but they are too general and therefore
provide no useful information. In other cases, they are simply not correct.

DWYER 9781781001288 Part I.indd 253 25/07/2012 11:07

 254 Handbook of research methods in tourism

33. A detailed account of TSAs inception can be found in Frechtling (1999, 2010) and Jones and Munday
(2008).
34. TSAs have also been constructed for regions (see Jones and Munday, 2008, 2010; Jones et al., 2003; Pham
et al., 2009). Canada and Australia have pioneered these undertakings.
35. Of course, there may be important differences across countries. The following comments apply only to
Spain and all figures are in million EUR and refer to 2005. First, the value of NRCT in the IO framework
(38.681) differs from the inbound tourism consumption figure in the TSA (42 217), a difference that may
be due to the inclusion of service exports (mainly transport fares) in the TSA (INE, 2002, p. 27). Second,
since total tourism demand (98 389.1) in the TSA is disaggregated into 13 characteristic products (totaling
80 609.9) and a single figure (17 779.2) for the rest of the economy, we cannot calculate individual product
shares. Third, for each of the characteristic products, let us say, lodging, total demand (27 554.2) includes
inbound (10 487.5), domestic (13 965.0), intermediate (3 039.5) and government (62.2) demand. Thus,
product shares for characteristic products are contaminated by the inclusion of intermediate and govern-
ment demand. Fourth, for some commodities (i.e., lodging) the sum of inbound (10 487.5) and domestic
(13 965) tourism consumption in the TSA is much greater than household lodging consumption expendi-
ture (11 829.9) in the symmetric table and even larger than production of the lodging branch (17 110.8).
The main explanation for this apparent inconsistency is that lodging household demand is masked as final
demand of other products (travel agencies) in the IO framework. For all these reasons, we can conclude
that the present format of the Spanish TSA does not provide the appropriate information to extract domes-
tic and inbound tourism from the households consumption expenditure vector in the IO framework.
36. Neither does it solve the problem of extracting domestic tourism expenditures from the household con-
sumption in the IO framework. Moreover, the lodging or even hotels characteristic product in the TSA
and IO tables is too heterogeneous (from 5-star hotels to bed and breakfast) to analyse tourism impacts,
a shortcoming that also affects other countries (Jones et al., 2003, pp. 27832785).
37. Another important shortcoming of the Spanish IO framework is the lack of information on the distribu-
tion of taxes (subsidies) on products across products both in the use and the symmetric tables. This defi-
ciency becomes especially relevant when the IO framework is embedded in a general equilibrium model
to analyse tax policies. The complementary tax tables have never been published and the symmetric table
is only available at basic prices.
38. Becken et al. (2001) estimated the direct energy consumption in the accommodation sector in New
Zealand but no indirect effects were calculated. They find that consumption patterns vary greatly among
different accommodation types and suggest disaggregating the accommodation sector and visitors by
types of accommodation.
39. These applications have been facilitated by the availability of environmental satellite accounts. Perch-
Nielsen et al. (2010) show using a NAMEA that emissions by the Swiss tourism characteristic sectors are
four times greater than average.
40. Lenzen (2003) also used the Leontief and Gosh models and Defourny and Thorbecke (1984) structural
path analysis to determine key sectors and important influence paths from an environmental perspective.
41. A more recent 2002 IO table was available but the authors could not use it because it did not include key
energy sectors such as petroleum refining and electricity generation.
42. The EFP provides an aggregate measure of resource consumption of a given population expressed in
units equivalent to an hectare with average productivity known as global hectare (gha). Gsling et al.
(2002) explain at length (pp. 201205) how they applied the EFP methodology to measure the environ-
mental impact of inbound tourism to the Seychelles. An obvious advantage of the EFP is that it tries to
take into account the footprint of built-up land in transportation, accommodation, etc.
43. Collins et al. (2009) discarded other alternatives such as environmental impact assessment, life-cycle
analysis, costbenefit analysis, etc., on the basis that the selected techniques are able to deal in various
degrees with the environmental consequences of travel outside the event areas, provide quantitative esti-
mates and are to some extent complementary.
44. There are discrepancies between the figures in Table 3 and the status quo column in Table 4 that are hard
to interpret. Moreover, curbing seasonality means equalizing summer and winter overnight stays to the
largest winter figure in the status quo column. This amounts to an increase in the summer season rather
than a reduction in seasonality.
45. In a recent paper, Filimonau et al. (2011) criticize both the EFP and EIO approaches and defend LCA as
a superior alternative. Their discussion of the literature indicates that out of the six papers that have used
the original LCA analysis, four combine it with the EIO model. They illustrate the approach estimating
CO2 emissions of a hypothetical weekend holiday trip from London to Poole, in our view a rather anec-
dotal contribution.
46. Many of those steps could have been avoided had the authors counted with a more recent and congruent
IO environmental data. In some cases, the adjustments made (adjusting emissions coefficients by the CPI
(step 10) the adjustments are very crude.

DWYER 9781781001288 Part I.indd 254 25/07/2012 11:07

 Inputoutput SAM models 255

47. It seems that the emissions caused by organization expenditures, intermediate consumption, are evalu-
ated using the multipliers derived from final demand.
48. Len et al. (2007) for instance claim that the optimal trajectory of tourist consumption increases when
the stock of natural resources is high and environmental attributes are preserved.
49. CGE models present a more complex structure that includes nested production and utility functions,
optimization behaviour and market-clearing conditions for all products and generally factors. In most
models, the nesting and production functions chosen at each level of the nest are to a large extent arbi-
trary, i.e., lack any empirical backing.
50. Fixed-price models are obviously inadequate to estimate the effects of fiscal policies that affect prices and
quantities. To analyse the impact of a change in VAT rates is preferable to employ a general equilibrium
model (Kehoe et al., 1988) than to feed in an IO model the quantity changes estimated with ad-hoc equa-
tions (Manente and Zanette, 2010).

REFERENCES

Alcntara, V. and E. Padilla (2007), Subsistemas Input-Output y contaminacin: Una aplicacin al Sector
Servicios y las Emisiones de CO2 en Espaa. II Jornadas Espaolas de Anlisis Input-Output Zaragoza:
Crecimiento, Demanda y Recursos naturales, Zaragoza: Libro de Comunicaciones.
lvarez, M.T. and C. Polo (2011), A general equilibrium assessment of external and domestic shocks in
Spain, in press.
Alward, G.S. and C.J. Palmer (1983), IMPLAN: an input-output analysis system for Forest Service planning,
in R. Seppala, C. Row and A. Morgan (eds), Forest Sector Models: Proceedings of the First North American
Conference on Forest Sector Modeling, pp. 131140. Williamsburg VA: USDA Forest Service.
Archer, B.H. (1976), The anatomy of a multiplier, Regional Studies, 10, 7177.
Archer, B.H. (1978), Domestic tourism as a development factor, Annals of Tourism Research, 5 (1),
126141.
Archer, B.H. (1982), The value of multiplier and their policy implications, Tourism Management, 3 (4),
236241.
Archer, B.H. (1984), Economic impact: misleading multiplier, Annals of Tourism Research, 12, 517518.
Archer, B.H. (1985), Tourism in Mauritius: an economic impact study with marketing implications, Tourism
Management, 6 (1), 5054.
Archer, B.H. (1995), Importance of tourism for the economy of Bermuda, Annals of Tourism Research, 22
(4), 918930.
Archer, B.H. and J. Fletcher (1996), The economic impact of tourism in the Seychelles, Annals of Tourism
Research, 23 (1), 3247.
Archer, B.H. and C.B. Owen (1971), Towards a tourist regional multiplier, Regional Studies, 5 (4), 289294.
Archer, B.H. and S. Wanhill (1980), Tourism in Bermuda: An Economic Evaluation, Hamilton, Bermuda:
Bermuda Department of Tourism.
Baster, J. (1980), Inputoutput analysis of tourism benefits. Lessons from Scotland, International Journal of
Tourism Management, 1 (2), 99108.
Becken, S., C. Frampton and D. Simmons (2001), Energy consumption patterns in the accommodation sector:
the New Zeland case, Ecological Economics, 39, 371386.
Bergstrom, J.C., H.K. Cordell, A.E. Watson and G.A. Ashley (1990), Economic impacts of state parks on
state economies in the South, Southern Journal of Agricultural Economics, 22 (2), 6977.
Berners-Lee, M., D.C. Howard, J. Moss, K. Kaivanto and W.A. Scott (2011), Greenhouse gas footprinting
for small businesses: the use of input-output data, Science of the Total Environment, 409, 883891.
Beynon, M., C. Jones and M. Munday (2009), The embeddedness of tourism-related activity: a regional analy-
sis of sectoral linkages, Urban Studies, 46, 21232141.
Blake, A., R. Durbarry, M.T. Sinclair and G. Sugiyarto (2001), Modelling tourism and travel using tourism
satellite accounts and tourism policy and forecasting models, unpublished manuscript, Nottingham.
Briassoulis, H. (1991), Methodological issues. tourism input-output analysis, Annals of Tourism Research, 19
(4), 700710.
Bulmer-Thomas, V. (1982), Input-output Analysis in Developing Countries: Sources, Methods, and Applications,
Chichester, NY: John Wiley.
Burgan, B. and T. Mules (1992), Economic impact of sporting events, Annals of Tourism Research, 19,
70010.
Burgan, B. and T. Mules (2001), Reconciling cost-benefit and economic impact assessment for event tourism,
Tourism Economics, 7 (4), 321330.

DWYER 9781781001288 Part I.indd 255 25/07/2012 11:07

 256 Handbook of research methods in tourism

Bushnell, R.C. and M. Hyle (1985), Computerized models for assessing the economic impact of recreation and
tourism, in D.B. Propst (compiler), Assessing the Economic Impact of Recreation and Tourism: Conference
and Workshop. Asheville, NC: Southeastern Forest Experiment Station.
Cardenete, A., P. Fuentes-Saguar and C. Polo (2011), CO2 decomposition in the Andalusian economy,
Journal of Industrial Ecology, in press.
Chenery, H.B. and T. Watanabe (1958), International comparisons of the structure of production,
Econometrica, 26 (4), 487521.
Cohen, S.I. (1989), Multiplier analysis in social accounting and input-output frameworks: evidence for several
countries, in R.E Miller, K.R. Polenske and A.Z. Rose (eds), Frontiers in InputOutput Analysis, Oxford:
Oxford University Press.
Collins, A., C. Jones and M. Munday (2009), Assessing the environmental impacts of mega sporting events:
two options?, Tourism Management, 30, 828837.
Cooper, M.J. and J.J. Pigram (1984), Tourism and the Australian economy, Tourism Management, 5 (1),
212.
Croes, R.R. and D.E. Severt (2007), Evaluating short-term tourism economic effects in confined economies:
conceptual and empirical considerations, Tourism Economics 13 (2), 289307.
Crompton, J.L. (2006), Economic impact studies: instruments for political shenanigans?, Journal of Travel
Research, 45 (1), 6782.
Crompton, J.L., S. Lee and T.J. Shuster (2001), A guide for undertaking economic impact studies: the
Springfest example, Journal of Travel Research, 40, 7987.
Daniels, M.J. (2004), Beyond input-output analysis: using occupation-based modelling to estimate wages
generated by a sport tourism event, Journal of Travel Research, 43 (1), 7582.
Daniels, M., W. Norman and M. Henry (2004), Estimating income effects of a sport tourism event, Annals of
Tourism Research, 31 (1), 180199.
Defourny, J. and E. Thorbecke (1984), Structural path analysis and multiplier decomposition within a social
accounting matrix framework, The Economic Journal, 94, 111136.
Dwyer, L. and P. Forsyth (1993), Assessing the benefits and costs of inbound tourism, Annals of Tourism
Research, 20, 751768.
Dwyer, L., P. Forsyth, L. Fredline, M. Deery, L. Jago and S. Lundie (2007), Yield measures for special-
interest Australian inbound tourism markets, Tourism Economics, 13 (3), 421440.
Dwyer, L., P. Forsyth and A. Papatheodorou (2011), Economics of tourism, Contemporary Tourism Reviews,
129.
Dwyer, L., P. Forsyth and R. Spurr (2004), Evaluating tourisms economic effects: new and old approaches,
Tourism Management, 25, 307317.
Dwyer, L., P. Forsyth and R. Spurr (2005), Estimating the impacts of special events on an economy, Journal
of Travel Research, 43, 351359.
Eurostat (2003), Handbook on Social Accounting Matrices and Labour Accounts, Leadership group SAM.
Luxembourg: Eurostat Secretariat Unit E3, European Commission.
Felsenstein, D. and D. Freeman (1998), Simulating the impacts of gambling in a tourist location: some evi-
dence from Israel, Journal of Travel Research, 37, 145155.
Filimonau, V., J.E. Dickinson, D. Robbins and M.V. Reddy (2011), A critical review of methods for tourism
climate change appraisal: life cycle assessment as a new approach, Journal of Sustainable Tourism, 19 (3),
301324.
Fletcher, J. (1989), Inputoutput analysis and tourism impact studies, Annals of Tourism Research, 16 (4),
514529.
Frechtling, D.C. (1999), The tourism satellite account: foundations, progress and issues, Tourism Management,
20, 163170.
Frechtling, D.C. (2010), The tourism satellite account. A primer, Annals of Tourism Research, 37 (1),
136153.
Frechtling, D.C. and E. Horvath (1999), Estimating the multiplier effects of tourism expenditures on a local
economy through a regional inputoutput model, Journal of Travel Research, 37, 324332.
Gelan, A. (2003), Local economic impacts. The British Open, Annals of Tourism Research, 30 (2), 406425.
Getz, D. (1989), Special events: defining the product, Tourism Management, 10 (2), 135137.
Gsling, S., C. Borgstrm Hansson, O. Hrstmeier and S. Saggel (2002), Ecological footprint analysis as a
tool to assess tourism sustainability, Ecological Economics, 43, 199211.
Hara, T. (2008), Quantitative Tourism Industry Analysis. Introduction to InputOutput, Social Accounting
Matrix Modelling and Tourism Satellite Accounts, Amsterdam: Elsevier.
Hara, T. and S. Naipaul (2008), Agritourism as a catalyst for improving the quality of the life in rural regions:
a study from a developed country, Journal of Quality Assurance in Hospitality and Tourism, 9 (1), 133.
Henderson, D.M. and R.L. Cousins (1975), The Economic Impact of Tourism: A Case Study in Greater Tayside.
Edinburgh: Scottish Tourist Board.

DWYER 9781781001288 Part I.indd 256 25/07/2012 11:07

 Inputoutput SAM models 257

Heng, T.M. and L. Low (1990), Economic impact of tourism in Singapore, Annals of Tourism Research, 17
(2), 246269.
Henry, E.W. and B. Deane (1997), The contribution of tourism to the economy of Ireland in 1990 and 1995,
Tourism Management, 18 (8), 535553.
Holtz-Eakin, D. (2001), Capital in a tourism satellite account, Tourism Economics, 7 (3), 215232.
Humphreys, J.M. and M.K. Plummer (1995), The Economic Impact on the State of Georgia of Hosting the 1996
Olympic Games, Georgia: Selig Center for Economic Growth.
IET (1997), Tabla intersectorial de la economa espaola 1992, Madrid: Instituto de Estudios Tursticos del
Ministerio de Comercio y Turismo.
INE (1985) Contabilidad Nacional de Espaa, Madrid: Instituto Nacional de Estadstica.
INE (2002), La cuenta satlite del turismo de Espaa. Metodologa y primeras estimaciones 19961999, Madrid:
Instituto Nacional de Estadstica.
Johnson, R.L. and E. Moore (1993), Tourism impact estimation, Annals of Tourism Research, 20, 279288.
Jones, C. and M. Munday (2008), Tourism satellite accounts and impact assessments: some considerations,
Tourism Analysis, 13, 5369.
Jones, C. and M. Munday (2010), Tourism satellite accounts for regions? A review of development issues and
an alternative, Economic Systems Research, 22 (4), 341358.
Jones, C., M. Munday and A. Roberts (2003), Regional tourism satellite accounts: a useful policy tool?,
Urban Studies, 40 (13), 27772794.
Jones, S. (2007), A economia de turismo em Moambique: tamanho, impacto, e implicaoes, DNEAP
Discussion Paper 55P, Direcao Nacional de Estudos e Anlise de Polticas, Ministrio de Planificaao e
Desenvolvimento, Repblica de Moambique.
Jones, S. (2010), The economic contribution of tourism in Mozambique: insights from a social accounting
matrix, Development Southern Africa, 27 (5), 679696.
Kehoe, T.J, A. Manresa, P. Noyola, C. Polo and F. Sancho (1988), A general equilibrium analysis of the 1986
tax reform in Spain, European Economic Review, 32 (23), 334342.
Keuning S.J. and W.A. de Ruijter (1988), Guidelines to the construction of a social accounting matrix, Review
of Income and Wealth, 34 (1), 71101.
Khan, H., C.F. Seng and W.K. Cheong (1990), Tourism multiplier effects on Singapore, Annals of Tourism
Research, 17, 408418.
Kim, S.S., K. Chon and K.Y Chung (2003), Convention industry in South Korea: an economic impact analy-
sis, Tourism Management, 24, 533541.
Konan, D. and H.L. Chan (2010), Greenhouse gas emisisions in Hawaii: household and visitor expenditure
analysis, Energy Economics, 32, 210219.
Kweka, J., O. Morrisey and A. Blake (2001), Is tourism a key sector in Tanzania? Input-output analysis of
income, output, employment and tax revenue, TTRI discussion paper 1001/1. Tourism and Travel Research
Institute, University of Nottingham, Nottingham, UK.
Kytzia, S., A. Walz and M. Wegmann (2011), How can tourism use land more efficiently? A model based
approach to land-use efficiency for tourist destinations, Tourism Management, 32, 629640.
Lee, C. and T. Taylor (2005), Critical reflections on the economic impact assessment of a mega-event: the case
of 2002 FIFA Worl Cup, Tourism Management, 26, 595603.
Leistritz, F.L., W. Ransom-Nelson, R.W. Rathge, R.C. Coon, R.A. Chase, T.A. Hertsgaard, S.H. Murdock,
N.E. Toman, R. Sharma and P.S. Yang. (1982), North Dakota Economic-Demographic Assessment Model
(NEDAM): Technical Description, Ag. Econ. Rpt. No. 158. Fargo: North Dakota State University.
Lenzen, M. (1998), Primary gases and greenhouse gases embodied in Australian final demand consumption:
an input-output analysis, Energy Policy, 26 (6), 265290.
Lenzen, M. (2001), A generalized input-output multiplier calculus for Australia, Economic System Research,
13 (1), 6692.
Lenzen, M. (2003), Environmentally important paths, linkages and key sectors in the Australian economy,
Structural Change and Economic Dynamics, 14, 134.
Lenzen, M., S.A. Murray, B. Corteand and C.J. Dey (2003), Environmental assessment including indirect
effects: a case study using inputoutput analysis, Environmental Assessment Review, 23, 263282.
Len, C.J., J.M. Hernndez and M. Gonzlez (2007), Economic welfare, the environment and the tourist
product lifecycle, Tourism Economics, 13 (4), 583602.
Leontief, W. (1936), Quantitative inputoutput relations in the economic system of the United States, Review
of Economics and Statistics, 18 (3), 105125.
Leontief, W. (1937), Interrelation of prices, output, savings and investment, Review of Economics and
Statistics, 19 (3), 109132.
Leontief, W. (1966), InputOutput Economics, New York: Oxford University Press.
Libreros, M., A. Massieu and S. Meis (2006), Progress in tourism satellite account implementation and devel-
opment, Journal of Travel Research, 45, 8391.

DWYER 9781781001288 Part I.indd 257 25/07/2012 11:07

 258 Handbook of research methods in tourism

Lin, T. and Y. Sung (1984), Tourism and economic diversification in Hong Kong, Annals of Tourism
Research, 11, 231247.
Liu, J. and T. Var (1982), Differential multipliers for the accommodation sector, Tourism Management, 3 (3),
177187.
Liu, J., T. Var and A. Timur (1984), Touristincome multipliers for Turkey, Tourism Management, 5 (4),
280287.
Lundie, S., L. Dwyer and P. Forsyth (2009), Environmentaleconomic measures of tourism yield, Journal of
Sustainable Tourism, 15 (5), 503519.
Manente, M. and M. Zanette (2010), Macroeconomic effects of a VAT reduction in the Italian hotels and
restaurants industry, Economic Systems Research, 22 (4), 407425.
Mansury, Y. and T. Hara (2007), Impacts of organic food agritourism on a small rural economy: a social
accounting matrix approach, The Journal of Regional and Policy Analysis, 37 (3), 213222.
Matheson, V.A. (2002), Upon further review: an examination of sporting event economic impact studies, The
Sport Journal, 5 (1).
Mazumder, M.NH., E. Musa Ahmed, Md. Wahid Murad and A. Quaem Al-Amin (2011), Identifying eco-
nomically inbound markets for Malaysian tourism, Journal of Vacation Marketing, 17 (1), 1731.
McCool, C., J. Thurlow and C. Arndt (2009), Documentation of social accounting matrix (SAM) develop-
ment in C. Arndt and F. Tarp (eds), Taxation in a Low Income Economy: The Case of Mozambique, London:
Routledge.
McGregor, P.G., J.K. Swales and Y.P. Yin (1996), A long-run interpretation of regional inputoutput analy-
sis, Journal of Regional Science 36 (3), 479501.
Mescon, T.S. and G.S. Vozikis (1985), The economic impact of tourism at the port of Miami, Annals of
Tourism Research, 12 (4), 515528.
Miernyk W.H. (1965), The Elements of InputOutput Analysis, New York: Random House.
Miller, R.E. and P.D. Blair (1985), InputOutput Analysis: Foundations and Extensions, Englewood Cliffs, NJ:
Prentice-Hall.
Miller, R.E. and P.D. Blair (2009), InputOutput Analysis: Foundations and Extensions, 2nd edition, Cambridge
New York: University Press.
Milne, S. (1987), Differentiation multipliers, Annals of Tourism Research, 14 (3), 498515.
Mules, T. (1999), Estimating the economic impact of an event on a local government area, region, state or
territory, in Valuing Tourism: Methods and Techniques. Occasional Paper, No. 28. Canberra: Bureau of
Tourism Research.
Mules, T. and Faulkner, B. (1996), An economic perspective on major events, Tourism Economics, 12 (2),
107117.
Oosterhaven, J. and T. Fan (2006), Impact of international tourism on the Chinese economy, International
Journal of Tourism Research, 8, 347354.
Perch-Nielsen, S., A. Sesartic and M. Stucki (2010), The greenhouse gas intensity of the tourism sector: the
case of Switzerland, Environmental Science and Policy, 13, 131140.
Prez-Viitez, H. and C. Polo (2007), Intensidades energticas en la economa de Catalua: un anlisis IO and
SAM, unpublished manuscript, Universidad Autnoma de Barcelona.
Pham, T.D., L. Dwyer and R. Spurr (2009), Constructing a regional tourism satellite account: the case of
Queensland, Tourism Analysis, 13, 445460.
Polo, C. and E. Valle (2002), Un anlisis inputoutput de la economa balear, Estadstica Espaola, 44 (151),
393444.
Polo, C. and E. Valle (2007), Un anlisis estructural de la economa balear, Estadstica Espaola, 49 (165),
227257.
Polo, C. and E. Valle (2008a), An assessment of the impact of tourism in the Balearic Islands, Tourism
Economics, 14 (3), 615630.
Polo, C. and E. Valle (2008b), A general equilibrium assessment of impact of a fall in tourism under alternative
closure rules: the case of the Balearic Islands, International Regional Science Review, 31 (1), 334.
Polo, C. and E. Valle (2009), Estimating tourism impacts using inputoutput and SAM model in the Balearic
Islands, in A. Matias, P. Nijkamp and M. Sarmiento (eds), Advances in Tourism Economics. New develop-
ments, Heildelberg: Pysica-Verlag, pp. 121143.
Polo, C. and E. Valle (2011), The weight of tourism in the Balearic Islands: 1983, 1987 and 2004, Estudios de
Economa Aplicada, 29 (3), 737754.
Polo, C., V. Ramos, J. Rey-Maqueira, M. Tugores and E. Valle (2006), Employment and added value effects
of upgrading hotels quality, Annals of Tourism Research, 33 (3), 574577.
Polo, C., V. Ramos, J. Rey-Maqueira, M. Tugores and E. Valle (2008), The potential effects of a change in the
distribution of tourism expenditure on employment, Tourism Economics, 14 (1), 709725.
Pratt, S. (2011), Economic linkages and impacts across the TALC, Annals of Tourism Research, in press.
Pyatt, G. (1988), A SAM approach to modeling, Journal of Policy Modeling, 10 (3), 327351.

DWYER 9781781001288 Part I.indd 258 25/07/2012 11:07

 Inputoutput SAM models 259

Pyatt, G. and J.I. Round (1977), Social accounting matrices for development planning, Review of Income and
Wealth, 23 (4), 239264.
Pyatt, G. and J.I. Round (1979), Accounting and fixed price multipliers in as social accounting matrix frame-
work, The Economic Journal, 89, 850873.
Pyatt, G. and E. Thorbecke (1976), Planning Techniques for a Better Future, Lausanne, Switzerland:
International Labour Office.
Pyatt, G., J. Bharier, R. Lindley, R. Mobro and Y. Sabolo (1972), A methodology for development planning
applied to Iran, mimeo, Warwick, England.
Pyatt, G., A.R. Roe, J.I. Round, R.M. Lindley and others (1977), Social Accounting for Development Planning,
Cambridge: Cambridge University Press.
Rasmussen, P.N. (1956), Studies in Intersectoral Relations, Amsterdam: North Holland.
Rickman D.S. and R.K. Schwer (1995), A comparison of the multipliers of IMPLAN, REMI, and RIMS II:
Benchmarking ready-made models for comparison, The Annals of Regional Science, 29 (4), 363374.
Ruz, A.L. (1985), Tourism and the economy of Puerto Rico: an inputoutput approach, Tourism
Management, 6 (1), 6165.
Saayman, M. and A. Saayman (2006), Estimating the economic contribution of visitor spending in the Kruger
National Park to the regional economy, Journal of Sustainable Tourism, 14 (1), 6781.
Smeral, E. (2006), Tourism satellite accounts: a critical assessment, Journal of Travel Research, 45, 9298.
SNA (1969), A System of National Accounts, Studies in Methods, Series F, No. 2 Rev. 3. New York: Statistical
Office of the United Nations.
Steenge, A. and A. Van de Steeg (2010), Tourism multipliers for a small Caribbean island state; the case of
Aruba, Economic Systems Research, 22 (4), 359384.
Stevens, B. and A. Rose (1985), Regional inputoutput methods for tourism impact analysis, in D.B. Propst
(ed.), Assessing the Economic Impacts of Recreation and Tourism, Asheville, NC: USDA Forest Service, pp.
1621.
Stone, R. (1961), InputOutput and National Accounts, Paris: Organisation for Economic Cooperation and
Development.
Stynes, D.J. (1999), Guidelines for measuring visitor spending, working paper, Michigan State University,
Department of Park, Recreation and Tourism Resources.
Stynes, D.J. (2000), Michigan Tourism Spending and Economic Impact Model (MITEIM), East Lansing, MI:
Department of Park, Recreation and Tourism Resources, Michigan State University.
Tabatchnaia-Tamirisa, N., M. Loke, P. Leung and K. Tucker (1997), Energy and tourism in Hawaii, Annals
of Tourism Research, 24 (2), 390401.
Thorbecke, E. (2000), The use of social accounting matrices in modelling, (revised version), Paper prepared
for the 26th General Conference of the International Association for Research in Income and Wealth.
Krakow, Poland.
Thorbecke, E. and J. Sengupta (1972), A consistency framework for employment output and income distribu-
tion projections applied to Colombia, World Bank, Washington.
Tohamy, S. and A. Swinscoe (2000), The economic impact of tourism in Egypt, working paper No. 40, The
Egyptian Centre for Economic Studies.
Treyz, G.I. (1981), Predicting the economic effects of state policy initiatives, Growth and Change, 12, 29.
Treyz, G.I., A.F. Friedlaender and B.H. Stevens (1980), The employment sector of a regional economic policy
simulation model, Review of Economics and Statistics, 62, 6373.
Treyz, G.I., D.S. Rickman and G. Shao (1991), The REMI economicdemographic forecasting and simula-
tion model, International Regional Science Review, 14 (3), 221253.
Tyrrell, T.J. and R.J. Johnston (2001), A framework for assessing direct economic impacts of tourist
events: distinguishing origins, destinations and causes of expenditures, Journal of Travel Research, 40,
94100.
United Nations Statistics Division, Statistical Office of the European Communities, Organisation for
Economic Co-operation and Development and World Tourism Organization (2008). 2008 Tourism Satellite
Account: Recommended Methodological Framework. Madrid: World Tourism Organization.
Van de Steeg, A.M. and A.E. Steenge (2008), Integrating the TSA and inputoutput methodology in tourism
impact studies, paper presented at the Intermediate InputOutput Meeting 2008, Sevilla, Spain, 810 July,
2008.
Vanhove, N. (1981), Tourism and employment, International Journal of Tourism Management, 2 (3), 162175.
Var, T. and J. Quayson (1985), The multiplier impact of tourism in the Okanagan, Annals of Tourism
Research, 12 (4), 497514.
Vaughan, D.R., H. Farr and R.W. Slee (2000), Estimating and interpreting the local economic benefits of
visitor spending: an explanation, Leisure Studies, 19 (2), 95118.
Wagner, J.E. (1997), Estimating the economic impacts of tourism, Annals of Tourism Research, 24 (3),
592608.

DWYER 9781781001288 Part I.indd 259 25/07/2012 11:07

 260 Handbook of research methods in tourism

Wanhill, R.C. (1988), Tourism multipliers under capacity constraint, The Service Industries Journal, 8 (2),
136142.
West, G.R. (1993), Economic significance of tourism in Queensland, Annals of Tourism Research, 20 (3),
490504.
West, G. and A. Gamage (2001), Macro-effects of tourism in Victoria, Australia: a nonlinear inputoutput
approach, Journal of Travel Research, 40 (1), 101109.
Wolff, E.N. (1985), Industrial composition, interindustry effect and the U.S. productivity slowdown, Review
of Economics and Statistics, LXVII (2), 268277.
WSDA (2007), Washington State Fairs and Other Fairground Activities Economic Impact Analysis, Washington
State: Northern Economics.

DWYER 9781781001288 Part I.indd 260 25/07/2012 11:07

 13 CGE modeling
Larry Dwyer and Tien Duc Pham

INTRODUCTION

There is now an extensive literature on evaluating the economic impacts of tourism. This
literature seeks to show how the impacts of changes in tourism expenditure, due perhaps
to improved destination marketing and promotion, development of special events or
external shocks such as terrorist attacks, can be evaluated in economic terms. An eco-
nomic impact analysis estimates the changes that take place in an economy due to some
existing or proposed project, action, event, or policy resulting in increased income and
expenditure for a range of different stakeholders, many of whom are not directly con-
nected with the tourism industry.
As typically employed in tourism research and policy analysis, economic impact anal-
yses trace the flows of spending associated with tourism activity in an economy through
business, households and government to identify the resulting changes in economic vari-
ables such as sales, output, government tax revenues, household income, value added,
and employment. A major objective of such estimates has been to inform policy makers
as to the appropriate allocation of resources both within the tourism sector itself and
between tourism and other industry sectors.
The effects of tourist expenditure on a tourism destination/region can be estimated
using economic models that identify and quantify the linkages between the different
sectors of the local economy and the linkages with other regions in the economy. Almost
every industry in the economy is affected to some extent by the indirect and induced
effects of the initial tourist expenditure. In CGE models, these flow-on effects determine
the impact of a shock on all industries including the tourism sector in the economy.
Important variables that represent the net impacts of shocks include Gross Domestic
Product (GDP) (or Gross Regional Product (GRP)), value added, factor incomes, and
employment. For any given injection of tourism expenditure, the increment to value-
added and employment in the region will vary according to several features of the
economy (Dwyer et al., 2010). These include:

the particular industries that are the recipients of the direct expenditure
strengths of the business linkages between tourism and other industry sectors
the structure of the model employed
the assumed factor constraints
the import content of consumer goods and inputs to production
the production and consumption relationships assumed
changes in the prices of inputs and outputs
changes in the exchange rate
the workings of the labor market, and
the government fiscal policy stance.

261

DWYER 9781781001288 Part I.indd 261 25/07/2012 11:07

 262 Handbook of research methods in tourism

CGE models represent world best practice in assessing economy-wide economic

impacts of changes in tourism expenditure. Proponents of CGE modeling for
tourismimpact analysis point out that the economy-wide, interactive effects must be
taken into account in determining the impacts of increased tourism expenditure on a
destination.

NATURE OF THE TECHNIQUE AND ITS EVOLUTION

The Structure of a CGE model

Computable general equilibrium (CGE) models combine the abstract Walrasian general
equilibrium structure, formalized by Arrow and Debreu (1954), Debreu (1959), and
Arrow and Hahn (1971), with realistic economic data to solve numerically for the levels
of supply, demand and price that support equilibrium across a specified set of markets.
CGE modeling involves a mathematical specification of simultaneous relationships
within the economy. Many features can be built into this framework, to tailor it to the
different conditions that characterize alternative real world circumstances. In this way,
the framework provides a widely encompassing means of evaluating the effects of policy
changes and exogenous shocks on resource allocation and can also permit assessment
of the distributional effects of such changes. CGE models treat an economy as a whole,
allowing for feedback effectsofone sector on another. They represent the economy as
a system of flows of goods and services between sectors (Bandara, 1991; Costa, 1998;
Shoven and Whalley, 1992).
CGE models consist of a set of equations characterizing the production, consumption,
trade, and government activities of the economy. There are four types of equations in a
typical CGE model that are solved simultaneously. These are (McDougall 1995):

Equilibrium conditions for each market ensure that supply is equal to demand for
each good, service, factor of production, and for foreign exchange. Assuming flex-
ible prices and wages, this enables factors of production, such as labor and capital,
imports, and exports to be modeled (although some sticky prices can be assumed
such as might occur in the labor market).
Incomeexpenditure identities ensure that the economic model is a closed system. All
earnings must be accounted for through expenditure or savings. These conditions
apply to all private households, the government, firms, and any other economic
agents that are modeled. These define various macroeconomic identities such as
aggregate employment and the components of gross domestic product.
Behavioral relationships state how economic agents (consumers, suppliers, investors,
and so on) acting in their own best interests can lead to changes in price and income
levels. For example, businesses will seek to maximize profits. Firms can retain a
share of their profits. Their financing can proceed through financial markets or
banks. Consumers will look for lowest prices for equivalent products. Households
can deduce their supply of labor from the maximization of their utility function.
The zero-pure-profits condition for production is assumed. Resource allocation is
via market forces where markets behave imperfectly unemployment may increase.

DWYER 9781781001288 Part I.indd 262 25/07/2012 11:07

 CGE modeling 263

The labor market can be segmented between industries, urban and rural areas, or
regions. Then, migration costs prevent perfect and instantaneous arbitrage between
the wages of workers with the same skill but working in different industries. Wages
can be bounded from under by a legal minimum wage. If it is above the equilib-
rium wages for some skills, the workers with these skills will partly be unemployed.
Increasing government expenditures are met either by raising taxes or borrowing,
with implications for the expenditure of other economic agents.
Production functions determine how much output is produced for any given level of
factor employment. With assumptions regarding market structure, these determine
what levels of labor employment, capital usage, and intermediate input usage are
required to satisfy a given level of output for a given set of prices. The production
assumptions allow substitution between intermediate inputs and factors of produc-
tion as prices and wages change. The production functions can include firm-specific
resources, which allows for decreasing returns to scale. Or they can include fixed
costs, which allows for increasing returns to scale.

A CGE model is a very versatile structure, which can easily be adapted to the
special features of any economy, to the questions that are asked, and to the data,
which are available. Its assumptions can easily be changed so that the results of a
simulation can be interpreted, for instance, as describing a short-term or long-run equi-
librium.

Model Database

Figure 13.1 is a schematic representation of the CGE cores inputoutput database,

revealing the basic structure of the CGE core. The exegesis herein is based on the study
by Peter et al. (1996). The model, known as MMRF divides the Australian economy
into eight regional economies representing the six States and two Territories. This model
has been used in various studies of the economic impact of tourism in Australia (Adams
and Parmenter, 1999; Dwyer et al., 2000, 2003).
The columns in Figure 13.1 identify the following agents:

domestic producers divided into J industries in Q regions;

investors divided into J industries in Q regions;
a single representative household for each of the Q regions;
an aggregate foreign purchaser of exports;
another demand category corresponding to Q regional governments; and
another demand category corresponding to Federal government demands in the
Q regions.

The rows show the structure of the purchases made by each of the agents identified in the
columns. Each of the I commodity types identified in the model can be obtained within
the region, from other regions, or imported from overseas. The source-specific com-
modities are used by industries as inputs to current production and capital formation,
are consumed by households and governments, and are exported. Only domestically

DWYER 9781781001288 Part I.indd 263 25/07/2012 11:07

 264 Handbook of research methods in tourism

ABSORPTION MATRIX

1 2 3 4 5 6
Producers Investors Household Export Regional Federal
Govt. Govt.

Size dJQS dJQS d Q S d 1 S d Q S d Q S

c
Basic
IS BAS1 BAS2 BAS3 BAS4 BAS5 BAS6
flows
T

c
Margins I S R MAR1 MAR2 MAR3 MAR4 MAR5 MAR6
T

c
Taxes IS TAX1 TAX2 TAX3 TAX4 TAX5 TAX6
T

c
Labour M LABR
T I = number of commodities

J = number of industries
c
Capital 1 CPTL
M = number of occupation types
T
Q = number of domestic regions
c
Land LAND R = number of commodities used as margins
1
T
S = 9:8 domestic regions plus 1 foreign import

c
Other
1 OCTS
Costs
T

Source: Peter et al. (1996).

Figure 13.1 The CGE core inputoutput database

produced goods appear in the export column. R of the domestically produced goods are
used as margin services (domestic trade and transport and communication) which are
required to transfer commodities from their sources to their users. Commodity taxes are
payable on the purchases. As well as intermediate inputs, current production requires
inputs of three categories of primary factors: labor (divided into M occupations), fixed
capital, and agricultural land. The other costs category covers various miscellaneous
industry expenses. Each cell in the inputoutput table contains the name of the corre-
sponding matrix of the values (in some base year) of flows of commodities, indirect taxes
or primary factors to a group of users. For example, MAR2 is a five-dimensional array

DWYER 9781781001288 Part I.indd 264 25/07/2012 11:07

 CGE modeling 265

showing the cost of the R margins services on the flows of I goods, both domestically and
imported (S), to I investors in Q regions.
The theoretical structure of MMRF includes: demand equations are required for type
of user; equations determining commodity and factor prices; market clearing equations;
definitions of commodity tax rates. The CGE core equations can be grouped according
to the following classification:

producers demands for produced inputs and primary factors;

demands for inputs to capital creation;
household demands;
export demands;
government demands;
demands for margins;
zero pure profits in production and distribution;
market-clearing conditions for commodities and primary factors;
indirect taxes;
regional and national macroeconomic variables and price indices.

Input Demands: Current Production

Figure 13.2 illustrates the structure of a production nest for an industry in a regional
economy. At the top level, the producer combines intermediate inputs, primary inputs,
and all other costs in the constant shares relationship, which is often referred to as the
Leontief technique in CGE modeling. For each intermediate input, the input can be
sourced from the domestic economy, or from overseas countries depending on which
source is relatively cheaper. This is governed by the constant elasticity of substitution in
the production nest between imports and domestic in Figure 13.2. In a similar approach,
the producer then chooses the cheapest region among all regions in the domestic
economy. This is represented at the bottom of the substitution level among region 1 to
region 8 in Figure 13.2.
The primary input bundle is also a combination of labor, capital, and land using the
CES functional form in order to minimize the cost of this bundle.

Investment Demands

Capital creators for each regional sector combine inputs to form units of capital. In
choosing these inputs they minimize costs subject to technologies similar to that in
Figure 13.2. Figure 13.3 shows the nesting structure for the production of new units of
fixed capital. Capital is assumed to be produced with inputs of domestically produced
and imported commodities. No primary factors are used directly as inputs to capital
formation. The use of primary factors in capital creation is recognized through inputs of
the construction commodity (service) (Peter et al., 1996, p. 13).

DWYER 9781781001288 Part I.indd 265 25/07/2012 11:07

 Output j

DWYER 9781781001288 Part I.indd 266

Leontief

Intermediate Intermediate Primary factors Other costs

inputs (i) inputs (n)

CES
CES

266
Capital Land Labour
Imports Domestic

CES CES

Occupation 1 Occupation 8
Region 1 Region 8

Source: Peter et al. (1996, Figure 2.2).

Figure 13.2 Production nest

25/07/2012 11:07
 CGE modeling 267

y(j)

Leontief

Intermediate Intermediate
inputs (i) inputs (n)

CES CES

Imports Domestic Imports Domestic

CES CES

Region 1 Region 8 Region 1 Region 8

Source: Peter et al. (1996, Figure 2.3).

Figure 13.3 Structure of investment demand

Household Demands

MMRF has eight representative household sectors, one for each state. Total house-
hold expenditure in a region is driven by the regional household disposable income (a
Keynesian approach) and the changes of the number of household in the region. Figure
13.4, household sector, indicates that the nesting in the household demands is nearly the
same as the nesting in the investment structure, except that MMRF adopts the Stone
Geary utility function at the top level instead of a Leontief function for investment and
current production (Adams, 2008; Peter et al., 1996).

Foreign Export Demands

Export demands are divided into two groups: traditional export and non-traditional
export. The traditional exports include industries such as agriculture and mining that

DWYER 9781781001288 Part I.indd 267 25/07/2012 11:07

 268 Handbook of research methods in tourism

Utility

StoneGeary

Good 1 Good n

CES CES

Imports Domestic Imports Domestic

CES CES

Region 1 Region 8 Region 1 Region 8

Source: Peter et al. (1996, Figure 2.4).

Figure 13.4 Household consumption nest

have large export shares in the total exports. Each traditional export commodity is
modeled to have a downward sloping demand curve, driven by individual constant
price elasticity for each commodity. Thus, each traditional export commodity is free
to respond to price signal independently. In contrast, all non-traditional export com-
modities are assumed to move together as a group to maintain fixed shares in the total
(aggregate) non-traditional export. The aggregate non-traditional export is modeled
to have a downward sloping demand curve with a constant price elasticity. Thus, each
non-traditional export commodity cannot respond freely to price signals as in the case
of traditional export commodities. However, the model does have a mechanism to allow
for changes to the composition of the non-traditional export commodities when needed.

Government Demands

There are two government demand groups in MMRF: one by the Federal Government
and one by the State Government. The modeling of government demands is straight

DWYER 9781781001288 Part I.indd 268 25/07/2012 11:07

 CGE modeling 269

forward. The Federal Government demands are driven by aggregate national household
consumption, while the regional government demands are driven by the regional aggre-
gate household consumption.

Demands for Margins

Two margin commodities are recognized in MMRF: transport and communication

and finance. These commodities, in addition to being consumed directly by the users
(e.g., consumption of transport when taking holidays or commuting to work), are
also consumed to facilitate trade (e.g., the use of transport to ship commodities from
point of production to point of consumption). The latter type of demand for transport
and communication and finance are the so-called demands for margins. The margin
demand equations in MMRF indicate that the demands for margins are proportional
to the commodity flows with which the margins are associated (Peter et al., 1996,
p.18).

Prices

The price system underlying MMRF is based on two assumptions: (1) that there are
no pure profits in the production or distribution of commodities, and (2) that the price
received by the producer is uniform across all customers.
The MMRF has two types of price equations: (1) zero pure profits in current pro-
duction, capital creation and importing and (2) zero pure profits in the distribution
of commodities to users. Zero pure profits in current production, capital creation,
and importing is imposed by setting unit prices received by producers of commodi-
ties (i.e.,the commodities basic values) equal to unit costs. Zero pure profits in the
distribution of commodities is imposed by setting the prices paid by users equal to the
commodities basic value plus commodity taxes and the cost of margins (Peter et al.,
1996, p. 18).

Market-clearing Equations for Commodities

These impose the condition that demand equals supply for domestically produced
margin and nonmargin commodities and imported commodities respectively. The
output of regional industries producing margin commodities must equal the direct
demands by the models six users and their demands for the commodity as a margin.
The outputs of the nonmargin regional industries are equal to the direct demands
of the models six users. Import supplies are equal to the demands of the users
excludingforeigners, i.e., all exports involve some domestic value added (Peter et al.,
p. 20).

Indirect Taxes

For each user, the sales-tax equations in MMRF allow for variations in tax rates across
commodities, their sources and their destinations.

DWYER 9781781001288 Part I.indd 269 25/07/2012 11:07

 270 Handbook of research methods in tourism

Regional Incomes and Expenditures

At the aggregate level, Gross Regional Product from the expenditure and income sides
is calculated as follows.

Gross Regional Product (expenditure side) 5

1 Total household consumption (C)
1 Total investment (I)
1 Total government consumption (G)
1 Total overseas export (E)
1 Total inter-state export (E_IS)
2 Total overseas imports (M)
2 Total inter-state imports (M_IS)

Gross Regional Product (income side) 5

1 Total wages
1 Total gross operating surplus
1 Total net commodity taxes
1 Total net production taxes

Model Closures

Simulation results from CGE models depend largely on the adopted assumptions that
are often referred to as a closure. For a comparative static CGE model, the solution path
over time is not known. Rather, it is assumed that the economy operates within a certain
timeframe either a long run or a short run, depending on the purpose of a simulation.
The two main assumptions that characterize a short-run closure are: (1) the economy
operates in a timeframe that wage rates are rigid, changes to labor demand will be
reflected by changes in employment, as often this is not long enough for sale contracts
to be renegotiated for higher commodity prices in order to take into account any neces-
sary changes to wage rates that producers have to pay to their employees; (2) there is not
enough time for capital stock to be adjusted; return to capital (or the rate of return) will
adjust to reflect changes in demand for capital. As investment is tied to capital stock,
in this short-run closure investment is mainly fixed. In addition, the short-run closure
also assumes that regional migration does not occur, therefore regional population
remained unchanged. Any changes to the demand for labor will be satisfied by changes
in the regional unemployment rates while labor participation rates are kept constant. In
order to isolate the impact of an increase in tourism demand, all unnecessary stimuli to
the regional economies are minimized thus regional government consumptions, techno-
logical changes and inventory (stocks) are fixed. A positive shock is imposed to increase
tourism demand for individual selected specific tourist. Total import is determined by the
domestic income level as well as how expensive domestic goods are, in relative to import
prices. These movements of export and import determine the net balance of trade. Figure
13.5 presents the macro assumptions of the short-run closure. Elements in the rectangles
are those which are fixed (i.e., exogenously set at zero) while the ovals are determined
during simulation.

DWYER 9781781001288 Part I.indd 270 25/07/2012 11:07

 Rate of return

DWYER 9781781001288 Part I.indd 271

Real wage rate Capital stock

Employment

271
Real household Real government International
Real GSP = + Real investment + +
consumption consumption trade balance

Regional Household
population income

Figure 13.5 A short-run closure

25/07/2012 11:07
 272 Handbook of research methods in tourism

The long-run closure adopts the opposite assumptions. For the labor market, real
wage rates are now fully flexible while employment is fixed. The whole economy cannot
have more than the total aggregate employment level that the labor market can offer.
Any demand higher than this will be reflected by a rise in the real wage. In an adverse
situation, the fixed employment implies that whoever wants a job will get employment
as long as they are prepared to take whatever wage rate that is affordable by the pro-
ducers. Implicitly, this assumes full employment in the economy. Thus, in a long-run
closure, the total employment at the national will not change under the impact of any
shocks introduced to the model. For the capital market, the economy-wide capital stock
is now allowed to change in response to the demand of capital in the economy, while the
national rate of return is fixed.

Static Versus Dynamic CGE Models

The majority of CGE models can be separated into two broad categories, comparative
static and dynamic.

Comparative static models focus on the reactions of the economy at only one point in
time. For policy analysis, results from such a model are often interpreted as showing the
reaction of the economy in some future period to some external shock or policy change
that the adjustment path is not known. Often the adjustment path is assumed to be in
either a short-run or a long-run scenario. In the short run, it is assumed that capital stocks
are fixed and the real wages are not flexible enough to settle the labor market. In contrast,
the long-run scenario allows enough time for capital to build up such that economy-wide
rate of return is restored; and that, real wages can adjust so that anyone who would like
to work will find a job. In a favorable condition, real wages will increase to reflect a higher
demand for labor; and, in an adverse condition, real wages will decrease to restore the
full employment level in the economy. Using the ORANI-F mode (Horridge et al.,
1993), Adams and Parmenter (1995) showed the effects on the Australian economy of
additional (ten percent) expansion of inbound tourism in a long-run scenario. In contrast,
Blake (2003b) used the Nottingham model in the short-run assumptions to measure the
impacts of foot and mouth disease on the UK economy. Short-run scenarios are assumed
to take place over 12 years. The long-run assumptions are applicable to adjustment over
35 years or longer. This is the approximate time scale under which all economic adjust-
ments will have been made following an exogenous change.

Dynamic CGE models explicitly trace each variable through time, often at annual inter-
vals so that the adjustment path of the economy can be examined. The path comprises
a series of simulations, throughout a sequence of years (often referred to as recursive
dynamic), linked by intertemporal equations describing investment decisions and capital
accumulation. The model explicitly generates a time path of the economy following the
change in the policy or introduction of a supply side or demand side shock. In a recursive
dynamic scenario, it requires two series of simulations to measure the impacts. The first
series portrays a business as usual case (baseline) and the second series incorporates the
policy shocks that take the economy away from the baseline scenario. The deviation of
the policy scenario from the baseline reflects the impacts of the policy on the economy.

DWYER 9781781001288 Part I.indd 272 25/07/2012 11:07

 CGE modeling 273

Dynamic models can be used to forecast the structure of the economy as well as to
assess the effects of policy and other shocks over a period of time. Blake (2005) uses
a forward looking dynamic model to project the economic impacts of the London
Olympics 2012, under different dynamic conditions. In particular, an unanticipated
shock differs significantly to an anticipated shock. He demonstrates that the benefits
of tourism are lower when a tourism boom is anticipated and this is of particular
concern when considering the economic benefits of future events that can be antici-
pated, such as major sporting and other special events. One area where a dynamic
approach is particularly useful is in modeling the effects of climate change mitigation
policies (Hoque et al., 2009). Dynamic CGE models can provide a richer set of infor-
mation than that available from static models, depending on the assumptions that are
made about the changes that are occurring in the economy. However, they are more
challenging to construct and solve since they require for instance that future changes
be predicted for all exogenous variables, not just those directly targeted by a possible
policy change.
The majority of case studies of CGE applications employed in a tourism context to
date are static. While dynamic models can certainly offer the time path of adjustment
over time, static models are essentially revealing the direction of impacts at the industry
and tourism destination level that are the major concerns of policy makers with less com-
plexity of requirement than a dynamic application. Thus, comparative static simulations
are often used in practice.
CGE models can be formulated at a number of spatial levels:

Single-country models such as ORANI (Dixon et al., 1982) or MONASH (Dixon

and Rimmer, 2002) with extensions for regional disaggregation using a top-down
approach.
Stand-alone models of regional economies (Meagher and Parmenter, 1990).
Multi-regional models such as MONASH-MRF (Adams, 2008).
Multi-country models (Hertel, 1997).

Models of one or more of these types would be suitable for assessing most tourism
impact issues.

BACKGROUND AND TYPES OF PROBLEMS THAT CGE

MODELING IS DESIGNED TO HANDLE

CGE models are a standard tool of empirical analysis. They are widely used to analyse
the aggregate welfare and distributional impacts of policies whose effects may be trans-
mitted through multiple markets, or contain mixtures of different tax, subsidy, quota, or
transfer instruments. CGE analysis is being employed to explore the economic impacts
of policy initiatives and frameworks, and broader changes as diverse as hazardous waste
management, trade liberalization, tariff protection, environmenteconomy interac-
tions, structural adjustment, agricultural stabilization programs, technological change,
labor market deregulation, financial market deregulation, fiscal reform, development
planning, macroeconomic reform, economic transition, international capital linkages,

DWYER 9781781001288 Part I.indd 273 25/07/2012 11:07

 274 Handbook of research methods in tourism

environmental regulation, public infrastructure, and industry sector studies (Dixon and
Parmenter, 1996; Gunning and Keyzer, 1995; Harrison et al., 2000).
CGE models are now increasingly used in tourism economics analysis and policy for-
mulation (Dwyer et al., 2004). Some advantages are:

Computable general equilibrium (CGE) models are well suited to tourism analy-
sis and policy, given their multi-sectoral basis. In contrast to partial equilibrium
approaches, CGE models can take account of the interrelationships between
tourism, other sectors in the domestic economy and foreign producers and con-
sumers.
CGE modeling can be tailored to alternative conditions, such as flexible or fixed
prices, alternative exchange rate regimes, differences in the degree of mobility of
factors of production and different types of competition.
CGE models can be used to quantify the effects of actual policies, such as changes
in taxation, subsidies or government borrowing, as well as predicting the effects of
a range of alternative policies or exogenous expenditure shocks. They can be used
to estimate the impacts of changes in tourism expenditure under a range of alter-
native macroeconomic scenarios; tailoring them to alternative conditions such as
flexible or fixed prices, various exchange rate regimes, differences in the degree of
mobility of factors of production, different government fiscal policy stances, and
different types of competition.
CGE models are helpful to tourism policy makers who seek to use them to provide
guidance about a wide variety of What if? questions, arising from a wide range of
domestic or international expenditure shocks or alternative policy scenarios.

APPLICATIONS OF CGE MODELING TO TOURISM

A good summary of the usefulness of CGE modeling in tourism analysis and policy
development is provided by Blake et al. (2006). To highlight some of the uses of CGE
modeling in tourism contexts, we classify them herein under the following headings:

Economic impacts of changes in inbound tourism

Economic impacts of tourism crises
Economic impacts of special events
Evaluation of economic policy

While the articles highlighted below comprise only some of the research by tourism
economists using CGE models, they indicate the types of issues that are being addressed
in the literature.

Economic Impacts of Changes in Inbound Tourism

CGE models provide information on the structural effects of tourism expansion. Adams
and Parmenter (1995, 1999) were among the first researchers to provide empirical evi-
dence to support Copelands (1991) theoretical argument that some sectors benefit and

DWYER 9781781001288 Part I.indd 274 25/07/2012 11:07

 CGE modeling 275

some lose as the result of tourism expansion. They used the ORANI-F, a 117-sector
general equilibrium model for Australia, to assess the effects on the Australian economy
of additional expansion of inbound tourism. To highlight the effects on the economy of
the increased international tourism, Adams and Parmenter compared the Australian
economy in a scenario with ten percentage point increase in inbound tourism higher than
the tourism growth in the base case projection of the Australian economy. The effects of
the additional tourism growth rate were reported as the differences between the amended
and bases case simulations.
The tourism expansion stimulates capital formation and generates an increase in the
rate of growth of investment. Holding both employment growth and technical change at
their base-case rates, there is a small increase in real GDP. Growth in private consump-
tion is reduced slightly due to an increase in income tax rates. To maintain a fixed Public
Sector Borrowing Requirement (PSBR), income tax rates must rise to offset rapid growth
in government investment expenditure. With additional growth in aggregate real domes-
tic absorption (consumption plus investment) exceeding additional growth in real GDP,
there is an increase in the balance of trade deficit. The import content of the induced
investment contributes directly to the deterioration in the trade balance. In addition, an
appreciation of the real exchange rate is required to make room for the increased level of
domestic demand. This generates substitution towards imports and reduces the traditional
exports of mining and agricultural commodities. This reduced export growth improves the
terms of trade and explains why growth in the aggregate volume of exports increases by
only a very small percentage. The improvement in the terms of trade and the reduction in
the activity levels of land intensive export industries allow an increase in the real wage rate.
At the sectoral level, there will be losers as well as gainers from the expansion in
inbound tourism. Four groups of industries are distinguished.

Service industries catering directly to international tourists (e.g., air transport, res-
taurants, and hotels). These are strongly stimulated by the additional expansion in
tourism. This strengthens further the industries base-case growth implied by the
strong growth assumed for tourism in the base forecasts.
Industries indirectly supplying tourism-related activities (e.g., aircraft maintenance
and construction). These are also stimulated by the additional expansion of tourism.
Except for construction, most of them also enjoy strong prospects in the base fore-
casts because of base-case growth of tourism. In the base forecasts, since invest-
ment growth is weak, especially residential investment, the additional expansion of
tourism eases the adjustment problems that the construction sector would otherwise
have experienced.
Non-tourism exporters (e.g., agriculture, mining, food and metals processing).
Growth prospects in these industries are reduced by the appreciation of the real
exchange rate produced by additional tourism expansion. However, the base fore-
casts are characterized by exchange rate depreciation, giving the export industries
good prospects. The adverse effects of the additional tourism expansion should thus
be easily accommodated.
Import-competing industries (e.g., transport equipment, chemicals, textiles, clothing
and footwear). Prospects in these industries are reduced by the tourism induced
appreciation of the exchange rate.

DWYER 9781781001288 Part I.indd 275 25/07/2012 11:07

 276 Handbook of research methods in tourism

The authors also explored the consequences for the different states within Australia. The
appreciation of the exchange rate adversely impacts on export-oriented activities includ-
ing mining and agriculture. Three heavily export-oriented states Queensland, Western
Australia and Tasmania are net losers from general tourism stimulation nationally.
Using a CGE model based on the ORANI model, Narayan (2003, 2004) simulated
the long-run impact of a 10 percent increase in visitor expenditures on Fijis economy.
The study indicates that, for an island developing country such as Fiji, an expansion
in inbound tourism can generate overall growth in real GDP but effects on the real
exchange rate, real wages, and the CPI imply that the gains to tourism-related sectors
are offset to some extent by losses in traditional export and import competing industries.
Zhou et al. (1997) analysed the impacts on the Hawaii state economy of a 10 percent
projected decrease in visitor expenditure. This is projected to result in: a reduction in
GSP; a small reduction in the general level of prices; output reductions in the industries
servicing tourist needs, and in traditional exports, manufacturing, construction and serv-
ices; a reduction in imports, particularly those associated with tourist related industries;
a fall in the balance of trade; and reduced employment in the hotel industry, restaurants
and bars, and transportation.
CGE modeling can be used to develop economy-wide yield measures of tourist
worth. Economy-wide yield measures indicate the bottom line for the economy from
tourist spending when all inter-industry effects have taken place (Dwyer et al., 2007).
Dwyer and Forsyth (2008) estimated three economy-wide yield measures associated
with 14 of Australias major tourism origin countries and for three regions. The yield
measures were Gross Value Added (GVA), Gross Operating Surplus (GOS), and
Employment. The economic impacts were then converted to yield measures by determin-
ing the economy wide effects of an additional tourist from each market. Yield measures
based on CGE modeling can inform organizations in both the private and public sector
about effective allocation of marketing resources and types of tourism development that
meet operator and destination manager objectives (Dwyer and Forsyth, 2008).

Economic Impacts of Tourism Crises

Blake et al. (2003b) investigated the economy-wide effects of foot and mouth disease
(FMD) in the UK using their Nottingham CGE model. The simulations showed that
FMD had considerable impacts not only on sectors directly related to tourism, but also
on other industries. The policy of maintaining FMD-free status supported meat exports,
but it cost the UK tourism industry substantially. The CGE tourism model of the UK
economy showed that the effects of the tourism decreases were to reduce GDP in the UK
by more than the loss of agricultural production. The authors concluded that a policy
geared towards supporting tourism would have been far less costly than the govern-
ments policy of supporting agricultural exports by means of slaughtering animals and
prohibiting access to many rural areas. The implication is that policy makers need to
adopt a whole of industry approach to policy making relating to FMD and in formulat-
ing agricultural policies.
Blake and Sinclair (2003) modeled the impacts on tourism in the USA of actual and
proposed policy responses of the US Government following the terrorist attacks of
September 11, 2001. A major finding was that an airline production subsidy outperforms

DWYER 9781781001288 Part I.indd 276 25/07/2012 11:07

 CGE modeling 277

all other types in each of the criteria with the exception of the number of accommodation
jobs lost. However, an airline subsidy fails to take account of job losses outside the sector
that are caused by the fall in demand from tourists who would have traveled by air. The
researchers also found that subsidies to accommodation establishments were reasonably
effective at boosting GDP and saving jobs in this sector. These subsidies are significantly
more effective than those allocated to catering and entertainment. In fact, subsidies to
the latter can have the effect of worsening GDP and labor and capital adjustment, as
they encourage workers to move out of the airline and accommodation sectors, thereby
increasing the job losses in these sectors.
In a context of uncertainty over traveler security, tourism experienced two major crises
in 2003 the Iraq War and SARS. Acknowledging that the relative impacts of a complex
array of impacts on travel decision making are almost impossible to dissect. Dwyer et
al. (2006c) explored the economic effects of the SARS crisis on tourism to Australia.
Although the crisis resulted in less inbound tourism, it also led to reduced outbound
tourism. The net economic impacts on the nation depend upon the extent to which
canceled or postponed outbound travel are allocated to savings, to domestic tourism, or
to the purchases of other goods and services. Using a computable general equilibrium
model of the Australian economy, simulations of the impacts of the events suggest that
the net effects were not as severe as were perceived by tourism stakeholders.
Pambudi et al. (2009) employed a multi-regional CGE model for Indonesia to esti-
mate the short-run effect of a decline in tourism following the 2002 Bali bombings on
the Indonesian economy. The results suggest that of Indonesias 26 provinces, GRP of
Bali is worst affected by a negative shock to tourism exports followed by other popular
tourist destinations, such as Jakarta and Yogyakarta. In Bali employment fell by 4.93
percent, household consumption by 4.68 percent, investment by 6.79 percent, exports
by 16.34 percent, and imports by 8.95 percent. Within Bali the tourism-related and
non-tradable sectors contain the worst affected industries while export-oriented indus-
tries, such as textiles, clothing, and footwear, and import-competing industries, such as
machinery and electronics expand. The researchers concluded that policy makers and
lending agencies should take into account not only the regional macroeconomic implica-
tions of the bombing, but also the sectoral results in allocating compensation packages.
The results indicate that while the sectors that are most closely tied to tourism are the
worst affected by the bombings, some sectors may grow even though they are in regions
that lose overall. For example, while the Balinese hotel and restaurant sector shrinks by
7.7 percent, machinery and electronics grow by 2.13 percent. Thus, to have most effect,
assistance should not just be allocated on a regional basis, but be sector specific.
Meng et al. (2010) studied the effects of the Global Financial Crisis (GFC) on
Singapore. The CGE simulation results demonstrate that at the macro level, the CFC
tends to have large negative effects on Singapores tourism and on the Singapore
economy. Although almost all economic variables are negatively affected, total exports
benefit greatly. At the industry level, a negative tourism shock impacts severely on the
tourism-related sectors, impacts only slightly on sectors weakly linked to tourism, but
tourism-competing sectors expand. In commodity market, prices and outputs decrease
for most products but real household consumption and exports increase. In the labor
market, low-skilled workers are adversely affected, but some occupational groups benefit
at the expense of others.

DWYER 9781781001288 Part I.indd 277 25/07/2012 11:07

 278 Handbook of research methods in tourism

Li et al. (2010) studied the effects of the GFC on China. They note that while headline
figures show that international tourism is suffering as a consequence of this economic
crisis, domestic tourism is larger than in many countries and impacts through this market
could be larger than through international tourism. In the context of Chinas tourism,
the authors test the common viewpoint that an increase in domestic tourism could com-
pensate a decline in inbound tourism, which supports the policy of focusing on the devel-
opment of domestic tourism. They evaluate the magnitude of economic impact of the
economic slowdown on Chinas tourism using computable general equilibrium modeling
and then bring forward some policy suggestions on the development of Chinas tourism.
Pham et al. (2010) used a sub-state CGE model of the Australian economy, known as
TERM (Horridge et al., 2003), and elaborated the model to incorporate the inbound and
domestic tourism sectors explicitly for five tourism destinations in Australia. The model
was then used to analyse the impacts of induced impacts of climate change via tourism
on the destination economies. The study emphasizes that while direct climate change
will have adverse impacts on the tourism destinations, the further induced impacts of
climate change on tourism will exacerbate the total impacts on the tourism destinations.
If the induced impacts are not taken into account, it most likely that the climate change
impacts on tourism destinations are underestimated.

The findings from the above studies suggest that government subsidies can be effective
in limiting the adverse effects of a major tourism crisis. However, the relative effec-
tiveness of the different policy responses varies considerably. The overall conclusion
is that directing subsidies to the sector that is most severely affected by the crisis is
the most efficient policy response in terms of both GDP and the total number of jobs
saved. Policy makers should be very careful in their decisions as to which sectors to
assist, as the provision of subsidies for those relatively unaffected can even be counter-
productive.

Economic Impacts of Special Events

Adam Blake has used a dynamic CGE model of the UK and London economies to
forecast the economic impacts of the London 2012 Olympics (Blake, 2005). While the
overall impacts on GDP and employment in the UK are positive, the simulations reveal
that there is a loss of GDP and employment in the areas outside London. Reasons for
this include: spending in London by UK residents from outside London visiting the
games; movement of workers, whether migrants, commuter, or temporary migrants, into
London because of higher wages in the capital; and the provision of Lottery funding,
which in effect transfers money to London. The study also shows that the impact of the
games will vary significantly across different sectors of the UK economy. Sectors that
expand include construction, passenger land transport, business services, hotels, and res-
taurants. Sectors that are not directly related to the games may contract in size indirectly
as a result of hosting the games. These include manufacturing, agriculture, fishing, and
other services. However, these results are relative to the No Games scenario in which
a substantial amount of growth takes place in all sectors of the economy. Thus, while no
sector is predicted to contract in the time span modeled, some will grow less because of
the impact of hosting the Olympics.

DWYER 9781781001288 Part I.indd 278 25/07/2012 11:07

 CGE modeling 279

The simulations indicate that any changes to the UK economy associated with the
Olympics 2012 will be comparatively small. Even in the Olympic year, the total economy
wide effect for the UK is only 0.066 percent of total UK GDP at 2004 prices. The results
of Blakes study indicate that a mega event, even of the size of the summer Olympics, is
unlikely to provide any substantial boost to either the national or host-region economy.
That the type of model used to estimate the economic impacts of special events has
a substantial effect on the assessment results became evident in a study by Dwyer et al.
(2005, 2006a, 2006b) which compared the results of using CGE and inputoutput mod-
eling to estimate the economic impacts of a special event held in Australia, namely, the
Qantas Formula One Grand Prix. The simulations reveal that inputoutput modeling
projects a much greater impact on real output for both New South Wales and Australia
(A$112.0 million and A$120.1 million), as compared to CGE modeling (A$56.70 million
and A$24.46 million). The output multiplier for New South Wales is 2.185 using the
inputoutput model but only 1.106 using the CGE model. For Australia as a whole, the
inputoutput model yields an output multiplier of 2.343 whereas the CGE model yields
a substantially smaller output multiplier of 0.477. The two models give different employ-
ment projections also. The projected increase in employment using an inputoutput
model is 521 (full-time equivalent) jobs in NSW and 592 jobs throughout Australia.
Using a CGE model the projected employment effects are 318 jobs and 129 jobs, respec-
tively.
Madden (2006) employed a multiregional CGE model to examine the effects of a mega
sporting event, the Sydney 2000 summer Olympics. Simulations were conducted for three
phases over a 12-year period. Careful attention was paid to the presence of constraints
on labor supply and capital and on the sources of savings funding Olympic investments.
Simulations were conducted under the assumption that no effects from the games remain
on the debt position of Australian governments or the nations external debt 5 years after
the Sydney Games. The results indicate that provided there is not too large a financial
loss on the games, inclusive of construction costs, a modest positive impact on the state
hosting the games can occur. The degree to which this positive impact comes at the
expense of other states depends crucially on assumed labor-market conditions.
A study by Li et al. (2011) takes the first step in applying CGE modeling to forecast-
ing the economic contribution of tourism generated by the Beijing Olympics. The paper
includes two types of estimations: ex ante and ex post. The ex-ante estimation was
conducted before the Beijing Olympics and thus predicted the impact of international
tourism based on historical data, such as previous literature and historical statistics. The
ex-post estimation was conducted several months after the Beijing Olympics and the
estimation was based on up-to-date statistics published by the China National Tourism
Administration. The economic impact generated from the two types of estimations is
compared. It was found that, while the economic impact of international tourism was
predicted to be positive in the ex-ante estimation, this impact was found to be negative
in the ex-post estimation.
Giesecke and Madden (2011) explored the issue of whether the large economic ben-
efits predicted for Olympics host countries actually materialize. They re-examined the
Sydney 2000 Olympics via historical modeling with a multiregional dynamic CGE model.
Their modeling encompasses the four key economic dimensions of the games: (1) games
operations; (2) construction of games facilities and associated financing; (3) spending by

DWYER 9781781001288 Part I.indd 279 25/07/2012 11:07

 280 Handbook of research methods in tourism

interstate visitors in the games year; and (4) spending by foreign visitors in the games
year. It was found that the Sydney Olympics generated a loss in Australian real private
and public consumption in present value terms of $2.1 billion. This does not necessarily
mean that Australia should not have hosted the games. An Olympic Games is a great
international sporting event that brings much enjoyment to a large number of people
around the globe. There is evidence (Atkinson et al., 2008) pointing to significant posi-
tive effects on welfare from an Olympic Games such as national pride, consumer surplus
on local ticket sales, and perceived improvements in various socially desirable goals (e.g.
greater cultural understanding, inspiration of children, and the promotion of healthy
living). Thus, although the Australian economy experienced a loss, whether Australians
gained or not in welfare terms from hosting the 2000 Olympics remains an open question.
Contrary to what might be implied in much of the economic impacts literature, esti-
mates of the economic impacts of events provide, in themselves, an imperfect basis for
decisions about resource allocation (Abelson, 2011). The problem arises from a failure to
distinguish clearly between the impacts and the (net) benefits of the event. A Cost-Benefit
Analysis (CBA) picks up a whole range of benefits and costs which would not be picked
up in a CGE model. This includes non priced effects which do not get included in the
markets which are modeled-noise from an event, the consumers surplus of home patrons,
loss of park amenity and traffic congestion associated with the event. The CGE modeling
technique, in contrast, picks up general equilibrium effects which the partial CBA is not
capable of detecting, including the increased income to households resulting from the
event. Studies commissioned by the Victorian Auditor-General (2007) found that, while
the Melbourne Formula One Grand Prix in 2005 increased Victorian State Product by
$62.4 million (CGE result), it generated a negative net social benefit (cost) to the State
of $6.7 million (CBA result). Since neither technique is completely comprehensive, both
have a role in a detailed evaluation of a project or event. Dwyer and Forsyth (2009)
outline a method for partially integrating the two techniques.

Evaluation of Economic Policy

CGE-related research has also considered the impact of tourism in developing countries.
CGE modeling can also provide important insights into policies relating to different
types and levels of taxation. Alavalapati and Adamowicz (2000) examined the interac-
tions between tourism, natural resources, and the environment in British Columbia,
paying specific attention to the effects of an environmental tax on each sector of the
economy. They considered two scenarios: one in which environmental damage occurs
because of natural resource extraction activities, notably the pulp industry, and the other
in which the environment deteriorates because of both natural resource extraction and
tourism. The simulated results indicated that the imposition of an environmental tax on
the resource sector is beneficial if the environmental damage results only from the extrac-
tion activities but that it has adverse effects if the damage results from both resource
extraction and tourism. The study indicates that the policy measure should be targeted
directly on the sector that is most directly affected.
Blake (2000) studied the impact of different types of taxation using a CGE model
that he developed for the Spanish economy. He showed that foreign tourism activities
in Spain are highly taxed relative to other sectors. In the case of domestic tourism, the

DWYER 9781781001288 Part I.indd 280 25/07/2012 11:07

 CGE modeling 281

tax system levies lower rates of taxation on tourism and subsidizes domestic transport.
Blake examined the issue of near-marginal tax incidence in the Spanish case and found
that marginal increases in taxes on foreign tourism are likely to result in higher domestic
welfare, since the effect of the increases is to reduce the pre-existing distortions in the
domestic economy that result from low levels of domestic taxation. The elimination
of indirect taxation would benefit foreign tourists but would decrease the quantity of
domestic tourism, indicating that the existing tax structure, in its entirety, has the net
effect of taxing foreign tourists but subsidizing domestic tourists. The removal of value
added tax from accommodation would have adverse effects on the economy. However,
an increase in the tax on accommodation would be beneficial over the long run, partly
because it would counteract the effects of the subsidies on domestic transport. Blake also
pointed out that both the short run transition costs and any externalities associated with
tax changes should also be taken into account in the process of policy formulation.
Mabugu (2002) applied a short-term computable general equilibrium model for
Zimbabwe to trace the direct and indirect effects of policy on the macroeconomy and
tourism. The results show that the main reason why benefits from tourism are bypass-
ing the country is because of poorly sequenced macroeconomic policies and a negative
political climate. Mabugu argues that as and when the national political situation sta-
bilizes and the economy begins to grow again, a credible macroeconomic stabilization
programme should be implemented, targeted to reduce fiscal deficits, and using flexible
foreign exchange markets and tight monetary policies to address inflation. However,
because Zimbabwe is in arrears, there can be no programmes or lending with the
International Monetary Fund and World Bank. Getting the budget in order without
aid money will be very difficult indeed, and the alternative (debt deflation by means of
hyperinflation) is worse.
Sugiyarto et al. (2003) used a CGE model of the Indonesian economy to examine the
effects of globalization via tariff reductions, as a stand-alone policy and in conjunction
with tourism growth. Indonesia is an interesting case study as it has experienced both
trade liberalization and tourism growth during the past decades The results show that
tourism growth amplifies the positive effects of globalization and lessens its adverse
effects. Production increases and welfare improves, while adverse effects on government
deficits and the trade balance are reduced. Globalization combined with tourism does
not necessarily have adverse effects on the domestic economy, in contrast to the past por-
trayal of the combination as a deadly mix. Globalization and foreign tourism growth
can, in fact, reduce the domestic price level and increase the amount of foreign trade and
availability of products in the domestic economy, thereby stimulating further produc-
tion. The end result in the Indonesian case is improved macroeconomic performance and
welfare, as domestic absorption, and household consumption increase. Foreign tourists
are also better off for they can consume more, given their spending level, and also benefit
from the greater availability of products. The ongoing growth of foreign tourism also
reduces the governments burdens as a result of embarking on globalization, by enabling
it to reduce its reliance on import tariffs and indirect taxation while, at the same time,
maintaining the level of income necessary to finance its expenditure. Tourism growth
would, therefore, enable the government to follow a fiscal policy of revenue neutral
globalization, allowing it to finance its expenditure without imposing higher taxes on the
Indonesian population.

DWYER 9781781001288 Part I.indd 281 25/07/2012 11:07

 282 Handbook of research methods in tourism

Gmez Gmez et al. (2003) formulated a dynamic GE model that examines the condi-
tions that are required for a tourism tax to contribute a double dividend of environmental
improvement and an increase in consumption. They highlight the role of the terms of trade
in giving rise to a double dividend in an economy specializing in tourism. Gillham (2004)
developed a dynamic model with imperfect competition that can take account of increas-
ing returns to scale in different sectors of the Spanish economy. He used the model to
evaluate the impact of foreign direct investment on tourism and other economic sectors in
Spain. He also used a CGE model developed specifically for the Canary Islands to examine
the interrelationships between tourism and trade and taxation policies in the islands.
Blake et al. (2003a) developed CGE models of the Maltese and Cypriot economies to
quantify and compare the impact of EU accession and changes in tourism on each of
the Mediterranean islands. Tourism is particularly important relative to other economic
activities in these island nations, so that changes in tourism demand are likely to have
considerable effects at both the macroeconomic and intersectoral levels. Results show
that EU accession is beneficial to both countries, although as a percentage of GDP Malta
benefits considerably more than Cyprus in part because EU funding is more substantial
in Malta when compared with GDP, but also because Malta trades a larger share of its
GDP with the EU than Cyprus does. The effects of accession on tourism are negative in
Malta and positive in Cyprus, because the greater effects from trade and funding allo-
cations lead to a greater demand for factors of production in Malta that increase wage
rates and take factors of production away from tourism. In Cyprus, effects that benefit
tourism outweigh such general equilibrium trade-off effects.
Gooroochurn and Sinclair (2005) employed a CGE model for Mauritius to explore
the effects of changes in production taxes and of sales taxes on different sectors of the
economy, in alternative contexts of exogenous and endogenous tourist arrivals. The
marginal economic benefit (MEB), a measure of the incremental welfare cost of raising
additional revenue from a tax that is already distortionary, was found to be higher for
almost all economic sectors in the case when tourist arrivals are exogenous. However, the
MEB was higher for restaurants and hotels when tourist arrivals are endogenous, as the
increase in welfare associated with a rise in domestic consumption outweighed the fall in
welfare associated with a smaller increase in government revenue. Consideration of the
distributional effects of an increase in sales tax on the tourism sector indicated that an
increase in the tax rate has smaller adverse effects on poorer than on richer households.
The overall results indicate that the structure of indirect taxation in Mauritius is not
optimal and that tourism-related sectors appear to be under-taxed, although any tax
increases should be considered in conjunction with estimates of the incidence of the tax
on tourists and their price elasticity of demand.
Gooroochurn and Sinclair (2005) found that directly taxing foreign tourists to
Mauritius increases domestic welfare. The authors also examined production and sales
tax reforms and compare the welfare cost of raising additional revenue from a tax on a
specific commodity while other taxes remain constant. They found that tourism-related
sectors (restaurants/hotels and transport/communication) were undertaxed, that the
cost of increasing the taxation of these sectors would be mainly born by foreign tourists,
and that increasing taxes in these sectors has smaller effects on poorer than on richer
households and reduces income inequality. The main driver of the last result is the higher
proportion of domestic consumption from the richer household groups for tourism-

DWYER 9781781001288 Part I.indd 282 25/07/2012 11:07

 CGE modeling 283

related sectors. A narrow policy, taxing the highly tourism-intensive sectors extracts
significantly more revenues from tourists than a broader policy where all tourism-related
sectors are taxed. The narrow policy has a higher beneficial term of trade effect than the
broader policy. Higher term of trade implies that the local economy can obtain more
imports (hence consumption) for a given amount of exports.
Blake et al. (2008) examined the issue of how tourism affects poverty. They used a
CGE model for Brazil to quantify the effects of increased inbound tourism on income
distribution and poverty relief that occur via changes in prices, earnings, and government
revenues. The CGE model is calibrated using a Social Accounting Matrix (SAM) that
shows the payments that take place among the different industries, products, factors,
households, firms, the government, and the rest of the world. The model incorporates the
earnings of different groups of workers within tourism, along with the channels by which
changes in earnings, prices, and the government affect the distribution of income among
rich and poor households. A 10 percent increase in demand by foreign tourists in Brazil
was assumed in order to investigate the effects on income distribution in the economy.
The results suggest caution when generalizing the effects of tourism growth on poverty
within a country. In the case of Brazil, there was a strong reinforcement effect whereby
the industries that reduced their output following an increase in tourism demand were
export industries that employ factors of production from the richer households. This
implies that the structure of earnings in non-tourism export sectors plays a significant
role in determining the net poverty effects of tourism. The authors emphasized that this
type of earnings structure may not apply in other countries. Hence, it would be impor-
tant to apply the model to tourism expansion in other countries, in order to investigate
the effects that would occur under different types of earnings structures.
To answer the question: is tourism growth pro-poor?, Wattanakuljarus and Coxhead
(2008) used a CGE model for Thailand to simulate the effects of tourism growth. The
simulations indicated that capital and labor in non-agriculture are the factors that gain
the most. In Thailand, corporations are the major owners of capital in non-agriculture,
and corporate income accrues mainly to wealthy non-agricultural households. Similarly,
since high-income non-agricultural households are the major owners of labor in non-
agriculture, they are the next biggest beneficiaries. As a result, given the distribution of
factor ownership across household groups, inbound tourism expansion raises incomes
across the board, but the main share of the gains accrues to the non-poor. The authors
conducted sensitivity analysis with different assumptions regarding factor constraints.
In every scenario, however, although tourism growth benefits all household classes, the
biggest gains accrue to high-income and non-agricultural households. It is concluded
that tourism growth in Thailand is not pro-poor or pro-agriculture. To address this
increased inequality, additional policy instruments are required to correct for the ine-
qualities occasioning tourism growth.
Macao has been witnessing robust economic growth in recent years. According to
Sheng and Tsui (2009) the ongoing boom is mainly driven by rapid tourism growth
reflected in massive tourist arrivals and foreign capital inflow. Although Macao is
praised as an economic wonder, serious externalities have emerged, raising concerns
about the sustainability of the citys long-term development. Using a modified simple
general equilibrium model, the authors show how economic, social, environmental, and
political externalities accompanying rapid tourism growth may possibly reduce the net

DWYER 9781781001288 Part I.indd 283 25/07/2012 11:07

 284 Handbook of research methods in tourism

welfare of host communities. The paper concludes that the government should keep
sustainability in mind when making tourism policies, carefully taking all related social,
environmental and political factors into consideration.
MMRF-GREEN (Adams et al., 2003) has been developed to estimate the greenhouse
gas emissions associated with economic activity. The development of green CGE
models is an important step towards identifying the extent of externalities associated
with tourism and other industries. Based on the MMRF-GREEN model, Hoque et al.
(2009) investigated the potential economic impacts of introduction by the Australian
government of its proposed Carbon Pollution Reduction Scheme (CPRS). The CPRS
is intended to introduce a cap and trade mechanism for reducing greenhouse gas emis-
sions in Australia. It was intended to commence in 2011 but passage of the legislation
has been delayed in Parliament. The modeling focused on a scenario called the CPRS5
with targets aimed to achieve greenhouse gas emissions reduction by 2020 of 5 percent
below 2000 level. Effects on the Australian economy and in particular the tourism sector
in Australia were examined. The authors concluded that under the proposed scheme, the
tourism sector will contract. Falls will occur in real tourism gross value added and tourism
employment. The largest falls are projected to be in accommodation, air and water indus-
tries, and in cafes, restaurants, and food outlets. While most tourism industries experi-
ence contraction in their real value added relative to baseline values, the rail transport
industry experiences an expansion because the emissions price causes substitution toward
this industry against the high-emissions transport industries such as air, water, and other
road transport industries. Overall, the gains experienced by some tourism industries will
be heavily outweighed by contractions in some of the tourism characteristic industries.
Forsyth et al. (2011) used a CGE model to assess the impacts of Australias Passenger
Movement Charge on key economic variables, such as GDP, GNI, and economic
welfare, and on tourism industry output and employment. The PMC is a tax on all
persons departing Australia by air. For Australia as a whole, a rise in the PMC is posi-
tive, though it is negative for the tourism industry (which forms part of the Australia-
wide effect). This effect on the tourism industry will be more negative if the reduction
in outbound tourism does not help stimulate domestic tourism. The net impact on the
tourism industry is small (if domestic tourism substitutes for outbound tourism) but this
can go either way. However there will be a net positive impact on the economy as a whole
in most cases. This comes about because of the tax effect Australia gains from foreign
tourists paying Australian taxes rather than Australian residents. This effect is sufficient
to outweigh other impacts. This suggests that if funds become available, reducing the
PMC would not be a cost-effective way of helping the tourism industry, and that other
ways, such as increasing promotion or measures directed to improving tourism industry
productivity, may prove to be more cost effective for the economy as a whole.

ADVANTAGES AND LIMITATIONS OF CGE MODELING

CONCEPTUALLY AND FOR POLICY FORMULATION

CGE models are sometimes said to be too difficult to use and too data demanding.
While it has been the case that the application of CGE models to tourism economics
has been much restrained until recently by the lack of data, the development of Tourism

DWYER 9781781001288 Part I.indd 284 25/07/2012 11:07

 CGE modeling 285

Satellite Accounts (TSA) has progressively reduced this difficulty, and we have seen more
and more of these applications in the recent years. This trend will certainly continue.
Moreover, CGE modeling techniques and software systems are now routinely available
globally. An increasing number of dynamic CGE models are now being developed, in
line with improvements in software and increasing computational power. In any case,
the data should be assessed in terms of its importance for the question to be investigated,
other than just in terms of the ease of data mobilization.
Some critics regard the assumption of market clearing in CGE modeling as unwar-
ranted, arguing that although there are forces pushing economies towards equilibrium,
there are also forces that prevent such equilibrium outcomes from being achieved. Croes
and Severt (2007) argue that the tourism product presents features difficult to align
with perfect equilibrium. Market distortions reduce the role of price signals and hence
affect the efficient allocation of resources. Furthermore, the presence of public goods
impacts the optimizing behaviour of economic agents. Finally, extreme cases of natural
hazards, such as in the case of hurricanes in Florida, can propel great external shocks to
the markets to the extent that the applicability of the CGE model is rendered deficient.
In response, as Blake et al. (2006) point out, this type of criticism can be leveled against
the great majority of economic models. All models provide a simplified representation
of reality but nevertheless provide an effective means of understanding and/or predict-
ing economic interrelationships and outcomes. CGE analysis allows the models to be
tailored to fit different real world circumstances. Thus, for example, constraints can be
imposed on the model to allow for rigidities in factor prices or mobility, exchange rates
or government borrowing. Similarly, alternative functional forms can be used to take
account of different types of market structure and competition.
CGE models can determine an industry-level impact and show the interactions
between industries (or sectors). However, like other models, CGE models usually do
not show what happens within an industry or market. In such circumstances, a market-
specific model is required.
CGE models provide useful guidance for policy formulation, as the above examples of
research indicate. Of course, the results obtained are particularly sensitive to some of the
parameter values (and assumptions) that are included in them. If such parameter values
are inaccurate, the results obtained from the models are also likely to be inaccurate, as
well as misleading for policy purposes. To overcome this problem, sensitivity analysis
can be conducted by including alternative parameter values in the model, to determine
the bounds within which the model results lie, for changes in the parameter values that
are deemed to be realistic. Results from CGE models can be explained clearly in terms
of straightforward economic mechanisms and properties of the data incorporated in the
model. It is the responsibility of the modeler to produce such explanations for the users
of the results.

FURTHER DEVELOPMENTS AND APPLICATIONS OF CGE

MODELING IN TOURISM RESEARCH

Future research on CGE tourism analysis is likely to focus on four main areas (Blake et
al., 2006). The first involves further research on dynamic CGE analysis. When there is an

DWYER 9781781001288 Part I.indd 285 25/07/2012 11:07

 286 Handbook of research methods in tourism

interest in the adjustment process, for example, how long it takes for a shift in tourism
flows to influence other variables in the economy, then a dynamic framework is required
As discussed above, dynamic models can be used to project the structure of the economy
as well as to assess the effects of policy and other shocks. This is currently at the frontier
of developments in tourism modeling, in terms of both theoretical and empirical contri-
butions.
The second area concerns the incorporation of more microeconomic information
into CGE models of tourism. This is an area that is in the forefront of research on CGE
modeling (Bourguignon et al., 2002; Hertel et al., 2001). The incorporation of detailed
information at the level of individual households consumption and firms production
behavior and their interactions with the macroeconomic representations of economic
behavior characterized by CGE models would improve the quantity and quality of data
as well as the accuracy and insights available from the analysis.
Econometric models can provide more accurate estimates of the parameter values
that are included in CGE models, relating to more disaggregated levels of analysis,
providing improved means of policy formulation. Improved data at the regional and
local levels would also assist more effective policy formulation, along with better coor-
dination between policy making at the local, regional, national, and international levels.
The building of SAM matrices including tourism activities and the application of CGE
models to tourism economics, have recently become easier with the ongoing development
of Tourism Satellite Accounts (TSAs) internationally. These have provided substantial
increases in the quantity and quality of the data that can be used in CGE modeling.
However, this information needs to be complemented by tourism modeling if they are to
provide industry and governments with effective guidance for dealing with the range of
events and policy decisions that have to be made on an ongoing basis.
A third area concerns the use of CGE models to estimate the welfare effects or net
benefits of some demand shock. Typically, CGE models focus on measures of economic
activity such as changes in GDP rather than on explicit measures of economic welfare.
However, if the economy wide effects of a positive demand shock are so large that there
is no change in national output then the net gains will be small. This is recognized by
Dixon and Rimmer (2002) and Blake et al. (2006, 2008a) who introduce welfare meas-
ures into their economic modeling of demand shocks. The measure favored by Blake
is the equivalent variation the amount of money that leaves a person as well off as
they would be after a change in economic activity. That is, it measures the amount of
money required to maintain a persons satisfaction, or economic welfare, at the level it
would be at after the change in economic activity. To overcome the limitations of relying
solely on GDP as a measure of the benefits accruing to Australians from policy changes,
Australias Productivity Commission uses adjusted Gross National Expenditure as a
welfare indicator. Adjusted GNE is an indicator of the real national income accruing to
the residents from the current value of production (GDP). It is the income accruing to
the residents from their labor and their savings invested at home and abroad. It is defined
as GDP less the net income paid to foreign investors. On this measure, changes in welfare
resulting from tourism demand shocks relate to changes in adjusted GNE. In contrast,
Dixon (2009) estimates welfare effects with reference to taxation revenues accruing from
foreign-financed capital expansion, efficiency effects, and effects of changes in the terms
of trade on net exports. Recently, Coleman (2008) has discussed the superiority of real

DWYER 9781781001288 Part I.indd 286 25/07/2012 11:07

 CGE modeling 287

gross domestic income over GDP as an indicator of welfare in the presence of changes
to the terms of trade. Analysis of the advantages and limitations of the different welfare
measures may be expected to increasingly occupy the attention of CGE modelers.
A fourth area concerns policy analysis. CGE modeling can be applied to a much
broader set of policy issues than at present. The agenda for future research in this area
should be to extend the analysis to different tourism destinations, and to include detailed
analyses of the appropriate behavioral characteristics of the economic agents that are
included in model specification and of the government policy settings that determine the
context for their behaviour. This could include the effects of foreign direct investment in
tourism, tourism productivity, and competitiveness, fiscal policies for tourism, policies
within wider international groupings such as the European Union, policies for transpor-
tation, the environment and related externalities, including the impacts of climate change
on tourism. CGE tourism modeling provides a versatile and effective means of examin-
ing the wide range of scenarios that can occur.
As a result of the development of more sophisticated modeling techniques in recent
years, the study of the economic impacts of tourism has become a more fertile ground
for research and policy evaluation.

REFERENCES

Abelson, P. (2011), Evaluating major events and avoiding the mercantilist fallacy, Economic Papers, 30 (1),
March, 4859.
Adams, P. (2008), MMRF: Monash Multi-Regional Forecasting Model: A Dynamic Multi-Regional Applied
General Equilibrium Model of the Australian Economy, Monash University: Centre of Policy Studies.
Adams, P.D. and B.R. Parmenter (1995), An applied general equilibrium analysis of the economic effects of
tourism in a quite small, quite open economy, Applied Economics, 27 (10), 985994.
Adams, P.D. and B.R. Parmenter (1999), General Equilibrium Models, Valuing Tourism. Methods and
Techniques, Canberra, Australia: Bureau of Tourism Research.
Adams, P., M. Horridge and G. Wittwer (2003), MMRF GREEN: a dynamic multi-regional applied com-
putable general equilibrium model of the Australian economy, based on the MMR and Monash model,
General Working Paper No. G-140, October, Centre of Policy Studies, Monash University, Australia.
Alavalapati, J.R.R. and W.L. Adamowicz (2000), Tourism impact modeling for resource extraction regions,
Annals of Tourism Research, 27 (1), 188202.
Arrow, K.J. and G. Debreu (1954), The existence of an equilibrium for a competitive economy, Econometrica,
XXII, 265290.
Arrow, K.J. and F.H. Hahn (1971), General Competitive Analysis, San Francisco: Holden-Day.
Atkinson, G., S. Mourato, S. Szymanski and E. Ozdemiroglu (2008), Are we willing to pay enough to back
the bid? Valuing the intangible impacts of Londons bid to host the 2012 Summer Olympic Games, Urban
Studies, 45, 419444.
Bandara, J.S. (1991), Computable general equilibrium models for development policy analysis in LDCs,
Journal of Economic Surveys, 5 (1), March, 369.
Blake, A. (2005), The economic impact of the London 2012 Olympics, Research Report 2005/5, Christel
DeHaan Tourism and Travel Research Institute, Nottingham University Business School.
Blake, A., T. Sinclair and J. Gillham (2006), CGE tourism analysis and policy modeling, in L. Dwyer and
P. Forsyth (eds), International Handbook of Tourism Economics, Cheltenham, UK and Northampton, MA,
USA: Edward Elgar.
Blake, A., J.S. Arbache, M.T. Sinclair and V. Teles (2008), Tourism and poverty relief , Annals of Tourism
Research, 35 (1), 107126.
Blake, A.T. (2000), The economic effects of tourism in Spain, Christel DeHaan Tourism and Travel Research
Institute Discussion Paper, available at http://www.nottingham.ac.uk/ttri/series.html, 2000/2.
Blake, A.T. and M.T. Sinclair (2003), Tourism crisis management: US response to September 11, Annals of
Tourism Research, 30 (4), 813832.

DWYER 9781781001288 Part I.indd 287 25/07/2012 11:07

 288 Handbook of research methods in tourism

Blake, A., M.T. Sinclair and G. Sugiyarto (2003a), Tourism and EU accession in Malta, Cyprus, TTRI
Discussion Paper 2003/7, University of Nottingham, Nottingham.
Blake A., M. Sinclair and G. Sugiyarto (2003b), Quantifying the impact of foot and mouth disease on tourism
and the UK economy, Tourism Economics, 9 (4), 449465.
Bourguignon, F., A. Robilliard and S. Robinson (2002), Representative versus real households in the macro-
economic modelling of inequality, paper presented at the Development Economics Study Group Annual
Conference, University of Nottingham, available at http://www.shef.ac.uk/uni/projects/desg/cinfo/cpapers/
robinson.pdf.
Coleman, W. (2008), Gauging economic performance under changing terms of trade: Real Gross Domestic
Income or Real Gross Domestic Product?, Economic Papers, 27 (4), 329342.
Copeland, B.R. (1991), Tourism, welfare and de-industrialization in a small open economy, Economica, 58,
515529.
Costa, M.L. (1998), General Equilibrium Analysis and the Theory of Markets, Cheltenham, UK, and Lyme,
NH, USA: Edward Elgar.
Croes, R. and D. Severt (2007), Evaluating short-term tourism economic effects in confined economies: con-
ceptual and empirical considerations, Tourism Economics, 13 (2), 289307.
Debreu, G. (1959), Theory of Value, New York: Wiley.
Dixon, P. and B. Parmenter (1996), Computable general equilibrium modelling for policy analysis and
forecasting, in H. Aman, D. Kendrick and J. Rust (eds), Handbook of Computational Economics, Vol. 1,
Melbourne: Elsevier Science B.V, pp. 485.
Dixon, P. and M. Rimmer (2002), Dynamic, General Equilibrium Modelling for Forecasting and Policy: a
Practical Guide and Documentation of MONASH, Amsterdam: North-Holland.
Dixon, P.B. (2009), Comments on the productivity commissions modelling of the economy-wide effects of
future automotive assistance, Economic Papers, 28 (1), March, 1118.
Dixon, P.B., B.R. Parmenter, J. Sutton and D.P. Vincent (1982), ORANI: A multisectoral model of the
Australian Economy, North-Holland, Amsterdam.
Dwyer, L. and P. Forsyth (2008), Economic measures of tourism yield: what markets to target?, International
Journal of Tourism Research, 10, 155168.
Dwyer, L. and P. Forsyth (2009), Public sector support for special events, Eastern Economic Journal, 35 (4),
481499.
Dwyer, L., P. Forsyth and W. Dwyer (2010), Tourism Economics and Policy, Cheltenham, UK: Channel View
Publications.
Dwyer, L., P. Forsyth, L. Fredline, L. Jago, M. Deery and S. Lundie (2007), Yield measures for Australias
special interest inbound tourism markets, Tourism Economics, 13 (3), 421440.
Dwyer, L., P. Forsyth, J. Madden and R. Spurr (2000), Economic impacts of inbound tourism under different
assumptions regarding the macroeconomy, Current Issues in Tourism, 3 (4), 325363.
Dwyer, L., P. Forsyth, R. Spurr and T. Van Ho (2003), The contribution of tourism to a state and national
economy: a multi-regional general equilibrium analysis, Tourism Economics, 9 (4), 431448.
Dwyer, L., P. Forsyth and R. Spurr (2004), Evaluating tourisms economic effects: new and old approaches,
Tourism Management, 25, 307317.
Dwyer, L., P. Forsyth and R. Spurr (2005), Estimating the impacts of special events on the economy, Journal
of Travel Research, 43 (May), 351359.
Dwyer, L., P. Forsyth and R. Spurr (2006a), Assessing the economic impacts of events: a computable general
equilibrium approach, Journal of Travel Research, 45, 5966.
Dwyer, L., P. Forsyth and R. Spurr (2006b), Assessing the economic impacts of special events, in L. Dwyer
and P. Forsyth (eds), International Handbook of Tourism Economics, Cheltenham, UK and Northampton,
MA, USA: Edward Elgar.
Dwyer, L., P. Forsyth and R. Spurr (2006c), Effects of SARS crisis on the economic contribution of tourism
to Australia, Tourism Review International, 10, 4755.
Forsyth, P., S. Hoque, L. Dwyer, Tien Duc Pham and R. Spurr (2011), The Impacts of the Passenger
Movement Charge on Tourism Output and the Economy, report commissioned by Tourism Research
Australia, The Centre for Economics and Policy, Sydney, Australia.
Giesecke, J.A and J.R. Madden (2011), Modelling the economic impacts of the Sydney Olympics in retrospect:
game over for the bonanza story?, Economic Papers, 30 (2), June, 218232.
Gillham, J. (2004), Economic interrelationships of tourism: a computable general equilibrium analysis, PhD
thesis, University of Nottingham.
Gmez Gmez, C.M., J. Lozano Ibez and J. Rey-Maquieira Palmer (2003), Tourism taxation in a dynamic
model of an economy specialised in tourism, Conference on Tourism Modelling and Competitiveness,
Paphos, Cyprus, October.
Gooroochurn, N. and C. Milner (2003), Assessing indirect tax reform in a tourism-dependent developing
country, Conference on Tourism Modelling and Competitiveness, Paphos, Cyprus, October.

DWYER 9781781001288 Part I.indd 288 25/07/2012 11:07

 CGE modeling 289

Gooroochurn, N. and T. Sinclair (2005), The economics of tourism taxation: evidence from Mauritius,
Annals of Tourism Research, 32 (2), 478498.
Gunning, J.W. and M. Keyzer (1995), Applied general equilibrium models for policy analysis, in J. Behrman
and T.N. Srinivasan (eds), Handbook of Development Economics Vol. III-A, Amsterdam: Elsevier, pp.
20252107.
Harrison, G., S. Jensen, L. Pedersen and T. Rutherford (2000), Using dynamic general equilibrium models for
policy analysis, in G. Harrison, S. Jensen, L. Pedersen and T. Rutherford (eds), Contributions to Economic
Analysis Vol. 248, Oxford and New York: Elsevier, North Holland.
Hertel, T.W. (1997), Global Trade Analysis: Modelling and Applications, Cambridge: Cambridge University
Press.
Hertel, T.W., P.V. Preckel, J.A.L. Cranfield and M. Ivanic (2001), Poverty impacts of multilateral trade
liberalization, unpublished research paper available from the Center for Global Trade Analysis, Purdue
University, USA.
Horridge, J.M., J. Madden and G. Wittwer (2003), Using a highly disaggregagted multi-regional single-
country model to analyse the impacts of the 200203 drought on Australia, Centre of Policy Studies,
General Working Paper No. G-141 Oct 2003, Monash University, Australia.
Horridge, J.M., B.R. Parmenter and K.R. Pearson (1993), ORANI-F: a general equilibrium model of the
Australian economy, Economic and Financial Computing, 3, 71140.
Hoque, S., P. Forsyth, L. Dwyer, R. Spurr and D. Pambudi (2009), Economic effects of an emissions trading
scheme on the Australian tourism industry: a dynamic CGE analysis, paper presented at conference of
International Association for Tourism Economics (IATE), Chiang Mai, Thailand, December 2009.
Li, S., A. Blake and C. Cooper (2010), Chinas tourism in a global financial crisis: a computable general equi-
librium approach, Current Issues in Tourism, 13 (5), 435453.
Li, S., A. Blake and C. Cooper (2011), Modelling the economic impact of international tourism on the Chinese
economy: a CGE analysis of the Beijing 2008 Olympics, Tourism Economics, 17 (2), 279303.
McDougall, R. (1995), Computable general equilibrium modelling: introduction and overview, Asia-Pacific
Economic Review, 1 (1), April, 8891.
Mabugu, R. (2002), Short-term effects of policy reform on tourism and the macroeconomy in Zimbabwe:
applied CGE analysis, Development Southern Africa, 19, 419430.
Madden, J. (2006), Economic and fiscal impacts of mega sporting events: a general equilibrium assessment,
Public Finance and Management, 6 (3), 346394.
Meagher, G. and B. Parmenter (1990), ORANI-NT: a multisectoral model of the Northern Territory
economy, The Australian National University, North Australia Research Unit, Darwin.
Meng, X., M. Siriwardana, B. Dollery and S. Mounter (2010), The impact of the 2008 world financial crisis on
tourism and the Singapore economy and policy responses: a CGE analysis, International Journal of Trade,
Economics and Finance, 1 (1), June.
Narayan, P.K. (2003), The long-run impact of coups on Fijis economy: evidence from a computable general
equilibrium model, Journal of International Development, 19, 149160.
Narayan, P.K. (2004), Economic impact of tourism on Fijis economy: empirical evidence from the comput-
able general equilibrium model, Tourism Economics, 10 (4), 419433.
Pambudi, D., N. McCaughey and R. Smyth (2009), Computable general equilibrium estimates of the impact
of the Bali bombing on the Indonesian economy, Tourism Management, 30.
Peter, M., M. Horridge, G.A. Meagher, Z. Naqvi and B.R. Parmenter (1996), The theoretical structure of
Monash-MRF, Preliminary Working Paper No. OP-85 April 1996, Centre of Policy Studies, Monash
University.
Pham Tien Duc, D.G. Simmons and R. Spurr (2010), Tourism: adapting to climate change and climate
policy, Journal of Sustainable Tourism Special Issue: 18 (3), 449473.
Sheng, L. and Y. Tsui (2009), A general equilibrium approach to tourism and welfare: the case of Macao,
Habitat International, 33, 419424.
Shoven, J.B. and J.L. Whalley (1992), Applying General Equilibrium, Cambridge: Cambridge University Press.
Sugiyarto, G., A. Blake and M.T. Sinclair (2003), Tourism and globalization: economic impact in Indonesia,
Annals of Tourism Research, 30 (3), 683701.
Victorian Auditor-General (2007), State Investment in Major Events. Melbourne: Victorian Printer.
Wattanakuljarus, A. and I. Coxhead (2008), Is tourism-based development good for the poor? A general
equilibrium analysis for Thailand, Journal of Policy Modeling, 30 (6), 929955.
Zhou, D., J.F. Yanagida, U. Chakravorty and P. Leung (1997), Estimating economic impacts of tourism,
Annals of Tourism Research, 24 (1), 7689.

DWYER 9781781001288 Part I.indd 289 25/07/2012 11:07

 14 Costbenefit analysis
Larry Dwyer

INTRODUCTION

Decision makers need a consistent basis for assessing competing proposals and to be
fully informed about the implications of using economic resources. The public sector
is a major user of a destinations available resources and, as such, should ensure that it
makes a significant positive contribution to the economy and society. More particularly,
no new proposal, programme, project or policy should be adopted without first answer-
ing questions such as:

What are the specific outcomes sought?

Do the gains to people exceed the costs the sacrifices required?
Are there better ways to achieve these outcomes?
Are there better uses for these resources?
Is this an appropriate area of responsibility for government?

The primary technique that should be used for the economic appraisal of actions or
proposals in terms of economic efficiency is CostBenefit Analysis (CBA). CBA is
a systematic process for identifying and assessing all costs and benefits of a policy,
project or programme in monetary terms, including those costs and benefits not usually
represented by dollar values, and then subtracting the costs from the benefits to show
the estimated net effect of that activity. Future costs and benefits are discounted rela-
tive to present costs and benefits in a net present value sum. The policy or project is
deemed to be socially acceptable if the sum of the benefits to society (including private
and social benefits) exceeds the sum of the costs to society (including private and social
costs).
CBA is the most comprehensive of the economic appraisal techniques. CBA is
particularly important in the context of evaluating tourism policy, programmes,
regulations, projects and developments. Specific examples might include regional
or local tourism plans; rezoning of land for tourism purposes; and major tourism
developments such as the creation of tourism shopping precincts, airport develop-
ment, resorts and hotels, nature reserves and sporting facilities. As such projects
generally have relatively wide economic, environmental and social implications for a
community that are not captured in the basic financial analysis undertaken by project
proponents.
CBA is especially appropriate to tourism because there is often a clear trade-off
between economic benefits and social costs of some programme, policy, project, invest-
ment or proposal (henceforth referred to as proposal). Tourism proposals that might
be economically beneficial may be rejected because of their adverse environmental and/
or social impacts.

290

DWYER 9781781001288 Part I.indd 290 25/07/2012 11:07

 Costbenefit analysis 291

NATURE OF CBA AND ITS EVOLUTION

CBA is a commonly applicable technique in public investment assessment developed by

Dupuit (1844). Throughout the last three decades, the focus on CBA as a vehicle for eco-
nomic efficiency appraisal of public projects has increased. Several studies (Boardman
et al., 2001; Jenkins and Harberger, 1995; Mishan, 2002) have shown the crucial role of
accurate valuation of benefits and costs in public projects evaluation. CBA is a system-
atic process for identifying and assessing all costs and benefits of a proposal in monetary
terms, as they are expected to occur through the life of a project. CBA is concerned with
measuring, the change in all sources of economic welfare, whether occurring in markets
or as implicit values.

Benefits and Costs

A CBA aims to estimate all the costs and benefits of an activity to a specified community.
It is an evaluation method that has been designed specifically to answer welfare or public
policy questions. Benefits are defined as increases in social welfare while costs are defined
as reductions in social welfare. A CBA provides an estimate of the worth of a proposal
relative to an accompanying estimate of what would happen in the absence of the pro-
posal (Dwyer et al., 2010, chapter 10).
In CBA, value or benefit is measured by willingness to pay (WTP) or willingness
to accept (WTA). Formally using the compensating variations principle, the net social
benefit is the maximum net amount that residents would be willing to pay for the pro-
posal and be just as well off with the proposal as without it. Using the equivalent varia-
tions principle, the net social benefit is the minimum amount that the community would
be willing to accept as compensation for not having the proposal. The social costs of a
project are measured in terms of opportunity costs that is, the value of the marginal
benefits foregone from the same resources in alternative uses. Using these two valuation
principles (for benefits and costs) the analyst can determine whether the value of con-
sumption gained is greater than the value of consumption that it has given up. The net
benefit is the sum of all welfare benefits less costs. Maximizing net welfare is the standard
policy objective implicit or explicit in costbenefit studies.
All costs and benefits, including social and environmental aspects, are assigned a
money value, allowing the calculation of the net benefits of different proposals as a basis
for evaluating alternatives. CBA are used to include, measure, weight and compare all
expected present and future benefits of a policy with all its expected present and future
costs. Future costs and benefits are discounted relative to present costs and benefits in
the net present value (NPV) sum. For a capital investment development, programme or
policy to be socially acceptable, the sum of the benefits to society (including private and
social benefits) must exceed the sum of the costs to society (including private and social
costs). These net benefits/costs can then be used to rank alternative proposals quantita-
tively. In principle, a CBA enables agencies to compare the relative merit of alternative
programs or projects in terms of their returns on the use of resources, public and private.
CBA is primarily designed to answer the question does the expenditure of money on this
particular project or programme provide a net benefit to the economy and the public,
given that these resources could be applied in an alternative use?

DWYER 9781781001288 Part I.indd 291 25/07/2012 11:07

 292 Handbook of research methods in tourism

By quantifying the net benefits of projects, programmes and policies in a standard

manner, CBA improves the information base for public sector decision making, thereby
assisting in the assessment of relative priorities.
We can highlight several issues involved in CBA.

Scope and Objectives of the Analysis

The initial step in a CBA is to outline the nature of the problem to be addressed: its back-
ground, context and rationale. Every proposal to spend money must have an underlying
objective. What is the programme, project or activity trying to achieve? Thus a destina-
tion manager may enquire as to the net benefits expected from an international tourism
promotion campaign or the construction of additional airport capacity. These objectives
should be defined, initially, in terms of the possible market failure or market imperfec-
tions that could warrant government intervention. A clear statement of objectives will
provide information on whose costs and benefits are being assessed.
CBA must also identify and specify a set of alternatives. A basic principle of any
type of project or proposal evaluation is the withwithout principle. In practice,
this means that the forecast state of the world with the proposal to be undertaken,
is compared with the state of the world that would have existed in the absence of
the project (the do nothing or status quo option). For example, government deci-
sion makers may assess the benefits of a tourism training programme, or bilateral air
services regulations by comparing scenarios with and without such schemes. A do
nothing option is generally required as a base because costs and benefits are always
measured as incremental to what would have happened had the project or policy not
been undertaken.
To ensure that all alternatives for meeting the identified objectives are feasible, various
constraints that may apply need to be considered, including financial, distributional,
managerial, environmental or other constraints.

Discounting

The costs and benefits flowing from a project will be spread over time. For purposes
of CBA, the selection of a discount rate introduces some complexities not relevant to
financial appraisal (Burgess and Zerbe, 2011). While debate continues at the theoretical
level, in reality analysts use the market interest rate to discount the costs and benefits
associated with different projects. The essential argument is that the government must
achieve a return on investment at least equivalent to what the money would earn if left
in the private sector to justify taxing the private economy to undertake public-sector
investments. If the government cannot achieve this by investing in some project it would
be better for the destination if the money is left untaxed in the private sector (Treasury
Board of Canada, 1998). A real discount rate should be used to discount constant dollar
or real benefits and costs. A real discount rate, eliminating the effects of expected infla-
tion, can be approximated by subtracting expected inflation from a nominal interest rate.
Benefits and costs should be expressed in real terms and adjusted for differential price
effects where a specific resource price is expected to move at a rate different from the
general inflation rate.

DWYER 9781781001288 Part I.indd 292 25/07/2012 11:07

 Costbenefit analysis 293

Decision Criteria

Various investment criteria are available to assist in reaching decisions on different

proposals. In CBA, the net social benefit (NSB), or the excess of total benefit over total
cost, is represented by the net present value (NPV) of the proposal. Under this deci-
sion rule, a project is potentially worthwhile (or viable) if the NPV is greater than zero,
that is, the total discounted value of benefits is greater than the total discounted costs.
If projects are mutually exclusive, the project which yields the highest NPV would be
chosen.
Net Present Value is the sum of the discounted project benefits minus the discounted
project costs. Formally, it can be expressed as follows:

aK b
T Bt T Ct
NPV 5 a 2 1 a (14.1)
t51 1 1 r
( ) t51 1 1 r)
(
t t

where:

S means summation over all the years of the project life (T years),
Bt stands for the benefits expected in the tth year,
(1 1 r)t represents the discounting factor by which values expected in the future are
turned into todays values.
K is the capital or construction costs (assumed to occur in the current year) and
Ct is the operation, maintenance, and repair costs expected in the tth year.
T is number of years that the benefits or costs are produced.

Under this decision rule, a project is potentially worthwhile (or viable) if the NPV is
greater than zero, that is, the total discounted value of benefits is greater than the total
discounted costs. If projects are mutually exclusive, the project which yields the highest
NPV would be chosen.
Sometimes the decision rule takes the form of a Benefit-Cost Ratio (BCR). This is the
ratio of the present value of benefits to the present value of costs. In algebraic terms, the
BCR can be expressed as follows:
T B 2 C
t t
a (1 1 r) t
t51
BCR 5 (14.2)
K
It is conventional to split costs into two types when calculating BCRs: initial capital
costs and ongoing costs. Ongoing costs in each year are subtracted from benefits in
that year to identify a net benefit stream, while initial capital costs are used as the
denominator.
A project is potentially worthwhile if the BCR is greater than 1, that is, the present
value of benefits exceeds the present value of costs. Thus:

If BCR > 1 implementation of the project is judged to be economically worthwhile.

If BCR 5 1 the project adds nothing to economic welfare.
If BCR < 1 the project reduces economic well being.

DWYER 9781781001288 Part I.indd 293 25/07/2012 11:07

 294 Handbook of research methods in tourism

If projects are mutually exclusive, this rule would indicate that the project with the
highest BCR should be chosen.

The BCR is a useful measure because, when there are a large number of proposals, there
may not be enough resources available to undertake them all, even if they all have high
net present values. As a rule of thumb, picking the projects with the highest BCR can
ensure maximum value for money in terms of contributing to outcomes, though only if
all the projects have similar time patterns of benefits and costs.
The choice of the discount rate can have a significant impact on the ranking of options/
projects and hence their choice. In general, short lived options are favoured by higher
discount rates relative to long-lived options.
As the discount rate rises:

The more are net benefits further into the future downgraded in present value
terms relative to net benefits closer to hand.
Projects with larger initial outlays become relatively less attractive compared with
projects with lower initial outlays.
Projects with lower ongoing outlays become relatively more attractive compared
to projects with higher ongoing outlays.

Distributional Impacts

In reality, it is highly likely that a change in resource allocation, such as the introduction
of a new tourism resort or new tourism policy or programme, will increase the welfare or
well being of some people but reduce the welfare of others. One approach to recognizing
this in a CBA is to identify the gainers and losers, and measure the extent of their gains
and losses in monetary terms. This information is then provided to the decision maker,
who then must make the distributional value judgments. Alternatively, others (Ray,
1984) have suggested that distributional weights be assigned to the benefits and costs
experienced by different groups in society (for example, with higher weights being given
to benefits and costs to poorer groups than benefits and costs to richer groups). These
weights would be provided by the government. The criterion for proceeding with the
project then depends on whether weighted benefits exceed weighted costs. While analyti-
cally this approach has its attractions, the problem of finding the appropriate weights
makes it difficult to implement in practice.

Investment Horizon

The investment horizon is the end of the period over which costs and benefits are com-
pared to ascertain whether an investment is acceptable. If costs and benefits can be iden-
tified over the useful life of the project and uncertainties are low, then this provides the
best investment horizon. This will be a long time for some tourism infrastructure projects
such as airports, highway systems and cruise shipping terminals. As a general rule, the
period of analysis should extend over the useful life of the project.
Residual values are also an important component of the total value of a proposal and
should be included in the assessment. The residual or terminal value of a project is its

DWYER 9781781001288 Part I.indd 294 25/07/2012 11:07

 Costbenefit analysis 295

estimated value at the end of the analysis period. While, in theory, an assets residual
value at the end of its economic life should be zero, in practice it may not be so.

Sensitivity Analysis

The values of future costs and benefits on which the NPV is based cannot be known
with certainty. While they should be forecast on expected values, it is important to test
the NPV for optimistic and pessimistic scenarios. This is known as sensitivity analysis
and is a critical component of any CBA. Sensitivity analysis can help to highlight those
factors (e.g., exchange rates, salary costs, demand drivers, project timings) that require
especially careful assessment or management. Sensitivity analysis is particularly useful in
assessing the effects of different discount rates on NPV calculations. Thus the net benefits
of developing a special event can be estimated for distinct cost scenarios based on differ-
ent allocation of stadium infrastructure costs (Access Economics, 2010).

Ex-post Evaluation

Ex-post evaluations involve:

Re-evaluation of the benefits and costs of the selected option to assess whether the
anticipated benefits were realized and the forecast costs were accurate;
Reconsideration of alternative options;
Examination of the project design and implementation to assess the scope for
improvement to the option adopted.

By re-examining these issues, ex-post evaluations will assist in the development and
evaluation of future projects. Flyvbjerg et al. (2003) demonstrate the value of the post
implementation review given the failures of megaprojects to perform according to expec-
tations.

VALUING COSTS AND BENEFITS

All economic evaluations in CBA are based on the incremental costs and benefits associ-
ated with a particular proposal. These can be classified usefully into three categories:

(1) Effects which can be Readily Identified and Valued in Money Terms

These could include ticket sales to residents for a special event, revenues generated from
restaurant meals, costs of waste disposal and so on. The relevant costs and benefits are
typically based upon market prices as they are the easiest to identify and usually reflect
implicit opportunity costs. Where competitive markets exist, prices reflect WTP at the
margin for goods and the opportunity costs of resources.
While market prices provide the starting point for measuring WTP, they sometimes
do not adequately reflect the true value of a good to society. Consumers are often willing
to pay more than the market price rather than go without a good they consume for

DWYER 9781781001288 Part I.indd 295 25/07/2012 11:07

 296 Handbook of research methods in tourism

example, a person might be willing to pay much more to attend a festival event than the
ticket price.
When market prices exist but are distorted for some reason, we need to estimate
what the prices would be in the absence of the distortions and then use these adjusted
market prices. These adjusted prices are called shadow prices (sometimes called social
prices or true prices). The use of shadow prices rather than market prices is most com-
monly advocated for the following market distorting situations: labour markets (when
unemployment and/or underemployment exists or wages are artificially high); foreign
exchange market (where the value of foreign exchange is different from the market rate);
market power (distortion of market prices by monopoly seller or monopsony buyer);
government regulations, taxes and subsidies. While theoretically interesting, the above
types of distortions may not have much effect on the CBA estimates. In many cases
given the potentially small changes that result to the net benefit calculation it may not
be worthwhile for the analyst to confront the difficult measurement problems associated
with shadow pricing.

(2) Effects which can be Identified and Measured in Physical Terms but which Cannot be
Easily Valued in Money Terms Because of the Absence of Market

These could include the value of preserving a wilderness area, the cost of noise from
airport or hotel construction, time savings to travellers from a new transport link or
greenhouse gas emissions from motor coaches and so. There are often cases where a
market does not exist for the goods and services flowing from a project, or market prices
are not directly observable or easy to estimate. This makes it difficult to estimate costs
and benefits (or even to determine to whom the costs and benefits accrue). For example,
undertaking a project may result in benefits received or costs incurred by others not asso-
ciated with the activity and for which payment is neither given nor received. Such spillo-
vers are termed externalities (also referred to as social effects or third party effects).
Externalities are impacts on third parties that are not the primary parties (producers
or consumers) in an economic exchange and hence are not required to pay for a cost
imposed or benefit received. This is particularly the case for policies and programmes
that generate social and/or environmental impacts. Examples of negative spillovers
abound in tourism. The construction of a golf course resort hotel that discharges residues
into a nearby river will have a negative impact on people who use the river downstream
of the resort. Similarly, the greenhouse gas emissions from aircraft affect the environ-
ment of people on the ground.
The two main general approaches to valuing externalities are the revealed preference
and stated preference methods.

Revealed preference
Revealed preference is based on consumer behaviour in related markets, and compares
situations where people have made trade-offs between a cost and some form of benefit.
This information indicates the extent to which people are prepared to pay for a given
benefit. Revealed preference techniques are based on preferences from actual, observed,
market-based information. These preferences for environmental goods and services are
revealed indirectly when individuals purchase marketed goods which are related to the

DWYER 9781781001288 Part I.indd 296 25/07/2012 11:07

 Costbenefit analysis 297

environmental good in some way. For example, there are markets for certain goods
to which environmental commodities are related, as either substitutes or complements
to the goods in question. In this way peoples purchasing behaviour in actual markets
reflect, to a certain extent, their preferences for environmental assets. Thus, people often
pay a higher price for a hotel room with a view of the ocean, or will take the time to
travel to a special spot for swimming or bird watching.These kinds of expenditures can
be used to place a lower bound on the value of the view or the recreational experience.
Two approaches often used in valuing tourism related benefits and costs are the Travel
Cost Method and Hedonic Pricing.

Travel cost method The travel cost method (TCM) estimates the economic values asso-
ciated with environmental attributes or sites that are used for recreation. Thus, peoples
WTP to visit the site can be estimated based on the number of trips that they make given
different travel costs. This is analogous to estimating peoples WTP for a marketed
good based on the quantity demanded at different prices. A basic assumption is that
the time and travel cost expenses that people incur to visit a site represent the price of
access to the site. The cost of travel is considered to be a proxy for WTP (Bedate et al.,
2004; Rosato, 2008; Ward and Beal, 2000). The TCM has been applied to demand esti-
mation in various tourism contexts including: hiking and biking (Hesseln et al., 2003);
fishing (Morey and Breffle, 2006); snorkeling (Park et al., 2002), visits to islands (Chen
et al., 2004), national parks (Liston-Heyes and Heyes, 1999), beach resorts (Bell and
Leeworthy, 1990), coral reefs (Carr and Mendelsohn, 2003); and the demand for rock
climbing (Shaw and Jakus, 1996).

Hedonic pricing The hedonic pricing method (HPM) assumes that people value the
attributes of a good, or the services it provides, rather than the good itself.The price that
a person pays will reflect the value of the set of attributes that he/she considers important
when purchasing the good. On this assumption, we can value any individual attribute of
any good by looking at how the price people are willing to pay for it changes when that
attribute changes. In valuing environmental attributes tourism researchers have used
HPM to estimate the value of aircraft noise (Espey and Lopez, 2000; Rahmatian and
Cockerill, 2004), scenic views (Baddeley, 2004; Monty and Skidmore, 2003), urban wet-
lands (Mahan et al., 2000), recreational and aesthetic value of water (Lansford and Jones,
1995), beach pricing (Wilman, 1981); and clean air (Harrison and Rubinfeld, 1978).

Stated preference
Stated preference techniques involve asking people about their preferences to identify the
trade-offs they are prepared to make as regards costs and benefits under certain hypo-
thetical scenarios. The approach simulates a market by estimating a consumers WTP for
the good or service, or WTA compensation to tolerate a negative or adverse economic
outcome. The two basic stated preference approaches that use surveys to elicit value
through peoples responses to given hypothetical situations are contingent valuation and
contingent choice.

Contingent valuation The contingent valuation method (CVM) is the most widely used
method for estimating non-use value. The basic principle of CVM is that people have

DWYER 9781781001288 Part I.indd 297 25/07/2012 11:07

 298 Handbook of research methods in tourism

preferences in relation to all goods including goods that are not available in any existing
market. CVM attempts to reveal these hidden preferences by means of questionnaires.
People are asked the maximum amount of money they are willing to pay (or willing to
accept as compensation) for a hypothetical change in the quantity or quality of a good.
It is assumed that this professed willingness to pay would equate to actual willingness if
a real market for the good did exist. The method is called contingent valuation because
people are asked to state their willingness to pay (accept), contingent on a specific hypo-
thetical scenario and description of the environmental amenity. CVM can be used to
put dollar values on non-market values of the environment, including everything from
the basic life support functions associated with ecosystem health or biodiversity, to the
enjoyment of a scenic vista or a wilderness experience, to appreciating the option to
go hiking, fishing or rafting in the future or the right to bequest those options to ones
grandchildren. It also includes the value people place on simply knowing that elephants,
whales and rainforests exist. CVM studies have been published on wildlife and the envi-
ronment (Chase et al., 1998; Giraud et al., 2002; Nunes and Nijkamp, 2009), the value
of coral reefs (Brander et al. (2006) and various outdoor recreation and ecotourism set-
tings :value of forests (Bostedt and Mattsson (1995); marine reserve management (Bhat,
2003); nature-based tourism resources (Lee, 1997); landcape (Willis and Garrod, 1993);
national parks (Bateman and Langford, 1997); protected areas (Dharmaratne et al.,
2000); ecotourism resources (Lee et al., 1998); wildlife (Solomon et al., 2004), and the
economic benefits of rare and endangered species (Loomis and White, 1996). CVM has
also been applied to valuing cultural heritage (Beltran and Rojas, 1996; Bille Hansen,
1997; Chambers et al., 1998; Lockwood et al., 1996; Pollicino and Maddison, 2001;
Noonan, 2003; Salazar and Marques, 2005).

Contingent choice The Contingent Choice Method (CCM) asks respondents to state a
preference between one group of environmental attributes, at a given price or cost to the
individual, and another group of environmental attributes at a different price or cost.
WTP (accept) can then be inferred from the hypothetical choices that include cost as an
attribute (Hanley et al., 2001; Louvier et al., 2000). CCM, also referred to as conjoint
analysis, measures preferences for different attributes of a multi-attribute choice. CCM
is similar to CVM, in that it can be used to estimate economic values for many environ-
mental services, but differs in that it does not directly ask individuals to state their values.
Instead, values are inferred from the hypothetical choices or trade-offs that people make
between sets of attributes. Because it focuses on trade-offs among scenarios with differ-
ent characteristics, CCM is especially suited to policy decisions where a set of possible
actions might result in different impacts on natural resources or environmental services
(Rahim, 2008).

(3) Effects that are Known to Exist but Cannot be Precisely Identified and Accurately
Quantified, Let Alone Valued

These could include crime prevention effects of tourist police programmes, outcomes
of a publically funded tourism and hospitality training programme, aesthetic effects
of streetscape beautification programmes, loss to humans from species extinction and
so on. Costs and benefits that cannot be valued in money terms are often described as

DWYER 9781781001288 Part I.indd 298 25/07/2012 11:07

 Costbenefit analysis 299

intangibles. Examples of intangibles in tourism projects include the development

of special themed events to improve destination image, branding or community
pride. Similarly the construction of a large-scale hotel adjacent to an attractive beach
might be aesthetically displeasing or socially alienating to local residents. Lost assets
as a result of tourism development may be considered to have heritage value. Effects
that cannot reasonably be quantified in monetary terms should not be ignored in the
CBA. If the externalities are unable to be quantified, they should at least be identified
and explained to decision makers. Intangible benefits and costs can sometimes be sig-
nificant in relation to the quantitative impacts, and heavily influence the final accept
reject decision. If they appear to be significant, they should be explicitly highlighted and
explained in the analysis so that decision makers are aware of the values underpinning
a particular option. This explanation can be quantitative, qualitative, descriptive, or
some combination.
While it is desirable to give a monetary value to as many benefits and costs as possible
within the resource and time constraints, often not all benefits and costs require valu-
ation in order to obtain relevant information from benefit cost analysis. For instance,
in some situations the valuation of only some benefits may be sufficient to indicate that
benefits exceed costs.

APPLICATIONS OF CBA TO TOURISM

CBA can be used to guide a wide range of decisions on types of tourism development,
especially within the following four broad contexts:

Analysing Capital Expenditure

Many projects involve capital expenditure for new or replacement facilities. Such
projects might include airport construction or expansion, the development of cruise ship-
ping terminals, a resort development or highway construction. Larger capital projects,
including buildings, equipment and other forms of infrastructure and productive invest-
ment, should be subjected to an analysis of their costs and benefits over their lifetime.
Key questions are whether or not to undertake the investment, whether to undertake
it now or later, and which option to choose. It is in the transportation area that most
CBA appear to have been conducted (Litman, 2010). Examples of tourism related CBA
studies include assessment of options for expanding capacity (runways and terminal
facilities) at Heathrow airport (Department of Transport, 2009); investment in high
speed rail infrastructure (De Rus, 2011); estimating the benefits of tourism development
in a National Park in Madagascar (Mercer et al., 1995); tourism development in South
Carolina (Hefner et al., 2001); economic evaluation of the Channel Tunnel (Anguera,
2006).

Analysing a Policy Option

In principle, any proposal or policy option can be subjected to CBA. A basic principle
of any type of project evaluation is the withwithout principle. In practice, this means

DWYER 9781781001288 Part I.indd 299 25/07/2012 11:07

 300 Handbook of research methods in tourism

that the forecast state of the world with the proposal to be undertaken, is compared
with the state of the world that would have existed in the absence of the project (the
do nothing or status quo option). Thus, a government may wish to estimate the costs
and benefits of some policy regarding requirements for tourist visas, bilateral aviation
agreements, tourism and hospitality training programmes or restrictions on the use of
migrant labour. Policies almost always confer benefits on some parties and impose costs
on others. CBA helps to uncover unanticipated costs and benefits of projects. It explicitly
estimates the size of gains and losses for affected individuals and groups. This informa-
tion is important in public sector decision making and should be made explicit because it
is important to identify those who stand to gain and lose from a programme or project.
These costs and benefits can be valued in the same way as costs and benefits arising from
capital expenditures. Recent tourism examples are CBA of protection of the northern
beaches of Australias Gold Coast (Raybould and Mules, 1999); road safety measures
(Elvik, 2001); and a smoking ban in the Dutch hotel and catering sector (Spreen and
Mot, 2008).

Retaining or Disposing of an Existing Asset

CBA can be undertaken where an agency is considering the retention or disposal of an

existing asset. A government agency may be considering closing a rail link to a tourist
region or closing a museum, cruise terminal facilities or zoo. CBA can address issues
such as whether or not to sell land, whether to relocate facilities, and whether to repair
an asset or to replace it. The potential benefit of an economic appraisal of assets is to
improve the allocation of public sector resources to ensure the governments objectives
are met to the maximum extent as are community benefits. Tourism-related examples
include assessing the economic benefits of removing dams and restoring rivers (Loomis,
1996) and the benefits to inter-urban rail passenger services of retaining a railroad (Joray
and Kochanowski, 1978).

Post-evaluation of a Project or Program

CBA provides a means of determining whether or not a particular programme or project

has generated a net benefit for the community. Thus, tax concessions offered to hotel
developers or tourism and hospitality training programmes may be reviewed to assess
the net benefits, and a government agency may investigate whether its decision to
provide financial support for a special event such as a Formula One Grand Prix event
or an Olympic Games generates an appropriate return on public funds. CBA is particu-
larly well suited to evaluating special events. Examples include CBAs of the Eurovision
Song Contest in Israel (Fleischer and Felsenstein, 2002), V8 Car Races in the Australian
Capital Territory (ACT Auditor General, 2002), the Vancouver Winter Olympics 2010
(Shaffer et al., 2003); a Formula One Grand Prix event in Melbourne (Victorian Auditor
General, 2007). On the basis of a CBA, it is possible for the decision maker to make a
judgment of whether the economic benefits of the event are greater than the costs, and
to also judge whether the event would represent the best use of the funds, when funds
are limited and alternative calls on funds exist. In the case of the Vancouver Winter
Olympics and both car races, the CBA revealed that net losses to society would occur

DWYER 9781781001288 Part I.indd 300 25/07/2012 11:07

 Costbenefit analysis 301

even though economic impact analyses projected positive effects on GDP and employ-
ment. Clearly, there are many potential effects of events that are often not accounted for
in a standard economic impact analysis. In order for the government to be more compre-
hensively apprised event assessments need to be broadened to take, where practicable,
a more comprehensive approach embracing not only economic but social and environ-
mental factors. Once these wider effects of events are acknowledged we can appreciate
why estimation of the economic impacts of events is only part of the evaluation story.
To determine the extent (if any) of government assistance to be provided, it is necessary
that the cost of these funds be compared to the wider costs and benefits from the event
(Abelson, 2011).

ADVANTAGES AND LIMITATIONS OF CBA

CBA has a number of strengths but also limitations.

Strengths

A systematic method
CBA is a systematic and consistent evaluation method that facilitates comparisons
between different options. It has a well-developed theoretical foundation neoclassical
welfare economics which is based on the individual being the best judge of their own
welfare and the welfare of society being the sum of the welfare of individuals. Its use pro-
motes the efficient use of resources. The template character of CBA permits the decision
maker to determine the adequacy of the information collected and see important infor-
mation is missing. This knowledge provides the decision maker with valuable insight into
the level of ignorance regarding important attributes of the policy. CBA makes propo-
nents consider costs and benefits that are external to the proponent. It forces the decision
maker to think in a rational way about the costs and benefits of alternative actions. It
can provide a clear focus on net benefits of a proposal without regard to who wins and
who loses from projects and programmes. CBA encourages clear consideration of the
true value added from a proposal by focusing on incremental net benefits. Its emphasis on
the quantification of costs and benefits on a comparable basis can provide a useful hard
edge to an evaluation strategy.

Transparent
The results of a well-executed CBA can be clearly linked to the assumptions, theory,
methods and procedures used in it. This transparency can add to the accountability of
public decisions by indicating where the decisions are at variance with the analysis. It can
also improve the political process by uncovering gains and losses which might otherwise
be neglected in the bargaining process between different stakeholders of some project,
programme or policy.

Comparability
CBA attempts to capture in a single index all the features of a policy decision that affects
the well being of society. The single-metric approach permits the comparison of policies

DWYER 9781781001288 Part I.indd 301 25/07/2012 11:07

 302 Handbook of research methods in tourism

that affect different attributes of well being differently, that is, it permits the decision
maker to compare apples and oranges on the basis of a single attribute (the index of
social welfare) common to both.

Limitations

Some values cant be monetized

CBA often includes subjective assumptions regarding non-economic values. While
the analysis usually highlights the various qualitative costs and benefits, assessing the
actual quantitative values can be quite contentious. Where the use of value judgments
and assumptions becomes unavoidable, the quantification of intangibles is particularly
susceptible to manipulation. Environmental protection is often desirable for reasons
that cannot be quantified social, spiritual, and psychological values that defy moneti-
zation. And what is the economic value of a countrys enhanced international reputa-
tion and standing following the hosting of a mega event, or the destruction of aspects
of cultural heritage through tourism development, or of increases in customer and
employee satisfaction? In sum, there are things that humans cannot put a price tag on.
The upshot is that CBA can yield dramatically different numbers, especially as differ-
ent methodologies for assigning economic values to non-economic benefits can vary
significantly.
By way of response, exponents of CBA should of course endeavour to value
whatever can be quantified and valued reliably within the resource constraints of the
situation. The remaining intangible effects should be listed and described as fully as
possible, and some consideration given to the impact of the intangible component
of the CBA on the results. Economic values are inferred from the choices made
by individuals. The worth of a rainforest might be inferred from the things that
one gives up to see the rainforest (e.g., the cost of travel to the location). To econo-
mists,the importance of things (tangible or intangible) is revealed by what a person
will give up to obtain them. The lower bound on the value of the item obtained is
equated to what is given up. It is sometimes suggested that some non-market effects
cannot be valued or have infinite values. However, finite values for non-market
effects are routinely explicit in government decisions and the behaviour of consum-
ers. Fordifficulttoquantifyvalues, it is therefore important that analysts make their
assumptions clear and that margins for error be identified or sensitivity testing be
undertaken.

Individual versus social welfare

Critics highlight the presumption of CBA that the well being of society can be
defined as some aggregation of the well being of individual members of that society,
and the methods by which the aggregation is performed. As Sagoff (1998) argues,
economists use the term social welfare as a proxy for the satisfaction of prefer-
ences and then trivially and speciously argue that the satisfaction of preferences
producessocialwelfare. On the contrary, common wisdom suggests that it is not the
satisfaction but the content and quality of desires that correlate with what people
mean by welfare or well being. The criticism here is relatively straightforward that
the economic value of something is not related to the well being that a person enjoys

DWYER 9781781001288 Part I.indd 302 25/07/2012 11:07

 Costbenefit analysis 303

as a result of that thing. Researchers do not seem to have adequately addressed this
type of criticism of CBA.

Future generations
It is sometimes argued that the activity of discounting implies that the needs of future
generations are given less weight than those of the current generation. Traditional CBA
tends to give little weight to costs that occur far in the future and overly emphasizes
short-term gain. Critics have argued that intergenerational equity is fostered if the dis-
count rate is low and some claim that this is appropriate when environmental sustain-
ability is an issue. In response, it should be noted that a low discount rate can make
projects seem profitable and can encourage their implementation even if they are at
odds with sustainability. Assuming no irreversible environmental effects are involved, a
project with a high rate of return when all of its costs and benefits are counted is better
for the present generation and, through reinvestment, better for future generations as
well. Only when benefits are non-renewable and consumed rather than reinvested is
there conflict across generations, with one generation paying and another benefitting
(Treasury Board of Canada, 1998).
It is important that intertemporal equity issues form an integral part of each decision-
making process in CBA and research is progressing in respect of this issue. Approaches
to selecting real discount rates fall into two broad groups, both of which have given rise
to a wide range of recommended rates a descriptive approach based on the opportu-
nity cost of drawing funds from the private sector; and a prescriptive approach that
derives from ethical views about intergenerational equity. Debate on this issue continues
(Harrison, 2010).

Equity considerations
It is often argued that CBA takes the existing distribution of income as given and does
not consider the equity implications of the policies that it seeks to evaluate. This criticism
emphasizes the anonymous manner in which the welfare changes of individuals are aggre-
gated to obtain estimates of the change in social welfare. Efficiency is measured without
regard to who would get the benefits and who would incur the costs. Questions related
to distribution of income are not taken up for consideration. The criticism is valid as far
as it goes. Anonymous weighting of individual welfare does not take equity into account.
However, distributional information can be supplied as an adjunct to CBA. One way in
which distributional aspects can be integrated into CBA is by assigning weights to benefits
received and costs borne by different socio-economic groups, e.g., giving higher weight
to the poorer categories. Another way of handling the distributional aspect is to establish
distributional constraint as an additional criterion. It is then up to decision makers and
the political process to identify the appropriate trade-off between equity and efficiency.

Complexity
CBA has been criticized for being complex and too onerous in its information require-
ments. Not surprisingly, there are many gaps in the scientific knowledge and many
deficiencies in the analytic techniques necessary to complete these analyses. However,
in some situations this may simply reflect the complexity of public decisions and, in
most cases, unnecessary complexity should and can be avoided. Use of CBA in tourism

DWYER 9781781001288 Part I.indd 303 25/07/2012 11:07

 304 Handbook of research methods in tourism

contexts may generally be less onerous on data collection and analysis than for environ-
mental decision making. Even simple applications of the technique can yield valuable
information for proponents and decision makers.

FURTHER DEVELOPMENTS AND APPLICATIONS OF CBA IN

TOURISM RESEARCH

The discussion indicates that there is substantial scope for greater use of CBA in
tourism contexts. To date, there has been relatively little attempt to undertake detailed
CBA studies of plans, policies or regulations that either directly or indirectly affect
tourism stakeholders. The preferred technique among tourism economists appears to
be economic impact analysis. However, estimates of the economic impacts of tourism
plans, policies, developments or programmes provide, in themselves, an imperfect
basis for decisions about resource allocation. While economic impact analysis (CGE
modelling) picks up general equilibrium effects which the partial CBA is not capable
of detecting, including changes in income to households, CBA picks up a whole range
of benefits and costs due to non priced effects arising from the absence of markets for
some goods and services affected which would not be included in economic impact
analysis.
CBA is not without its challenges as the discussion of limitations highlights. Each type
of limitation is being addressed in the research literature and CBA remains the primary
technique for informing policy makers of the net benefits of different strategies to achieve
objectives associated with the public good.
One area of tourism in which CBA is expected to play an increasingly important role
is in the assessment of special events. The success or contribution of a particular event
should not be measured only by its direct financial contribution but needs to consider
the wider economic, social and environmental impacts that are invariably associated
with it. There is a growing awareness among researchers that event assessment which
focuses only on economic impacts is too narrow in scope to provide sufficient informa-
tion to policy makers and government funding agencies and that, where practical, a
more comprehensive approach should be employed to embrace the importance of social
and environmental impacts in addition to economic impacts. Similar reasoning supports
greater use of CBA generally in tourism research.

REFERENCES

Abelson, P. (2011), Evaluating major events and avoiding the mercantilist fallacy, Economic Papers, 30 (1),
4859.
Access Economics (2010), Cost benefit analysis of the FIFA 2022 World Cup, report by Access Economics
Pty. Ltd for Department of Resources, Energy and Tourism, Canberra, Australia.
ACT Auditor General (2002), ACT Auditor Generals Office Performance Audit Report V8 Car Races in
Canberra: Costs and Benefits, Canberra, ACT.
Anguera, R. (2006), The Channel Tunnel: an ex post economic evaluation, Transportation Research Part A:
Policy and Practice, 40 (4), 291315.
Baddeley, M.C. (2004), Are tourists willing to pay for aesthetic quality? An empirical assessment from Krabi
Province, Thailand, Tourism Economics, 10 (1), 4561.

DWYER 9781781001288 Part I.indd 304 25/07/2012 11:07

 Costbenefit analysis 305

Bateman, I. and I. Langford (1997), Non-users willingness to pay for a national park: an application and
critique of the contingent valuation method, Regional Studies, 31 (6), 571578.
Bedate, A., L.C. Herrero and J.. Sanz (2004), Economic valuation of the cultural heritage: application to
four case studies in Spain, Journal of Cultural Heritage, 5 (1), 101111.
Bell, F. and V. Leeworthy (1990), Recreational demand by tourists for Saltwater Beach days, Journal of
Environmental Economics and Management, 18, 189205.
Beltran, E. and M. Rojas (1996), Diversified funding methods in Mexican archeology, Annals of Tourism
Research, 23 (2), 463478.
Bhat, M. (2003), Application of nonmarket valuation to the Florida Keys Marine Reserve management,
Journal of Environmental Management, 67, 315325.
Bille Hansen, T. (1997), The willingness-to-pay for the Royal Theatre in Copenhagen as a public good,
Journal of Cultural Economics, 21, 128.
Boardman, A.E., D.H. Greenberg, A.R. Vining and D.L. Weimer (2001), CostBenefit Analysis. Concepts and
Practice, 2nd edition, Upper Saddle River, NJ: Prentice Hall.
Bostedt, G. and L. Mattsson (1995), The value of forests for tourism in Sweden, Annals of Tourism Research,
22 (3), 671680.
Brander, L.M., P van Beukering and S.J. Cesar (2006), The recreational value of coral reefs: a meta-analysis,
IVM Working Paper: IVM 06/07, Institute for Environmental Studies, Amsterdam.
Burgess, D.F. and R.O. Zerbe (2011), Appropriate discounting for benefit-cost analysis, Journal of Benefit-
Cost Analysis, 2 (2), 118.
Carr, L. and R. Mendelsohn (2003), Valuing coral reefs: a travel cost analysis of the Great Barrier Reef ,
Ambio, 32 (2), 353357.
Chambers, C., P. Chambers and J. Whitehead (1998), Contingent valuation of quasipublic good: validity, reli-
ability, and application to valuing a historic site, Public Finance Review, 26 (2), 137154.
Chase, L., D. Lee, W. Schulze and D. Anderson (1998), Ecotourism demand and differential pricing of
national park access in Costa Rica, Land Economics, 74 (4), 466482.
Chen, W., H. Hong, Y. Liu, L. Zhang, X. Hou and M. Raymond (2004), Recreation demand and economic
value: an application of travel cost method for Xiamen Island, China Economic Review, 15, 398406.
Department for Transport (2009), Adding capacity at Heathrow Airport: impact assessment, DfT
Publications, London, January.
De Rus, G. (2011), The BCA of HSR: should the government invest in high speed rail infrastructure?, Journal
of Benefit-Cost Analysis, 2 (2), 126.
Dharmaratne, G.S., F. Yee Sang and L.J. Walling (2000), Tourism potentials for financing protected areas,
Annals of Tourism Research, 27 (3), 590610.
Dupuit, A.J..J. (1844), De la mesure de lutilit des travaux publics, Annales des ponts et chausses, Second
series, 8. Translated by R.H. Barback as On the measurement of the utility of public works, International
Economic Papers, 1952, 2, 83110.
Dwyer, L., P. Forsyth and W. Dwyer (2010), Tourism Economics and Policy, Cheltenham, UK: Channel View
Publications.
Elvik, R. (2001), Cost-benefit analysis of road safety measures: applicability and controversies, Accident
Analysis and Prevention, 33, 917.
Espey, M. and H. Lopez (2000), The impact of airport noise and proximity on residential property values,
Growth and Change, 31, 408419.
Fleischer, A. and D. Felsenstein (2002), Cost-benefit analysis using economic surpluses: a case study of a tel-
evised event, Journal of Cultural Economics, 26 (2), 139156.
Flyvbjerg, B., N. Bruzelius and W. Rothengatter (2003), Megaprojects and Risk: An Anatomy of Ambition,
Cambridge: Cambridge University Press.
Giraud, K., B. Turcin, J. Loomis and J. Cooper (2002), Economic benefit of the protection program for the
Steller sea lion, Marine Policy, 26, 451458.
Hanley, N., S. Maurato and R. Wright (2001), Choice modelling approaches: a superior alternative for envi-
ronmental valuation?, Journal of Economic Surveys, 15 (3), 435462.
Harrison, D., Jr., and D.L. Rubinfeld (1978), Hedonic housing prices and the demand for clean air, Journal
Environmental Economics and Management, 5, 81102.
Harrison, M. (2010), Valuing the future: the social discount rate in cost-benefit analysis, Visiting Researcher
Paper, Productivity Commission, Canberra.
Hefner, F., J. Crotts and J. Flowers (2001), The costbenefit model as applied to tourism development in the
state of South Carolina, Tourism Economics, 7 (4), 163175.
Hesseln, H., J.B. Loomis, A. Gonzalez-Caban and S. Alexander (2003), Wildfire effects on hiking and biking
demand in New Mexico: a travel cost study, Journal of Environmental Management, 69, 359368.
Jenkins, G. and A.C. Harberger (1995), Cost-Benefit Analysis of Investment Decisions, Cambridge, MA:
Harvard Institute for International Development.

DWYER 9781781001288 Part I.indd 305 25/07/2012 11:07

 306 Handbook of research methods in tourism

Joray, P. and P. Kochanowski (1978), Inter-urban rail passenger service: social benefits of retaining the South
Shore Railroad (Chicago), Logistics and Transportation Review, 14 (1), 8189.
Lansford, N.H. Jr. and L.L. Jones (1995), Recreational and aesthetic value of water using hedonic price analy-
sis, Journal of Agricultural and Resource Economics, 20 (2), 341355.
Lee, C. (1997), Valuation of nature-based tourism resources using dichotomous choice contingent valuation
method, Tourism Management, 18, 587591.
Lee, C., J. Lee and S. Han (1998), Measuring the economic value of ecotourism resources: the case of South
Korea, Journal of Travel Research, 36 (Spring), 4047.
Liston-Heyes, C. and A. Heyes (1999), Recreational benefits from the Dartmoor National Park, Journal of
Environmental Management, 55, 6980.
Litman, T. (2010), Transportation cost and benefit analysis, Victoria Transport Policy Institute (www.vtpi.
org).
Lockwood, M., K. Tracey and N. Klomp (1996), Analyzing conflict between cultural heritage and nature
conservation in the Australian Alps: a CVM approach, Journal of Environmental Planning and Management,
39 (3), 357370.
Loomis, J. (1996), Measuring the economic benefits of removing dams and restoring the Elwha River: results
of a contingent value survey, Water Resources Research, 32, 441447.
Loomis, J. and D. White (1996), Economic benefits of rare and endangered species: summary and meta-
analysis, Ecological Economics, 18 (3), 197206.
Louvier, J., D. Henscher and J. Swait (2000), Stated Choice Methods: Analysis and Applications, Cambridge:
Cambridge University Press.
Mahan, B.L., S. Polasky and R.M. Adams (2000), Valuing urban wetlands: a property price approach, Land
Economics, 76 (1), 100113.
Mercer, E., R. Kramer and N. Sharma (1995), Rainforest tourism: estimating the benefits of tourism develop-
ment in a new national park in Madagascar, Journal of Forest Economics, 1 (12), 239269.
Mishan, E.J. (2002), On the conceptual underpinning of a cost-benefit analysis, The Singaporean Economic
Review, 47 (1), 116.
Monty, B. and M. Skidmore (2003), Hedonic pricing and willingness to pay for bed and breakfast amenities
in Southeast Wisconsin, Journal of Travel Research, 42, 195199.
Morey, E.R. and W.S. Breffle (2006), Valuing a change in a fishing site without collecting characteristics data
on all fishing sites: a complete but minimal model, American Journal of Agricultural Economics, 88 (1),
150161.
Noonan, D. (2003), Contingent valuation and cultural resources: a meta-analytic review of the literature,
Journal of Cultural Economics, 27 (34), 159176.
Nunes, P. and P. Nijkamp (2009), Contingent valuation method, Encora Coastal Portal, available at http://
www.coastalwiki.org.
Park, T., J.M. Bowker and V.R. Leeworthy (2002), Valuing snorkeling visits to the Florida Keys with stated
revealed preference models, Journal of Environmental Management, 65, 301312.
Pollicino, M. and D. Maddison (2001), Valuing the benefits of cleaning Lincoln Cathedral, Journal of Cultural
Economics, 25, 131148.
Rahim, K.A. (2008), Non-market valuation techniques, Economic Valuation of the Goods and Services
of Coastal Habitats The Regional Training Workshop, March 2428, 2008, Samut Songkram Province,
Thailand.
Rahmatian, M. and L. Cockerill (2004), Airport noise and residential housing valuation in Southern
California: a hedonic pricing approach, International Journal of Environmental Science & Technology, 1 (1),
1725.
Ray, A. (1984), Cost Benefit Analysis, Washington DC: The World Bank.
Raybould, M. and T. Mules (1999), A cost-benefit study of protection of the northern beaches of Australias
Gold Coast, Tourism Economics, 5 (2), 121139.
Rosato, P. (2008), Travel cost method, Encora Coastal Portal, http://www.coastalwiki.org.
Sagoff, M. (1998), Aggregation and deliberation in valuing environmental public goods: a look beyond con-
tingent pricing, Ecological Economics, 24, 213230.
Salazar, S. and J. Marques (2005), Valuing cultural heritage: the social benefits of restoring an old Arab
tower, Journal of Cultural Heritage, 6, 6977.
Shaffer, M., A. Greer and C. Mauboules (2003), Olympic costs and benefits, Canadian Centre for Policy
Alternatives Publication, February.
Shaw, W. and P. Jakus (1996), Travel cost models of the demand for rock climbing, Agricultural and Resource
Economics Review, 25, 133142.
Solomon, B., C. Corey-Luse and K. Halvorsen (2004), The Florida Manatee and ecotourism: toward a safe
minimum standard, Ecological Economics, 50, 101115.
Spreen, M. and E. Mot (2008), Een rookverbod in deNederlandse horeca, Een kosten-batenanalyse [A smoking

DWYER 9781781001288 Part I.indd 306 25/07/2012 11:07

 Costbenefit analysis 307

ban in the Dutch hotel and catering sector, a cost-benefit analysis], Netherlands Bureau for Economic
Analysis, Document 159, February (English summary, main report only in Dutch).
Treasury Board of Canada (1998), Benefit-cost analysis guide, Treasury Board of Canada Secretariat Ottawa.
Victorian Auditor General (2007), State investment in major events, Victorian Government Printer, Victoria.
Ward, F.A. and D. Beal (2000), Valuing Nature with Travel Cost Method: A Manual, Cheltenham, UK and
Northampton, MA USA: Edward Elgar.
Willis, K. and G. Garrod (1993), Valuing landscape: a contingent valuation approach, Journal of Environmental
Management, 37 (1), 122.
Wilman, E. (1981), Hedonic prices and beach recreation values, in V. Smith (ed.), Advances in Applied
Microeconomics, Volume 1, Greenwich, CT: JAI Press.

DWYER 9781781001288 Part I.indd 307 25/07/2012 11:07

DWYER 9781781001288 Part I.indd 308 25/07/2012 11:07
 PART II

QUALITATIVE RESEARCH
METHODS
Gayle R. Jennings

QUALITATIVE RESEARCH: ITS NATURE AND EVOLUTION

Qualitative research is a well-established approach to researching phenomena in the

social sciences. Comparatively, its application in the fields of tourism studies and man-
agement is a more recent occurrence dating from the late 1970s and 1980s. Increasingly,
however, in the early decades of the twenty-first century, qualitative research is gaining
broader acceptance within those fields. This acceptance is due to the ability of qualitative
research to provide rich, in-depth knowledge from multiple viewpoints along with its
emphasis on verstenhen, empathetic understanding, especially, with regard to the how
and why of tourism related phenomena and experiences.
In the past, the fields of tourism studies and management have tended to emphasize
quantitative research. That emphasis still continues in the early 2010s. Quantitative
research places an emphasis on erklren, causal or nomethetic (rule or law related) expla-
nation, particularly, who, when, what, where, along with how and why. Qualitative
research distinguishes itself from quantitative research by nature of the formers emic
approach to research. Emic approaches acknowledge the subjectivity of the researcher
within research processes as well as shape research to include the researcher as a subjec-
tively embedded being in the research process. Emic research has also been described as
insider research since the researcher usually brings background experiences that have
complementarity and/or empathy with the focus of the study. Rather than distort the
research, such background commonality facilitates the interpretive processes of the
research act and adds to the depth and diversity of interpretations.
On the other hand, quantitative research is associated with etic approaches to
research. Specifically, etic approaches are objectively situated and may best be framed as
research conducted from a neutral stance, that is, an outsider perspective. Such a stance
is assumed to remove or reduce bias from and distortions to analyses and findings. In
discussing emic and etic approaches, we are starting to consider aspects associated with
axiological positions upon which research may be situated.
Axiology is the study of values and ethics (Jennings, 2009), in this case, with respect
to research processes. Axiology is one of four frames that guide research processes
(Guba 1990; Jennings, 2009). The other frames are ontology (world view), epistemology

309

DWYER 9781781001288 Part II.indd 309 25/07/2012 11:07

 310 Handbook of research methods in tourism

(science of knowledge generation as well as the nature of the relationship between the
researcher and that which is to be known (Jennings, 2009, p. 674; see also Guba, 1990)
and methodology (principles for guiding the conduct of research). Each of these four
frames enables us to understand the differing suites of paradigms used in tourism studies
and management research.

PARADIGMS INFORMING TOURISM STUDIES RESEARCH

Broadly speaking, there are two clusters of paradigms. Those associated with qualitative
research that use qualitative methodologies and those linked to quantitative research,
that predominantly use quantitative methodologies. Within the latter set, there are
approaches that use mixed methods by drawing on both quantitative and qualita-
tive methods of research. Quantitative research is associated with the paradigms of
positivism and post-positivism.
Qualitative research is generally associated with (social) constructivism, interpretivism
critical theory orientation and participatory paradigms (Jennings 2009, 2010; Lincoln et
al. 2011; Schwandt, 2000). Examples of use of social constructivism in tourism studies are
found in the works of Gurney and Humphreys (2006) and Jennings (1999). Discussion of
social constructivism also occurs in OGormans (2005) work. An interpretive approach
is demonstrated in the work of Gurung (2008) and Jennings and Stehlik (2009). Critical
theory orientation as a research tool is used by Fox (2007) and Kensbock (2012/13) as
well as considered in the writings of Chambers (2007). An example of a participatory
approach is provided in the research of Bennett (2004) and Jennings et al. (2010). The
approach is also discussed in the writings of Darcy (2006) and Westwood (2007).
Amongst social science and tourism scholars, debates remain as to what should or
should not be included within the cluster of paradigms associated with qualitative
research. For example, feminist perspectives, cultural studies, queer theory, critical race
theory and postmodern paradigms are variously considered as part of the cluster of qual-
itatively associated paradigms. Others argue regarding the inclusion of mixed methods
approaches. Refer to Denzin and Lincoln (2011), Jennings (2009), Lincoln et al. (2011)
for further consideration of this debate.
Despite the preceding debates, within the fields of tourism studies and management,
the dominant, hegemonic research paradigms are positivism and post-positivism. The
cluster of paradigms bearing complementarity with qualitative research continues in
the meantime to grow in application within tourism studies and management research.
The difference between post/positivistic paradigms and paradigms that associate around
social constructivism are now considered further.
What distinguishes post/positivistic paradigms from social constructivism? The pre-
ceding discourse has hinted at some differences etic versus emic approaches; and quan-
titative versus qualitative methodological approaches. There are other differences arising
from the respective paradigms attitudes to ontology, epistemology, methodology and
axiology. Generally, post/positivistic paradigms adopt:

an ontology that views truth and laws as universal (positivism) or immutable and
shaped by historical and social circumstances (post-positivism);

DWYER 9781781001288 Part II.indd 310 25/07/2012 11:07

 Part II Qualitative methods 311

an epistemology that is objective (positivism) or objective but recognizes the

potential for researcher bias (post-positivism);
primarily a quantitative methodology. Some post-positivistic research may use
mixed methods; and
an axiology predicated on value-free attitudes and a focus on the extrinsic value of
research processes and outcomes. Some post-positivistic research may embrace an
emancipatory role for research.

As already noted, there is a suite of paradigms that constitute those associated with a
qualitative methodological research approach. For comparative purposes with post/
positivism, social constructivism has been selected from that suite. Social constructivism
adopts:

an ontology that embraces multiple realities;

an inter-subjective epistemology;
a focus on qualitative methodologies; and
a value-laden axiology, which emphasizes the intrinsic value of research processes
and research outcomes.

Table II.1 provides an overview of positivism and post-positivism, the hegemonic para-
digms, along with critical theory orientation, social constructivism and participatory
paradigms, the burgeoning paradigms. The table does not present all paradigms in use.
For fuller discussion, readers are referred to Denzin and Lincoln, 2011, Jennings (2009,
2010) and Lincoln et al. (2011).
Which paradigm should a researcher use? The answer should be determined by the
nature of the research questions, issues and tourism/leisure phenomena of interest. It
should not be determined based only on hegemonic research practice with regard to par-
adigm selection. Cohen (1988), Goodson and Phillimore (2004), Hollinshead (1996), and
Jennings (2010), Riley and Love (2000), Walle (1997) advocate that qualitative meth-
odologically informed research can contribute to understanding tourism phenomena
and should be used. Qualitative research serves to complement quantitative researchs
focus on explanation by emphasizing understanding. Both types of research have and do
make significant contributions to our development of knowledge in the fields of tourism
studies and management. The first section of this book has provided readers with insights
into the various research approaches and methods associated with positivism and post-
positivism. This, the second section of the book, considers a range of qualitative research
methods and approaches connected with social constructivism and related paradigms.

A COMMENTARY ON TOURISM STUDIES AND

MANAGEMENT RESEARCH

Before moving on to the background and research foci of qualitative research, as tourism
researchers, we need to take stock of significant criticism that has been directed at global
research efforts. Tourism studies and management research, like other disciplinary
and fields of studies research, is predicated on western-centric research processes. This

DWYER 9781781001288 Part II.indd 311 25/07/2012 11:07

 Table II.1 Comparison of major paradigms used in tourism studies and management research

Examples of paradigms Positivism Post-Positivism Critical Theory Constructivism/ Participatory

DWYER 9781781001288 Part II.indd 312

Interpretivism
Ontology (world view, Universal Fallible Realities are Realities and Realities are
perspective on the nature truths and truths that are socially and perspectives are constructed
of reality) laws. influenced by social and historically framed multiple in nature. collectively via
historical circumstances. and reflect power engagements
relations. between individuals
and others.
Epistemology (science of Objective, Objective, Subjective Intersubjective Reflexive,
knowledge construction; etic etic whilst recognizing objective (emicetic) (emic) situated and
relationship between potential biases inherent unless post-positivist embodied. Critical
researcher and resear- in researcher decision- critical theory subjectivity.
ched or participants making processes. (objective). Emic and etic.

312
Hermeneutic.
Methodology (research Quantitative Quantitative Qualitative Qualitative Qualitative.
guidelines) (Use of mixed Some Quantitative.
methods) quantitative Mixed method.
Axiology (study of values Value free. Value free Value laden Value laden. Value laden.
and ethics) Research emphasizes albeit research agendas. Research projects Transformation.
projects fallibility of researcher Intrinsic research associated with
purposes recognized. Extrinsic foci, political intrinsic foci.
are nature of research agendas, emanci-
extrinsic in purposes albeit some patory and transfor-
nature. approaches lean toward mative in nature.
emancipatory roles of
research.

Source: Derived from Denzin and Lincoln (2011), Guba (1990), Jennings (2009, 2010), Lincoln and Guba (2000), Lincoln et al. (2011), and Schwandt (2000).

25/07/2012 11:07
 Part II Qualitative methods 313

predication has been critiqued in various disciplinary and fields of studies extant lit-
erature, see, for example, Bourdieu (1990), Bourke (1995), Foley (2003), Ivanitz (1999),
Laclau (1990), Scheurich (1997), Smith (1999, 2005), Stanfield II (1994) and Urry (1990,
1996, 2002). Within tourism literature, the key critiques are associated with:

ignoring tourism phenomena external to western cultural and social knowledge

constructions (Berno, 1996; Hollinshead, 1992);
limited use of a variety of culturally bounded sign systems and their meanings to
collect, interpret and/or analyse empirical materials and/or data beyond western-
centric systems and meanings (Echtner, 1999);
paucity of integrated multicultural studies (Echtner, 1999);
need for emic, context-based and temporally situated research designs when com-
paring tourist settings and tourism cultural differences (Cohen, 1979; Dann et al.,
1988; Jennings 2007a; Sofield, 2000; Tribe, 2006);
need to acknowledge indigenous epistemologies and use indigenous methodologies
(Denzin et al., 2008; Kuokkanen, 2003; Schaper et al., 2007);
need for inclusion of differing ontologies, epistemologies, methodologies and axi-
ologies in tourism research beyond western-centric ones (Jennings, 2003, 2009).

As researchers we need to be cognizant of these critiques as well as address them in

our research enterprises. This is especially important when considering that qualita-
tive research seeks to achieve in-depth, empathetic understandings. In this book, Carla
Santos provides readers with her thoughts on cross-cultural research.

QUALITATIVE RESEARCH: BACKGROUND AND RESEARCH

FOCI

As already noted, qualitative research is predicated on verstenhen, empathetic under-

standing. It aims to gain detailed insights into phenomena by adopting worldviews
(ontologies) that recognize the multiplicity of realities and the manner in which these
realities are constructed. Knowledge (epistemology) is generated variously through
subjectiveobjective orientations in the case of critical theory orientation, intersubjec-
tivity in social constructivism, and reflexive and hermeneutic processes in participatory
research. In each case, the embodied and situated nature of participants and researchers
contribute to the co-construction of knowledge in interactions that are founded on tenets
of reciprocity (mutual exchange). Methodologically, as would be expected, the para-
digms that associate under the umbrella of qualitative research predominantly utilize
qualitative methodologies. Qualitative methodologies are focused on amassing empirical
materials (data) drawn from the empirical world.
A methodology is a set of guidelines for conducting research, in this case qualitative
research. The guidelines for qualitative methodologies are (Jennings, 2010):

use of an inductive approach to explore the nature of reality in real world contexts;
assumption of an ontological perspective that advocates multiple world views and
realities;

DWYER 9781781001288 Part II.indd 313 25/07/2012 11:07

 314 Handbook of research methods in tourism

emphasis on intersubjective relationships between research participants and

researcher(s);
embracing an axiology, which values propositional, transactional, experiential
knowledge that will facilitate social change and/or emancipation;
adoption of an intersubjective epistemology, which results in value laden ethics;
perception of researchers as insiders by study participants;
flexible, emergent research designs to enable researcher responsiveness to field set-
tings;
study-specific research designs grounded in the setting being studied;
researcher(s) focus/foci on emblematic themes arising during study processes and
stages;
non-probability sampling methods, where every person in the study population
does not have an equal chance of participation unless complete participation is
the focus;
collection of written, oral and visual empirical materials that are amassed as
textual units, instead of numeric representations;
interpretation/(re)construction of empirical materials to elicit key themes and
motifs synergistic with study participants world experiences;
representation of the findings using narrative-based written text forms, use of first
person active voice by the researcher(s); along with visual and performative texts,
as well as practices and praxes for participatory paradigms;
acknowledgment that the capture and representation of a slice of life from the
study setting is specific to only that study setting.

These guidelines in turn inform the methods of empirical material collection used. For
the purpose of this chapter, a method is deemed to be the tools for empirical material
collection and interpretation/(re)construction (Jennings, 2010).

METHODS OF EMPIRICAL MATERIAL COLLECTION

Qualitative research has a wide range of methods that may be used for empirical mate-
rial generation. These include fieldwork, diaries, journals, memo-ing and note-making,
participant observation, interviewing, visual methods, focus groups, documentary and
archival analysis and computer-mediated methods of empirical material collection
(Jennings, 2010).

METHODS OF EMPIRICAL MATERIAL INTERPRETATION

Qualitative research draws on a suite of methods for empirical material interpreta-

tion. The most used forms tend to be successive approximation, constant compari-
son, grounded theory interpretations, content analysis and deconstruction (Jennings,
2010).

DWYER 9781781001288 Part II.indd 314 25/07/2012 11:07

 Part II Qualitative methods 315

QUALITATIVE EMPIRICAL MATERIAL COLLECTION AND

INTERPRETATION

While the ensuing chapters address a variety of methods for empirical material collec-
tion and interpretation, in this chapter, I want to highlight five aspects of which qualita-
tive researchers must be cognizant during qualitative research processes. The particular
aspects to which I refer are qualitative vocabulary, qualitative sampling, reciprocity,
triangulation/crystallization and reflexivity.
I emphasize here that these are my perspectives informed by the perspectives and
discourses of others drawn from the broader area of social sciences research. Readers
should note that these five aspects may not be considered or addressed by the authors
of the following chapters. In fact, the authors may write using a vocabulary which is
contrary to the one I articulate below. It is for them to justify the use of their chosen
vocabulary and/or terminology just as I intend to do.

Qualitative Vocabulary

Qualitative research tends to use a different vocabulary than quantitative research

to describe elements and aspects of the research processes. Rather than use the term
data, often you will see the term empirical materials used. The terms interpretation
and (re)construction may be applied instead of analysis. Participants, co-researchers
rather than subjects, respondees or interviewees will be applied. Why this different
usage. Essentially, the terms data, analysis and subjects imply a passive role within
research processes. They resonate with actions occurring over which that which is to
be researched has no power. The research act is heavily imbued with power. Without
our participants or access to empirical materials we have no basis for research. Power is
vested in our participants or gatekeepers who hold access keys to other empirical mate-
rials, such as, documents and records. The research process is far from being a passive,
objective process.
Moreover, the qualitative research act is a social process of meaning making where
that which is to be researched actively engages with the researcher(s). Specifically, the
act is a co-construction of knowledge arising from intersubjective relationships between
participants and researcher(s). What is being created as knowledge in the qualitative
research act is not data but empirical world knowledge and realities originating in and
arising from empirical world materials. Such knowledge and realities are subjectively
and actively co-constructed.
You also will find that terms such as trustworthiness, authenticity, groundedness,
participant checks, resonance, originality, credibility, dependability, usefulness, transfer-
ability, confirmability, goodness of fit and boundedness are used to discuss the outcomes
of qualitative research processes. Terms used in quantitative research, such as, reliability,
validity and levels of confidence, and limitations, for example, are not used as qualitative
research adopts differing ontological, epistemological, methodological and axiological
principles. Being different does not mean of lesser value or less rigorous in research proc-
esses. It just means using different principles and attendant vocabulary to match those
principles.

DWYER 9781781001288 Part II.indd 315 25/07/2012 11:07

 316 Handbook of research methods in tourism

Qualitative Sampling

Qualitative research uses the principles of non-probability sampling. Qualitative

research is not about a specific set of numbers, cases or participants. It is about tar-
geted research, which includes participants and other empirical materials, which have
relevance to the research focus or foci. Such sampling is referred to as purposive
sampling. Selection of participants occurs bearing several points in mind: size, nature
and accessibility of the population of interest (Jennings, 2010). While various authors
advocate a specific number of participants for qualitative research, in reality, there is no
standard. Qualitative researchers continue to amass empirical material while simultane-
ously interpreting/(re)constructing the empirical materials until they reach a qualitative
informational isomorph (Lincoln and Guba, 1985, pp. 2334). This is also referred to as
theoretical saturation. At this stage, further amassment of empirical materials will yield
no further insights into the study phenomena. The interpretations are then participant
checked for authenticity with participant lived experiences of the research phenomena.

Reciprocity

Qualitative research is an intersubjective experience. Reciprocity is giving something

back in the research exchange. This means that during empirical material collection and
interpretation/(re)construction researchers actively engage in the research process. For
instance, interviews and focus groups are exchanges and rightfully, participants can ask
researchers about their opinions. Principles of reciprocity require researchers to share
their thoughts and insights with the participants.
Other forms of reciprocity may include gift giving as a thank you for the participants
time. Such gift giving could include chocolates, cakes, food items, books, gift vouchers
and a chance to win cash prizes. Careful consideration of gift giving needs to be under-
taken, the researcher wants participants to participate because they are interested in
sharing their insights with the researcher rather than as a means to receive a gift.

Triangulation/Crystallization

Previously, with regard to qualitative research, the term triangulation was used. In fact,
it is still a term used today. Triangulation is associated with establishing a fixed point as
an established reality. From a qualitative research perspective, such a reality is question-
able given an ontological worldview of multiple realities. Consequently, a contrary term
has gained more currency. That term is crystallization. The term emanates from the work
of Richardson (1994). Crystallization refers to the practice of using various methods to
illuminate as much of study phenomena as possible. Richardson (1994) tells us that when
viewing any phenomena we will only achieve partial views of it, somewhat like trying to
view the many facets of a crystal at the one time in a snapshot.
Relatedly, there cannot be a singular fixed reality within qualitative research.
Knowledge and reality are constantly being constructed and reconstructed and inter-
preted and reinterpreted. Realities and knowledges are always in flux and are part of
social processes, which are heavily value-laden because of the subjective and embodied
beings, who participate in knowledge construction. Subsequently, due to multiple reali-

DWYER 9781781001288 Part II.indd 316 25/07/2012 11:07

 Part II Qualitative methods 317

ties and perspectives, a better term for use is crystallization rather than triangulation (see
Jennings, 2010).

Reflexivity

Reflexivity refers to constantly looking back on the researchers roles and presence in
the field and the research process. Qualitative researchers are constantly questioning
themselves about the course of their research and the empirical materials that they gather
and the interpretations that unfold. When writing qualitative research texts, research-
ers should incorporate reflections of reflexivity in their work. They should also address
aspects of their researcher social situatedness with regard to the research phenomena
and/or participants. Readers need to be informed of the background, knowledge and
experiences that the researcher(s) bring(s) to the research setting. This enables readers
to become co-interpreters of the research during the course of reading research texts,
whether the text is an article, conference paper, thesis or technical report. It also enables
readers to be informed upon what bases the researchers proximity to the research focus
has informed interpretations, which as mentioned before should always be reflexively
articulated in qualitative texts.
In considering these five aspects of qualitative research, I have set the stage for con-
sideration of what elements or criteria should be used to judge the quality of qualitative
research.

CRITERIA FOR JUDGING QUALITATIVE RESEARCH

The following list provides a series of points, which can be used to determine the overall
quality of qualitative research. The list is intended primarily for evaluations of written
works. The list also may have applicability to oral and performative texts. The degree to
which each of these criteria is addressed will be influenced by:

(1) Audience, for example, academic, practitioner, general public, stakeholder group,
community, examiners, reviewers;
(2) Language, specifically, level of language, style of language, cross-cultural consid-
erations;
(3) Genre of the text, report, article, conference paper, poster, thesis, news-media
report/article, play, film, multimedia presentation;
(4) Context of the presentation, such as, conference, congress, workshop, performance,
gallery, stage, multimedia settings;
(5) Time limits in the case of oral presentations and performance;
(6) Word limits in the case of written texts.

Needless to say, in academic written contexts, the majority of the following criteria
should be addressed.

Research aim is clearly articulated;

Research design complements research aim;

DWYER 9781781001288 Part II.indd 317 25/07/2012 11:07

 318 Handbook of research methods in tourism

Researcher appears informed concerning area being researched (background lit-

erature, theoretical constructs, methodology and methods of empirical material
collection and interpretation);
Theoretical paradigm has been articulated and supported with lucid arguments
regarding relationship with research aim/focus;
Unpacking of ontology, epistemology, methodology and axiology has occurred;
Approach to ethics has been articulated;
Methods of empirical material collection clearly described, justified, supported
with examples of field notes, memos and text units (visual, aural, oral and print
based);
Methods of empirical material interpretation clearly described, justified and exem-
plified;
Theoretical sampling explained;
Theorizing clearly outlined and supported beyond initial first level of codings;
Evidence that the theory has been grounded with participants or has goodness of
fit, authenticity and trustworthiness;
Theory considered in regard to related extant literature, especially literature
beyond tourism and hospitality studies;
Researcher social situatedness explicitly stated so readers can follow interpreta-
tions made;
Reflexive processes identified and discussed;
Research setting and context clearly described and considered;
Recruitment of participants explained, access, ethics;
Numbers of participants explained and justified;
Reciprocity explained;
Writing follows a narrative style befitting qualitative research;
Writing adopts a first person active voice stance;
Reader hears the voices of participants or feels close to the phenomena being
shared.

In addressing the above points, researchers should provide readers with more than just
a checklist style of writing. They should incorporate a narrative style, which simultane-
ously fully integrates these elements and informs the reader.

Limitations

The discussion of limitations is a hangover from post/positivistic research. In choos-

ing qualitative research, the researcher has determined that it is the best approach for
illuminating the research focus or achieving insights into a research studys aim(s).
Consequently, there are no limitations to the research. There are, however, boundaries.
Each qualitative researcher explains to readers and stakeholders the bounded nature of
their study. The study is what the study is qualitative research does not masquerade as
being more than it is. Qualitative research focuses on providing a slice of life and offers
multiple perspectives and does not profess to identify a truth for all.
The bounded nature of qualitative research is influenced by research topic, researcher
insider-ness and social situatedness with regard to the research topic, connectedness of

DWYER 9781781001288 Part II.indd 318 25/07/2012 11:07

 Part II Qualitative methods 319

researcher(s) to participants and empirical materials, access to research participants

and materials, research scope and resources such as time, human, financial and mate-
rial resources. Obviously with regard to resources, qualitative research experiences
similar exposure to elements that bound research when using quantitative and mixed
methods research: time, human financial and material resources. When using the term,
material resources, I refer to technological equipment, audio, video recorders, cameras,
computer-related programs, applications and diverse databases.

QUALITATIVE RESEARCH: APPLICATIONS IN TOURISM

STUDIES

At the beginning of this chapter, I presented examples of applications of qualitative

research used in tourism studies based on paradigms. The studies that I selected were
chosen because of their demonstration of commensurability with qualitative research.
Readers are directed to each of the references associated with those examples for further
information. In the course of writing this chapter, I have also referred to seminal works
beyond the field of tourism studies as evidenced in the reference list. As researchers, we
are all required to familiarize ourselves with seminal and past works within and beyond
our field of study. There are no short cuts. To convince readers that you are an informed
qualitative researcher, you need to think, write, talk and act like one. So ensure that you
familiarize yourself and understand the related methodological literature as well as the
works that use qualitative research in tourism studies.
Following this chapter, a range of studies that have used qualitative research in
tourism studies are presented. In this section of the book, the authors of chapters
highlight their perspectives on a number of qualitative methods that may be utilized
in the course of qualitative research. A number of these authors focus on qualitative
research approaches, notably: archival research, Dallen Timothy; community case
studies,Dianne Dredge and Rob Hales; and cross cultural approaches,Carla Santos and
Changsup Shim. Others provide insights into the use of a range of qualitative methods
of empirical material collection: ethnographic approach, Kathleen Adams; participant
observation,Nuno Ribeiroand Eric Foemmel; interview techniques,Nancy McGehee;
and focus groups,Carl Cater and Tiffany Low. Olga Junek and Les Killion, focus on
qualitative research empirical material collection and interpretation in their chapter on
grounded theory.
Beyond these chapters, tourism and hospitality journals provide additional models,
demonstrations and applications of qualitative research. As you read such works, be crit-
ical and do not take all articles at face value. There are articles that have been published
that profess to be qualitative research which are not. Use the information contained in
this chapter and the following chapters to critically read and evaluate such works.
Additionally, critiques of published qualitative and quantitative research are useful to
iterate here. Bear these critiques in mind when you are preparing to share your research
with others. As you draft your writing, use these critiques as well as the previous criteria
for judging the quality of qualitative research as useful checks of your own writing. The
critiques note that authors of articles do not provide insights into their epistemological
stance or methodological processes (Weed, 2005), or paradigmatic positioning (Botterill,

DWYER 9781781001288 Part II.indd 319 25/07/2012 11:07

 320 Handbook of research methods in tourism

2001). Others do not familiarize readers with the theoretical relevance of their analy-
ses and interpretations (Dann et al., 1988). Still other research has been critiqued for
conceptual weakness and fuzziness (Cooper, 2003, p.1). Ensure that you address these
critiques and when conducting your methodological literature review, remember, not
all work published is quality work. Endeavour to make sure your work is quality work.

QUALITATIVE RESEARCH AND POLICY FORMULATION

The first decades of the twenty-first century have continuously highlighted the complexi-
ties of a globalized connected world, in which unexpected and unprecedented events are
becoming the norm and are constant challenges for governments, communities, environ-
ments, tourism experience consumers and providers.
Unpacking the how and why of tourism and hospitality phenomena and experi-
ences with in-depth insights grounded in the multiple realities of the participants and the
complexity of the worlds in which they live are critical for policy formulation and main-
tenance of quality of life of stakeholders now and for the future. Qualitative research
provides a slice of life, which enables policy makers to clearly understand the lived expe-
riences, expectations, needs and wants of their constituents. Qualitative research enables
creative developments of empirical material collection and interpretation. Creativity and
innovation is part and parcel of the qualitative research process because of the emergent
design embedded in qualitative research processes. Creativity and innovation is an area
that has been underdeveloped in educational systems and so qualitative research enables
researchers to develop these skill sets and capacities, which in turn enable creative and
innovative responses and strategies to be developed for planning and policy purposes.

QUALITATIVE RESEARCH: FURTHER DEVELOPMENTS AND

APPLICATIONS FOR TOURISM RESEARCH

Essentially, here I wish to emphasize the need for two particular developments with
regard to tourism research. First, the development and use of emic, holistic approaches
to research and, second, the facilitation of wider worldviews in tourism research enter-
prises.

Emic and Holistic Approaches

Given the complexity of the opening decades of the twenty-first century, questions
have to be raised regarding the use of etic (outsider/objective) models that are based
on dimensional delimitation. Such delimitation tends to focus on uni- (Pearce, 1993),
two-dimensional, or selected-dimensional perspectives and/or unicausal analysis (Ryan,
1991). This results in fragmented understandings of tourism phenomena a holistic
perspective is and will not be achieved. To achieve holistic perspectives, an emic (insider/
subjective) approach is required. Holistic-oriented approaches are able to deconstruct
the complexity (Lee et al., 1994) inherent in tourism phenomena. Qualitative research
assists with achieving holistic views.

DWYER 9781781001288 Part II.indd 320 25/07/2012 11:07

 Part II Qualitative methods 321

Over time, tourism research has variously supported an emic approach (see for
example Banner and Himmelfarb, 1985; Graburn, 1983, p. 28; Jennings, 1999, 2009;
Ryan and Kinder in Ryan, 1997). Yet wide-scale adoption of emic approaches lags
behind the unquestioning adherence to hegemonic tourism research practices premised
on etic practices. Research in tourism studies needs to adopt holistic and emic based
research in order to move beyond limited dimensional studies. The chapters in this
section of this book provide a number of examples to inform readers regarding how to
advance this agenda. Research courses and programmes need to be reviewed and revised
to equally privilege qualitative and quantitative research processes and to ensure emer-
gent researchers are versed in both approaches to research.

Wider World Views

As already noted, our reflexive lenses on tourism phenomena have been heavily founded
in western developed world contexts. The application of western-centric concepts and
dimensions assumes that these are universal and that different cultures and peoples of
the world use the same lenses to understand and interpret the world. Such universaliz-
ing is akin to imperialism and colonialism (Bourdieu, 1990; Laclau, 1990; Smith, 2005).
Western epistemologies are insufficient to support alternate knowledge bases with regard
to tourism phenomena. Tourism research needs to incorporate and promote ways of
knowing and researching that are not just predicated on western worldviews (Jennings,
2007b). Finally, tourism research needs to embrace multicultural research teams and
multicultural studies, which are emically focused and are contextually and temporally
situated.

REFERENCES

Banner, D.K. and A. Himmelfarb (1985), The work/leisure relationship toward a useful typology, Leadership
and Organization Development Journal, 6 (4), 2255.
Bennett, J. (2004), Indigenous entrepreneurship, social capital and tourism enterprise development: lessons
from Cape York, unpublished PhD thesis, La Trobe University, Australia.
Berno, T. (1996), Cross-cultural research methods: content or context? A Cook Island example, in R.
Butler and T. Hinch (eds), Tourism and Indigenous Peoples: Issues and Implications, London: International
Thomson Business Press, pp. 376395.
Botterill, D. (2001), The epistemology of a set of tourism studies, Leisure Studies, 20, 199214.
Bourdieu, P. (1990), The Logic of Practice, R. Nice (trans.), Cambridge: Polity Press.
Bourke, E. (1995), Dilemmas of integrity and knowledge: protocol in Aboriginal research, Paper presented at
the Indigenous Research ethics, Townsville. September.
Chambers, D. (2007), Interrogating the Critical in critical approaches to tourism research, in I. Ateljevic,
N. Morgan and A. Pritchard (eds), The Critical Turn in Tourism Studies: Innovative Research Methodologies,
Amsterdam: Elsevier.
Cohen, E. (1979), Rethinking the sociology of tourism, Annals of Tourism Research, 6, 1835.
Cohen, E. (1988), Traditions in the qualitative sociology of tourism, Annals of Tourism Research, 15,
2946.
Cooper, C. (2003), Progress in tourism research, in C. Cooper (ed.), Classic Reviews in Tourism, Aspects of
Tourism Series, Number 3, Bristol, UK: Channel View Press.
Dann, G., D. Nash and P. Pearce (1988), Methodology in tourism research, Annals of Tourism Research, 15,
128.
Darcy, S. (2006), Setting a research agenda for accessible tourism, Technical Report, Australia: CRC for
Sustainable Tourism.

DWYER 9781781001288 Part II.indd 321 25/07/2012 11:07

 322 Handbook of research methods in tourism

Denzin, N.K. and Y.S. Lincoln (2011), Introduction: the discipline and practice of qualitative research, in
N.K. Denzin and Y.S. Lincoln (eds), The Sage Handbook of Qualitative Research, 4th edition, Thousand
Oaks, CA: Sage Publications, pp. 120.
Denzin, N.K., S. Lincoln and L.T. Smith (eds) (2008), Handbook of Critical and Indigenous Methodologies,
Thousand Oaks, CA: Sage.
Echtner, C.M. (1999), The semiotic paradigm: implications for tourism research, Tourism Management, 20,
4757.
Foley, D. (2003), Indigenous epistemology and Indigenous standpoint theory, Social Alternatives, 22 (1), 4452.
Fox, K. (2007), Critical theory and Katrina, City, 11 (1), 8199.
Goodson, L. and J. Phillimore (2004), The inquiry paradigm in qualitative tourism research, in L. Goodson
and J. Phillimore (eds), Qualitative Research in Tourism: Ontologies, Epistemologies and Methodologies,
London: Routledge, pp. 3045.
Graburn, N.H.H. (1983), The anthropology of tourism, Annals of Tourism Research, 10, pp. 933.
Guba, E. (1990), The alternative paradigm dialog, in E. Guba (ed.), The Paradigm Dialog, Newbury Park,
CA: Sage.
Gurney, P.M. and M. Humphreys (2006), Consuming responsibility: the search for value at Laskarina holi-
days, Journal of Business Ethics, 64, 83100.
Gurung, H.B. (2008), Fusioning, PhD thesis, Griffith University, Gold Coast Australia.
Hollinshead, K. (1992), White gaze, red people-shadow visions: the disidentification of Indians in cul-
tural tourism, Leisure Studies, 11 (1), 4364.
Hollinshead, K. (1996), The tourism researcher as bricoleur: the new wealth and diversity in qualitative
inquiry, Tourism Analysis, 1, 6774.
Ivanitz, M. (1999), Culture, ethics, and participatory methodology in cross-cultural research, Australian
Aboriginal Studies, 2, 4658.
Jennings, G.R. (1999), Voyages from the centre to the margins: an ethnography of long term ocean cruisers,
PhD thesis, Murdoch University, Murdoch, Australia.
Jennings, G.R. (2003), Tourism research: theoretical paradigms and accountability, Targeted Research: The
Gateway to Accountability: TTRA 34th Annual Conference Proceedings [CD Rom], 1518 June, St Louis,
Missouri.
Jennings, G.R. (2007a), Tourism perspectives towards regional development and Asias Values, paper pre-
sented at Dong-Eui University, 2223 October, Busan, South Korea.
Jennings, G.R. (2007b), Advances in tourism research: theoretical paradigms and accountability, in A.
Matias, P.A. Neto and P. Nijkamp (eds), Advances in Modern Tourism Research, Economic Perspectives,
Heidelberg: Springer, Physica-Verlag, pp. 935.
Jennings, G.R. (2009), Methodologies and methods in T. Jamal and M. Robinson (eds), Handbook of
Tourism Studies, Thousands Oaks, CA: Sage Publications, pp. 672692.
Jennings, G.R. (2010), Tourism Research, 2nd edition, Brisbane: John Wiley.
Jennings, G.R. and D. Stehlik (2009), Farm stay enterprises: (re)interpreting public/private domains
and home sites and sights, in P. Lynch, A. McIntosh and H. Tucker (eds), The Commercial Home:
International Multidisciplinary Perspectives, London: Routledge, pp. 5059.
Jennings, G.R., S. Kensbock and U. Kachel (2010), Enhancing education about and for sustainability in a
tourism studies enterprise management course: an action research approach, Journal of Teaching in Travel
& Tourism, 10 (2), 163191.
Kensbock, S. (2012/13), The employment experiences of hotel room attendants at five star hotels on the Gold
Coast of Queensland, Australia, PhD thesis, Griffith University, Gold Coast Australia.
Kuokkanen, R. (2003), Toward a new relation of hospitality in the academy, American Indian Quarterly, 27
(1&2), 267295.
Laclau, E. (1990), New Reflections on the Revolution of our Time, London: Verso.
Lee, Y., J. Datillo and D. Howard (1994), The complex and dynamic nature of leisure experience, Journal of
Leisure Research, 26, 195211.
Lincoln, Y.S. and E.G. Guba (1985), Naturalistic Inquiry, Newbury Park, CA: Sage.
Lincoln, Y.S. and E.G. Guba (2000), Paradigmatic controversies, contradictions, and emerging confluences,
in N.K. Denzin and Y.S. Lincoln (eds), Handbook of Qualitative Research, 2nd edition, Thousand Oaks,
CA: Sage, pp. 163188.
Lincoln, Y.S., S.A. Lynham and E.G. Guba (2011), Paradigmatic controversies, contradictions, and emerg-
ing confluences, revisited, in N.K. Denzin and Y.S. Lincoln (eds), Handbook of Qualitative Research, 4th
edition, Thousand Oaks, CA: Sage, pp. 97128.
OGorman, K.D. (2005), Modern hospitality: lessons from the past, Journal of Hospitality and Tourism
Management, 12 (2), 141151.
Pearce, P.L. (1993), Fundamentals of tourist motivation, in D.G. Pearce and R.W. Butler, Tourism Research,
London: Routledge.

DWYER 9781781001288 Part II.indd 322 25/07/2012 11:07

 Part II Qualitative methods 323

Richardson, L. (1994), Writing: a method of inquiry, in N. Denzin and Y.S. Lincoln (eds), Handbook of
Qualitative Research, 2nd edition, Thousand Oaks, CA: Sage, pp. 516529.
Riley, R.W. and L.L. Love (2000), The state of qualitative tourism research, Annals of Tourism Research, 27
(1), 164187.
Ryan, C. (1991), Recreational Tourism, a Social Sciences Perspective, London: Routledge.
Ryan, C. (ed.) (1997), The Tourist Experience, a New Introduction, Studies in Tourism Series, London: Cassell.
Schaper, M., J. Carlsen and G. Jennings (2007), Reflections on researching Indigenous enterprises, in J.
Buultjens and D. Fuller (eds), Striving for Sustainability: Case Studies in Indigenous Tourism, Lismore,
Australia: Southern Cross University Press, Chapter 2, pp. 3766.
Scheurich, J.J. (1997), Research Methods in the Postmodern, London: The Falmer Press.
Schwandt, T.A. (2000), Three epistemological stances for qualitative inquiry: interpretivism, hermeneutics,
and social constructionism in N.K. Denzin and Y.S. Lincoln (eds), Handbook of Qualitative Research, 2nd
edition, Thousand Oaks, CA: Sage, pp.189213.
Smith, L.T. (1999), Decolonizing methodologies: Research and Indigenous Peoples, London: Zed Books.
Smith, L.T. (2005), On tricky ground: researching the native in the age of uncertainty, in N.K. Denzin and
Y.S. Lincoln (eds) Handbook of Qualitative Research, 2nd edition, Thousand Oaks, CA: Sage, pp. 85107.
Sofield, T.H.B. (2000), Re-thinking and re-conceptualising social and cultural issues of tourism develop-
ment in South and East Asia, Institute for Sustainability and Technology Policy, Case Study. Murdoch
University, Australia, available at http://sustainability.murdoch.edu.au/casestudies/Case_Studies_Asia/
toursea/toursea.htm.
Stanfield II, J.H. (1994), Ethnic modelling in qualitative research, in N.K. Denzin and Y.S. Lincoln (eds),
Handbook of Qualitative Research, 1st edition, Newbury Park, CA: Sage, pp. 175188.
Tribe, J. (2006), The truth about tourism, Annals of Tourism Research, 33 (2), 360381.
Urry, J. (1990), The Tourist Gaze, Leisure and Travel in Contemporary Societies, London: Sage.
Urry, J. (1996), Sociology of time and space, in B.S. Turner (ed.), The Blackwell Companion to Social Theory,
Oxford: Blackwell, pp. 369395.
Urry, J. (2002), The Tourist Gaze, 2nd edition, London: Sage.
Walle, A.H. (1997), Quantitative versus qualitative tourism research, Annals of Tourism, 24 (3), 524536.
Weed, M. (2005), Sports tourism, theory and method: concepts, issues and epistemologies, European Sport
Management Quarterly, 5 (3), 229242.
Westwood, S. (2007), What lies beneath? Using creative, projective and participatory techniques in qualitative
tourism inquiry, in I. Ateljevic, N. Morgan and A. Pritchard (eds), The Critical Turn in Tourism Studies:
Innovative Research Methodologies, Advances in Tourism Research, Amsterdam: Elsevier, pp. 293316.

DWYER 9781781001288 Part II.indd 323 25/07/2012 11:07

DWYER 9781781001288 Part II.indd 324 25/07/2012 11:07
 15 Grounded theory
Olga Junek and Les Killion

INTRODUCTION

Grounded theory (GT) as a research approach seeks to induce theory from empiri-
cal material through the ongoing interpretation of that material. In the development
of a theory or theoretical framework, continual integration between the participants,
the empirical materials, the researcher and the interpretation takes place (Strauss and
Corbin, 1994). Since its origin in 1965 (Glaser and Strauss, 1965, 1967), and following
a number of methodological and philosophical arguments and contested meanings,
grounded theory has undergone a number of substantive changes.
In the last decade, grounded theory has become a popular and increasingly important
research methodology in tourism studies, thus acknowledging the multi-disciplinary
nature of tourism and the moving away from a previously dominant business and quan-
titative focus. Using grounded theory in tourism studies provides researchers with a lens
into tourist behaviour at a specific time, and, in particular, behaviour that may not have
been studied before from a qualitative perspective (Birks and Mills, 2011). Grounded
theory is also used when there is no a priori theory to explain an area or phenomenon
of interest, or when the existing theory does not adequately take account of the social or
temporal contexts of tourist behaviour (Hobson, 2003; Junek, 2004; Mehmetoglu and
Olsen, 2003).
In briefly noting some of the debates surrounding its development, this chapter
draws the distinction between grounded theory as an alternative qualitative para-
digm, and grounded theory methods which now form the basis of methodologies in
a broad range of approaches to social enquiry. The chapter sees grounded theory as
being founded upon the premise that emergent explanatory theories can be formulated
during the ongoing interplay between gathering and analysing empirical materials.
Tourism research, once founded firmly upon positivist perspectives and paradigms
and the testing of a priori theory, has, in contrast to other disciplines, only recently
embraced the principles and precepts of grounded theory. The chapter examines some
of the more recent tourism-related investigations in which grounded theory approaches
have been implemented. One such area is that of tourism policy formulation and
implementation and the chapter argues that grounded theory approaches offer the
potential to achieve collaborative research outcomes reflecting the inputs of the wide
spectrum of industry stakeholders whose performance within the industry are influ-
enced both directly and indirectly by tourism policies and the wider ambit of public
policy decisions. Such collaborative research efforts have the potential to generate
Mode 2 Knowledge characterised by the constant flow back and forth between the
fundamental and the applied, between the theoretical and the practical (Gibbons et
al., 1994, p. 19). Through the adoption of grounded theory methods and approaches it
is considered that trans-disciplinary research and the generation of Mode 2 Knowledge

325

DWYER 9781781001288 Part II.indd 325 25/07/2012 11:07

 326 Handbook of research methods in tourism

provides a way forward in seeking insights into the issues that challenge tourism in the
twenty-first century.

NATURE OF GROUNDED THEORY

While adopted by a diverse range of disciplines, grounded theory approaches (and

qualitative research more generally) remain somewhat undervalued in certain areas of
enquiry. Within interpretivist social sciences paradigms, grounded theory approaches
focus on the realities of peoples social experiences (Bryant and Charmaz, 2007;
Charmaz, 2006). Grounded theory approaches aim to develop a situation specific emer-
gent theory founded upon the construction and interpretation of the experiential stories
of informants who are at the centre of the investigation. Formulating theory through
inductive reasoning was a response to the critique that quantitative positivist paradigms
did not provide in-depth understanding of social phenomena.
While accepted as a popular and robust methodology amongst social science research-
ers, many researchers in other fields of study such as in business, especially novice
researchers and those doing their doctoral dissertations, are still often confronted with
scepticism when using grounded theory methodologies. Trepidation, fear, misunder-
standing and misgivings about using grounded theory are common issues confronting
researchers using this methodology.
As is the case with any methodology, grounded theory has inherent strengths and
weaknesses. Its use depends on both the paradigmatic considerations of the researcher
and objectives of the research. The strengths of grounded theory lie in the closeness of
the researcher to the empirical materials throughout the information collection and
analysis stages of research. In grounded theory approaches, the two stages of collection
and analysis are inseparable and often referred to as the process of data generation. They
occur concurrently allowing for an openness and creativity for the researcher in fore-
grounding the emergent theory. This is not to say that the analysis is not systematic. On
the contrary, to be able to validate and support the emerging theory, grounded theory
researchers need to convince themselves and others that the emergent theory is founded
upon the classification, categorization and conceptualization of the empirical materials.
A major difference between grounded theory and other methodologies, and often the
cause of much confusion and consternation amongst novice researchers is the role of the
literature review. Since Glaser and Strauss (1967) postulated that the grounded theory
researcher should not review the literature relevant to the field of study before collecting
and analysing empirical materials, many grounded theory researchers have argued the
place of the literature review and preconceived theories emanating from the literature.
The notion that a researcher enters the field with as little knowledge as possible had
been initially strongly defended by Strauss and Corbin (1990). Their stand changed
as grounded theory evolved, with their subsequent advocacy of some literature being
reviewed in the early stages of research (Dunne, 2011; Wiener, 2007).
Contrariwise, Glaser (1998) and many of his followers still argue that no extensive lit-
erature review should be undertaken (Nathaniel, 2006). Faced with this rather prescrip-
tive requisite of purist grounded theory methodology, doctoral students often faced the
prospect of not fulfilling candidature requirements if a substantial literature review was

DWYER 9781781001288 Part II.indd 326 25/07/2012 11:07

 Grounded theory 327

not submitted when seeking confirmation of candidature. To some degree this has now
changed. Through discussion, and informed justification, several prominent grounded
theory advocates have turned away from the prescriptive Glaserian guidelines and have
arrived at a middle ground regarding literature review (Dunne, 2011, p. 117). Grounded
theory researchers have found their own ways of using the literature review while still
ensuring they enter the research field with openness without prejudging their research
results. In relation to the place of literature Dunne (2011), for instance, discusses several
important positions, aspects and ideas as how best to make proper use of previous
knowledge (Struebing, 2007, p. 587).
It is important at this point to draw attention to the terminology used to distinguish
between grounded theory, that is, the theory that emerges from the empirical materials,
and grounded theory method (GTM), the methods and processes employed to gather,
analyse and interpret empirical materials that can lead to the development or emergence
of a grounded theory. GTM today exists in many different versions, based on a number
of similarities but just as many differences among researchers, the disciplines in which
they are working, and their research objectives (Hood, 2007; Locke, 2001; Stern, 2007;
Wiener, 2007). Bryant and Charmaz (2007) see these similarities and differences as a
family of methods (p. 11) based on Wittgensteins (1953) ideas about family resem-
blances.
As a first step, investigators who base their research on a grounded theory method
generally gather the necessary information in naturalistic settings through collabora-
tion with those participating in the study using informal, conversational interviews
based upon open-ended, minimally structured questions to focus the discussion. The
information thus gained is often referred to as empirical materials as distinct from
data, a term that is seen to have positivistic, quantitative overtones. In general, empiri-
cal materials gathered through a grounded theory method take the form of transcribed
records of such interviews and conversations. Theorizing and finding meaning within
the empirical materials then adopts an inductive (as opposed to deductive) approach in
order to generate a theory grounded in those materials. In order to establish distinct
units of meaning (Goulding, 2002, p. 74), and subsequently to generate theory in this
way, the empirical materials must be broken down through a system of codification.
Based upon her insights into the seminal works of Glaser and Strauss (1967) and
Glaser (1992), Goulding (2002) provides a useful summary of the codings achieved
through constant comparison. That is, by sampling (theoretical sampling) from within
the empirical materials that research participants have provided, comparing incident
with incident, and incident with concept, to emerge more categories and their proper-
ties (Glaser, 1992, p. 38). Through such theoretical sampling, similarities and differ-
ences across incidents found within the empirical materials can be identified. Initially,
open coding is employed to open up the interview data by fragmenting it, identifying
concepts and [using] constant comparison to scrutinize the data for every meaning
(Stevenson et al., 2008, p. 9). The initial codes are labelled with descriptive labels
to generate concepts that, as Goulding (2002) explains, can then be aggregated into
descriptive categories which, in turn, can be further grouped into higher order concepts
(Strauss and Corbin, 1998). Importantly, throughout the coding processes, memos
(memo-ing) are written as a basis for highlighting comparisons between incidents
contained in the empirical materials with those recalled from experience and from the

DWYER 9781781001288 Part II.indd 327 25/07/2012 11:07

 328 Handbook of research methods in tourism

literature. In depicting the further steps of a grounded theory method, Goulding (2002,
p. 169) then notes that axial coding is required which seeks to identify incidents which
have a relationship to each other. Through axial coding a higher order of abstraction is
sought as the basis for understanding the relationships between identified concepts. In
that sense, axial coding can be seen as reassembling the empirical materials that were
initially broken down in the process of open coding. As Goulding (2002, p. 169) has
explained, the descriptive codes are subsumed into a higher order category that unites
the theoretical concepts to offer an explanation or theory of the phenomenon under
investigation.
It is important to acknowledge that the procedures of constant comparison, theo-
retical sampling, coding, categorizing and memo-ing form the basis of GTM. However,
while based on the use of these fundamental approaches, not all research using GTM
results in a grounded theory. GTM may just as readily lead to the formulation of hypoth-
eses, propositions, theoretical frameworks or conceptual themes for subsequent investi-
gation. In contrast, only when a theory that is grounded in the empirical materials of the
research emerges can it be claimed that grounded theory (as opposed to grounded theory
approaches) as a methodology has been used.

BACKGROUND AND EVOLUTION OF GROUNDED THEORY

Grounded theory is now considered to be one of the most used methodologies within
qualitative research (Morse et al., 2009). Since its origins in 1967 as a response to
the hegemony of quantitative research and the positivist paradigm, grounded theory
and GTM have evolved from the original foci on illness and its effects on caregivers
(Charmaz, 2007) to encompass a wide number of other social science areas. The origi-
nal writings of Glaser and Strauss (1965) focussed heavily on the GTM that is how
to conduct research using these methods and as such provided a rather dogmatic
approach to conducting research. However, the foundations of grounded theory were
laid and many of the original characteristics still remain despite the well-documented
arguments between the two authors. There now exist multiple interpretations of the tra-
dition (Jennings and Junek, 2009, p. 199) led by the new, second-generation of authors
who have adapted grounded theory according to their own research paradigms and their
particular research aims.
Amongst the second generation of authors who have progressed the use of grounded
theory and have used this methodology as a launching pad for their own iterations
(Birks and Mills, 2011, p. 3) while also providing seminal writings, and pragmatic guid-
ance, Charmaz (1997, 2000, 2003, 2006), Corbin (2008, 2009) and Clarke (2009) are
prominent. Corbin (2008, 2009) sees methodology as a dynamic, living thing (2008, p.
37) and in light of that dynamic process her research reflects changes in her approaches
to grounded theory whilst staying rooted in the original ideas and processes of Glaser
and Strausss (1967) grounded theory. Charmazs (2006, 2007, 2009) constructivist
grounded theory assumes a relativist epistemology, sees knowledge as socially pro-
duced, acknowledges multiple standpoints ... and takes a reflexive stance toward our
actions, situations and participants in the field setting and our analytic constructions of
them (2009, p. 129). Epistemologically, in the constructivist grounded theory approach

DWYER 9781781001288 Part II.indd 328 25/07/2012 11:07

 Grounded theory 329

theorists and informants are co-creators of knowledge. It is the emergent, constructivist

characteristics of GTM stressed by Charmaz (2003) that are central to the constructivist
grounded theory.
Clarke (2005, 2009), the third of the three major influential members of the second
generation of authors on grounded theory, took yet another direction with situational
analysis, grounded in symbolic interactionist sociology . . . to be understood as a
theory/methods package (Clarke, 2009, p. 197). In situational analysis using grounded
theory, so Clarke argues (2009), human element is not enough and attention needs to be
paid to non-human objects to fully research the situation or problem.
It was this second generation of writers who approached grounded theory with distinct
philosophical frameworks of epistemological, ontological, axiological and methodologi-
cal paradigms (Birks and Mills, 2011) in their quest to use grounded theory as a method-
ology as well as a set of interrelated methods and approaches.
The further development of grounded theory has not only been about the evolution
of methodology and methods. It has also been about an expanding range of areas of
research inquiry that have embraced grounded theory. Originally used in the areas of
health care, particularly palliative care and nursing, grounded theory has moved into
other areas of social research including human development, education, social justice,
indigenous studies and to a lesser, although steadily growing extent, business studies, in
particular organizations, tourism and marketing.

APPLICATION TO TOURISM AND TOURISM RELATED

STUDIES

Several reasons exist to explain the relatively recent adoption of grounded theory in
tourism studies. Tourism as a field of study has traditionally been placed within the
business schools of universities where quantitative and positivistic paradigms tend to
predominate as the guiding methodologies (Cohen, 1988; Jennings, 2001; Jennings
and Junek 2009; Riley and Love, 2000; Walle, 1997). Furthermore, even within quali-
tative research there has been a noticeable lack of researchers using grounded theory
to progress the field of tourism studies from a social studies interpretivist paradigm.
However, since the 1990s more and more research has been undertaken using grounded
theory and grounded theory methods.
Some of the early grounded theory researchers in the area of tourism explored
travel prestige (Riley, 1995); travel experiences of cruisers (Jennings, 1997, 1999,
2005); heritage site visitors (Daengbuppha, et al., 2004; Goulding, 1999); backpacker
travellers in Australia (Hillman, 2001); visitor perceptions of theme parks (Johns and
Gyimothy, 2002); commodification (Goulding, 2000) and authenticity in the tourist
experience (Mehmetoglu and Olsen, 2003). The common research characteristics of
these studies were the in-depth explanations and understandings of the phenom-
ena and experiences being studied. As well as an increase in the use of grounded
theoryresearch in tourism studies, there has also been a recognition of the need for
more qualitative research methodologies in tourism studies (including grounded
theory) as discussed by Connell and Lowe (1997); Hobson (2003) and Junek (2004);
Riley (1995).

DWYER 9781781001288 Part II.indd 329 25/07/2012 11:07

 330 Handbook of research methods in tourism

In the last decade grounded theory and GTM have continued to play an important
part in tourism studies research and have become even more important, as tourism
studies and tourism researchers embrace a more interdisciplinary approach. Some
of these studies include Belhassen et al.s (2007) research on the use of cannabis
amongst vacationers; village residents attitude towards tourism (Lepp, 2007); tourist
behaviour (Drew and Woodside, 2008); volunteer tourism (Alexander, 2009); tourism
jobs (Jiang and Tribe, 2009); slow travel (Lumsdon and McGrath, 2011) and online
holiday reviews (Papathanassis and Knolle, 2011). As tourism as a field of studies
has increasingly incorporated a number of different perspectives beyond the com-
mercial and business perspective, the listed studies contribute to the widening areas of
the social, the personal and the psychological perspectives. In terms of the personal
and the psychological perspectives in particular, grounded theory offers an in-depth
exploration of the lived experiences of tourists and tourism, an exploration that has
broadened our understanding of the complexities and inter relationships amongst
tourism stakeholders.
It has thus become evident that research using grounded theory makes important con-
tributions to the field of tourism studies. In summary these include:

Theory building as opposed to theory testing (Alexander, 2009; Drew and

Woodside, 2008)
Greater insight and in-depth understanding of many different types of tourists and
tourism, whether this be backpackers, cruising, slowtravellers, volunteer tourists
(Alexander, 2009; Lumsdon and McGrath, 2011) or memorable tourism experi-
ences (Tung and Ritchie, 2011)
Understanding behaviour and social processes within a particular context
(Belhassen et al., 2007; Jiang and Tribe, 2009).

The legitimacy of grounded theory as a robust and widely accepted research method-
ology is reflected in the number of research articles founded on such approaches that
have been published in prestigious, peer-reviewed tourism journals such as Annals of
Tourism Research, Tourism Management and Journal of Sustainable Tourism. This has
been a significant change from the 1990s and early 2000s when researchers often had
to resort to conference publications and lower-ranked journals to have their grounded
theory studies published.
Tourism, tourists and tourist behaviour exist within many social contexts and
embrace a number of different stakeholders. It is therefore important to understand the
multi-faceted nature of tourism as a business but also as a sociological phenomenon.
Grounded theory, through its methods and as an approach resulting in an emergent
theory can provide a holistic picture of tourist behaviour and tourism but it can also be
used to add deeper insight to studies using quantitative methodology, whether this be
in the form of mixed methods or using existing studies as a basis for a more qualitative
understanding of the research topic or area. The areas of policy formulation, implemen-
tation and evaluation, for instance, offer an emergent field of enquiry in which grounded
theory approaches offer potential contributions.

DWYER 9781781001288 Part II.indd 330 25/07/2012 11:07

 Grounded theory 331

ADVANTAGES AND LIMITATIONS OF GROUNDED

THEORY APPROACHES: CONCEPTUALLY AND FOR POLICY
FORMULATION
Founded upon the central precepts and paradigms of qualitative research, grounded
theory approaches offer a number of advantages in tourism research projects aimed at
achieving insights into unexplored questions, or complementing quantitative data with
the richness that a qualitative approach can contribute to further explanation and/or
confirmation of previously investigated social phenomena. Despite these acknowledged
advantages, like many qualitative research approaches, grounded theory research can
sometimes be seen as limited through its focus on situation-specific phenomena.

The Advantages and Contributions of Grounded Theory Approaches to Tourism Research

Congruent with the purpose and aims of an investigation, grounded theory approaches
may contribute to tourism research in several ways:

1. Exploratory: where no a priori theories exist, grounded theory approaches may be

employed to explore the situation from the perspectives of those involved and gener-
ate emergent theory by constructing and interpreting of empirical materials obtained
through informal interviews and observations. For example, in a host community
such as the Sisters of St Joseph residing in Mount Street, North Sydney, Australia,
who were engaging with mass tourism for the first time following the Beatification of
Saint Mary MacKillop, the communitys responses to visitor impacts were explored
through grounded theory approaches (Killion, 2003). Earlier, Jennings (1999)
employed an ethnographic methodology to theorize about the social realities and
lived experiences of long-term ocean cruisers.
2. Explanatory: where phenomena are already known, grounded theory approaches
may facilitate new explanations, for example, of the responses made by tourism
operators seeking to recover from the impacts of the 2011 floods in Queensland,
Australia. Radel (2010) employed a grounded theory approach in order to account
for factors which have contributed to the successful operation of the Dreamtime
Cultural Centre, an Indigenous tourism enterprise located in Rockhampton,
Australia.
3. Confirmatory: finally, grounded theory approaches, through their application to
specific research contexts, may confirm (or refute) established knowledge as found in
the literature, mindful that in grounded theory methodologies engagement with per-
tinent literature occurs after the initial formulation of emergent explanatory theory.

The Limitations of Grounded Theory Approaches

Grounded theory approaches have been seen as freeing researchers from the tyranny of
verifying theory and instead instruct[ing] them to focus on generating theory (Wagenaar,
2003) founded upon the construction of empirical materials and reflexivity to produce
thick description of social reality. Such a postpositivist paradigm provided systematic
inductive guidelines for collecting and analyzing data to build middle-ground theoretical

DWYER 9781781001288 Part II.indd 331 25/07/2012 11:07

 332 Handbook of research methods in tourism

frameworks that explain the collected data (Charmaz, 2000, p. 509). The view of Glaser
and Strauss (1967) from the genesis of grounded theory approaches was, and is, that
theory is not a final statement about some social phenomenon or activity, but a strategy
for handling data informed by real-world social realities.
The limitations of grounded theory approaches reflect the critiques of much postposi-
tivist, qualitative research that the emergent theory, supported and justified by the gener-
ated empirical materials, is project-specific and is inherent to these approaches. Critics
then question the extent to which such emergent theory can be generalized to other
contexts. Eichelberger (1989), however, argues that through subsequent engagement
with relevant literature, the findings derived from one investigation can be integrated and
compared with others who have studied the same, or similar experiences and phenomena
as the basis for substantive theory. That is, as described by Flick (2002), a preliminary
and relative version of the world. In a similar vein, Jafari (1987) and Kerr et al. (2001)
recommend case specific studies to develop thick description and improve understanding
in a specific context rather than attempting to develop universal models.
Project and temporal specificity make replication of findings in the positivist tradition
difficult for grounded theorists. This has also engendered criticism of such approaches.
Such criticism fails to appreciate the nature of interpretivist qualitative paradigms.
Stevenson et al. (2008) explain that grounded theory approaches provide systematic pro-
cedures and rigorous methodologies akin to those of positivist persuasion for collecting
and analysing qualitative data, and enable investigators to build theory from the bottom
up from the actions, words and behaviour of the research informants. Grounded theory
approaches emphasize the need to get out in the field (Glaser, 1998) to study phenom-
ena using the perspectives or voice of those studied, collect and analyse data simultane-
ously and refine theory using a wider range of data (Goulding, 2002). The systematic
and constant comparison of data through open and then axial coding allows grounded
theorists to identify incidents that have a relationship to each other, and to achieve a
level of abstraction that leads to appreciation of dynamic relationships between concepts
(Goulding, 2002). Subsequent engagement with the wider literature then aims for further
substantiation or refutation of the emergent theory.

Grounded Theory Approaches and Policy Making

Policy makers and those tasked with evaluating the impacts and consequences of policy
decisions, favoured process models of policy making and were more inclined to follow
positivist paradigms and quantitative data to measure their efforts. In investigating
policy formulation and implementation, grounded theory approaches have made only
relatively recent contributions. The reasons for this exist at two interrelated levels,
underscoring the need for greater knowledge of policy dynamics at all stages. First, less
than perfect understanding of policy-making processes driven by diverse stakeholder
groups within and beyond government; and second, the apparent lack of success of
policy decisions in achieving desired policy outcomes. In both instances, grounded
theory approaches offer levels of understanding extending beyond those gleaned only
from quantitative approaches.
In considering the applications of grounded theory approaches to policy formulation
in their study of Leeds city, Stevenson et al. (2008) emphasize that policy making is

DWYER 9781781001288 Part II.indd 332 25/07/2012 11:07

 Grounded theory 333

essentially a social process, involving communication and negotiation between people in

the context of wider change and between many people in a variety of different organi-
sations. Policy making can be considered a social conceptualization or social construct
and further research should investigate the communications involved in producing
policy (Stevenson et al., 2008). Noting that policy means different things to different
people, Dredge and Jenkins (2007, p. 14) also describe policy making and planning as:

dynamic, socially constructed activities that involve a wide range of agents and organisations
characterised by varying degrees of interest and commitment to tourism. Tourism planning and
policy development takes place in multi-actor settings, and it is important to understand how
different conceptions of tourism and different values and ideas are mediated in planning and
policy processes.

Policy is formulated in a dynamic environment. The consequences of change may

provide the rationale for the formulation of policy positions in the first place. Whether
in public or private sectors, in formulating policy, policy makers respond to the facts
of a given situation. Whatever these may be, facts reflect the social realities of those
engaged in policy making, and the intended recipients of policy provisions. Facts are
social constructs formed on the basis of interactions between policy observers, policy
advocates, the communities of which they are also members, and the wider world. The
ways in which facts are configured and perceived reflect the social realities and expe-
riences of all parties. Likewise, and as Stevenson et al. (2008) observe, the making of
policy is a complex social process that involves interaction and communication between
a large number of people from different backgrounds and organizations political
parties, lobby groups, industry stakeholders, consumers, community members any
or all of whom may facilitate or constrain the formulation and implementation of
policy. Unsurprisingly, multiple perspectives and multiple social realities find their way
into the policy processes as policy is developed, enacted, implemented and evaluated.
Understanding how these policy actors (Dredge and Jenkins, 2007; Hall, 2010) relate,
interact and communicate makes a fertile field for the development of emergent theories
founded upon grounded theory approaches.
Weeds (2005) study of sport and tourism policy responses further exemplifies the
application of grounded theory approaches in a longitudinal investigation of tensions in
the policy process at three levels: the macro- (or context level); the meso- (or setting and
situated activity level); and, the micro- (or self level). The grounded theory model based
upon empirical materials collected through the discussion of policy issues among key
stakeholders, points to the value of supplementing quantitative data with the depth and
richness of understanding that grounded theory approaches can contribute.
While critical of the questionable quality of much qualitative work in European policy
research, Wagenaar (2003, pp. 12) considers much policy research abounds in grandi-
ose, abstract, a priori theoretical statements (often with a strong normative ring to it) and
small, badly executed empirical studies that have the tacked-on quality that Glaser and
Strauss warn us for. In these terms, Wagenaar is critical of iterative models of the policy
process such as that offered by Althaus et al. (2007), for example. He criticizes such
process models and frameworks because they largely neglect the wider range of actors
who directly and indirectly influence the development and implementation of policy, but
whose interpretations and social realities are often not considered (Nagel, 1999).

DWYER 9781781001288 Part II.indd 333 25/07/2012 11:07

 334 Handbook of research methods in tourism

Wagenaar (2003) presents three reasons for applying grounded theory approaches
in such investigations. His reasons reflect some of the advantages which such method-
ologies offer policy researchers. First, policy, claims Wagenaar, is a good subject for
grounded theory, especially in studies aiming to show how policies work and/or what
the social impacts of a policy might be. In this Wagenaar agrees with Goulding (2002)
in outlining the advantages of grounded theory approaches in business, management
and marketing research. Second, Wagenaar notes the return of policy studies to the
contextual analysis advocated by Lasswell and Lerner (1951) in encouraging a holistic
understanding of policy processes, consequences and impacts. Third, and importantly,
Wagenaar argues grounded theory approaches force the development of an intimate
and detailed familiarity with the world that is the object of particular policy (p. 4). He
argues that while those directly engaged in political action may not require such intimate
familiarity with social realities, it enables advisors to policy decision makers and those
evaluating policy outcomes:

to grasp the meaning that policy measures have for those affected by them, and thereby to
better understand how they will adapt to them in their everyday practices. In the final analysis,
it allows him [sic] to grasp the value of lay, experiential knowledge as it intersects with the more
abstract knowledge that informs policy programs, and to involve this in the understanding of
complex, evolving policy situation. (Wagenaar, 2003, p. 4.)

FURTHER DEVELOPMENTS AND APPLICATIONS OF

GROUNDED THEORY APPROACHES IN TOURISM RESEARCH

The tourism industry comprises complex networks of key stakeholders including public
and private sector and not-for-profit owners and operators of all elements of the des-
tination mix; tourists; and, the host community. Each of these functions and interacts
within an industry well known for its competitiveness and the dynamics and turbulence
of change processes. Some of these can be reasonably anticipated while others, such as
the hurricanes and floods of Australias summer of 2011, less so. Knowledge of how
stakeholders function, interact and behave is less than complete. Abstract extant theories
seemingly provide only limited explanations.
Adopting the central precepts of a grounded theory approach, in his study of tourism
development in Malta, Bramwell (2006) advocated the actor perspective to investigate
interactions between different stakeholder groups. The future applications of grounded
theory approaches will continue to focus on the interpretation of lived social realities
as the basis for inductive theorizing. Moreover, as Bramwell and Meyer (2007) suggest
with reference to policy making and policy research, grounded theory forms the basis of
a relational approach that is holistic in its intentions to understand the contexts and
realities of informants and participants. By placing people at the centre of the investiga-
tion, grounded theory approaches emphasize the social realities and social interactions
that are fundamental to the process. Referring to the tourism policy of the city of Leeds
(UK), Stevenson et al. (2008, p. 748) note that policy making is a soft intuitive human
process rather than a rational scientific process. Grounded theory approaches provide a
direct focus on the vagaries of human intuition.

DWYER 9781781001288 Part II.indd 334 25/07/2012 11:07

 Grounded theory 335

Grounded Theory Approaches, Transdisciplinary Research and Mode 2 Knowledge

Grounded theory approaches to tourism research carry the possibility of generating

Mode 2 Knowledge through the integrated transdisciplinary collaborative efforts of
academics and industry stakeholders. Transdisciplinary research embraces approaches
such as grounded theory to integrate the knowledge, epistemologies and methods of
established academic disciplines with real world experiences and situations to gener-
ate Mode 2 Knowledge that combines scientific knowledge derived from established
disciplines with extra-scientific real world (or life-world) experiences and practices in
problem-solving (Mittelstrass, 2001). Knowledge production is not confined to academic
disciplines but also emanates from various organizations, communities, groups and indi-
viduals outside of academia as people undertake their day-to-day functions and interac-
tions. Mode 2 Knowledge is generated at the intersections between academic disciplines
(the horizontal plane) and the life-world experiences of people and groups outside
academia (the vertical plane) (Mittelstrass, 2001).

CONCLUSION

While direct findings in the form of emergent theory are context specific, grounded
theory approaches provide further insights and understandings of the complex social
phenomena associated with tourism. The lived social experiences of tourists themselves,
tourism operators and other stakeholders within the industry, and members of host
communities, provide extended insights into and understandings of an industry that is
highly competitive, dynamic and subject to the influences of wider social change proc-
esses. Quantitative approaches founded upon the traditions of positivist paradigms have
provided important baseline data. The insights gained through the applications of GT
and GTM provide fresh information and the prospects of transdisciplinarity capable of
generating Mode 2 Knowledge. This is of significance to those who formulate, imple-
ment and evaluate tourism policy to resolve contemporary issues and prepare for the
future.

REFERENCES

Alexander, Z. (2009), Understanding voluntourism: a Glaserian grounded theory, The voluntourism effect:
case studies and investigations, 1820 March, 2009, San Diego, CA.
Althaus, C., P. Bridgeman and G. Davis (2007), The Australian Policy Handbook, 4th edition, Sydney,
Australia: Allen & Unwin.
Belhassen, Y.S., A.S. Santos and N. Uriely (2007), Cannabis usage in tourism:a sociological perspective,
Leisure Studies, 23 (3), 303319.
Birks, M. and J. Mills (2011), Grounded Theory: A Practical Guide, London: Sage.
Bramwell, B. and D. Meyer (2007), Power and tourism policy relations in transition, Annals of Tourism
Research, 34 (3), 766788.
Bramwell, W. (2006), Actors, power and discourses of growth limits, Annals of Tourism Research, 26 (2),
392415.
Bryant, A. and K. Charmaz (2007),Introduction grounded theory research: methods and practices, in A.
Bryant and K. Charmaz (eds), The SAGE Handbook of Grounded Theory, Thousand Oaks, CA: Sage, pp.
128.

DWYER 9781781001288 Part II.indd 335 25/07/2012 11:07

 336 Handbook of research methods in tourism

Charmaz, K. (1997), Identity dilemmas of chronically ill men, in A. Strauss and J. Corbin (eds), Grounded
Theory in Practice, Thousand Oaks, CA: Sage, pp. 3563.
Charmaz, K. (2000), Grounded theory: objectivist and constructivist methods, in N.K. Denzin and Y.S.
Lincoln (eds), Handbook of Qualitative Research, Thousand Oaks, CA: Sage, pp. 509535.
Charmaz, K. (2003), Grounded theory: objectivist and constructivist methods, in N.K. Denzin and Y.S.
Lincoln (eds), Strategies of Qualitative Inquiry, Thousand Oaks, CA: Sage, pp. 249291.
Charmaz, K. (2006), Constructing Grounded Theory: A Practical Guide through Qualitative Analysis, Thousand
Oaks CA: Sage.
Charmaz, K. (2007), Constructionism and grounded theory, in A.J. Holstein and J.F.Gubrium (eds),
Handbook of Constructionist Research, New York: Guilford Publishers, pp. 397412.
Charmaz, K. (2009), Shifting the grounds: constructivist grounded theory methods, in J.M. Morse, P.N.
Stern, J. Corbin, B. Bowers, K. Charmaz and A.E. Clarke, Developing Grounded Theory: The Second
Generation, Walnut Creek, CA: Left Coast Press, pp. 127192.
Clarke, A. (2005), Situational Analysis: Grounded Theory after the Postmodern Turn, Thousand Oaks, CA:
Sage.
Clarke, A. (2009), From grounded theory to situational analysis: whats new? Why? How?, in J.M. Morse,
P.N. Stern, J. Corbin, B. Bowers, K. Charmaz and A.E. Clarke, Developing Grounded Theory: The Second
Generation, Walnut Creek, CA: Left Coast Press, pp. 194234.
Cohen, E. (1988), Traditions in the qualitative sociology of tourism, Annals of Tourism Research, 15, 2946.
Connell, J. and A. Lowe (1997), Generating grounded theory from qualitative data: the application of
inductive methods in tourism and hospitality management research, Progress in Tourism and Hospitality
Research, 3 (2), 165173.
Corbin, J. (2008), Basics of Qualitative Research: Techniques and Procedures for Developing Grounded Theory,
3rd edition, Thousand Oaks, CA: Sage.
Corbin, J. (2009), Taking an analytic journey, in J.M. Morse, P.N. Stern, J. Corbin, B. Bowers, K. Charmaz
and A.E. Clarke, Developing Grounded Theory: The Second Generation, Walnut Creek, CA: Left Coast Press,
pp. 3554.
Daengbuppha, J., N. Hemmington and K. Wilkes (2004), Using grounded theory approach: theoretical and
practical issues in modelling heritage visitor experience, in K.A. Smith and C. Schott (eds), Proceedings
of the New Zealand Tourism and Hospitality Research Conference 2004, 810 December, Wellington, New
Zealand, pp. 6478.
Dredge, D. and J. Jenkins (2007), Tourism Planning and Policy, Milton, Queensland: John Wiley and Sons
Australia Ltd.
Drew, M. and A. Woodside (2008), Grounded theory of international tourism behaviour, Journal of Travel
and Tourism Marketing, 24 (4), 245258.
Dunne, C. (2011), The place of the literature review in grounded theory research, International Journal of
Social Research Methodology, 14 (2), 111124.
Eichelberger, R.T. (1989), Discipline Inquiry: Understanding and Doing, Educational Research, New York:
Longmans.
Flick, U. (2002), An Introduction to Qualitative Research, 2nd edition, London: Sage Publications.
Gibbons, M., C. Limoge, H. Nowotny, S. Schwartzman, P. Scott and M. Trow (1994), The New Production of
Knowledge: The Dynamics of Science and Research in Contemporary Societies, London: Sage Publications.
Glaser, B. (1992), Basics of Grounded Theory Analysis: Emergence vs Forcing, Mill Valley, CA: Sociology Press.
Glaser, B. (1998), Doing Grounded Theory: Issues and Discussions, Mill Valley, CA: Sociology Press.
Glaser, B.G. and A.L. Strauss (1965), Awareness of Dying, Chicago, IL: Aldine.
Glaser, B.G. and A.L. Strauss (1967), The Discovery of Grounded Theory Strategies for Qualitative Research,
New York: Aldine Publishing Company.
Goulding, C. (1999), Heritage, nostalgia and the grey consumer, Journal of Marketing Practice: Applied
Marketing Science, 5 (6/7/8), 177199.
Goulding, C. (2000), The commodification of the past, postmodern pastiche, and the search for authentic
experiences at contemporary heritage attraction, European Journal of Marketing, 34 (7), 835853.
Goulding, C. (2002), Grounded Theory: A Practical Guide for Management, Business and Market Researchers,
Thousand Oaks CA: Sage.
Hall, C.M. (2010), Politics and tourism: interdependency and implications in understanding change, in R.
Butler and W. Suntikul (eds), Tourism and Political Change, Oxford: Goodfellow Publishers, pp. 712.
Hillman, W. (2001), Searching for authenticity and experience: backpackers travelling in Australia, TASA
Conference, University of Sydney.
Hobson, J.S.P. (2003), The case for more exploratory and grounded tourism research, Martin Oppermann
Memorial Lecture 2001, Pacific Tourism Review, 6 (2), 7381.
Hood, J. (2007), Orthodoxy vs. power: the defining traits of grounded theory, in A. Bryant and K. Charmaz
(eds), The SAGE Handbook of Grounded Theory, London: Sage, pp. 151164.

DWYER 9781781001288 Part II.indd 336 25/07/2012 11:07

 Grounded theory 337

Jafari, J. (1987), Tourism models: the socio-cultural aspects, Tourism Management, 9 (1), 8284.
Jennings, G.R. (1997), The travel experience of cruisers, in M. Oppermann (ed.), Pacific Rim 2000: Issues,
Interrelations, Inhibitors, London: CABI International, pp. 94105.
Jennings, G.R. (1999), Voyages from the centre to the margins: ethnography of long term ocean cruisers, PhD
thesis, Murdoch University, Murdoch, Australia.
Jennings, G.R. (2001), Tourism Research, Brisbane, Australia: John Wiley.
Jennings, G.R. (2005), Caught in the irons: one of the lived experiences of cruising women, Tourism Research
International, 9 (2), 177193.
Jennings, G.R. and O. Junek (2009), Grounded theory: innovative methodology or a critical turning from
hegemonic methodological praxis in tourism studies?, in I. Ateljevic, A. Pritchard and N. Morgan (eds),
The Critical Turn in Tourism Studies: Innovative Research Methodologies, Oxford: Elsevier, pp. 197210.
Jiang, B. and J. Tribe (2009), Tourism jobs short lived professions: student attitudes towards tourism careers
in China, Journal of Hospitality, Leisure, Sport and Tourism Education, 8 (1), 419.
Johns, N. and S. Gyimothy (2002), Mythologies of a theme park: an icon of family life, Journal of Vacation
Marketing, 8 (4), 320331.
Junek, O. (2004), A qualitative inquiry into leisure and travel patterns of international students. Part 1: back-
ground and methodology, The 2nd Asia-Pacific CHRIE (APacCHRIE) Conference & the 6th Biennial
Conference on Tourism in Asia, 2004, Conference Proceedings. 2729 May, Phuket, Thailand, 2004.
Kerr, B., G. Barron and R. Wood (2001), Politics, policy and regional tourism administration: a case examina-
tion of Scottish Area Tourist Board Funding, Tourism Management, 22 (6), 649657.
Killion, E.L. (2003), More than a miraculous journey an interpretivist study of the Sisters of the
Congregation of St. Joseph and their experiences of visitor impacts following the beatification of Blessed
Mary MacKillop, PhD thesis, Rockhampton CQUniversity Australia.
Lasswell, H.D. and D. Lerner (eds) (1951), The Policy Sciences, Palo Alto, CA: Stanford University Press.
Lepp, A. (2007), Residents attitude towards tourism in Bigodi village, Uganda, Tourism Management, 28,
876885.
Locke, K. (2001), Grounded Theory in Management Research, London, Sage.
Lumsdon, L. and P. McGrath (2011), Developing a conceptual framework for slow travel: a grounded theory
approach, Journal of Sustainable Tourism, 19 (3), 265279.
Mehmetoglu, M. and K. Olsen (2003), Talking authenticity: what kind of experiences do solitary travelers
in the Norwegian Lofoten Islands regard as authenticity?, Tourism, Culture and Communication, 4 (3),
137152.
Mittelstrass, J. (2001), On trans-disciplinarity, in Science and the Future of Mankind, Vatican City: Pontifical
Academy of Sciences, 495500, available at http://www.vatican.va/roman_curia/pontifical_academies/acd-
scien/documents/sv%2099(5of5).pdf.
Morse, J.M., P.N. Stern, J. Corbin, B. Bowers, K. Charmaz and A.E. Clarke (2009), Developing Grounded
Theory: The Second Generation, Walnut Creek, CA: Left Coast Press.
Nagel, S. (ed.) (1999), Policy Analysis Methods, New York: New Science Publishers.
Nathaniel, A. (2006), Thoughts on the literature review and GT, Grounded Theory Review, 5 (2/3), 3541.
Papathanassis, A. and F. Knolle (2011), Exploring the adoption and processing of online holiday reviews: a
grounded theory approach, Tourism Management, 32, 215224.
Radel, K. (2010), The Dreamtime Cultural Centre: A Grounded Theory of Doing Business in an Indigenous
Tourism Enterprise, PhD thesis, Rockhampton CQUniversity Australia.
Riley, R.W. (1995), Prestige-worthy behaviour, Annals of Tourism Research, 22 (3), 630649.
Riley, R.W. and L.L. Love (2000), The state of qualitative tourism research, Annals of Tourism Research, 27
(1), 164187.
Stern, P. (2007), On solid ground: essential properties of growing grounded theory in A. Bryant and K.
Charmaz (eds), The SAGE Handbook of Grounded Theory, London: Sage, pp. 114126.
Stevenson, N., D. Airey and G. Miller (2008), Tourism policy making: the policy-makers perspective, Annals
of Tourism Research, 35 (3), 732750.
Strauss, A.L. and J. Corbin (1990), Basics of Qualitative Research Grounded Theory Procedures and Techniques,
Newbury Park, CA: Sage.
Strauss, A.L. and J. Corbin (1994), Grounded theory methodology: an overview, in N.K. Denzin and Y.S.
Lincoln, Handbook of Qualitative Research. Thousand Oaks, CA: Sage, pp. 273285.
Strauss, A.L. and J. Corbin (1998), Basics of Qualitative Research Techniques and Procedures for Developing
Grounded Theory. Thousand Oaks, CA: Sage.
Struebing, J. (2007), Research as pragmatic problem-solving: the pragmatist roots of empirically-grounded
theorising, in A. Bryant. and C. Charmaz (eds), The SAGE Handbook of Grounded Theory, London: Sage,
pp. 580601.
Tung, V. and J. Ritchie (2011), Exploring the essence of memorable tourism experiences, Annals of Tourism
Research, 38 (4), 13671386.

DWYER 9781781001288 Part II.indd 337 25/07/2012 11:07

 338 Handbook of research methods in tourism

Wagenaar, H. (2003), The (re-)discovery of grounded theory in postpositivist policy research, paper presented
for the ESF Workshop Qualitative Method for the Social Sciences, Vienna.
Walle, A.H. (1997), Quantitative versus qualitative tourism research, Annals of Tourism Research, 24 (3),
524536.
Weed, M. (2005), A grounded theory of the policy process for sport and tourism, Sport in Society: Cultures,
Commerce, Media, Politics, 8 (2), 356377.
Wiener, C. (2007), Making teams work in conducting grounded theory, in A. Bryant (ed.), The SAGE
Handbook of Grounded Theory, London: Sage, pp. 293310.
Wittgenstein, L. (1953), Philosophical Investigations, Oxford: Blackwell.

DWYER 9781781001288 Part II.indd 338 25/07/2012 11:07

 16 Ethnographic methods
Kathleen M. Adams

INTRODUCTION

Although ethnographic methods derive from the discipline of sociocultural anthropol-

ogy, because of their potential for producing insights into human actions and behaviors
they have come to be embraced by sociologists, psychologists, and other social scientists
interested in gaining insights into human behavior. Ethnographic methods fall into the
broader category of qualitative methodologies and are aimed at understanding cultural
practices, human beliefs and behaviors, and sociocultural changes over time. As such,
ethnographic methods are particularly apt for tourism-related research and for tourism
policy planning, as noted by a number of recent tourism scholars (Cole, 2005; Graburn,
2002; Nash, 2000; Palmer, 2001, 2009; Salazar, 2011; Sandiford and Ap, 1998). This
chapter (1) introduces the key elements of ethnographic methodologies, (2) examines the
types of tourism research problems for which this research strategy is suited, (3) reviews
challenges entailed in using ethnographic methods in tourism settings, (4) surveys key
tourism studies grounded in ethnographic research, and (5) closes with a discussion of
new visions for this technique in tourism research.

NATURE AND EVOLUTION OF ETHNOGRAPHIC METHODS

Anthropologists use the term ethnography in two senses. In the first sense, an ethnog-
raphy is a written account of the socio-cultural dynamics animating a particular human
population. In the second sense, doing ethnography (or ethnographic research) entails
long-term fieldwork and draws on a mixture of qualitative research techniques all aimed
at generating insights into sociocultural relationships and the native[s]s point[s] of
view (Geertz, 1976; Malinowski, 1922). As Fife notes, in this context native refers not
to a concept of aboriginality but rather to anyone who has grown up within a specific
cultural milieu (Fife, 2005, p. 71). In broad terms, ethnographic research methods are
ideally suited for describing or analysing cultural practices, beliefs and behaviors, includ-
ing encounters between different groups. Most ethnographic research does not entail
theory-testing per se. Instead, it draws on grounded theory, which entails conducting
field research on broad themes and drawing on the data collected to develop compelling
theories.
Foremost among the techniques that comprise the ethnographic approach is that
of participant observation (to be explained below). But ethnographic research draws
upon a variety of other research techniques, including in-depth unstructured interviews,
structured interviews, questionnaires, focus groups, mapping, photography, and video-
documentation. Ethnographic researchers also collect and analyse various materials
produced by, for, and about the group being studied, including brochures, newsletters,

339

DWYER 9781781001288 Part II.indd 339 25/07/2012 11:07

 340 Handbook of research methods in tourism

visitor and locally produced blogs, web pages, historical archival material, handicrafts,
museum exhibits, and so forth.
Perhaps more than any other research strategy, the roots of ethnographic methods lie
in travel and displacement. The earliest ethnographic researchers all lived amongst the
people under study and participated to varying degrees in their daily activities. Franz
Boas (18581942), a German-born scholar considered one of the fathers of American
anthropology, spent a year in the mid-1880s living on Baffin Island where he researched
the Inuits adaptation to their physical environment (Boas, 1888). In keeping with the
German tradition, Boas data-collection strategy stressed the importance of direct obser-
vation (Ellen, 1984, p. 42). His subsequent fieldwork trips were generally shorter and
often team projects wherein assistants adept in the local languages gathered, transcribed,
and interpreted local community members explanations of their cultural practices.
Although Boas tended to make repeat research trips to the same areas, long-term immer-
sion in the field and participation in local activities were not yet essential components of
ethnographic methods (Eriksen and Nielsen, 2001, pp. 3940).
Participant observation, the cornerstone of ethnographic methods, was not developed
until anthropologist Bronislaw Malinowskis (18881942) landmark research amongst
the Trobriand Islanders in the South Pacific. Although Malinowski had not initially
envisioned a 2-year immersion in his research setting, as a Pole from Austria-Hungary
visiting British-controlled Melanesia when World War I erupted, Malinowski opted for
exile in the Trobriand Islands over internment for the duration of the war. Partially as
a result of this inadvertent long-term fieldwork, Malinowski came to believe that the
most effective approach to researching cultures was that of participant observation. He
laid out the principles underlying this methodology in his classic ethnography of the
Trobriand Islanders (Malinowski, 1922).
Revolutionary in its day, one of the key ideas underlying Malinowskis vision of par-
ticipant observation was that researchers armed with real scientific aims reside with
the people being studied for an extended period of time (usually at least a year to enable
observation of the annual cycle of events). As Malinowski wrote, Living in the village
with no other business but to follow native life, one sees the customs, ceremonies and
transactions over and over again, one has examples of their beliefs as they are actually
lived through, and the full body and blood of actual native life fills out soon the skeleton
of abstract constructions (Malinowski, 1922, p. 18). Malinowski believed that it was
only through participating in and observing everyday life over a long period of time that
anthropologists could hope to begin to see and experience the world through the eyes
of those whose lives one seeks to understand. Likewise, he reasoned that as the com-
munity under study becomes more accustomed to the anthropologists presence, the less
likely the anthropologist is to inadvertently alter community members normal activities.
Malinowski also emphasized that long-term participant observation enabled anthropol-
ogists to come to recognize the differences between societal rules for behavior and actual
behavior (what has been dubbed the difference between ideal culture and real culture).
Also, in a departure from the then-standard anthropological practice of relying
heavily on interpreters, Malinowski stressed the need for researchers to become profi-
cient in the local language. In his view, one could not possibly achieve ethnographys
aim to grasp the natives point of view, his relations to life, to realize his vision of his
world (Malinowski, 1922, p. 25) without mastery of the local lingua franca. Today,

DWYER 9781781001288 Part II.indd 340 25/07/2012 11:07

 Ethnographic methods 341

particularly for those who study global processes such as tourism, this means that eth-
nographic researchers (ethnographers) must often be multi-lingual, with proficiency in
a local language, national language, and some of the languages of international visitors.
As Malinowski and others have stressed, ethnographers culturally contextualized trans-
lations of quotations drawn from informal and formal conversations are a key form of
ethnographic data, and are heavily used in ethnographies.
For Malinowski, one of the main points of the ethnographic methods was to collect
concrete data over a wide range of facts, recording it meticulously in a fieldwork
log. As he emphasized, the obligation is not to enumerate a few examples only, but to
exhaust as far as possible all the cases within reach (Malinowski, 1922, pp. 1314). He
directed anthropologists to gather data on local activity patterns, institutions, folklore,
and even what he termed the inponderabilia of everyday life which included things
such as the routine of a mans working day, the details of the care of his body, the manner
of taking food and preparing it; the tone of conversation and social life ..., the existence
of strong friendships or hostilities, and of passing sympathies or dislikes between people;
the subtle yet unmistakable manner in which personal vanities and ambition are reflected
in the behaviour of the individual and in the emotional reactions of those who surround
him (Malinowski, 1922, pp. 1819). All of these, he stressed, were things which could
not possibly be recorded via interviews or via documents, but only through participant
observation (Malinowski, 1922, p. 18). These facts, Malinowski emphasized, should
not be simply recorded as detailed listings, but with an effort at penetrating the mental
attitude expressed in them (Malinowski, 1922, p. 19). That is, ethnographic methods
should entail a rich and diverse array of data and cases that, taken together, reveal the
emic or cultural insiders perspective (Pike, 1967). Thus, with Malinowskis invention
of participant observation, it was no longer acceptable for anthropologists to conduct
their research as distant observers working with translators and relying solely on inter-
views (Eriksen and Nielsen, 2001, p. 42). Rather, ethnographic methods became an
enterprise that centered on immersion, language mastery, participation, and cultural
interpretation.
By the early 1970s ethnographic research began to be envisioned as entailing thick
description, a concept that was popularized by anthropologist Clifford Geertz (1973).1
For Geertz, thick description involves unearthing and explicating the multiple,
nuanced, and layered webs of meaning of cultural behaviors, relationships, activi-
ties, rituals, and productions. This contrasts with thin description which tends to
be favoured by low-in-context conventional linear styles of research, and which are
prone to focusing exclusively upon the cause-and-effect power of single variables
(Hollinshead, 2004, p. 92). As Geertz tells us, From the point of view ... of the text-
book, doing ethnography is establishing rapport, selecting informants, transcribing
texts, taking genealogies, mapping fields, keeping a diary, and so on. But it is not these
things, techniques and received procedures that define the enterprise. What defines it is
the kind of intellectual effort it is: an elaborate venture in ... thick description (Geertz,
1973, p. 6).2
Dating back to the days of Malinowski, a major concern in ethnographic research
methods has been with finding ways to ensure the reliability of data collected via eth-
nographic research methods. Malinowskis early directive for field researchers to con-
stantly check ones data with multiple sources, rather than rely on a single informant

DWYER 9781781001288 Part II.indd 341 25/07/2012 11:07

 342 Handbook of research methods in tourism

(the term used in Malinowskis day: today anthropologists prefer the terms mentor,
collaborator, or consultant) was echoed and elaborated upon by Raymond Firth,
who cautioned that individuals assertions should not simply be accepted as truths, but
instead should be approached as reflections of the position and interests of the people
who give them (Firth, 1965, p. 3, cited in Duranti, 1997, p. 91). Many researchers draw
on the strategy of respondent validation, a process of presenting ones interpretations
and write-ups of findings to various participants for checking (for a tourism study that
draws on and discusses this strategy, see Cole, 2005, p. 64).
The mid-1980s blossoming of the post-modernist movement brought new attention
to the ways in which the researchers racial, class, national and gender identity, and
personality shape the data-collection and interpretation processes, producing not objec-
tive scientific reports but what some termed partial truths, a mixture of objective and
subjective observations (see Clifford and Marcus, 1986). Since this crisis moment in the
evolution of ethnographic methods, most who embrace ethnographic methods have
sought to compensate for these biases in various ways. Many ethnographers adopt a
self-reflexive stance, revealing the particulars of their identity and the circumstances of
their field research and embracing the first person singular in their writings to signal their
positionality. For an example of effective use of this strategy in tourism research, see
Causeys (2003) penetrating study of Toba Batak carvers and tourist interactions in the
handicraft market. In this study, Causey shares his experiences of his informal appren-
ticeship with a Toba Batak carver and in so doing readers come to better understand
the challenges local artisans face and the ways in which ways in which they adjust their
stories and products to the tourist market. As Simonds et al. (2012) summarize, in the
end, ethnographers are content to recognize that their observations are at once subjective
interpretations of a mere fragment of their own constrained perceptions of social experi-
ences and at the same time sensitive and intensely engaging but fleeting glimpses of life
as it is lived by others.
A discussion of ethnographic research methods would not be complete without
addressing ethics. It is essential to ensure that those we are studying are not com-
promised by our research, or made unwitting participants. Researchers are ethically
required to secure permission for their research from the relevant gate-keepers (human
subjects boards, national and local governments, etc.), and to reveal their research aims
to those whose communities they are studying. Additionally, informed consent forms
are the norm (Spradley, 1980, pp. 2025). Generally, ethnographers are also careful
to ensure the anonymity of community members, and use coded names in field notes
and other writings, unless they determine that participants prefer to be named in their
research publications.

BACKGROUND AND TYPES OF PROBLEMS THAT

ETHNOGRAPHY IS DESIGNED TO HANDLE

As many have noted, the clearest value of ethnography is in terms of its relationship to
developing theories pertaining to social life (see Hammersley and Atkinson, 1991, p.
23). Engaging in ethnographic methods enables researchers to recognize their own mis-
leading preconceptions concerning the motives underlying human behavior. Long-term

DWYER 9781781001288 Part II.indd 342 25/07/2012 11:07

 Ethnographic methods 343

immersion in the research setting inevitably forces the researcher to reassess his or her
prior understandings of the dynamics at play in the cultural setting under study. As the
researchers understanding shifts over time, he or she can begin to develop theory in a
way that provides much more evidence of the plausibility of different lines of analysis
than is available to the armchair theorist, or even the survey researcher or experimental-
ist (Hammersley and Atkinson, p. 24).
Ethnographic research methods are also ideally suited for tapping into local points of
view. Ethnography can effectively unearth local peoples perceptions regarding changes,
challenges and triumphs in their worlds. Local voices particularly those of minority
groups who may speak different languages, be illiterate, impoverished and relatively
disenfranchised from their nations are notoriously difficult to access via classic survey
methods and questionnaire-based interviews. Fear of giving the wrong answer, mis-
trust of outsiders, reluctance to speak honestly (and critically) to unknown temporarily
visiting researchers, and power imbalances can compromise the validity of other research
methodologies. As Berno observed in her discussion of her research on tourism in the
Cook Islands, due to a lack of familiarity with structured questionnaires and social
science research in general, subjects tended to acquiesce, and a positive response bias
was evident on questions that dealt with satisfaction with tourism and tourists (Berno,
1996, p. 384, cited in Cole, 2004, p. 295). Long-term participant observation, however,
enables researchers to gain the trust of the community members whose lives, perceptions,
and attitudes they aspire to understand. Describing her ethnographic research in Eastern
Indonesia, Cole underscores that spontaneous, indoor fireside chats were a more suc-
cessful technique than attempting to carry out questionnaire-based interviews ... [and]
disclosed information on topics that were not openly discussed at other times (Cole,
2004, pp. 2956).
This methodology is especially valuable for offering insights into how people concep-
tualize themselves vis--vis others, as well as into how groups are organized and how
they perceive newcomers, be they migrants, tourists, new businesses, or new development
projects. These attributes make ethnographic research particularly useful for gaining
understandings of the dynamics underlying conflict situations. Ethnographic research
can offer fuller data on the issues lurking behind the scenes of questionnaires and quan-
tified surveys (Simonds et al., 2012). Members of disgruntled groups may be unwilling
to fully air their critiques in surveys or formal interviews (particularly when they are
unsure about the extent to which confidentiality will be maintained), but more likely to
convey their frustrations to a long-term field researcher who has gained the trust of the
community. The experience of participant observation enables the ethnographer to gain
a first-hand appreciation of the basis of the conflict. As both insider and outsider, the
ethnographer is well situated to identify possible avenues for conflict resolution between
local groups and external agencies, be they private or governmental (see Adams, 2011).

APPLICATIONS OF ETHNOGRAPHIC METHODS TO TOURISM

STUDIES

As noted above, ethnographic research strategies yield rich and nuanced understand-
ings of how people make sense of and engage with the world. In this regard, they are

DWYER 9781781001288 Part II.indd 343 25/07/2012 11:07

 344 Handbook of research methods in tourism

valuable for understanding tourists perspectives on their travel experiences, as well

as on the broader ritualized meanings of their voyages vis--vis the rest of their lives.
Nelson Graburn observes that this methodological emphasis on meaning might not
reveal much about ... [tourists] spending, demonstration effects or other measures of
impact; nor does it necessarily relate to their satisfaction or the likelihood of their return,
though it might. More importantly, meaning has to do with the feelings experienced and
the values expressed by temporary travelers, almost always in relation to their whole ...
lives (Graburn, 2002, p. 30). In this regard, ethnographic methods have been used by
tourism scholars to illuminate how travel experiences can impact tourists perceptions of
other nations as well as prompt reassessments, reaffirmations or transformations of their
own personal, religious, or national identities. For example, Shaul Kelners (2010) study
of Israeli birthright tourism draws on his participant observation on homeland tours for
Jewish young adults that are sponsored by North American Jewish organizations and
the State of Israel. These free tours are designed to foster diaspora Jewish engagement
with Israel. In addition to participating in these tours, Kelner also spent time with guides
and observed guiding program activities. As Kelner concludes, although these diaspora
tours may not necessarily change young travelers identities, they can foster emotionally
compelling, embodied experiences that re-ground these young visitors process of self-
creation. The on-tour activities encourage non-Israeli Jews to strike symbolic roots in
a place where they do not live ... through their actions they inscribe place onto self and
self onto place, recreating their understandings of themselves and of Israel in the meeting
of the two (Kelner, 2010, p. 198).
Tourism scholars have also drawn on ethnographic methods to examine on-the-
ground dimensions of tourisms relationship to nation-building and nationalism (see
Adams, 1998; Babb, 2011; Bruner, 2001; Cohen, 2010; Edensor, 1998; Sanchez and
Adams, 2008). For example, the differences between the practices and orientations of
domestic and international tourists in Asia were seldom addressed by scholars and
planners until several researchers engaged in participant observation and chronicled
their varying interpretations of what they were seeing. Nelson Graburns field research
in Japan (1983) prompted the realization that the language and framing of domestic
tourism in Japan was quite different from that of international tourism. Likewise, my
participant observation in Indonesias ethnic tourism destination of Tana Toraja identi-
fied and examined the different framings and meanings underlying domestic and inter-
national tourists visits to this Indonesian hinterland region. Whereas the Indonesian
government frames domestic tourism in terms of reaffirming regional and national inter-
ethnic ties (contrasting with the ways in which international travelers visits are framed),
my participant observation revealed that government visions of building cross-cutting
ties via domestic tourism were often complicated and confounded by prior inter-ethnic
or religious tensions. In short, despite governmental framing of tourism as an avenue for
strengthening the ties between the diverse groups within the nation, domestic tourism
does not automatically build inter-ethnic bridges: sometimes it reignites long-simmering
tensions (Adams, 1998).
In a similar vein, several tourism scholars have engaged in participant observation
at Indonesias Taman Mini ethnic theme park and elaborated on the complex ways in
which this park, designed to foster nation-building and nationalism, challenges some of
our key tourism theories (Bruner, 2005; Hitchcock, 1998). For instance, Bruner notes

DWYER 9781781001288 Part II.indd 344 25/07/2012 11:07

 Ethnographic methods 345

that when Toba Batak tourists visit the Batak section of this ethnic theme park in their
nations capital, they do not go with Urrys tourist gaze seeking experiences different
from those of their everyday life, but rather they go to see what they know they are.
They do not discover the other but rather witness a performance of themselves in a dif-
ferent context (Bruner, 2005, p. 227). In this sense, then, ethnographic methods have
effectively enabled us to better understand how tourism can foster self-construction.
More broadly, ethnographic field methods are well suited for shedding light on tour-
isms role in shaping particular communities and cultures, insights that are less likely
to be culled from questionnaires and surveys. For instance, Hazel Tuckers (2003)
long ethnographic research in Goreme, Turkey revealed some of the new cultural
forms arising out of encounters between tourists and locals: her findings challenge the
dominant stereotype of tourism as a force that uniformly disempowers locals: rather,
relationships between insiders and outsiders are constantly negotiated. With regards to
tourism and gender relations, Tucker notes that romantic relationships between tourist
women and local men are creating new expanded ideas about gender relations in the
village (Tucker, 2003). Likewise, Maribeth Erbs long-term participant observation on
the Eastern Indonesian island of Flores enabled her to observe, document, and interpret
local debates surrounding the rebuilding of traditional houses with an eye to the poten-
tial to attract tourists. This methodological approach led Erb to understand that these
processes were not instances in which the tourism was destroying traditional culture,
but rather instances in which interactions with foreign tourists (as well as foreign priests
and the state) were introducing new yardsticks for authenticity and enabling local people
to make claims for different kinds of authenticity in different contexts (Erb, 1998, 2009).
Still other scholars have used ethnographic methods to document the reorganization of
local labor, artistic meanings, and conceptions of traditions when ethnic crafts and music
are marketed to tourists (see Causey, 2003; Cohen, 2000; Forshee, 2001).
Ethnographic data gathering methods have also offered rich illustrations of how
tourism transforms or reaffirms ethnic and racial hierarchies (see Adams, 2006;
Guerrn-Montero, 2006; Stronza, 2008; van den Berghe, 1994). For example, Pierre
van den Berghes pioneering ethnographic research in Peru enabled us to recognize
that tourism is always superimposed on pre-existing structures of ethnic hierarchies
(van den Berghe, 1980). Van den Berghes more recent fieldwork on ethnic tourism
in San Cristobal, Mexico, revealed a tri-partite tourism arena comprised of wealthier
foreign and urban Mexican tourists, predominantly ladino middlemen who tend to reap
the greatest financial gains from tourism and indigenous Maya tourees who are the
objects of tourists attention (van den Berghe, 1994). Ultimately, his fieldwork under-
scores the ways in which ethnic divisions of labor can underlie and sustain tourism in
various destinations.
Finally, ethnographic methods are also particularly well suited for gathering data on
delicate subjects such as money. For instance, in her study of how tourism employment
has transformed Bulgarian womens lives, Kristen Ghodsee recounts how her attempts
to gather income information via surveys yielded worthless data. It was only via par-
ticipant observation and informal interviews that she came to fully appreciate the inap-
propriateness of the questions that framed her earlier survey (Ghodsee, 2005, pp. 667).
Likewise, while Ghodsee did not feel comfortable asking direct questions about illicit
sources of tourism-related income, participant observation and informal conversations

DWYER 9781781001288 Part II.indd 345 25/07/2012 11:07

 346 Handbook of research methods in tourism

enabled her to learn about income-supplementing strategies of over-charging and short-

changing foreign tourists (Ghodsee, pp. 723).
One last advantage of ethnographic research methods is worth underscoring before
turning to review the limitations of this methodology. As some scholars have noted,
the multi-stranded nature of ethnographic research (which combines long-term
participant observation with interview, archival and other data) forms a solid basis
for triangulation and enables researchers to systematically compare data culled from
various sources (Hammersley and Atkinson, 1991, p. 24). Hammersley and Atkinson
underscore that this avoids the risks that stem from reliance on a single kind of data:
the possibility that ones findings are method-dependent (Hammersley and Atkinson,
1991, p. 24).

ADVANTAGES AND LIMITATIONS OF ETHNOGRAPHIC

RESEARCH METHODS

Ethnographic research methods have potential value for policy formulation and plan-
ning that has yet to be fully tapped. First, ethnographic work is particularly adept at
giving voice to minorities and at mapping out competing regimes of power at tourism
sites. Tapping into this sort of data early can result in the formulation of policies that
are more likely to circumvent inequities and conflict at tourism sites slated for develop-
ment. More generally, ethnography produces a more textured appreciation of the varied
responses to prospective tourism development projects in any given community.
Second, ethnographic research methods can also yield insights into the underlying
dynamics of on-going clashes at established tourism destinations and this information is
valuable for wise policy formulation. Ethnographic research demands that one be atten-
tive to the diverse groups interacting in and beyond the tourist sites, be they domestic
tourists, foreign tourists, guides, hoteliers, crafts-makers, souvenir sellers, restaurant and
bar employees, local residents, planners, tour organizers, transport workers, or travel
writers. Ethnography can shed light on the different visions each of these groups has of
a tourist site, and how these visions may clash (see Bruner, 1993, 1996; Notar, 2006).
For instance, Bruners study of tourism at Ghanas slave castles revealed dramatic con-
trasts between Ghanaian visions of the slave castles as a route to development, African-
American tourists conception of the slave castles as sacred pilgrimage space that should
not be commercialized, and other foreign or domestic tourists interest in the slave castles
due to their historic roles as other things (past Dutch and British colonial headquarters,
site where an important Asante king was imprisoned in 1896, etc). As Bruner (1993)
observed, conflicts and frustrations are aroused when different claims about the signifi-
cance of the space were made simultaneously. His study points towards the challenge
faced by planners: how to represent dynamic spaces with long histories to vastly different
audiences in such a way that all (or at least many) are satisfied? Via the ethnographic
data, however, planners can gain clues as to how to avoid some of the pitfalls inherent
in this challenge.
Third, ethnographic research can potentially point towards avenues for bridge-
building between diverse stake-holders in touristic settings. Ethnographys orientations
towards the broader cultural and political contexts in which tourism unfolds means we

DWYER 9781781001288 Part II.indd 346 25/07/2012 11:07

 Ethnographic methods 347

engage in participant observation not only in the arenas of tourism enterprise, but well
outside the areas directly tied to the tourism industry. And, sometimes, we find that
avenues for tourism conflict resolution lie in spaces and institutions that, on the surface,
have little connection to tourism, per se. For example, as part of my field research in
the California mission tourism town of San Juan Capistrano (Adams, 2011) I began
attending a weekly coffee chat that had begun as a journalists short-cut for gather-
ing local news stories. These open caf gatherings soon blossomed into an arena that
drew into dialogue both powerful and minority community members and called to
mind Habermass observations on the historic role of British coffee houses in fostering
democratic civic engagement. On several occasions, the coffee chat became a forum for
constructive tourism problem resolution. It is worth underscoring that the pivotal role
of the coffee chat as an arena for bringing together diverse stakeholders in dialogue and
ultimately resolving potentially explosive tourism-related conflicts could not have been
uncovered easily without ethnographic fieldwork, since the coffee chat forum did not
have any obvious relation to tourism. It is in these sorts of ways that the broader com-
munity and beyond scope of ethnographic research can help steer policy makers and
planners towards unexpected new avenues for fostering constructive dialogue between
diverse stakeholders in tourism sites.
There are, however, some limitations of this technique that are worth noting.
First, ethnographic research is tremendously time-consuming. Policy makers cannot
always wait a year for ethnographic data to be gathered and still more months for the
researcher to analyse the massive amount of data produced by ethnographic research
(for a fuller discussion of this critique, see Yin, 1989, pp. 212). Second, as with other
qualitative methods, some feel that the danger of researcher bias inherent in ethno-
graphic methods render them less appealing than statistical material or survey responses
(Sandiford and Ap, 1998, p. 11). In short, some have critiqued ethnographic methods
as a soft methodology (Phillimore and Goodson, 2004, pp. 34).3 Third, perhaps
ironically, ethnographic researchs ability to produce rich, nuanced accounts of peoples
understandings of tourist sites has also been envisioned by some as one reason for its lim-
itations for policy formulation (Filippucci, 2009, p. 321). As Paola Filippucci notes, the
diverse narratives we collect in tourist settings can often reveal contradictory perspec-
tives that are impossible to translate into recommendations to public policy agencies
charged with developing local heritage (Filippucci, 2009, p. 319). Finally, still others
have critiqued ethnographic research methods as too specific to be replicated or applied
generally (Alder and Alder, 1994, cited in Hollinshead, 2004).

FUTURE DEVELOPMENTS

While efforts to accurately predict the future are notoriously difficult due to the many
variables on which such predictions are based (Prideaux, 2002, p. 221), some recent
developments have indicated new directions in the evolution of ethnographic methods
that are pertinent to tourism studies. These include social scientists growing recognition
of the ways in which local sites are tethered to distant regions of the globe, new computer
technology, new tweakings of ethnographic methods themselves, and increasing expec-
tations on the part of local stakeholders to have a say in tourism development policies.

DWYER 9781781001288 Part II.indd 347 25/07/2012 11:07

 348 Handbook of research methods in tourism

This section briefly addresses the potential ramifications of these developments for eth-
nographic field research methods in tourism studies.
In recent years, scholars have become increasingly cognizant of the ramifications of
accelerated globalization for tourism studies: the communities tourism ethnographers
study are tethered to global flows, and that tourists are not the only ones to be moving in
and out of these destinations: local guides go for training sessions outside the region (see
Salazar, 2010), romances with foreign tourists may result in tourates relocating to distant
nations (Brennan, 2004), traders move ethnic art wares and tourist souvenirs across great
distances (see Forshee, 2001), and even tourism campaigns and slogans travel the globe.
In tandem with these developments, tourism ethnographers have grappled with new
methodological challenges and have begun what some have dubbed a post-local era,
casting their ethnographic lens to multiple locales to foster richer, more complex under-
standings of the workings of tourism in our hyper-global era. One pioneering example
of this new development in qualitative research methods is Noel Salazars multi-sited
ethnography, which draws on two years of what he terms glocal ethnography (Salazar,
2011) to examine tour guide training and performance in Yogyakarta, Indonesia, and
Arusha, Tanzania. Salazars ethnography is deeply rooted in the destinations he con-
siders and offers rich documentation of how tour guides practices, narratives, and
aspirations are informed by globally circulating imaginaries of the past and the future
(Salazar, 2010). Drawing on multi-sited ethnographic research techniques to study the
local-to-global dynamics of tourism processes, as well as the power trail of touristic
systems is clearly a productive avenue for future research.
New technological developments, in tandem with a global economic and ecological
crisis, also have implications for the evolution of ethnographic research methods suitable
for studying new tourism trends. As Prideaux (2002) has noted, future tourists may be
alternating between real and cyber experiences. In his words, harsh political and/or eco-
nomic realities may force a retreat of the tourist to more confined areas, and the taste for
real-time travel may be restricted or substituted by cybertourism experiences (p. 324).
Such a scenario poses new challenges for adapting ethnographic research techniques
to the highly individualized and personalized experiences of this new genre of tourism.
Some scholars have begun exploring the methodological terrain of virtual ethnogra-
phy, or netography (see Hall, 2011; Kozinets, 2010). In addition, one possible avenue
for adapting ethnographic methods to fit these new forms of tourism is that of auto-
ethnography. Auto-ethnography is a newer form of ethnographic research that draws on
the researchers own experience as a source of data. This approach has already been pro-
ductively applied to tourism topics by various scholars (see Morgan and Pritchard, 2005;
Walsh and Tucker, 2009). For instance, Morgan and Pritchard (2005) have examined
their uses and narratives of their own collections of tourist arts amassed over lifetimes of
travel, ultimately shedding light on the relations between material things, tourism, and
constructions of self-identity. It is likely that future tourism researchers will be increas-
ingly adding auto-ethnography to their methodological toolkit.
Finally, in an era where local communities are increasingly expecting to have a say
in tourism planning decisions that will impact their lives, a growing number of tourism
researchers are drawing on ethnographic research methods to facilitate this process (see
Cohen, 2010; Jamal and Stronza, 2009; Stronza, 2005, 2010). For instance, Amanda
Stronza (2005) has engaged in participatory ethnographic research (what some term

DWYER 9781781001288 Part II.indd 348 25/07/2012 11:07

 Ethnographic methods 349

collaborative ethnographic methods) in a Peruvian Amazon ecotourism destination to

shed light on and help steer planning in the direction that local stakeholders envision for
ecotourism development. This is a promising path for future tourism scholars, as insights
culled from collaborative ethnographic research methods can lead us to find ways to
skirt the classic problems by tourism development, yielding more contented stakeholders
and tourists.

NOTES

1. This expression derives from the British philosopher Gilbert Ryles lectures (1971). Norman Denzin (1989)
further explicated the scope of thick description in qualitative ethnographic research, delineating and offer-
ing examples of 11 genres of thick description. For a full exploration of the origins and evolution of this
concept in qualitative research, see Ponterotto (2006).
2. Hollinshead (1991) offers a discussion and adjustment of Geertzs vision of ethnography for tourism schol-
ars.
3. However, qualitative researchers would rebut that surveys are also susceptible to researcher bias and that
fully objective research is a cultural myth, not a reality.

REFERENCES

Adams, K.M. (1998), Domestic tourism and nation-building in South Sulawesi, Indonesia, Indonesia and the
Malay World, 26 (75), 7797.
Adams, K.M. (2006), Art as Politics: Re-crafting Identities, Tourism and Power in Tana Toraja, Indonesia,
Honolulu: University of Hawaii Press.
Adams, K.M. (2011), Public interest anthropology, political market squares, and re-scripting dominance:
from swallows to race in San Juan Capistrano, CA, Journal of Policy Research in Tourism, Leisure and
Events, 3 (2), 147169.
Alder, P.A. and P. Alder (1994), Observation technique, in N.K. Denzin and Y.S. Lincoln (eds), Handbook of
Qualitative Research, Thousand Oaks, CA: Sage.
Babb, F. (2011), The Tourism Encounter: Fashioning Latin American Nations and Histories, Stanford, CA:
Stanford University Press.
Berno, T. (1996), Cross-cultural research methods: content or context? A Cook Island example, in
R.Butlerand T. Hinch (eds), Tourism and Indigenous Peoples, London: International Thompson Business
Press
Boas, F. (1888), The Central Eskimo, Sixth Annual report of the Bureau of American Ethnology, Smithsonian
Institution, Washington DC.
Brennan, D. (2004), Whats Love Got to do with It? Transnational Desire and Sex Tourism in the Dominican
Republic, Durham, NC, and London: Duke University Press.
Bruner, E. (1993), Tourism in Ghana: the representation of slavery and the return of the black diaspora,
American Anthropologist, 98 (2), 290304.
Bruner, E. (1996), Lincolns New Salem as a contested site, Museum Anthropology, 17 (3), 1425.
Bruner, E. (2001), The Maasai and the Lion King: authenticity, nationalism and globalization in African
tourism, American Ethnologist, 28 (4), 881908.
Bruner, E. (2005), Taman Mini: self constructions in an ethnic theme park in Indonesia, in E. Bruner, Culture
on Tour, Chicago, IL: University of Chicago Press, pp. 211230.
Causey, A. (2003), Hard Bargaining in Sumatra: Western Travelers and Toba Bataks in the Marketplace of
Souvenirs, Honolulu: University of Hawaii Press.
Clifford, J. and G.E. Marcus (1986), Writing Culture: The Poetics and Politics of Ethnography, Berkeley, CA:
University of California Press.
Cohen, B.C. (2010), Take Me to My Paradise: Tourism and Nationalism in the British Virgin Islands,
Piscathaway, NJ: Rutgers University Press.
Cohen, E. (2000), The Commercialized Crafts of Thailand: Hill Tribes and Lowland Villages, London: Curzon
Press and Honolulu: University of Hawaii Press.
Cole, S. (2004), Shared benefits: longitudinal research in Eastern Indonesia, in J. Phillimore and L. Goodson

DWYER 9781781001288 Part II.indd 349 25/07/2012 11:07

 350 Handbook of research methods in tourism

(eds), Qualitative Research in Tourism: Ontologies, Epistemologies and Methodologies, Abingdon, UK:
Routledge, pp. 292310.
Cole, S. (2005), Action ethnography: using participant observation, in B.W. Ritchie, P. Burns and C. Palmer
(eds), Tourism Research Methods: Integrating Theory With Practice, Wallingford, UK, and Cambridge, MA:
CABI Publishing, pp. 6372.
Denzin, N.K. (1989), Interpretive Interactionism, Newbury Park, CA: Sage.
Duranti, A. (1997), Linguistic Anthropology, Cambridge and New York: Cambridge University Press.
Edensor, T. (1998), Tourists at the Taj: Performance and Meaning at a Symbolic Site, London: Routledge.
Ellen, R.F. (1984), Ethnographic Research, London: Academic Press Inc.
Erb, M. (1998), Tourism space in Manggarai, Western Flores, Indonesia: The house as a contested place,
Singapore Journal of Tropical Geography, 19 (2), 177198.
Erb, M. (2009), Tourism as glitter, in P. Toe, T.C. Chang and T. Winter (eds), Asia on Tour: Exploring the
Rise of Asian Tourism, London: Routledge, pp. 170182.
Eriksen, T.H. and F.S. Nielsen (2001), A History of Anthropology, London and Sterling, VA: Pluto Press.
Fife, W. (2005), Doing Fieldwork: Ethnographic Methods for Research in Developing Countries and Beyond,
New York and Hampshire, UK: Palgrave Macmillan.
Filippucci, P. (2009), Heritage and methodology: a view from social anthropology, in M.L.S. Srensen and J.
Carman (eds), Heritage Studies: Methods and Approaches, Hoboken, NJ: Routledge, pp. 319325.
Firth, R. (1965), Primitive Polynesian Economy, New York: Norton.
Forshee, J. (2001), Between the Folds: Stories of Cloth, Lives and Travels from Sumba, Honolulu: University
of Hawaii Press.
Geertz, C. (1973), The Interpretation of Cultures: Selected Essays, New York: Basic Books.
Geertz, C. (1976), From the natives point of view: on the nature of anthropological understanding, in K.
Basso and H. Selby (eds), Meaning in Anthropology, Albuquerque, NM: University of New Mexico Press,
pp. 221238.
Ghodsee, K. (2005), The Red Riviera: Gender, Tourism and Postsocialism on the Black Sea, Durham, NC: Duke
University Press.
Graburn, N. (1983), To Pray, Pay, and Play: The Cultural Structure of Japanese Domestic Tourism, Aix-en-
Provence, France: Centre des Hautes Etudes Touristiques.
Graburn, N.H.H. (2002), The ethnographic tourist, in G. Dann (ed.), The Tourist As Metaphor of the Social
World, Wallingford, UK and Cambridge, MA: CABI Publishing, pp. 1938.
Guerrn-Montero, C. (2006), Cant beat me own drum in me own native land: Calypso music and tourism in
the Panamanian Atlantic coast, Anthropological Quarterly, 79 (4), 633663.
Hall, C.M. (2011), In cyberspace can anybody hear you scream? Issues in the conduct of online fieldwork,
in C.M. Hall (ed.), Fieldwork in Tourism: Methods, Issues and Reflections, Abingdon, UK, and New York:
Routledge, 267288.
Hammersley, M. and P. Atkinson (1991), Ethnography: Principles in Practice, London and New York:
Routledge.
Hitchcock, M. (1998), Tourism, Taman Mini, and national identity, Indonesia and the Malay World, 26 (75),
12435.
Hollinshead, K. (1991), The scientific nature of anthropology, Annals of Tourism Research, 18, 653660.
Hollinshead, K. (2004), Ontological craft in tourism studies: the productive mapping of identity and image
in tourism settings, in J. Phillimore and L. Goodson (eds), Qualitative Research in Tourism: Ontologies,
Epistemologies and Methodologies, Abingdon, UK: Routledge, pp. 83101.
Jamal, T. and A. Stronza (2009), Collaboration theory and ecotourism practice in protected areas, Journal of
Sustainable Tourism, 17 (2):169189.
Kelner, S. (2010), Tours That Bind: Diaspora, Pilgrimage and Israeli Birthright Tourism, New York and
London: New York University Press.
Kozinets, R. (2010), Netography: Doing Ethnographic Research Online, London: Sage.
Malinowski, B. (1922), Argonauts of the Western Pacific: An Account of Native Enterprise and Adventure in the
Archipelagoes of Melanesian New Guinea, London: Routledge and Kegan Paul.
Morgan, N. and A. Pritchard (2005), On souvenirs and metonymy: narratives of memory, metaphor and
materiality, Tourist Studies, 5 (1), 2953.
Nash, D. (2000), Ethnographic windows on tourism, Tourism Recreation Research, 25 (3), 2935.
Notar, B. (2006), Displacing Desire: Travel and Popular Culture in China, Honolulu: University of Hawaii
Press.
Palmer, C. (2001), Ethnography: a research method in practice, International Journal of Tourism Research, 3
(4), 301312.
Palmer, C. (2009), Reflections on the practice of ethnography within heritage tourism, in M.L.S. Srensen
and J. Carman (eds), Heritage Studies: Methods and Approaches, Hoboken, NJ: Routledge, pp. 123139.
Phillimore, J. and L. Goodson (2004), Progress in qualitative research in tourism: epistemology, ontology,

DWYER 9781781001288 Part II.indd 350 25/07/2012 11:07

 Ethnographic methods 351

and methodology, in J. Phillimore and L. Goodson (eds), Qualitative Research in Tourism: Ontologies,
Epistemologies and Methodologies, Abingdon, UK: Routledge, pp. 329.
Pike, K.L. (1967), Etic and emic standpoints for the description of behavior, in D.C. Hildum (ed.), Language
and Thought: An Enduring Problem in Psychology, Princeton, NJ: D. Van Norstrand Company, pp. 3239.
Ponterotto, J. (2006), Brief note on the origins, evolution, and meaning of the qualitative research concept
thick description, Qualitative Research, 11 (3), 538549.
Prideaux, B. (2002), The cybertourist, in G. Dann (ed.), The Tourist As Metaphor of the Social World,
Wallingford, UK, and Cambridge, MA: CABI Publishing, pp. 317339.
Ryle, G. (1971), Collected Papers, Volume II, Collected Essays 19291968, London: Hutchingson.
Salazar, N. (2010), Envisioning Eden: Mobilizing Imaginaries in Tourism and Beyond, New York and Oxford:
Berghahn Books.
Salazar, N. (2011), Studying local-to-glocal tourism dynamics through glocal ethnography, in C.M. Hall
(ed.), Fieldwork in Tourism: Methods, Issues and Reflections, Abingdon, UK and New York: Routledge, pp.
177187.
Sanchez, P. and K. Adams (2008), The Janus-faced character of tourism in Cuba: ideological continuity and
change, Annals of Tourism Research, 35 (1), 2746.
Sandiford, P.J. and J. Ap (1998), The role of ethnographic techniques in tourism planning, Journal of Travel
Research, 37 (1), 311.
Simonds, L.M., P.M. Camic and A. Causey (2012), Using focused ethnography in psychological research, in
H. Cooper, P.M. Camic, D.L. Long, A.T. Panter, D. Rindskopf and K. Sher (eds), American Psychological
Association Handbook of Research Methods in Psychology: Vol. 2: Research designs: Quantitative, Qualitative,
Neuropsychological, and Biological, Washington DC: American Psychological Association.
Spradley, J.P. (1980), Participant Observation, New York: Holt, Rinehart, and Winston.
Stronza, A. (2005), Trueque Amaznico: Lessons in Community-based Ecotourism, Washington DC: Critical
Ecosystem Partnership Fund.
Stronza, A. (2008), Through a new mirror: reflections on tourism and identity in the Amazon, Human
Organization, 67 (3), 244257.
Stronza, A. (2010), Commons management and ecotourism: ethnographic evidence from the Amazon,
International Journal of the Commons, 4 (1), 5677.
Tucker, H. (2003), Living With Tourism: Negotiating Identities in a Turkish Village, London: Routledge.
van den Berghe, P. (1980), Tourism as ethnic relations: a case study of Cuzco, Peru, Ethnic and Racial Studies,
3 (4), 375392.
van den Berghe, P. (1994), The Quest for the Other: Ethnic Tourism in San Cristobal, Mexico, Seattle, WA:
University of Washington Press.
Walsh, N. and H. Tucker (2009), Tourism things: the travelling performance of the backpack, Tourist
Studies, 9 (3), 223239.
Yin, R.K. (1989), Case Study Research: Design and Methods (revised edition), London: Sage.

DWYER 9781781001288 Part II.indd 351 25/07/2012 11:07

 17 Focus groups
Carl Cater and Tiffany Low

NATURE OF THE TECHNIQUE AND ITS EVOLUTION

Focus groups are one of the core methods in the qualitative researchers toolkit, for their
ability to canvass opinions of a range of stakeholders in a relatively efficient format. In
their simplest form they take the format of a group of people discussing a particular
issue, with the researcher acting as facilitator. Although there is an issue or topic under
debate, the method is relatively unstructured, relying on the interaction of participants to
keep the discussion moving. Despite the fact that they are quick to complete, Bosco and
Herman (2010) argue that focus groups are one of the most engaging research methods
available. In this chapter we explore the development of focus groups, and their poten-
tial for collaborative research. Applications of the technique within tourism research
are discussed, alongside some of the advantages and disadvantages of the technique. It
seems apparent, however, that focus groups have been underutilized in tourism research,
particularly in their critical form, and yet they have much to offer.
Focus groups are particularly useful when used in combination with other methods.
They may be useful in the early stages of research for providing a broad overview of a
topic and generating questions that can be tested by other methods. Alternatively they
may be used to test generalizations and theories that are generated by other methods.
As Goss (1996, p. 113) demonstrates, they are of use throughout the research process,
from orienting the researcher to a new field; generating hypotheses that can be tested by
quantitative research; identifying appropriate concepts for the development of question-
naires; following up surveys and helping in the interpretation of quantitative data; and
presenting results to the community for validation.
Focus groups have their modern origin as an evaluative method developed to assess
audience reactions to radio programs, although they may have existed informally much
earlier. The structuring of the method developed to assess US Army training and propa-
ganda films during World War II (Goss, 1996). The focus here was a specific piece of
media which the researcher had already analysed, and discussion was moderated by a
trained facilitator working from a guide. Continuing from its inception, focus group
research is extensively used in marketing and media studies, although this format tends
to be of a more structured style. In consumer research storyboards or other scenarios are
used to evaluate the effectiveness of media messages. In the movie industry focus groups
are often used with test screenings of films to assess audience reaction, and feedback may
often result in significant changes to film storylines prior to general release. However,
these marketing-oriented focus groups have come under some critique for their Fordist
approach to data collection, particularly where they are used in support of business and
commercial pursuits (Bosco and Herman, 2010;).
In critical academic work, focus groups have been most developed within sociology,
with the pioneer of the technique being recognized as Robert K. Merton (Bosco and

352

DWYER 9781781001288 Part II.indd 352 25/07/2012 11:07

 Focus groups 353

Herman; 2010 Goss, 1996). Building on work conducted with the homeless during the
Great Depression, Merton refined his technique as the focussed interview, developed
with the backdrop of an interest in the sociology of knowledge, and a growing interest
in phenomenological approaches. In later years Merton had concerns about the appli-
cation of his technique without such a theoretical background, lacking the ability to
extend prediction of audience response to examination of human behaviour (Bosco and
Herman, 2010). Nevertheless, Merton was keen to stress the wide application of focus
groups, emphasizing that the focussed interview is a generic technique, that could and
would be applied in every sphere of human behavior and experience, rather than largely
confined to matters of interest in marketing research (Merton, 1987, p. 551).

BACKGROUND AND TYPES OF PROBLEMS THAT FOCUS

GROUPS ARE DESIGNED TO HANDLE

It is clear that there is a division between traditional focus groups commonly used in
marketing studies, and in-depth groups more useful for critical research (Bosco and
Herman, 2010). The ability for the latter to present a more grounded form of research
is highly important. As Weber (2001) suggests, many previous approaches to the study
of tourism have been dominated by etic research, imposing theoretical conceptualiza-
tions of the experience upon the study community. Historically there have been fewer
attempts at more emic research, which would seek to uncover tourism from the view of
the subject, and in particular recognize the variety in individuals experiences. Similar to
interviews, focus groups are qualitative in nature, and provide rich material for analysis.
However, their more unstructured nature does allow the respondents to impart their
own reality . . . cataloguing the socially constructed knowledge of informants rather
than the hypothesising of the investigator (Riley, 1995, p. 636). In this sense, the voices
of respondents in focus groups are representative of the emic approach that Weber
(2001) supports. By combining the academics perception with an insiders view of the
way of life under consideration, the researcher can thus get behind the statistical shapes
and patterns ... studying them from the value terms of the people themselves, in their
own terms and on their own ground (Pryce, 1979, p. 279). This fits with the theory of
symbolic interaction which treats theory as something that should be brought into line
with the empirical world of lived experience, rather than the other way round (McCall
and Becker, 1990).
The dynamic nature of focus groups is grounded in an interactionist tradition which
pays attention to the intersubjective ways in which people understand their lived experi-
ence (Cloke et al., 1991). The goal of focus groups is not to create one single meaning, but
to develop multiple meanings from shared discussion (Finn et al., 2000). In a focus group
there is the opportunity for the researcher to interact directly with respondents and each
other in a social setting. This further justifies an interactionist approach, where any
human event can be understood as the result of the people involved continually adjust-
ing what they do in the light of what others do, so that each individuals line of action
fits into what the others do (McCall and Becker, 1990, p. 3). Indeed, the conversations
among participants are the distinguishing elements of focus groups (Bosco and Herman,
2010). This acknowledgement of change attempts to connect mobile, moving, shifting

DWYER 9781781001288 Part II.indd 353 25/07/2012 11:07

 354 Handbook of research methods in tourism

minds (and their representations) to a shifting, external world (Denzin, 1997, p. 32).
Following Bodens observations on conversational analysis, focus groups must therefore
be centrally concerned with temporality, with duration, with action, and with, as it were,
the pulse (1990, p. 265) of the conversation.
Focus groups therefore are a collaborative research method, as the outcome is depend-
ent on the interactions of group members as well as interactions with the facilitator.
However, this should be seen as a strength, not a weakness, as interviewers are them-
selves implicated in the construction of meanings with their interviewees. Such intertex-
tuality is crucial and unavoidable, and the data which results is essentially collaborative
(Cloke, 2002). Indeed, experience suggests that respondents tend to enjoy focus groups
as they feel that they are part of the project, assisting in collaborative and generative
enquiry. There are also opportunities for all stakeholders to engage in learning through
focus group participation.
In line with their collaborative nature, focus groups tend to be useful for engaging
with a defined community of interest. For example, focus groups are useful when the
study group of interest is small and would not be represented fairly in a larger survey
(Veal, 2006). In addition, they are also important for their ability to engage with minor-
ity groups, and indeed can form part of a broader process of empowering marginalized
communities. Onyx and Benton (1995, p. 50) define empowerment as essential in com-
munity development and connected to concepts of self-help, participation, networking
and equity. These principles are embedded within critical focus group approaches. As
focus groups involve the sharing of knowledge, education is a critical consideration in
research design, as education theory argues that knowledge is power, and advocates of
community empowerment contend that increased levels of community and individual
awareness about tourism lead to levels of all types of empowerment (Friedmann, 1996,
cited in Timothy, 2007, p. 209). Therefore there should be consideration of reciprocity
in focus groups, following the work of Cole (2006), who advocates a research approach
that ensures adequate contribution to the capacity of the researched. Focus groups can
form part of such a strategy, although they cannot be relied upon as a sole empowerment
method.
A number of researchers in the critical social sciences have called for greater flexibility
in the focus group method, and are wary of the rules that specify the ideal form of the
focus group discussion, including: group composition, that is the number, gender, age,
social status and life experience of participants; the communication skills and personal-
ity of the moderator; the nature of topic and questions under discussion; and the timing,
setting, seating arrangements and provision of refreshments (Goss, 1996). Therefore,
any guidelines should be treated only as such, and any researcher using the focus group
method should employ their own reflexivity in their design.
One of the main barriers to effective focus groups can be arranging them in the first
instance. Whilst the discussion itself and data generation are relatively swift, their
organization may take many months. Finn et al. (2000, p. 79) suggest that this is behind
a greater use of focus groups in academic research, but less in undergraduate study, due
to the requirements for cooperation, organization, time and resources that few students
have access to. A major part of this is having access to communities of interest, through
gatekeepers, who control avenues of opportunity (Hammersley and Atkinson, 1995).
Following access issues there are concerns with sampling and recruitment, which are

DWYER 9781781001288 Part II.indd 354 25/07/2012 11:07

 Focus groups 355

often overlooked in qualitative research. Sampling here is just as important as in other

methods, even if the community of interest is very specific. There has been debate within
the literature with regard to the homogeneity of focus group members. Conradson (2005)
explains that a general design principle in setting up focus groups is one of intra-group
homogeneity (in terms of gender, race, age, class or occupational sector). This is echoed
in much of the literature, with experienced users of the technique such as Burgess (1996)
recommending this strategy, with the reasoning behind this being so that respondents
feel relaxed in an atmosphere of similar views. In some cases it may be appropriate to
have members from specific marginalized groups together so that their voices can be
heard. However, there has been some recent critique of this strategy, for it assumes a
shared identity, a view which contrasts with the focus group approach where it is sur-
mized that identities are relational (Bosco and Herman, 2010). Furthermore, although
Finn et al. (2000) suggest that the atmosphere needs to be relaxed, in some cases a more
charged atmosphere may generate fruitful discussion, although one would wish to avoid
outright conflict, and this should be considered in risk assessments.
There is no fixed number for a focus group, although one needs to be cognizant of the
need to have enough people to generate discussion, without so many that it becomes dif-
ficult for the facilitator to involve all participants. Experience suggests that between five
and ten participants is a workable number. Consideration also needs to be given to the
location, and discursive locales in which focus groups take place (Pryce, 1979). As focus
groups are attempting to create as open and comfortable an environment as possible,
this should be considered in the design of the interview space. Again there are paral-
lels with community planning strategies, for example those used by Gill and Williams
(1994) in their living room approach to assessing community concerns. In this study 18
focus groups or living-room meetings were held with 187 Whistler residents who met in
small groups ranging from 6 to 15 individuals in the informal settings of private homes
to discuss community growth issues. Such informal settings have also been used in the
digital villages project described below. Similarly there also needs to be consideration of
timing of focus groups, and how this may influence recruitment and participation. Time
of day, other commitments, or timing in relation to an event or issue under consideration
are all elements worthy of acknowledgement, not just for logistical reasons, but perhaps
more importantly for the influence these factors may have on data generated.
The format of the focus group itself is largely up to the individual, and the priori-
ties of the research project, although it is hoped that the examples provided below will
provide some food for thought. Some form of structure is normal, with an introduction
to the research, as well as ensuring that all participants are adequately informed as to the
purpose of the study. Discussion may then be framed around specific subtopics, and the
researcher may wish to impose a rough time limit on these if it is necessary that all areas
are to be covered. One of the major roles of the facilitator is to ensure that everyone in
the focus group has adequate opportunity to participate. It is important to note that it is
not necessarily equal time on the floor that all participants should share, rather that there
is equal opportunity to share ideas and opinions. Individuals differ greatly in their ability
and desire to articulate, and the facilitator needs to be mindful of this in steering conver-
sations. As Timothy (2007) points out, it is rare all stakeholders in a community will have
homogenous views about tourism, the important thing is that everyone enjoys an equal
opportunity to participate. This may also involve politely steering the conversation away

DWYER 9781781001288 Part II.indd 355 25/07/2012 11:07

 356 Handbook of research methods in tourism

from topics of limited interest when required. Of course, deciding what is of interest, and
what is of relevance is an important but difficult determination to make, and one which
can undoubtedly change the outcome of the focus group.
It may not be solely conversations with purpose (Bosco and Herman, 2010) that are
generated from the focus group process. There has been a growing interest in using visual
techniques in tourism (Garrod, 2008) which can be used in combination with the focus
group method. Photographs, either researcher or respondent generated, can be used
as an additional prompt for discussion. In community tourism planning in the Rinjani
National Park on Lombok, Indonesia, focus groups generated a range of stakeholder
maps which were combined to form a planning and tourism information resource (Cater
2012). These maps reflected the beliefs, interests and issues of most concern to these dif-
ferent sectors, for example, indigenous female residents were most interested in where
their crops and water sources were located, whereas parks authority representatives took
a much more scientific approach to the landscape.
It is normal for focus groups to be recorded, either with tapes or more usually with
digital media. The researcher should ensure that the microphones are appropriately
located and all participants are recorded clearly. Given the other demands made on
the facilitator, it would be problematic for them to attempt to record the discussion on
paper at the time. Nevertheless, the facilitator may wish to keep a paper record of the
focus group to note any other observations from the discussion. These may be body
language, assessments of emotions or relationships or any ancillary points to follow up
outside of the discussion. As Denzin (1997, p. 38) suggests, it is often the unsaid, the
assumed, and the silences in any discourse (which) provide the flesh and bone the back-
drop against which meaning is established. Indeed, it may be helpful to have a second
researcher involved assisting in the administration of the group, acting as observers and
to note material which is supplementary to the text. This would be particularly appro-
priate if for some reason recording was not possible. Burgess (1996, p. 133) emphasizes
the importance of a debrief session with facilitators and observers following the focus
group to discuss and record what was going on, how individuals responded, the themes
and topics that came up (or did not come up), our initial ideas about the significance of
particular issues.
There are some concerns that in the transcribed focus groups the physical body of
respondents is absent, along with their self image and speaking repertoire (Denzin,
1997). Hammersley and Atkinson (1995) suggest that recordings may present a distorted
sense of the field, by focussing data collection on what can be recorded, and concentrat-
ing attention on the analysis of spoken action. Therefore consideration may be given to
the use of video cameras, whilst recognizing the no less representational shortcomings of
this technique. There should also be adequate consideration of the ethical aspects of this
technology, as well as inhibitions that this may or may not place on participants.

APPLICATIONS OF THE TECHNIQUE TO TOURISM

Focus groups came relatively late to tourism research, even in their less critical market-
ing form. Indeed, for Peterson (1994) general marketing methods were underutilized in
travel and tourism marketing research until the 1990s. Research that presented focus

DWYER 9781781001288 Part II.indd 356 25/07/2012 11:07

 Focus groups 357

groups as a conversation were also lacking, as tourism scholarship has historically

neglected the personal voice of either the subject or the analyst in the analysis (Swain,
1995). One of the earliest applications of the focus group method was by Simmons (1994)
who incorporated it in a participatory approach to understanding the role of community
participation in tourism planning. In this instance, the focus groups were preceded first
by a series of informal interviews and subsequently by a postal survey. It was seen that
the adoption of these three complementary methods allowed the researcher to continu-
ally refine the research focus as the project developed. Specifically, the researchers focus
groups consisted of small groups of people (n 5 812) who met the prerequisite of having
completed the second-stage postal survey. Focus groups were held at three locations
within the defined geographical area of interest of the project in order to represent a spec-
trum of tourism and opportunity (Simmons, 1994). The focus groups themselves took on
a three-phase structure, where participants were first given the opportunity to review and
interpret the results from the aforementioned postal survey. It was felt this provision of
information provided a quantitative framework from which participants could generate
specific tourism development proposals. Following this, more definitive tourism devel-
opment options were developed by focus group participants and subsequently ranked
using the nominal group technique (NGT). This technique involves every member of
the group giving their view of the solution, in this case to tourism development, and then
any duplicate solutions are eliminated from the list of all solutions. Group members then
proceed to rank these solutions, first, second, third, fourth, and so on. The third and
final phase of the focus groups involves a discussion around key planning constraints
which might shape future tourism developments in the area. As the focus of this study
was primarily concerned with community participation in the formation of tourism
planning decisions, the outcomes of the study noted that of all three stages of investiga-
tion, focus groups were seen as the most satisfactory mechanism to encourage participa-
tion amongst the community. Importantly, the results of this study demonstrated that
through the provision of information (survey results) to focus group participants, the
group came to resemble that of an advisory group which was found to be one of the
preferred mechanisms for participation identified in the survey.
Despite a few examples, the utilization of focus groups as a research tool from the
mid-1990s in tourism was scarce. Not until recently and particularly from the second
half of the 2000s has the adoption of focus groups become more widespread. Examples
of recent research topics using focus groups in tourism include destination marketing in
the context of major UK cities, namely London and Edinburgh (Fyall and Leask, 2006),
resident attitudes towards tourism in Lao Peoples Democratic Republic (Suntikul et al.,
2010), evaluation of the development and management of rural tourism projects in Great
Britain and South Africa (Briedenhann, 2009), experiences of eating-out for Norwegian
solo female travellers (Heimtun, 2010) and cross-cultural analysis of sport tourism
between Chinese-Canadians and Anglo-Canadians (Hudson et al., 2010). In her study
examining perceptions of international air travels impact on global climate change,
Becken (2007) used focus groups as a means of exploring a consensus on this topic. In
this instance, five focus groups were held (n 5 48) lasting on average for two hours. Her
focus groups were facilitated by one researcher, and observed by another, and in addi-
tion were audio and video taped for later analysis. Focus group participants were also
provided with snacks and refreshments as an incentive to participate. As in the earlier

DWYER 9781781001288 Part II.indd 357 25/07/2012 11:07

 358 Handbook of research methods in tourism

example, a structured approach for convening the focus groups was adopted with the
use of a quiz, role-play and a general discussion. At the outset, participants were briefed
with an introduction to climate change and how tourism relates to it, with consideration
given to choice of language and content so as not to influence or prime participants.
The purpose of a climate and travel quiz was to introduce facts on climate change and
tourism at the start of the discussion and to stimulate debate about the issues (Becken,
2007). The quiz was seen as a way of testing participant knowledge on the subject area
in a relaxed environment, as well as providing an opportunity for discussion through
the provision of multiple choice answers. The second phase of the session involved the
use of role-play. Scenarios which formed the basis for the role-plays were based on con-
temporary policy discussions, recent studies on emissions trading, emission charges and
energy or carbon budgets. Participants were asked to choose pre-defined roles, defined
through a few key points. The role-play was seen as a means of reducing the impact of
response desirability where participants were seen to be more comfortable presenting
answers/arguments as their hypothetical alias than they might otherwise be. The final
phase of these focus groups involved a general discussion element. Becken (2007) notes
that at the conclusion of the role play activity participants were comfortable giving their
own opinions on the topic, and as such only loose facilitation and guidance was needed.
In conclusion, the researcher noted that because of the specific nature of participants and
the structured format of the focus groups that the findings do not allow for wider conclu-
sions about tourists in general to be made. Findings from this research generally pointed
towards a desire for more information on the environmental impacts of international air
travel by tourists.
A further recent study concerning customer-derived value in the timeshare industry
in Australia (Sparks et al., 2007) also used focus groups as one of its core methods of
data collection. Sparks and her colleagues employed the use of focus groups because
of their ability to canvass a range of opinions at one time, as well as allowing for the
collection of a broad range of information from two major sub-groups simultaneously.
They note that this qualitative group interviewing process enabled the collection of rich
data about the derived customer value of timeshare ownership, whilst also allowing
other participants to add cumulatively to the data. As such, this cumulative approach
also had the effect of permitting group members to challenge and verify what other
participants were saying, thus providing an iterative yet interactive dialogue (Sparks
et al., 2007, p. 31). In their study, Sparks et al. (2007) recruited three focus groups (n
5 16, consisting of eight couples) with the help of an industry body, as well as three
major timeshare firms. For each focus group, a pre-prepared interview schedule was
used, which had been devised in consultation with the aforementioned industry panel
members. They noted that this structure consisted of both structured and unstruc-
tured interviewing approaches, and in doing so, attempts were made to maintain equal
contributions from all participants. Each focus group lasted for between 60 and 75
minutes, and, as in the previous example, was also audio-taped and later transcribed.
The analysis of results for this study was undertaken with the use of a software program
(NVivo2.0, QSR International, 2002), where the first stage involved two researchers
assessing the transcribed audio for comments related to customer-derived value and
used an iterative process to differentiate group explanations of consumer value. From
this, a codebook was developed and another researcher then recoded text based on final

DWYER 9781781001288 Part II.indd 358 25/07/2012 11:07

 Focus groups 359

classifications, with an inter-rater reliability of 90 per cent realized. The results from
this analysis were then presented to the expert industry panel. In the main, consum-
ers opinions were found to be in line with Holbrooks (1999) relativistic-preferential
definition of consumer value, where participants often compared the value of timeshare
against other products they could have purchased, or another leisure experience they
had undertaken.
Jennings et al. (2009) carried out focus groups in their study of quality in adventure
tourism experiences from the perspectives of adventure travellers and providers on
Australias Gold Coast. The specific aims of this study were to determine the nature of
quality adventure tourism experiences for adventure travellers; to determine the nature
of quality tourism experiences for adventure travel providers and to develop a quality
adventure tourism experience evaluation tool. The research was conducted in partner-
ship with a group of hostels and activity providers serving this sector. Focus groups were
identified as being a suitable method for investigation as it would allow rich discussion
on issues of quality contrasting with existing quantitative methods such as SERVQUAL
(Parasuraman, 1991). The research took place within two specific time periods, a peak
summer season, and a quieter off peak period, to assess the relationship between capacity
and quality. Given that the adventure youth travel market was defined as people aged
between 1829 years, who were travelling in Australia outside of family units, not for
business, and not primarily to visit friends or relatives, whose travel included at least one
overnight stay and purchase of adventure travel products, services or experiences, this
formed the sampling frame for the focus groups. As many of the industry partners were
hostels, they agreed access to their guests and provided facilities in which to hold discus-
sions. The focus groups for the adventure travel providers were held at a central location
which was neutral in regards to offering adventure travel experiences. Breakfast and
afternoon tea were supplied depending on the time of day that the focus group was being
held. The timing of the focus group was based on industry preferences and convenience
either early morning or late afternoon.
As there was a team of researchers involved in this project, it was important to create
guidelines for the administration of the focus groups. This included the introduction
given to focus group participants, so that all of the participants had the same prior
knowledge about the project. Incentives were used to increase traveller participation.
The focus groups included the provision of food and beverages, and tourism souvenirs
(for example, t-shirts and caps). The focus groups were held around the late afternoon
at the accommodation venues when the travellers were between the end of day activities
and start of evening activities. As such these focus groups could be loosely described as
a conversation-based working evening-meal (Jennings, 2009). The focus groups used
in this project were augmented with the use of individual interviews and travel diaries.
From interpreting (analysing) the information derived from the focus groups, interviews
and traveller diaries, the research team identified five higher order concepts, which
highlighted and determined the nature of a quality adventure tourism experience. Those
concepts are (1) personal connectivity; (2) social connectivity; (3) experience delivery; (4)
combining experiences; and (5) inter-connectivity of the entire adventure travel experi-
ence (Jennings, 2009). These concepts were derived by constant comparison of themes
generated from the initial and successive analysis of dominant themes.
Focus groups can also be useful in teaching feedback and development, as in a project

DWYER 9781781001288 Part II.indd 359 25/07/2012 11:07

 360 Handbook of research methods in tourism

carried out at Griffith University, Australia, on the teaching of ethics and responsible
practice to tourism students (Hornby et al., 2007). The purpose of this project was to
develop a resource to assist and facilitate teaching staff to train and educate students in
fieldwork courses in ethics and responsible practice. A driver for this project was antici-
pation that the release of the 2007 National Statement on Ethical Conduct in Human
Research would designate an institutional responsibility for the training and educating
of students carrying out research. Thus this project aimed to develop good practice in
delivery of training and education for students doing fieldwork courses, by developing a
resource to assist and facilitate that delivery. To achieve best practice, a team was formed
that included members experienced in running fieldwork courses, and members experi-
enced in training, advising and administering research ethics and responsible practice in
the university.
After the 2006 delivery of the resources in the undergraduate and postgraduate
courses, students were given the opportunity to anonymously participate in focus groups
and/or surveys regarding the resource and delivery of materials from the resource. A
total of 34 students participated in the survey, while three focus groups were held. These
focus groups found that delivery of the materials was successful in increasing awareness
and understanding of ethics and responsible practice issues, and this assisted students
in complying. However, they also identified that there was a need to demonstrate more
application to the students projects; a need for more class time and depth on the topic;
and a need to include issues of subjectivity and cultural influences. Thus a second round
of focus groups was conducted following a second implementation of the resource. Such
a longitudinal application of the focus group method is particularly useful when examin-
ing change.

ADVANTAGES AND LIMITATIONS OF FOCUS GROUPS

Given the usual low overall numbers of participants in focus groups and the rich tran-
scripts that result, it would normally be inappropriate for there to be any quantitative
analysis of focus group material. Indeed, one possible challenge of focus groups is the
potentially large amounts of material that are generated and seeking meaning within
this. This emphasizes the importance of keeping some degree of focus during the discus-
sion phase! The initial approach to analysis is to produce a transcript of the focus group
discussion, although as Veal (2006) suggests there is a requirement not just to identify
interviewer and interviewee, but also different members of the focus group, which may
be difficult from the recording alone. Some researchers advocate producing a summary
of the focus group identifying salient themes directly from the recording. If this is done
as the notes are being taken then the analysis is effectively proceeding at the same time
... since the note-taker must be drawing out key elements of the discussion in light of
the objective of the study (Veal, 1997, p. 139). Care should therefore be taken as to
the accurate representation of themes, and it is recommended that there are multiple
listenings to the recording or another person should be used to validate the analysis.
Technological developments including software developments in programs such as
ATLAS Ti (Scientific Software Development, 1993) and Nvivo (QSR International,
1999) have also enabled researchers to code directly from audio, still images and even

DWYER 9781781001288 Part II.indd 360 25/07/2012 11:07

 Focus groups 361

video. However, developments such as these have applied quantitative techniques to the
evaluation of data, a path which some researchers have criticized.
One of the main concerns regarding focus groups, and indeed one of the main reasons
for their use is their ability to uncover power relations. However, in its application one
must ensure that there is not simply a replication of existing power structures, or assump-
tions about where they might exist. Geographical examinations of the scalar nature of
power by, for example, Allen (2003) and Herod and Wright (2002) stress that there is
no certainty to power or the scales at which it manifests itself. Thus there needs to be
consideration both within focus groups and outside as to the application of the process.
The former is relatively easier by ensuring that discussion is not dominated by one or two
vociferous members of the group (Veal, 2006). However, the balanced external applica-
tion of focus groups in a policy context is an altogether trickier proposition. For example
in the case of the community mapping exercise discussed earlier (Cater, 2012), there has
been concern by observers that despite a notionally inclusive process, participation in the
benefits of tourism is still limited. As Schellhorn (2010) shows, indigenous residents have
limited position of power in the tourism industry, with outsiders and migrants control-
ling the business of tourism. Nevertheless, focus groups do provide the opportunity to to
revise and re-work theories and concepts from the ground up, together with participants,
providing a more transparent lens to knowledge construction. In this focus groups
provide an opportunity to challenge traditional divisions between theory and methodol-
ogy and between data and interpretation (Bosco and Herman, 2010, p. 195).

FURTHER DEVELOPMENTS AND APPLICATIONS OF FOCUS

GROUPS IN TOURISM RESEARCH

There is certainly opportunity for more use of focus groups within critical tourism
studies, particularly in longitudinal research applications. Here there may be parallels
with the Delphi method which uses expert opinion in order to answer specific research
questions. The Delphi method relies on an expert panel being questioned over a number
of rounds, with each member being given feedback on the answers provided by the other
panellists between each round. In that sense the Delphi method is also a collaborative
research strategy. However, Garrod (2012) highlights a lack of rigour often applied by
those using Delphi, which could be shared by focus group applications. Indeed, studies
using the Delphi method can be very sensitive to the manner in which the experts are
recruited, the expertise that is then represented on the panel and how this expertise is
exploited (Garrod, 2012). In particular Garrod points to the problem of panel attrition
in Delphic methods, and questions the validity of studies where attrition is too high.
Validity in focus group research comes through rigour in design and analysis. Thus the
call for flexibility in focus group design described in this chapter should not be mistaken
for a lack of process.
Technology also offers significant opportunity for development of the focus group
method. Software such as Skype offers the ability to have group conversations remotely
with up to ten participants (Skype, 2011). The company suggests that video makes com-
munications richer helping you to develop more collaborative relationships (Skype,
2011, p. 2), and there are clear opportunities for focus groups here. Recruitment and

DWYER 9781781001288 Part II.indd 361 25/07/2012 11:07

 362 Handbook of research methods in tourism

participation may be increased as there is no need for all participants to be physically

located in the same space. Digital video recording of the focus group may also be easier.
However, there are some issues of concern, for example, how the facilitator maintains
control over the discussion, as the group dynamics may be less clear. A lack of pres-
ence may mean that there is less potential to shift the balance of power away from the
researcher towards the participants and allow for supportive and reflexive encounters
(Bosco and Herman, 2010, p. 194). Although video is a step up from just speech as
discussed above, such technology may miss much of the non-verbal expressions such
as silences, attitudes and emotions (Bosco and Herman, 2010), central to the qualita-
tive method. One also should be wary of possible exclusion of respondents who are not
able, for whatever reason, to engage with this technology. Some of the issues in using
Skype for focus groups were encountered by the digital villages project which targeted
Information and Communication Technology (ICT) use in the home and community
in villages in Cleveland, northern England (Digital Villages Research Network, 2011).
Issues encountered were space requirements, sound and picture quality and broadband
speeds. Nevertheless there were felt to be significant advantages in providing a relaxed
home environment as well as flexible catering opportunities (Digital Villages Research
Network, 2011).
It would appear that the use of focus groups in tourism research is now accepted,
although there is much potential for their future refinement. A review of the use of
focus groups in the tourism literature indicates greatest satisfaction when used with
other complementary methods and we would stress that methodological triangulation is
important for any research project. However, focus groups are certainly unusual in their
ability to generate a great deal of information relatively quickly, and one of their greatest
strengths is in the collaborative nature of the outputs. Whilst this chapter has presented
a number of cases where focus groups have been employed, we have shied away from a
list of focus groups dos and donts. In line with concerns in the social sciences, we feel
that it is more important to employ a degree of reflexivity and rigour in methodological
design appropriate to the project in question. Nevertheless, we do also contend that it
is perhaps time for tourism researchers to do less borrowing from other areas of study,
and create their own innovations in focus group methodology. Many of the cases in this
chapter exemplify this innovation. Lastly, it is worth remembering that the skills gained
from employing focus groups are useful in many other areas of inquiry. Indeed, as aca-
demics we may often find ourselves in informal focus groups with tourism professionals
at industry or networking functions. The insights from these interactions contribute to
shaping our understanding of the tourism phenomenon, and are part of the collaborative
knowledge generation that is the pillar of focus group research.

REFERENCES

Allen, J. (2003), Lost Geographies of Power, Oxford: Blackwell.

Becken, S. (2007), Tourists perceptions of international air travels impact on the global climate and potential
climate change policies, Journal of Sustainable Tourism, 15 (4), 351368.
Boden, D. (1990), People are talking: conversation analysis and symbolic interaction, in M. McCall and
H. Becker (eds) Symbolic Interaction and Cultural Studies, Chicago, IL: University of Chicago Press, pp.
244273.

DWYER 9781781001288 Part II.indd 362 25/07/2012 11:07

 Focus groups 363

Bosco, F.J. and T. Herman (2010), Focus groups as collaborative research performances, in D. DeLyser, S.
Herbert, S. Aitken, M. Crang and L. McDowell (eds) The SAGE Handbook of Qualitative Geography, New
York: Sage, pp. 220236.
Briedenhann, J. (2009), Socio-cultural criteria for the evaluation of rural tourism projects: a Delphi consulta-
tion, Current Issues in Tourism, 12, (4), 379396.
Burgess, J. (1996), Focusing on fear, Area, 28 (2), 130136.
Cater, C. (2012), Community involvement in trekking tourism: the Rinjani Trek Ecotourism Programme,
Lombok, Indonesia, in B. Garrod and A. Fyall. (eds) Contemporary Cases in Tourism Volume 1, Oxford:
Goodfellow, pp. 191212.
Cole, S. (2006), Information and empowerment: the keys to achieving sustainable tourism, Journal of
Sustainable Tourism, 14 (6), 629644.
Conradson, D. (2005), Focus groups, in R. Flowerdew and D. Martin (eds), Methods in Human Geography:
A Guide for Students Doing a Research Project, Harlow, UK: Pearson pp. 111126.
Cloke, P., C. Philo and D. Sadler (1991), Approaching Human Geography, London: Paul Chapman.
Cloke, P.J. (2002), Deliver us from evil? Prospects for living ethically and acting politically in human geogra-
phy, Progress in Human Geography, 26 (5), 587604.
Cloke, P.J., I. Cook, P. Crang, M. Goodwin, J. Painter and C. Philo (2004), Practising Human Geography,
London: Sage.
Denzin, N.K. (1997), Interpretative Ethnography: Ethnographic Practices for the 21st Century, London: Sage.
Digital Villages Research Network (2011), http://digitalvillage.org.uk/using-a-skype-link-for-a-small-focus-
group/.
Finn, M., M. Elliott-White and M. Walton (2000), Tourism and Leisure Research Methods: Data Collection,
Analysis, and Interpretation, Harlow, UK: Pearson.
Fyall, A. and A. Leask (2006), Destination marketing: future issues strategic challenges, Tourism and
Hospitality Research, 7 (1), 5063.
Garrod, B. (2008), Exploring place perception a photo-based analysis, Annals of Tourism Research, 35 (2),
381401.
Garrod, B. (2012), Applying the Delphi method in an ecotourism context: a response to Deng et al.s
Development of a point evaluation system for ecotourism destinations: a Delphi method, Journal of
Ecotourism, forthcoming.
Gill, A. and P. Williams (1994), Managing growth in mountain tourism communities, Tourism Management,
15 (3), 212220.
Goss, J.D. (1996), Introduction to focus groups, Area, 28 (2), 113114.
Hammersley, M. and P. Atkinson (1995), Ethnography: Principles in Practice, 2nd edition, London: Routledge.
Heimtun, B. (2010), The holiday meal: eating out alone and mobile emotional geographies, Leisure Studies,
29 (2), 175192.
Herod, A. and M.W. Wright (eds) (2002), Geographies of Power: Placing Scale, Oxford: Blackwell.
Holbrook, M. (1999), Introduction to consumer value in M. Holbrook (ed.), Consumer Value: A Framework
for Analysis and Research, London: Routledge, p. 5.
Hornby, G., G. Jennings, G. Allen, C. Cater and K. Toohey (2007), Educational resource to assist with
delivery of training in ethical principles and responsible practise, Unpublished Report, Circulation, Griffith
University.
Hudson, S., T. Hinch, G. Walker and B. Simpson (2010), Constraints to sport tourism: a cross-cultural analy-
sis, Journal of Sport & Tourism, 15 (1), 7188.
Jennings, G.R., Y.-S. Lee, C. Cater, A. Ayling, C. Ollenburg and B. Lunny (2009), Quality tourism experi-
ences: reviews, reflections, research agendas, Journal of Hospitality and Leisure Marketing, Special Issue:
Experience Marketing, 18 (23), 294310.
McCall, M.M. and H.S. Becker (eds) (1990), Symbolic Interaction and Cultural Studies, Chicago, IL: University
of Chicago Press.
Merton, R.K. (1987), The focussed interview and focus groups: continuities and discontinuities, Public
Opinion Quarterly, 51, 550665.
Merton, R.K., M. Fiske and P.L. Kendall (1956), The Focussed Interview, Glencoe, IL: Free Press.
Onyx, J. and P. Benton (1995), Empowerment and ageing: toward honoured places for crones and sages, in G.
Craig and M. Mayo (eds), Community Empowerment: A Reader in Participation and Development, London:
Zed Books.
Parasuraman, A.A., L.L. Berry and V.A. Zeithaml (1991), Refinement and reassessment of the SERVQUAL
scale, Journal of Retailing, 67 (4), 420450.
Peterson, K.I. (1994), Qualitative research methods for the travel and tourism industry, in J.R. Ritchie and
C.R. Goeldner (eds), Travel, Tourism and Hospitality Research, New York: John Wiley, 7592.
Pryce, K. (1979) Endless Pressure: A Study of West Indian Lifestyles in Bristol, London: Penguin.
Riley, R. (1995), Prestige-worthy tourism behaviour, Annals of Tourism Research, 22, 630649.

DWYER 9781781001288 Part II.indd 363 25/07/2012 11:07

 364 Handbook of research methods in tourism

Schellhorn, M. (2010), Development for whom? Social justice and the business of ecotourism, Journal of
Sustainable Tourism, 18 (1), 115135.
Simmons, D. (1994), Community participation in tourism planning, Tourism Management, 15 (2), 98108.
Skype (2011), Group video calling product datasheet, available at http://download.skype.com/share/business/
guides/gvc-product-datasheet.pdf.
Sparks, B., K. Butcher and G. Pan (2007), Understanding customer-derived value in the timeshare industry,
Cornell Hotel and Administration Quarterly, February 2007, 2845.
Suntikul, W., T. Bauer and Song, H. (2010), Towards tourism: a Laotian perspective, International Journal
of Tourism Research, 12, 449461.
Swain, M.B. (1995), Gender in tourism, Annals of Tourism Research, 22 (2), 247266.
Timothy, D.J. (2007), Empowerment and stakeholder participation in tourism destination communities, in A.
Church and T. Coles (eds), Tourism, Power and Space, London: Routledge, pp. 199216.
Veal, A.J. (1997), Research Methods for Leisure and Tourism: A Practical Guide, 2nd edition, Harlow, UK:
Prentice Hall.
Veal, A.J. (2006), Research Methods for Leisure and Tourism: A Practical Guide, 3rd edition, Harlow, UK:
Pearson.
Weber, K. (2001), Outdoor adventure recreation. A review of research approaches, Annals of Tourism
Research, 28 (2), 360377.

DWYER 9781781001288 Part II.indd 364 25/07/2012 11:07

 18 Interview techniques
Nancy Gard McGehee

NATURE OF THE TECHNIQUE AND ITS EVOLUTION

Throughout history, people have attempted to make sense of the human condition
through narratives, putting words to their experiences. Some argue that the original
paradigm of human inquiry and the archetypical research method is, in fact, the act of
conversation between two people (Heron, 1981). Interviewing as a research technique
has roots in anthropology and sociology (Seidman 2006) and has been well-regarded in
those disciplines for decades.
The term interviewing includes a broad range of techniques, spanning from highly
structured, standardized, closed questions, to unstructured, open-ended conversations.
The focus for this chapter will be upon the latter, most often referred to as in-depth
interviewing. In-depth interviewing uses individuals as the point of departure for the
research process, assumes that individuals have unique and important knowledge about
the social world that is ascertainable, which can be shared through verbal communica-
tion (Hesse-Biber and Leavy, 2011, p. 94). With this approach, the researcher typically
loosely follows an interview schedule of open-ended, broad questions, with the goal of
guiding the informant toward her/his reconstruction and interpretation of the topic of
study.
Interviewing is primarily located within the qualitative epistemology of research
methods, particularly the critical and interpretive paradigms (Goodson and Phillimore,
2004). Those who identify with these paradigms and engage in the interview technique
believe that all research is influenced by the philosophical position of the researcher, the
nature of the project, and its intended audience (Jordan and Gibson, 2004), and view
research as a unique, context-specific product, co-created between the researcher and the
informant. It is important to note, however, that more positivist, quantitative research-
ers sometimes utilize the interview technique as well.

BACKGROUND AND TYPES OF PROBLEMS THAT THE

INTERVIEW TECHNIQUE IS DESIGNED TO HANDLE

The effectiveness of any research technique is dependent on its fit with the research ques-
tion. The interview technique is useful for a variety of research scenarios. As with most
qualitative techniques, interviewing is especially effective when the research question at
hand requires depth and specificity. This method is well-suited for issue-oriented research
questions or problems, or when the researcher wants to learn about the experience and
perceptions of the informant. Interviewing is valuable when the researcher wants to
capture an informants ideas, thoughts, and experiences in their own words. Perhaps
the most important role of interviewing is its ability to give voice to the experiences of

365

DWYER 9781781001288 Part II.indd 365 25/07/2012 11:07

 366 Handbook of research methods in tourism

persons who are often marginalized in traditional, survey-based quantitative studies.

The interview method is also flexible in that it can yield descriptive, explanatory, and/or
exploratory data.

APPLICATIONS OF THE INTERVIEW TECHNIQUE TO

TOURISM

In many ways, the tourism experience facilitates opportunities for interviews, particu-
larly with the tourists themselves. Persons actively engaged in the act of tourism are often
in a position to share their discretionary time with researchers while on vacation. In
other words, informants can afford the time to talk. Additionally, the subject matter of
tourism is one that many tourists are eager to discuss; they enjoy analysing or recounting
their travel experiences.
On the supply-side, tourism industry stakeholders are often very busy people, but they
also tend to be immersed and involved in their work, and as such are eager to share their
thoughts. From a practical perspective, the relative youth of the industry, along with
a lack of clean, systematic, quantitative data measuring the tourism industry also con-
tributes to the usefulness of the interview technique in tourism. As researchers begin to
gain an appreciation of the complexity surrounding the relationship between the tourism
industry and the rest of the community, the interview technique offers a way to capture
residents varied experiences and perceptions of tourism. Finally, this is an era where
researchers are finding average survey response rates to be plunging (Sheehan, 2001),
often due to survey fatigue. The interview technique may incite a more welcome response
from potential tourism study subjects of all types.
The in-depth interview technique has been applied in a number of ways to research
in tourism. Some utilize it as a sole method of study, while others triangulate it with a
variety of qualitative techniques, either sequentially or simultaneously (multi-method).
Still others utilize it as a precursor to quantitative research methods (mixed methods).
Buzinde et al. (2010) provide an excellent example of the application of interviews as
the sole data collection method. They utilize the technique known as funneling (Patton,
1990) as part of the interview process surrounding their work that targets visitors per-
ceptions of beach reclamation efforts at a coastal resort in Mexico. Buzinde et al.s use of
the funneling process begins with broad questions about the landscape and then follows
up with more specific queries about particular circumstances or situations as a means
to explore the contested meanings that emerge from the humanenvironment relation-
ship. White and White (2007) also utilize interviews solely in their study of the changing
notions of home and away amongst tourists in the context of todays internet con-
nected world. They find that tourists who stay in contact with family and friends at home
through inexpensive fixed telephone services and/or internet cafs experienced a blurring
of lines between their everyday environment of home and life on the road. Informants
reported both positive and negative elements of connectivity: positive in that they could
maintain established relationships, but negative in that communication in this context
could be upsetting and lead to re-assessment of contentious relationships.
While these are excellent examples of the use of the interview technique as a stand-
alone method for a study, tourism researchers also utilize interviews as one component

DWYER 9781781001288 Part II.indd 366 25/07/2012 11:07

 Interview techniques 367

of a multi-method study, triangulating the interview findings with other qualitative data
collection techniques (Aas et al., 2005; Brown, 2009; Muzaini, 2006; Palmer, 2005). This
is commonly utilized in the ethnographic approach, whereby researchers combine par-
ticipant observation, field diaries, conversations, systematic lurking, and/or other quali-
tative methods with in-depth interviews (Fetterman, 1998). For example, Muzaini (2006)
utilizes interviews in concert with field diaries in a fascinating study of backpackers in
Southeast Asia and their strategies for achieving a high level of cultural immersion in
order to have what they perceive as an authentic experience. This study provides an espe-
cially good example of how to classically merge interview results with other techniques.
The mixed method approach is similar to the multi-method approach, but utilizes a
mix of both qualitative and quantitative methods. One very innovative use of the inter-
view method as part of a mixed-method study may be found with Pan and Fesenmaiers
(2006) exploration of the online information search as part of the vacation planning
process. Interviews were used as a way to follow-up on a vacation planning exercise con-
ducted with a group of 15 individuals that included the audio and video recording of the
online planning process as well as computer and internet activity monitoring using spe-
cialized quantitative software. The interviews were extremely valuable as they provided
insight and clarity into the perspectives of the research subjects.
There is certainly a more traditional positivistic area of the literature where in-depth
interviewing is utilized as a way to develop quantitative variables, scales and survey items
(Hung and Petrick, 2010; Rittichainuwat, 2011; Yang, 2011) or as a form of reconnais-
sance to determine if a tourist site is appropriate for the research problem (Williams and
Soutar, 2009). In other words, the interview is used to lay the groundwork or act as a
precursor to the primary quantitative study. Hung and Petrick (2010) provide a classic
example of this in their development of a measurement scale for constraints to cruising,
as does Rittichainuwat (2011) in an interesting study that uses interviews of a travel
writer, journalist, tour operator, and tsunami survivors to develop a scale to measure
possible deterrents from visiting tsunami-hit destinations.
These aforementioned studies provide rigorous examples of the various uses of the
interview technique in the area of tourism. But what do we know about overall trends
in the use of the interview technique in tourism research? As with qualitative research
overall, the interview method has gained considerable ground in the study of tourism
over the past 20 years (Goodson and Phillimore, 2004). At the time this chapter was
written, over 130 peer-reviewed articles from the top three tourism journals (Journal of
Travel Research, Annals of Tourism Research, and Tourism Management) were found
to have utilized the interview method within the last decade (20002010). In the decade
previous to that, less than half that number were found. The majority of recent interview-
focused research seems to be located within the visitor behavior/consumer psychology
and heritage/culture areas of study, but the method has also been utilized in the areas of
destination development, policy and planning, tourism industry management, resident
attitudes, and hostguest relations.
Much work that utilizes the interview technique is being conducted in the broad area
of visitor behavior/consumer psychology, including park usage (Cochrane, 2006), risk
(Elsrud, 2001; Uriely and Belhassen, 2006), tourist satisfaction (Moyle and Croy, 2007;
Rodriguez Del Bosque and San Martin, 2008), constraints to travel (Hung and Petrick,
2010; Rittichainuwat, 2011), study abroad (Brown, 2009), niche and/or marginalized

DWYER 9781781001288 Part II.indd 367 25/07/2012 11:07

 368 Handbook of research methods in tourism

groups (Andriotis, 2009; Gibson, 2002; Hecht and Martin, 2006; Heuman, 2005; Lane
and Waitt, 2007; Sin, 2009; White and White, 2007), motivation (Maoz, 2007), and the
postmodern tourist (Maoz and Bekerman, 2010). One of the most interesting studies
from this area of tourism research is Uriely and Belhassens (2006) work in the area of
drugs and risk-taking in tourism. By utilizing in-depth interviews, the researchers are able
to delve deeply into the rationale used by tourists who commonly seek out destinations
based on the availability of drugs. Findings reveal that the informants are very aware of
the risks associated with drug use when they are far from home, but informants also have
a sense that in some ways being a tourist provides greater protection from those risks.
Given the sensitivity of this subject matter, and the difficulty of finding a large sample of
informants, the interview technique is a very appropriate method.
A variety of issues surrounding the interface of heritage and culture with tourism have
been analysed using the interview method within the past decade: the social and political
construction of heritage (Chronis, 2005; Park, 2010; Santos and Yan, 2008), authenticity
(Kim and Jamal, 2007; Martin, 2010), commodification (Cole, 2007), identity (Goulding
and Domic, 2009; Light, 2007; Palmer, 2005), tourism and performance (Mordue, 2005),
and cross-cultural relations (Raymond and Hall, 2008), to name a few. Lights (2007) use
of the interview technique in his work focusing on the social construction of a Dracula-
centered cultural identity in Romania is an excellent example of the value of the inter-
view method in this area of tourism research. Light is interested in how tourism planning
and policy making was conducted during the socialist era, and how it in turn constructed
and supported the myth of Transylvania. As very little documentation is accessible to
him from that secretive time in Romanian history, he utilizes the interview method as
a way to gain knowledge from former policy makers during that era. Thus he is able to
access information and explore the contradictions between the myth of Dracula and the
image that Romania was attempting to portray to the world during that time that other-
wise would not have been accessible.
In the area of destination development, interview based research is being conducted
in the topics of eco-tourism (Cruz et al., 2005; Jones, 2005), rural tourism (Cawley
and Gillmore, 2008; Nepal, 2007; Saxena and Ilbery, 2009), race/ethnicity/gender and
tourism development (Buzinde and Santos, 2009; Carr, 2007; Gentry, 2007), dependency
(Lacher and Nepal, 2010), conflict (Lee et al., 2010), and community participation (Li,
2006; Matarrita-Cascante, 2010; Sebele, 2010). As mentioned previously, the interview
technique is especially useful for capturing the perspectives and complexities of mar-
ginalized groups normally excluded from the conversation. Gentry (2007) utilizes the
method to explore Belizean womens experiences in a wide range of employment (both
tourism and non-tourism), and by doing so is able to tease out the complex interac-
tions between development and culture in the context of gender. For example, while the
Belizean cultural cornerstones of gender segregation and housewifization of labor are
often reinforced and magnified in the workplace, employment opportunities in tourism
do, in some cases, reduce the reliance of women on the male heads of households as the
sole financial decision maker and give them greater economic independence. Gentry is
able to focus on women (in this case a marginalized group) through the interview tech-
nique and capture these differences and contradictions that might otherwise be missed in
a male-dominated culture.
Some work utilizing the interview technique is found in the area of policy, although

DWYER 9781781001288 Part II.indd 368 25/07/2012 11:07

 Interview techniques 369

not as prominent as in the three previous topics. Categories include stakeholders (Aas et
al., 2005), politics (Airey and Chong, 2010; Altinay and Bowen, 2006) power (Bramwell
and Meyer, 2007), social policy (Stevenson et al., 2008) and governance (Yuksel et al.,
2005). Bramwell and Meyer (2007) interview key decision makers and utilize discourse
analysis to explore the networks, strategies, and power relations involved in the tourism
development of an island in the former East Germany. The interview process allows the
researchers to conduct multiple interviews of a purposive sample of 21 key informants,
and triangulate that data with informal conversations with 53 local residents, published
and unpublished reports, internal documents, and newspaper articles. From the data,
Bramwell and Meyer discovered unique tensions and difficult transitions, most of which
are a result of the political changes brought about by a formerly socialist region being
absorbed into a capitalist democratic state. Bramwell and Meyer also exposed the dia-
lectics of power relations that existed during the process of tourism development on the
island.
Interview-based research targeting the area of tourism industry management is dif-
ficult to categorize. Ethics (Lahdesmaki, 2005; Lovelock, 2008), sustainability (Vernon,
2000; Vernon et al., 2005), product development (Lumsdon, 2006), performance meas-
urement (Phillips and Louvieris, 2005), risk (Wang et al., 2010), knowledge transfer
(Weidenfeld et al., 2010; Wong and McKercher, 2011), and business networks (OBrien,
2006; Paget et al., 2010; Tinsley and Lynch, 2001) are only a few of the categories discov-
ered while conducting a literature review in this area. While all of these areas are impor-
tant, one of the most important (yet difficult to study) is knowledge transfer. Weidenfeld
et al. (2010) use the interview technique to examine innovation and knowledge transfer
within the under-studied attractions segment of the tourism industry in the region of
Cornwall, England. Perhaps the most interesting element of this study is the way in
which framework analysis is used to analyse 32 interviews of a variety of stakeholders,
which involved the

familiarization, classification, and indexation that allowed the identification of different themes
and their coding, using Non-numerical Unstructured Data Indexing Searching and Theory-
building. ... Each knowledge transfer or innovation process identified was named, classified
as product innovation, process innovation or other form of knowledge transfer. Each pair
(knowledge provider and receiver) was classified as neighbours when both enterprises were
located in the same cluster, and distant when not. Attractions were classified thematically as
product-similar or product-different in terms of the attractions that they received or supplied
knowledge to. (Weidenfeld, Williams & Butler, 2010, p. 613.)

Findings from the study indicate that at the local and regional level, spatial proximity,
product similarity, and market similarity generally facilitate knowledge transfers and
innovation spillovers.
While some quantitative studies of resident attitudes toward tourism (McGehee and
Andereck, 2004; Vargas-Snchez et al., 2011) conclude with a call for more qualitative
work in the area, very little has been published in the top-tier journals within the last
decade. In fact, only one resident attitude article is found: a study of residents attitudes
toward tourism in a small village in Uganda by Lepp (2008), which provides an excel-
lent example of the usefulness of the interview technique in locales where literacy rates
may be low, survey research may be unfamiliar to the targeted research population,

DWYER 9781781001288 Part II.indd 369 25/07/2012 11:07

 370 Handbook of research methods in tourism

and/or resources needed for survey research are scarce or unavailable. Lepp utilizes the
method of constant comparison to analyse the interview data, which is the process origi-
nally developed by Glaser and Strauss (1967) as the methodology for grounded theory.
Constant comparison consists of (1) collecting and analysing data, (2) developing
tentative conclusions, hypotheses, and themes, (3) collecting and analysing additional
data, (4) testing against the initial hypotheses and themes, (5) purposefully seeking out
additional new perspectives and data sources, and (6) continuing the process until you
reach data saturation (you are not learning anything new from your data collection).
As a result of his use of constant comparison, Lepp is able to capture the complexity of
some seemingly simple responses regarding tourism. For example, while many farmers
interviewed express frustration with the issue of crop raiding by protected wildlife from
a nearby preserve, they view that as a by-product of conservation rather than tourism,
and that the revenue from tourism is actually a positive compensation for the inevitable
crop raiding. This perspective has not been captured in previous research in this area of
study.
In the area of hostguest relations publications in top-tier journals are limited (de
Albuquerque, 2001; Maoz, 2007; Salazar, 2005). Given the legacy of Smith (1989), this
is quite surprising. Maoz (2007) conducted a long-term study of Israeli backpackers in
India that included in-depth interviews and participant observation to present the notion
of the mutual gaze (Maoz, 2007, p. 221). She posits that not only is there a tourist gaze
whereby tourists objectify local residents, the opposite occurs among the Indian hosts of
the large numbers of Israeli backpackers.
As conclusion to this section, it is important to note that while the focus was on
citing works from top-tier journals, a great deal of valuable work utilizing the interview
technique is situated in more topic-specific journals, book chapters, and monographs.
Focusing on the top-tier journals is simply utilized as a way to manage the large quantity
of work that exists in this area.

ADVANTAGES AND LIMITATIONS OF INTERVIEW

TECHNIQUE CONCEPTUALLY AND FOR POLICY
FORMULATION

As with any research method, there are advantages and limitations to the use of in-depth
interviews. Jordan and Gibson (2004, pp. 2223) provide a thorough overview of these
in their study of solo women travel experiences, as does Seidman (2006) in his guide of
interviewing techniques targeting researchers in education. To summarize its advan-
tages, interviewing:

is adaptable to a wide range of themes and topics

enables researchers to account for body language and non-verbal forms of com-
munication often missed in other forms of data collection
can provide background for studies using multiple methods
can create rich, descriptive data and illustrative examples of the human
experience
the iterative nature of interviewing can ensure its validity and accuracy.

DWYER 9781781001288 Part II.indd 370 25/07/2012 11:07

 Interview techniques 371

Limitations to the interviewing technique include:

poor data quality if the possible power differentials and status differences (includ-
ing class, gender, race and ethnicity) between the researcher and the informant are
not taken into consideration
the potential for failure and misunderstandings if a partnership is not established
early on between the researcher and informant
the time-consuming nature of the method, both the interviewing process itself
as well as the pre-interview process, which includes conceptualizing the research
problem, gaining entre with informants, scheduling (and often re-scheduling)
interviews, and the post-interview transcription and analysis
the investment of training, experience, and confidence to implement
the monetary expense, including researcher travel and compensation of informants
for their time
political pitfalls for the researcher in choosing the paradigmatic road less
traveled, and for the informant if the subject is sensitive to family, friends, co-
workers, and her/his community
the dependency on the honesty of the informant to the success of the interview and
the research project overall.

There are a number of strategies that can be employed to maximize the advantages
and minimize the disadvantages of the interview technique. They include:

appropriate selection of the sample or panel

the role of triangulation and other methods of trustworthiness
finding the appropriate length and structure for the interview
conducting an ethical study that accounts for the respect of and confidentiality for
the informants
applying active listening techniques and taking advantage of the iterative nature
of interviewing.

First, selection of informants is extremely important. Who you choose to interview will
have a very powerful impact on your research. It is vital that the researcher recognize and
appreciate why she/he wants to study the selected sample. Researchers should be cogni-
zant of a number of possible issues, including bias (both positive and negative), prejudice,
familiars versus strangers as informants, and snowball versus purposive sampling. One
example of how these issues may be addressed is found in Zahra and McGehees (in press)
study of volunteer tourism in the Philippines and its impact on community capital. Zahra
has over 20 years experience working with the communities and organizations purpose-
fully selected for the study, which provided a very strong level of access and trust, but while
the primary researcher was aware that her familiarity and entre into the community was
vital to the success of the study, she also was aware of any potential bias that she might
introduce given her strong ties with the organizations and the community. To combat this,
she immersed herself in the volunteer tourism literature (particularly the area of critical
theory), worked hard to allow for a wide spectrum of voices in the interviews, and inten-
tionally sought out a co-researcher whom she felt would be as unbiased as possible.

DWYER 9781781001288 Part II.indd 371 25/07/2012 11:07

 372 Handbook of research methods in tourism

Second, the inclusion of criteria for trustworthiness is vital to all forms of qualitative
research, including interviews (Lincoln and Guba, 1985). Special care should be taken in
the development of the research design to assure Maxwells (2005) parameters for valid-
ity in qualitative research, which is also in alignment with Lincoln and Gubas (1985)
typology. This includes four criteria for qualitative inquiry: credibility, transferability,
dependability, and confirmability (DeCrop, 2004, p. 159). Credibility, which relates to
the quantitative criterion of internal validity (DeCrop, 2004, p. 159), may be enhanced
using techniques of prolonged engagement, persistent observation, and member checks.
Transferability (much like external validity) may be accounted for through purposive
sampling. Dependability mirrors quantitative reliability. Dependability may be increased
through the development of a detailed research plan which can include an audit trail of
the transcripts, a personal journal of the research process kept by the researcher, pro-
longed engagement, and/or the inclusion of a research auditor. The research audit process
is also useful to assess confirmability, e.g., assurances that a variety of explanations about
the phenomenon are being studied. Additionally, the role of triangulation is important
to any type of qualitative research; the interviewing technique is no exception. This may
include data, method, investigator, and/or theoretical triangulation (DeCrop, 2004).
Triangulation simply means the use of more than two data sources (data triangulation),
methodological approaches (method), researchers (investigator), or theoretical perspec-
tives (theoretical) to approach a problem as a way to increase confidence in the results.
Third, determining the appropriate length and structure of the interview format is
also an important means of maximizing the advantages while minimizing the limitations
of the method. In general, 90 minutes is seen as a typical length for interviews, but as
with much of qualitative research there are no cookbook rules. Informants notice,
appreciate, and respond to researchers who respect their time and resources by keeping
the interview on task but not regimental, structured but not rigid, and appropriate but
not constrained. Similarly, the fourth element of conducting an ethical research program
and assuring the respect toward and confidentiality of informants is tantamount to the
success of the interview techniques. Hesse-Biber and Leavy (2011, p. 83) provide a practi-
cal and insightful Checklist of Questions for Conducting an Ethical Research Project
that includes such important elements as the importance of explaining the purpose of
your study to your informants, any risks that the informants may incur by participating
in the project, exactly the parameters and expectations of confidentiality, who will have
access to the data, and what ethical framework informs the research project.
Fifth, the role of the researcher as active listener is crucial to gaining the trust of the
informant and obtaining honest answers, but must be carefully approached (Hesse-
Biber and Leavy, 2011). Active listeners are interviewee focused, supportive but non-
interruptive, non-judgmental, accepting of difference, allowing for and listening to the
importance of silences, and resisting the need to fill in the blanks by putting words into
the informants mouth. Active listeners keep asking questions until they fully understand
the informant, probing and exploring, taking advantage of the iterative nature of the
interview format. For fledgling researchers, this can be difficult. A few starter ques-
tions are always valuable to keep nearby as a guide, should the interview begin to lose
momentum. These include: tell me a story about ______; tell me more about ________;
draw a picture that reflects _______; can you explain that a bit more for me?; and is there
an example you can think of?

DWYER 9781781001288 Part II.indd 372 25/07/2012 11:07

 Interview techniques 373

FURTHER DEVELOPMENTS AND APPLICATIONS OF THE

INTERVIEW TECHNIQUE IN TOURISM RESEARCH
It is hoped that tourism research as a whole will continue to become more sophisticated
and multi-faceted, reaching beyond the sometimes overused survey method to include a
wider variety of techniques. In this manner tourism researchers can work toward richer,
more robust solutions to the myriad problems of tourism development, to better under-
stand the behavior of the tourist, and to give voice to the under-served and marginalized
groups on both sides of the hostguest dichotomy.
In the specific area of in-depth interviews, while some have utilized a pre- and post-
event interview technique (Maoz, 2007), there is very little research in tourism that
utilizes the classic three interview structure (Schuman, 1982; Seidman, 2006) quite
often used in interpretivist phenomenological research (Schutz, 1967). Phenomenology
is the study of structures of experience or consciousness. Phenomenologically based
interviewing combines life history interviewing with in-depth interviewing. While
a number of studies in tourism have utilized the phenomenological perspective
(Andriotis, 2009; Pernecky and Jamal, 2010; Wang, 1999), the author could find
no tourism studies where the three interview structure was employed. Designed by
Schuman (1982), this technique includes a first interview, whereby the participants
life history is placed in the context of the research topic; a second interview, whose
purpose is to focus on the concrete details of the informants present, lived experi-
ence; and a third interview, which focuses on meaning-making, or the intellectual
and emotional connections between the topic at hand and the persons life. Each
interview is important, in that it facilitates putting the informants life experience into
language (Seidman, 2006) as well as creating a rapport between the researcher and the
informant.
In addition to the need for interview-based tourism research that includes multiple
interviews with each informant, there also seems to be a paucity of practice and pilot
interviews utilized as a way to refine and improve the interview quality. Just as research-
ers conduct pilot surveys, pilot interviews are also valuable.
The interview technique allows tourism researchers to approach issues in the field in
a deep, rich way not possible within many quantitative methods. As noted earlier, the
interview method may act as a stand-alone technique, or it can work in concert with
other methods, both qualitative and quantitative. As with any method, paying careful
attention to the appropriate ways to utilize the technique will minimize its limitations
and maximize its strengths.

REFERENCES
Aas, C., A. Ladkin and J. Fletcher (2005), Stakeholder collaboration and heritage management, Annals of
Tourism Research, 32 (1), 2848.
Airey, D. and K. Chong (2010), National policy-makers for tourism in China, Annals of Tourism Research,
37 (2), 295314.
Altinay, L. and D. Bowen (2006), Politics and tourism interface: the case of Cyprus, Annals of Tourism
Research, 33 (4), 939956.
Andriotis, K. (2009), Sacred site experience: a phenomenological study, Annals of Tourism Research, 36 (1),
6484.

DWYER 9781781001288 Part II.indd 373 25/07/2012 11:07

 374 Handbook of research methods in tourism

Bramwell, B. and D. Meyer (2007), Power and tourism policy relations in transition, Annals of Tourism
Research, 34 (3), 766788.
Brown, L. (2009), The transformative power of the international sojourn: an ethnographic study of the inter-
national student experience, Annals of Tourism Research, 36 (3), 502521.
Buzinde, C. and C. Santos (2009), Interpreting slavery tourism, Annals of Tourism Research, 36 (3),
439458.
Buzinde, C., D. Manuel-Navarrete, E. Yoo and D. Morais (2010), Tourists perceptions in a climate of
change: eroding destinations, Annals of Tourism Research, 37 (2), 333354.
Carr, A. (2007), Mori nature tourism businesses: connecting with the land, in R. Butler and T. Hinch (eds),
Tourism and Indigenous Peoples: Issues and Implications, Oxford: Elsevier, pp. 113127.
Cawley, M. and D. Gillmore (2008), Integrated rural tourism: concepts and practice, Annals of Tourism
Research, 35 (2), 316337.
Chronis, A. (2005), Coconstructing heritage at the Gettysburg storyscape, Annals of Tourism Research, 32
(2), 386406.
Cochrane, J. (2006), Indonesian national parks: understanding leisure users, Annals of Tourism Research, 33
(4), 979997.
Cole, S. (2007), Beyond authenticity and commodification, Annals of Tourism Research, 34 (4), 943960.
Cruz, R., E. Baltazar, G. Gomez and E. Lugo (2005), Social adaptation: ecotourism in the Lacandon Forest,
Annals of Tourism Research, 32 (3), 610627.
de Albuquerque, K. (2001), Tourist harassment: Barbados survey results, Annals of Tourism Research, 28 (2),
477492.
DeCrop, A. (2004), Trustworthiness in qualitative tourism research, in J. Phillimore and L. Goodson (eds),
Qualitative Research in Tourism: Ontologies, Epistemologies, and Methodologies, London: Routledge, pp.
156169.
Elsrud, T. (2001), Risk creation in traveling: backpacker adventure narration, Annals of Tourism Research,
28 (3), 597617.
Fetterman, D. (1998), Ethnography, London: Sage.
Gentry, K. (2007), Belizean women and tourism work: opportunity or impediment?, Annals of Tourism
Research, 34 (2), 477496.
Gibson, H. (2002), Busy travelers: leisure-travel patterns and meanings in later life, World Leisure, 44 (2),
1120.
Glaser, B.G. and A. Strauss (1967), The Discovery of Grounded Theory: Strategies for Qualitative Research, 3rd
edition, Madison, WI: Sociology Press.
Goodson, L. and J. Phillimore (2004), The inquiry paradigm in qualitative tourism research, in J. Phillimore
and L. Goodson (eds), Qualitative Research in Tourism: Ontologies, Epistemologies, and Methodologies,
London: Routledge, pp. 3046.
Goulding, C. and D. Domic (2009), Heritage, identity and ideological manipulation: the case of Croatia,
Annals of Tourism Research, 36 (1), 85102.
Hecht, J.A. and D. Martin (2006), Backpacking and hostel-picking: an analysis from Canada, International
Journal of Contemporary Hospitality Management, 18 (1), 6677.
Heron, J. (1981), The philosophical basis for a new paradigm, in P. Reason and J. Rowan (eds), Human
Inquiry, New York: Wiley, pp. 1935.
Hesse-Biber, S.N. and P. Leavy (2011), The Practice of Qualitative Research, 2nd edition, Thousand Oaks,
CA: Sage.
Heuman, D. (2005), Hospitality and reciprocity: working tourists in Dominica, Annals of Tourism Research,
32 (2), 407418.
Hung, K. and J. Petrick (2010), Developing a measurement scale for constraints to cruising, Annals of
Tourism Research, 37 (1), 206228.
Jones, S. (2005), Community-based ecotourism: the significance of social capital, Annals of Tourism Research,
32 (2), 303324.
Jordan, F. and H. Gibson (2004), Let your data do the talking: researching the solo travel experiences of
British and American women, in J. Phillimore and L. Goodson (eds), Qualitative Research in Tourism:
Ontologies, Epistemologies, and Methodologies, London: Routledge, pp. 215235.
Kim, H. and T. Jamal (2007), Touristic quest for existential authenticity, Annals of Tourism Research, 34 (1),
181201.
Lacher, R.G. and S. Nepal (2010), Dependency and development in Northern Thailand, Annals of Tourism
Research, 37 (4), 947968.
Lahdesmaki, M. (2005), When ethics matters: interpreting the ethical discourse of small nature-based entre-
preneurs, Journal of Business Ethics, 61 (1), 5568.
Lane, R. and G. Waitt (2007), Inalienable places: self-drive tourists in Northwest Australia, Annals of
Tourism Research, 34 (1), 105121.

DWYER 9781781001288 Part II.indd 374 25/07/2012 11:07

 Interview techniques 375

Lee, T., M. Riley and M. Hampton (2010), Conflict and progress: tourism development in Korea, Annals of
Tourism Research, 37 (2), 355356.
Lepp, A. (2008), Tourism and dependency: an analysis of Bigodi village, Uganda, Tourism Management, 29
(6), 12061214.
Li, W. (2006), Community decision-making: participation in development, Annals of Tourism Research, 33
(1), 132143.
Light, D. (2007), Dracula tourism in Romania: cultural identity and the State, Annals of Tourism Research,
34 (3), 746765.
Lincoln, Y.S. and E. Guba (1985), Naturalistic Inquiry, Beverly Hills, CA: Sage.
Lovelock, B. (2008), Ethical travel decisions: travel agents and human rights, Annals of Tourism Research, 35
(2), 338358.
Lumsdon, L. (2006), Factors affecting the design of tourism bus services, Annals of Tourism Research, 33 (3),
748766.
McGehee, N.G. and K. Andereck (2004), Factors influencing rural residents support of tourism, Journal of
Travel Research, 43 (2), 131140.
Maoz, D. (2007), Backpackers motivations: the role of culture and nationality, Annals of Tourism Research,
34 (1), 122140.
Maoz, D. and Z. Bekerman (2010), Searching for Jewish answers in Indian resorts: the postmodern traveler,
Annals of Tourism Research, 37 (2), 423439.
Martin, K. (2010), Living pasts: contested tourism authenticities, Annals of Tourism Research, 37 (2), 537554.
Matarrita-Cascante, D. (2010), Beyond growth: researching tourism-led development, Annals of Tourism
Research, 37 (4), 11411163.
Maxwell, J.A. (2005), Qualitative Research Design: An Interactive Approach, 2nd edition, Thousand Oaks, CA:
Sage.
Mordue, T. (2005), Tourism, performance and social exclusion in Olde York, Annals of Tourism Research,
32 (1), 179198.
Moyle, B. and G. Croy (2007), Crowding and visitor satisfaction during the off-season: Port Campbell
National Park, Annals of Leisure Research, 10 (3/4), 518531.
Muzaini, H. (2006), Backpacking Southeast Asia: strategies of looking local, Annals of Tourism Research,
33 (1), 144161.
Nepal, S. (2007), Tourism and rural settlements: Nepals Annapurna region, Annals of Tourism Research, 34
(4), 855875.
OBrien, D. (2006), Event business leveraging: the Sydney 2000 Olympic Games, Annals of Tourism Research,
33 (1), 240261.
Paget, E., F. Dimanche and J. Mounet (2010), A tourism innovation case: an actor-network approach, Annals
of Tourism Research, 37 (3), 828847.
Palmer, C. (2005), An ethnography of Englishness: experiencing identity through tourism, Annals of Tourism
Research, 32 (1), 727.
Pan, B. and D. Fesenmaier (2006), Online information search: vacation planning process, Annals of Tourism
Research, 33 (3), 809832.
Park, H. (2010), Heritage tourism: emotional journeys into nationhood, Annals of Tourism Research, 37 (1),
116135.
Patton, M. (1990), Qualitative Evaluation and Research Methods, 2nd edition, Newsbury Park, CA: Sage.
Pernecky, T and T. Jamal (2010), Phenomenology in tourism studies, Annals of Tourism Research, 37 (4),
10551075.
Phillips, P. and P. Louvieris (2005), Performance measurement systems in tourism, hospitality, and
leisure small medium-sized enterprises: a balanced scorecard perspective, Journal of Travel Research, 44
(2), 201.
Raymond, E. and C. Hall (2008), The development of cross-cultural (mis)understanding through volunteer
tourism, Journal of Sustainable Tourism, 16 (5), 530543.
Rittichainuwat, B. (2011), Ghosts: a travel barrier to tourism recovery, Annals of Tourism Research, 38 (2),
437459.
Rodriguez Del Bosque, I. and H. San Martin (2008), Tourist satisfaction a cognitive-affective model, Annals
of Tourism Research, 35 (2), 551573.
Salazar, N. (2005), Tourism and glocalization: local tour guiding, Annals of Tourism Research, 32 (3),
628646.
Santos, C. and G. Yan (2008), Representational politics in Chinatown: the ethnic other, Annals of Tourism
Research, 35 (4), 879899.
Saxena, G. and B. Ilbery (2009), Integrated rural tourism: a border case study, Annals of Tourism Research,
35 (1), 233254.
Schuman, D. (1982), Policy Analysis, Education, and Everyday Life, Lexington, MA: Health.

DWYER 9781781001288 Part II.indd 375 25/07/2012 11:07

 376 Handbook of research methods in tourism

Schutz, A. (1967), The Phenomenology of the Social World, G. Walsh and F. Lenhert (trans.), Chicago, IL:
Northwestern University Press.
Sebele, L.S. (2010), Community-based tourism ventures, benefits and challenges: Kham Rhino Sanctuary
Trust, Central District, Botswana, Tourism Management, 31 (1), 136146.
Seidman, I. (2006), Interviewing as Qualitative Research: A Guide for Researchers in Education and the Social
Sciences, New York: Teachers College Press.
Sheehan, K. (2001), E-mail survey response rates: a review, Journal of Computer-Mediated Communication,
6 (2).
Sin, H.L. (2009), Volunteer tourism: Involve me and I will learn?, Annals of Tourism Research, 36 (3),
480501.
Smith, V. (1989), Hosts and Guests Revisited: The Anthropology of Tourism, Philadelphia, PA: University of
Pennsylvania Press.
Stevenson, N., D. Airey and G. Miller (2008), Tourism policy making: the policymakers perspectives, Annals
of Tourism Research, 35 (3), 732750.
Tinsley, R. and P. Lynch (2001), Small tourism business networks and destination development, International
Journal of Hospitality Management, 20 (4), 367378.
Uriely, N. and Y. Belhassen (2006), Drugs and risk-taking in tourism, Annals of Tourism Research, 33 (2),
339359.
Vargas-Snchez, A., N. Porras-Bueno and M.A. Plaza-Mejia (2011), Explaining residents attitudes toward
tourism: is a universal model possible?, Annals of Tourism Research, 38 (2), 460480.
Vernon, J. (2000), Barriers to sustainability in tourism-related businesses in South-East Cornwall: results
of discussion groups with tourism business owners, Working Paper No.1. University of Plymouth:
Department of Geographical Sciences.
Vernon, J., S. Essex, D. Pinder and K. Curry (2005), Collaborative policymaking: local sustainable projects,
Annals of Tourism Research, 32 (2), 325345.
Wang, K., P. Jao, H. Chan and C. Chung (2010), Group package tour leaders intrinsic risks, Annals of
Tourism Research, 37 (1), 154179.
Wang, N. (1999), Rethinking authenticity in tourism experience, Annals of Tourism Research, 26 (2), 349370.
Weidenfeld, A., A. Williams and R. Butler (2010), Knowledge transfer and innovation among attractions,
Annals of Tourism Research, 37 (3), 604626.
White, N. and P. White (2007), Home and away: tourists in a connected world, Annals of Tourism Research,
34 (1), 88104.
Williams, P. and G.N. Soutar (2009), Value satisfaction and behavioral intentions in an adventure tourism
context, Annals of Tourism Research, 36 (3), 413438.
Wong, C. and B. McKercher (2011), Tourist information center staff as knowledge brokers: the case of
Macau, Annals of Tourism Research, 38 (2), 481498.
Yang, L. (2011), Ethnic tourism and cultural representation, Annals of Tourism Research, 38 (2), 561585.
Yuksel, F., B. Bramwell and A. Yuksel (2005), Centralized and decentralized tourism governance in Turkey,
Annals of Tourism Research, 32 (4), 859886.
Zahra, A. and N. McGehee (2012), Host perceptions of volunteer tourism: a community capital perspective,
Annals of Tourism Research.

DWYER 9781781001288 Part II.indd 376 25/07/2012 11:07

 19 Participant observation
Nuno F. Ribeiro and Eric W. Foemmel

NATURE OF PARTICIPANT OBSERVATION AND ITS

EVOLUTION

A cornerstone research method of cultural anthropology and sociology, and widely

used in a number of other social sciences, participant observation falls under the broad
umbrella of ethnographic fieldwork (Bernard, 2006). Participant observation consists
of engaging in regular and prolonged interaction with the people the researcher seeks
to study immersing oneself in the participants local environment in order to learn
more about obvious and not-so-obvious aspects of their lives and culture. DeWalt and
DeWalt (2011, p. 1) define it as a method in which an observer takes part in the daily
activities, rituals, interactions, and events of the people being studied as one of the means
of learning the explicit and tacit aspects of their culture. Moreover, by observing and
participating in daily life as much as possible, the researcher is able to overcome partici-
pants natural distrust towards outsiders, becomes part of the living landscape and as a
result is able to collect a myriad of data that would be impossible to obtain otherwise.
Thus the researcher is saddled with a dual task: she is at the same time the investigator
and the data collection instrument.
Both distinct from and complementary to other means of data collection such as semi-
structured and structured interviewing, questionnaire research, focus groups, and so
forth, the nature of participant observation is primarily qualitative (Jennings, 2001; but
see Schensul et al., 1999), although the data produced through participant observation
can be analysed by both qualitative and quantitative means (Bernard, 2006; LeCompte
and Schensul, 1999; see also Oakes, 1998). Participant observation data arises primarily
from field notes of what the researcher observes and participates in, as well as his inter-
pretation of the events that occurred (Emerson et al., 1995). The field notes are then
compiled, analysed, and contrasted with data acquired through other means (e.g., photo-
graphs, interview transcripts, survey data), with the end result being an ethnography, or
written account of a given culture or people (LeCompte and Schensul, 1999).
While the origins of participant observation can be traced back to the work of early
sociologists such as Beatrice Webb (1926), William Foote Whyte (1943), and others at
the then emerging Chicago School in the 1920s (e.g., Park et al., 1925; for an historical
overview of the Chicago School, see Bulmer, 1986), Bronislaw Malinowski is generally
credited with encoding the principles and merits of participant observation (Bernard,
2006). In his seminal Argonauts of the Western Pacific (1922[1984]), Malinowski
described the proper conditions for ethnographic work thus:

cutting oneself off from the company of other white men, and remaining in as close contact
with the natives as possible, which can only be achieved by camping right in their villages ...
by means of this natural intercourse, you learn to know him, and you become familiar with

377

DWYER 9781781001288 Part II.indd 377 25/07/2012 11:07

 378 Handbook of research methods in tourism

his customs and beliefs far better than when he is a paid, often bored, informant (...) There is
all the difference between a sporadic plunging into the company of natives, and being really in
contact with them. What does the latter mean? On the Ethnographers side, it means that his life
in the village, which at first is a strange, sometimes unpleasant, sometimes intensely interesting
adventure, soon adopts a natural course very much in harmony with his surroundings. (pp. 67)

In spite of its colonialist and racist overtones, Malinowskis description of fieldwork

in the Trobriands contains several of the characteristics of modern participant observa-
tion: (1) prolonged immersion in the field; (2) gradual acquaintance and building of
trust between researcher and informant; (3) familiarization with the local language and/
or dialect(s); and (4) the acquisition of data in as much detail as possible. Such char-
acteristics have not changed in almost a century, and they remain hallmarks of sound
participant observation.
If the core of the method itself has known little change from the days of Malinowski
(Bernard, 2006), participant observation has nonetheless evolved, with a growing
number of fields of research such as tourism adopting it as a valid research method-
ology. While space limitations preclude us from delving into the evolution and growing
popularity of participant observation in detail (a propos, see Agar, 1996; Atkinson
and Hammersley, 1994; Becker, 1958; Delamont, 2004; DeWalt and DeWalt, 2011;
Jorgensen, 1989; Platt, 1983), it is nonetheless necessary to identify four moments in the
history of the use of this research method.
The first period, comprising the early attempts of ethnographers to participate in
local communities daily life in order to better understand them and record their char-
acteristics before they disappear (i.e., salvage ethnography see Barnard, 2000),
lasted almost a century from the late 1800s until the early 1980s. In fact, the term
participant observation did not appear until the 1920s (DeWalt and DeWalt, 2011).
With a few notable exceptions (e.g., Cushing and Green, 1981 [1879]; Mead, 1928
[1961]; Lindeman, 1924; McCall and Simmons, 1969), during this period little effort
was devoted by researchers towards systematizing the methodological knowledge nec-
essary in order to conduct participant observation. During this period, the researcher
was supposed to go out into the field and simply do it. The apocryphal and outdated
tales of dissertation advisors counseling the young researcher about to go into the field
to take plenty of quinine and stay away from village women in lieu of rigorous
training in participant observation methods find a parallel in todays lack of formal
coursework in ethnographic field methods in a number of graduate programs in the
social sciences.
The second period spanned roughly a decade (the 1980s) and saw the emergence
of a number of important treatises on the topic of participant observation, such as
Spradleys (1980) and Jorgensens (1989) volumes both entitled Participant Observation,
Polkinghornes Methodology for the Human Sciences (1983), and Van Maanens edited
volume on Qualitative Methodology (1983). The contribution of these works towards the
legitimization of participant observation as a reputed method of data collection in the
social sciences was considerable. Issues such as gaining entry to the field, the researchers
positioning along a continuum ranging from complete outsider to complete insider,
covert and overt observation, the importance of key informants and gate keepers, partic-
ipant observation ethics, and many other topics which until then were privy to a handful
of veteran researchers and accordingly passed from one generation of field workers to

DWYER 9781781001288 Part II.indd 378 25/07/2012 11:07

 Participant observation 379

Complete Observer Observer as Participant Participant as Observer Complete Participant

Source: Gold (1969), Jorgensen (1989).

Figure 19.1 The participant observation continuum

the next were finally discussed, systematized, and brought to the attention of a wider
audience of scholars.
Among these topics, the participant observation continuum (Jorgensen, 1989) is par-
ticularly worthy of discussion (Figure 19.1). Depending on the goals of the research and
the researchers intentions/skills, the researcher will adopt one of four postures in the
field that correspond to different levels of involvement between the researcher and the
research participants. Thus, the researcher may choose to exclusively observe, taking
no part in what is going on; or he may participate to a small extent; he may also choose
to participate to a greater degree; or lastly, the researcher may become a complete par-
ticipant and thus virtually indistinguishable from those he seeks to study, as Sally Cole
(1991) did when she worked alongside fishermens wives in a Portuguese coastal commu-
nity dealing with the early impacts of tourism development in the Mediterranean basin.
In regard to tourism research, the 1980s also marked the beginning of the legitimiza-
tion of tourism as a reputable field of inquiry in the social sciences, and coincidentally or
not, many of tourisms most prominent advocates came from backgrounds where par-
ticipant observation was an important, if not crucial, method of data collection, such as
sociology, anthropology, and human geography (e.g., Cohen, 1984; Crick, 1989; Dann
et al., 1988; Dumont, 1984; Graburn, 1983; Hartmann, 1988). Moreover, a number of
seminal tourism studies using participant observation were published in this period, such
as Boyntons (1986) study of tourism impacts on Amish communities, Esmans (1984)
analysis of tourism as an ethnic preservation tool, Fosters (1986) study on short-lived
societies, and Kemper et al.s (1983) analysis of tourism as a cultural domain.
The third period (1990s2000s) was marked by the interpretivist/postmodern challenge
(for an overview, see Barnard, 2000), which forced researchers in the social sciences and
elsewhere to adopt a more reflexive stance regarding the communities they were study-
ing on the one hand, and on themselves and the product of their research, on the other.
With respect to participant observation, there was an explosion of scholarly literature
on the topic, with an increasing number of researchers publishing the insights garnered
from the encounter between the researcher and the other (i.e., the researched). Earlier
works such as Paul Rabinows Reflections on Fieldwork in Morocco (1977) had set the
tone for a host of studies on post-modernity, colonialism, semiotics, representation, and
power with participant observation as the preferred method of research, and with an
increasing emphasis on self-reflection by the researcher, in what Tedlock (1991) dubbed
the observation of participation.
This vigorous publication period had obvious effects on the use of participant observa-
tion in tourism research, which in turn reflected the then dominant self-reflexive stance of
ethnographic fieldwork. First, participant observation was used to dissect the encounter
between the tourist and the other (the host) in all its nuances, with particular emphasis
being put in the dnouement of power dynamics between the (often wealthy) tourists

DWYER 9781781001288 Part II.indd 379 25/07/2012 11:07

 380 Handbook of research methods in tourism

and the (often impoverished) hosts, and paying also close attention to the symbolic
nature of the tourism activity (e.g., Abram et al. 1997; Bruner, 1996; Chambers, 1997;
Hoskins, 2002; Smith and Brent, 2001). Second, there was a surge in the interest in
research methods by tourism researchers, resulting in the publication of several how-
to research manuals, which included participant observation as an important method
of data collection (e.g., Jennings, 2001; Veal, 1992). And third, it forced the question
of whether the researchers were all that different from the tourists they were studying
(Errington and Gewertz, 1989).
The last period, ranging from the early 2000s to the present, saw the coming of age
of multi-sited ethnographic fieldwork, which had emerged in the mid-1990s (Marcus,
1995). With the accelerated impacts of globalization and the increased connectivity of
local events and global processes, fieldworkers no longer could work within the narrow
confines of secluded villages, and were forced to place their findings within broader
economic, political, and social contexts (Durrenberger, 2003; Vandegriff, 2008). Nelson
Graburns (2002) notion of the ethnographic tourist stands out as a perfect example of
the effects of multi-sited ethnography on tourism research, as does Noel Salazars (2010)
glocal ethnographic study of tourism in Indonesia and Tanzania.

BACKGROUND AND TYPES OF ISSUES PARTICIPANT

OBSERVATION IS DESIGNED TO HANDLE

Observing and participating in a groups activities (e.g., a group tours behavior while
on vacation), a researcher gains intimate knowledge of the group as the researcher goes
from being a complete outsider to a complete insider (Creswell, 1997, p. 123; see also
Figure 19.1). This intimate knowledge provides the researcher with the context needed to
explain differences in behavior that may be influenced by the groups assembly.
When tourism researchers immerse themselves in the field to conduct participant
observation, they begin to understand the context and the associated interactions
of natural surroundings (Riley and Love, 2000, p. 5). This understanding gives the
researcher the information needed to posit theories and also suggest ways to mitigate
conflicts in communities that may be experiencing negative impacts caused by tourism.
Wallace and Diamente (2005) argue that the dialectic dynamics between economics,
ecotourism, and conservation efforts may have detrimental effects on local communities,
and that anthropology may be the voice of reason in addressing the conflicting agendas
of tour operators, conservationists and local residents. As stated earlier, participant
observation is a useful method in conducting research that may offer reasonable solu-
tions to issues in tourism management when in-depth, detailed knowledge of local condi-
tions is of consequence.
Participant observation requires that the researcher do more than simply hang out and
observe the group she is studying. Gaining access to the population under study is criti-
cal. The researcher is a virtual outsider when entering the field for the first time, and will
definitely be noticed by the individuals of a group as he takes notes and asks questions.
It is useful to identify and befriend a gatekeeper who will assist the researchers entry
into the field or introduce them to people who will be essential for successful data collec-
tion. Depending on the group and the situation, the researchers may want to formally

DWYER 9781781001288 Part II.indd 380 25/07/2012 11:07

 Participant observation 381

introduce themselves to the gatekeepers and follow up the meeting with a detailed, pro-
fessional letter of introduction explaining their research goals and methods. Connecting
a face with a letter explaining the research and asking the informants consent to par-
ticipate in the research may help instill a familiarity between the researcher and the key
informants. In other cases, such formalities are not necessary and may even be detrimen-
tal to the research project, such as when the researcher is interested in marginal/illegal
tourism activities, such as sex tourism (e.g., Brennan, 2004).
During her time in the field, the researcher takes notes on the surroundings, interac-
tions, conversations, habits, customs, and daily routines of the group she is studying.
These notes are the raw data the researcher uses to write a detailed and thick description
of the group. The researcher should write a faithful description of the surroundings,
events, people and interactions that will give the reader a vivid understanding of this
social group within the context of their environment.
Direct observation will also give the researcher valuable data needed to confirm or
disprove hypotheses. After an extensive literature review, the researcher should have
relevant research questions and an excellent understanding of theories that pertain to his
study. Prolonged participant observation gives the researcher the opportunity to watch a
group and determine the veracity of current literature or theories. As Sandiford and Ap
(1998) remark, although ethnography is often considered most appropriate in the early
stages of theory development, hypothesis testing can also be carried out by field-workers
at a later stage using focused observations (p. 8).
It is not unforeseeable that two researchers will have different accounts after con-
ducting participant observations at the same field site. Although both accounts may be
valid, participant observation is not a method that can be easily replicated by numer-
ous researchers and yield the exact same results. The most famous example is Derek
Freemans (1999) refutation of Margaret Meads descriptions of Samoa in her seminal
work Coming of Age in Samoa (1928 [1961]) as a stress free utopia complete with free
love and little conflict. Margaret Mead concluded that adolescent Samoan woman
raised in this environment do not have the tumultuous teenage experiences of American
adolescents. After extensive participant observation in Meads same research setting,
Freeman interviewed Meads informants, who confessed they had lied to her. Freeman
also pointed out that there were cultural and social norms that place limitations on the
sexual relationships of adolescent Samoan women (Freeman, 1999). The difference in
their accounts has led to heated debates (e.g., Tcherkzoff, 2001).
In tourism research, we are unaware of controversies equivalent to the MeadFreeman
debate, but nonetheless there are points of discord which participant observation has
helped resolve (or in some cases exacerbated). For instance, the panacea-for-all-evils
approach to tourism as an engine of economic prosperity common in post-World War II
Europe was challenged by ethnographical accounts of the impacts of unbridled tourism
development in communities on the receiving end of mass tourism in Crete (Andriotis,
2001), Goa (Wilson, 1997), Portugal (Cole, 1991), and Spain (Rozenberg, 1990), thus
shedding light on a much more nuanced tourism reality.
Participant observation is a craft in the sense that the researcher must develop excel-
lent note-taking skills, record events in a field journal, and diligently write about her
observations and experiences each time she exits the field. It is best to write ones notes
right after leaving the field as details will be easier to recall (Bernard, 2002). Moreover,

DWYER 9781781001288 Part II.indd 381 25/07/2012 11:07

 382 Handbook of research methods in tourism

taking field notes (or using more intrusive methods such as audio and visual recordings)
may predispose those the researcher is attempting to study against him. Nonetheless,
with time these constraints will fade as the researcher gains the trust of the group under
study. In addition to the tenacity and skill needed to conduct participant observation,
the researcher must develop the social skills needed to establish rapport with the group
she is studying. These skills, combined with rigorous methods of analysis of ethno-
graphic data (LeCompte and Schensul, 1999), can make a profound contribution to the
study of the topic at hand. Although as an outsider it may be possible to observe a group
of individuals, the researcher will only be able to collect superficial data. Participant
observation allows the researcher to become immersed in the group and collect data
that are not apparent to the casual observer. This gives the researcher the background
she needs to explain the structure, culture, and behavior of a group. The researcher con-
ducting participant observation has a more in-depth understanding of a phenomenon
with his analysis than findings from a standard questionnaire or survey, which may have
problems with accuracy and internal validity (Bernard, 2002).

APPLICATIONS OF PARTICIPANT OBSERVATION IN

TOURISM RESEARCH

From Dennis ORourkes Cannibal Tours (1988) to the successive editions of Hosts
and Guests (Smith, 1977[1989]; Smith and Brent, 2001), and to the more recent surge in
ethnographic accounts of postmodern encounters between the tourist and the host (e.g.,
Bruner, 2005; Edensor, 1998; Ness, 2003; Salazar, 2010; Urry, 1990; Yamashita, 2003),
uses of participant observation in tourism research have been plentiful. Participant
observation has been used to study topics as diverse as the impacts of tourism on an
Australian indigenous community (Dyer et al., 2003), vacation cruising (Foster, 1986;
Yarnal and Kerstetter, 2005), trader tourism in Eastern Europe (Konstantinov et al.,
1998), community involvement in tourism development in Nepal and China (Nyaupane
et al., 2006; see also Stonich, 2005), music tourism in Goa (Saldanha, 2002), effects of
tourism development in post-revolutionary Nicaragua (Babb, 2004), tourist consumer
satisfaction (Bowen, 2002), dark tourism in Alcatraz and Robben Island (Strange and
Kempa, 2003), ecotourism (Wallace and Diamente, 2005; see also Weil, 1997), and sport
tourism (Costa and Chalip, 2005).
The vast majority of tourism studies that have used participant observation, however,
have used this technique to look into the impact of tourism development in host com-
munities, on the one hand; and investigate the complexities of the hostguest relation-
ship, on the other. In the case of the former, there is a substantial body of research on
the impact of tourism development on traditional economic activities, such as fishing
(Cole, 1991; Kottak, 1983 [2005]; Johnson and Orbach, 1990), weaving (Boynton, 1986),
and agriculture (Davis and Morais, 2004). In regard to the dynamics between hosts and
tourists, a number of recent studies have brought to light the increasing complexities of
the encounter between local communities and the tourists (e.g., Gmelch, 2004[2009]).
For instance, Kottak (2005) conducted ethnographic fieldwork in Arembepe (north-
eastern Brazil) over the course of more than 30 years (19622004). Making extensive
use of participant observation, Kottak and his research assistants (2005) were able to

DWYER 9781781001288 Part II.indd 382 25/07/2012 11:07

 Participant observation 383

chronicle the changes that occurred in Arembepe from a subsistence economy in the
1970s to an international tourism destination with a sustainable diversified economy
in the early 2000s (Kottak, 2004, p. 501). Kottaks successive revisits to the research site
allowed an almost-unparalleled longitudinal and detailed study of the conditions under
which international tourism impacts local communities and, moreover, how these com-
munities are able to minimize its potential harmful aspects (e.g., commodification of
culture, excessive urban development, mono-employment) and maximize its advantages
(e.g., attraction of foreign currency, decreased unemployment rates, economic diver-
sification). Moreover, the extended permanence in the field over the course of almost
four decades forced the author to reflect on tourism as both an agent and a product of
change; if the first edition of Kottaks study (1983) decried the commodifying and de-
characterizing effects of tourism as an agent of social change, its later editions (1999,
2005), place tourism within the nexus of global economic forces, and acknowledge the
multiple advantages that tourism brought to the people of Arembepe.
Other studies that used participant observation have been more critical of tour-
isms overall impacts in local communities and economies (e.g., Chambers, 2000;
Crick, 1989; Smith 1977[1989]; Stocker, 2007). In an oft-quoted article, Greenwood
(1977[1989]) decried the selling of culture by the pound that tourism activities
had brought to Hondarribia, Spain. Set out to study the Alarde ritual, Greenwood
was able to describe in minute detail how the sudden interest by outsiders (tourists)
had commercialized and, in his opinion, devalued what was an emblematic part
of Hondarribian character and culture. Yet the same author, a few decades later,
revised his earlier findings and debated whether or not his desire to protect a culture
he admired from the effects of tourism had not played a part in his discussion of the
phenomenon (Greenwood, 2004).
Participant observation has also been useful in deconstructing travel imagery, from
glossy magazine adverts of tourism resorts to discount group travel brochures that
portray tourism destinations as idyllic and unidimensional landscapes where the local
community is all but willing to satisfy the tourists every whim, and contrasting it with
factual and participatory accounts of hostguest interactions in tourism destinations
(e.g., Bruner, 2005; Peaslee, 2010; Roland, 2010). Perhaps the most famous of these
studies is Edward Bruners (1996) account of the postmodern encounter between US and
European tourists with a Ghanaian tourism attraction (Elmina Castle) that emphasized
the brutal conditions under which African slaves were kept before being sent to the colo-
nies of the New World. Through participant observation (as a tourist, ethnographer,
and consultant for one of the agencies in charge of preserving and managing the attrac-
tion), in-depth interviews with key informants, and focus groups, Bruner shed light on
the many layers of meaning that the castle and its past held for the Ghanaian popula-
tion and the tourists that originated from several countries, all of which with a colonial
past (United States, United Kingdom, the Netherlands). Their often conflicting views
and actions were made visible by the use of participant observation, and the use of this
research method also provided a rich context imbedded in meaning that would have been
difficult to obtain otherwise.
Overall, tourism studies that have used participant observation as a research method
were helpful in revealing the multifaceted and ambiguous nature of tourism as a human
activity. As Wallace (2005) rightly points out, whereas the early ethnographic tourism

DWYER 9781781001288 Part II.indd 383 25/07/2012 11:07

 384 Handbook of research methods in tourism

studies of the 1970s and 1980s concentrated on the negative impacts of tourism on host
communities and cultures (e.g., Smith, 1977 [1989]), the last 20 years gave way to a host
of research highlighting the positive aspects of tourism development (e.g., Lanfant et
al., 1995; Smith and Brent, 2001). This ambiguous character of tourism activity was evi-
denced thanks in no small part to the use of participant observation as a research meth-
odology. In turn, the use of ethnographic research methods has contributed not only to
move the field forward theoretically (Jafari, 2001), but has also been extremely useful in
the elaboration and implementation of tourism policies.

ADVANTAGES AND LIMITATIONS OF PARTICIPANT

OBSERVATION CONCEPTUALLY AND FOR POLICY
FORMULATION

Witnessing the impacts of tourism, researchers can voice concerns and steer policies to
address issues seen in the field. Participant observation offers tourism researchers the
experience and insight needed to make valid statements based on locally produced (i.e.,
emic) knowledge. Moreover, participation observation gives the researcher the oppor-
tunity to obtain a general understanding of how any social institution of organization
works (Bernard, 2002, p. 355). Participating and writing field notes on the interactions
and behaviors of a group and their culture, the researcher collects valuable data from
within the field. The researcher uses this data to provide a thick description of the
group and their social norms (Geertz, 1973), and also to communicate to the reader
how a culture operates. This first-hand experience, knowledge and understanding of
the groups culture, coupled with an intimate knowledge of the research site, are critical
components of the foundation needed to obtain an expert knowledge of the field. Within
the context of tourism, valid statements and findings about successful tourism opera-
tions, a groups attitudes about tourism, and positive policy developments in tourism
stem from this expert knowledge.
Researching an innovative tourism operator, Paget et al. (2010) used participant
observation methods to study tourists consumer behavior and the operations of Delta,
a tourism operator providing alternative winter activities in France such as dinners
in tepees in remote areas. The success of Deltas programs depended on using existing
resources in unconventional ways. To understand the complexity of this research site
and Deltas operations, these researchers observed the entire tourismscape. Not only
did the researchers observe the interactions between the tourists and tour operators, they
collected data on the natural resources or quality of the ski slopes, the infrastructure of
the town and its ability to host tourists, the amenities of ski resorts and their lifts, and
the promotion and advertizing of the area as a winter destination (Paget et al., 2010).
They reported that the constant presence of the researchers in the field made it possible
for them to establish a rapport with the participants and gain access to previously inac-
cessible data and participate in several strategic company meetings (Paget et al., 2010,
p. 835). This level of interaction using participant observation as a method provided
opportunities to better explain complex processes and allowed the team of researchers
to understand how Delta was able to offer and successfully provide tourists with winter
activities other than skiing.

DWYER 9781781001288 Part II.indd 384 25/07/2012 11:07

 Participant observation 385

In another study, Mason (2005) commented on the benefits of using participant

observation when assessing tourist destinations. In the development of the Erie Canal
National Heritage Corridor in the United States, participant observation gave the plan-
ners the information they needed to reallocate resources to improve facility use. Mason
(2005) reports the canal was primarily used for industrial shipping in the nineteenth
century, but was revitalized as a heritage park offering recreational activities in the
twenty-first century. By observing and interviewing the visitors, the researchers discov-
ered that inadequate information about waterway facilities was proving frustrating to
visitors. With this data, a tourism developer at the Erie Canal Heritage Corridor would
have the information needed to allocate resources to meet the actual needs of the land-
use visitors, and to effectively inform visitors about existing water-based recreational
resources that were not being used to their fullest potential.
Participant observation as a research method offers the researcher an opportunity to
spend a prolonged amount of time in the field, become accepted as a participant, become
acquainted with the host communitys perspective, and understand their beliefs and
attitudes. All of these factors have a positive influence on data collection and analysis
as the researcher will gain an insight into the groups behavior and understand subtle-
ties in the development of attitudes that may challenge existing theories. Understanding
(and ultimately predicting) human behavior is a critical goal of tourism research. It is
imperative to understand the behavior of tourists, and the behavior of the host com-
munity. Interactions between the tourists and hosts, whether harmonious or volatile,
are better understood when researchers identify the attitudes and intent of both parties.
Lepp (2007) studied the residents attitudes about tourism in a Bigodi village in Uganda,
and he felt the use of participant observations and other qualitative methods gave him
the data that improved his understanding of their attitudes and the prediction of their
behavior. After observing participants in the field, Lepp identified the economic benefits
tourism brought to this rural Ugandan community, and he observed that everyone in the
community seemed to benefit from tourism. The tourists brought an outside source of
income when they paid for food. They paid for lodging and other services from farmers
and merchants, who then spent their earnings in the community. What is more, Lepp
(2007, p. 883) suggested that positive attitudes toward tourism would increase inten-
tions to behave in a manner conducive to tourism and this would lead to an increase in
protourism. Lepps findings demonstrate how participant observation can be used to
provide an insightful explanation of human behavior in tourism settings.
The usefulness of participant observation in collecting pertinent data does not end
with making valid statements about the success of tourism programs or the ability to
effectively predict and explain human behavior. Data collected by an objective researcher
using participant observation gives planners the information needed to formulate and
implement policies that consider the values and concerns of the community hosting tour-
ists, and the needs and expectations of tourists visiting a destination. Considering both
parties, ethnocentric biases of the host community and the tourists are mitigated. The
cultural values or social norms of all parties are considered and none are privileged at
the expense of others.
Using participant observation and other qualitative research methods typically asso-
ciated with anthropology to tourism development, Lalone (2005) has coined the term
anthro-planning, which has benefited communities developing heritage parks and

DWYER 9781781001288 Part II.indd 385 25/07/2012 11:07

 386 Handbook of research methods in tourism

tourism sites. Anthro-planning facilitates design projects that plan with and for
people, groups, and communities (Lalone, 2005, p. 136). In contrast, developers may
forgo this type of research and formulate policies and ignore the need to consult with
anyone in the community. In a sense, developing policy in this manner is primarily
developercentric. According to Lalone (2005) a successful research design will use
anthropological techniques to study the social, economic, and political patterns and
groupings within the target community and will apply that knowledge in ways that seek
representative, participatory involvement from the public (p. 137). Objective observa-
tions require the researcher to suspend personal values and judgments based on cultural
norms that may not apply to the community being observed. Planners developing
tourism policies with data collected from participant observation can create seamless
connections between relevant research and the formulation of policy.
On the other hand, exclusive reliance on participant observation methods instead of
more holistic/mixed methods approaches may hinder the success of tourism planning
and development initiatives. Because participant observation data is seldom generaliz-
able beyond the sample of individuals/groups studied (but see Bernard, 2006), it would
be a crucial error to suggest and/or enforce tourism policies based solely on data
obtained via the researchers observations and actions in the field. Moreover, exces-
sive permanence in the field may result in significant emotional and institutional ties,
which may compromise the objectivity of tourism policies based on participant obser-
vation. Consciously or inadvertently, the researcher may be more prone to suggest
policies that benefit his people, as opposed to more equitable tourism development
guidelines (on the ethical quagmires of conducting participant observation in tourism
studies, see Adams, 2005). DeWalt et al. (1998) recommend that self-reflection and
use of additional research methods should be employed to reduce potential researcher
bias.
Nonetheless, participant observation is an effective method that puts the tourism
researcher in the field. This offers the researcher a vantage point to collect data from key
informants, observe subtle nuances in behavior, test theories, and gain insight needed
to develop inclusive policies that acknowledge both the host community and the tour-
ists (Nyaupane et al., 2006). Participant observation is an exciting and effective way to
collect detailed emic data, with the added benefit of necessitating the active involvement
of the researcher with the object of study. In turn, such involvement often generates
additional insights which are helpful in the formulation of tourism policy and tourism
product development and appraisal (Bowen, 2002; Bowie and Chang, 2003).

FURTHER DEVELOPMENTS AND APPLICATIONS OF

PARTICIPANT OBSERVATION IN TOURISM RESEARCH

It seems unlikely that participant observation will experience any dramatic developments
in the near future. Its main precepts (prolonged immersion in the field, direct contact
with the participants, emphasis on emic models of knowledge, focus on meaning and
interpretation) are unlikely to change or be replaced by others. The deceptive simplicity
of this research technique is one of its greatest strengths, and one that makes it particu-
larly suitable for tourism research.

DWYER 9781781001288 Part II.indd 386 25/07/2012 11:07

 Participant observation 387

Conversely, recent developments in the application of participant observation are

noteworthy. First, in recent years there has been a push towards rapid assessment (RAP)
research (i.e., quick ethnography Handwerker, 2002; see also Beebe, 2001), whereby
ethnographic methods such as participant observation are still desirable, but time and
budget constraints limit their application over the course of typical time frameworks
(i.e., a year or more). Thus researchers have developed variations of a wide range of qual-
itative methods such as participant observation which can be used in a shorter time
span while still retaining many of its strengths (Handwerker, 2002). Thus far, RAP uses
in tourism research have been limited to environmental impact assessment of tourism
activities (e.g., Ervin, 2003; Kikuchi et al., 2003), but it is likely that, particularly in times
of budget constraints and fiscal austerity, RAP will see an increase in its popularity in
tourism research.
Second, there has long been an interest in increasing the external validity (i.e., the pos-
sibility of generalization of results) of ethnographic methods such as structured obser-
vation, which involves collecting participant observation data at pre-defined, regular
intervals (DeWalt and DeWalt, 2011; Kawulich, 2005). Techniques such as instanta-
neous spot sampling and random spot checks (Bernard, 2006) have also been used to
bolster the possibility of inference from small, non-randomly sampled data. To our
knowledge, only our most recent work (Ribeiro and Foemmel, 2012; Ribeiro et al., 2011)
has made use of these techniques within a tourism setting, but hopefully more tourism
studies will employ them. The use of recent technological advances, such as hand-held
computers, is also likely to play a role in future uses of participant observation when
external validity is a concern (Ice, 2004).
Lastly, mirroring a growing trend in the social sciences towards the blurring
of the divide between quantitative and qualitative research (Creswell and Clark,
2007; Johnson and Onwuegbuzie, 2004; Tashakkori and Teddlie, 1998), there has
been an increase in the adoption by tourism scholars of mixed methodologies of
research(MacKay and Campbell, 2004; Lemelin, 2006; Pansiri, 2005; Riley and Love,
2000). Taking into account the multi and inter-disciplinary nature of tourism research,
one expects the use of mixed-methods research methods to increase among tourism
scholars in years to come, with participant observation playing a crucial role in the
process.

REFERENCES

Abram, S., J., Waldren and D.V. Mcleod (eds) (1997), Tourists and Tourism: Identifying with People and Places,
Oxford: Berg.
Adams, K.M. (2005), Generating theory, tourism, and world heritage in Indonesia: ethical quandaries for
anthropologists in an era of tourist mania, NAPA Bulletin, 23, 4559.
Agar, M. (1996), The Professional Stranger: An Informal Introduction to Ethnography, 2nd edition, Beverly
Hills, CA: Sage Publications.
Andriotis, K. (2001), Tourism planning and development in Crete: recent tourism policies and their efficacy,
Journal of Sustainable Tourism, 9 (4), 298316.
Atkinson, P. and M. Hammersley (1994), Ethnography and participant observation, in N.K. Denzin and Y.S.
Lincoln (eds), Handbook of Qualitative Research, Los Angeles, CA: Sage, pp. 248261.
Babb, F.E. (2004), Recycled Sandanistas: from revolution to resorts in the new Nicaragua, American
Anthropologist, 106 (3), 541555.
Barnard, A. (2000), History and Theory in Anthropology, New York: Cambridge University Press.

DWYER 9781781001288 Part II.indd 387 25/07/2012 11:07

 388 Handbook of research methods in tourism

Becker, H.S. (1958), Problems of inference and proof in participant observation, American Sociological
Review, 23 (6), 652660.
Beebe, J. (2001), Rapid Assessment Process: An Introduction, Thousand Oaks, CA: AltaMira Press.
Bernard, H.R. (2002), Research Methods in Anthropology: Qualitative and Quantitative Approaches, 3rd
edition, Lanham, MD: AltaMira Press.
Bernard, H.R. (2006), Research Methods in Anthropology: Qualitative and Quantitative Approaches, 4th
edition, Lanham, MD: AltaMira Press.
Bowen, D. (2002), Research through participant observation in tourism: a creative solution to the measure-
ment of consumer satisfaction/dissatisfaction (CS/D) among tourists, Journal of Travel Research, 41 (1),
414.
Bowie, D. and J.C. Chang (2003), Tourist satisfaction: a view from a mixed international guided package
tour, Journal of Vacation Marketing, 11 (4), 303322.
Boynton, L.L. (1986), The effect of tourism on Amish quilting design, Annals of Tourism Research, 13,
451564.
Brennan, D. (2004), Whats Love Got to do With it? Transnational Desires and Sex Tourism in the Dominican
Republic, Durham, NC: Duke University Press.
Bruner, E.M. (1996), Tourism in Ghana: the representation of slavery and the return of the Black diaspora,
American Anthropologist, 98 (2), 290304.
Bruner, E.M. (2005), Culture on Tour: Ethnographies of Travel, Chicago, IL: The University of Chicago Press.
Bulmer, M. (1986), The Chicago School of Sociology. Chicago, IL: University of Chicago Press.
Chambers, E. (1997), Tourism and Culture: An Applied Perspective, Albany, NY: SUNY Press.
Chambers, E. (2000), Native Tours: An Applied Perspective, Albany, NY: SUNY Press.
Cohen, E. (1984), The sociology of tourism: approaches, issues, and findings, Annual Review of Sociology,
10, 373392.
Cole, S.C. (1991), Women of the Praia: Work and Lives in a Portuguese Community, Princeton, NJ: Princeton
University Press.
Costa, C.A. and L. Chalip (2005), Adventure sport tourism in rural revitalization: an ethnographic evalua-
tion, European Sport Management Quarterly, 5 (3), 257279.
Creswell, J.W. (1997), Qualitative Inquiry and Research Design: Choosing Among Five Traditions, Thousand
Oaks, CA: Sage.
Creswell, J.W. and V.L. Clark (2007), Designing and Conducting Mixed Methods Research, Thousand Oaks,
CA: Sage.
Crick, M. (1989), Representations of international tourism in the social sciences: sun, sex, sights, savings, and
servility, Annual Review of Anthropology, 18, 307344.
Cushing, F.H. and J. Green (eds) (1981) [1879], Zuni: Selected writings of Frank Hamilton Cushing, Winnipeg,
Manitoba: Bison Books.
Dann, G., D. Nash and P. Pearce (1988), Methodology in tourism research, Annals of Tourism Research, 15,
128.
Davis, J.S. and D.B. Morais (2004), Factions and enclaves: small towns and socially unsustainable tourism
development, Journal of Travel Research, 43 (1), 310.
Delamont, S. (2004), Ethnography and participant observation, in G. Gobo, J.F. Gubrium and D. Silverman
(eds), Qualitative Research Practice, London: Sage, pp. 205217.
DeWalt, K.M. and B.R. DeWalt (2011), Participant Observation: A Guide for Fieldworkers, 2nd edition,
Lanham, MD: AltaMira Press.
DeWalt, K.M., B.R. DeWalt and C.B. Weyland (1998), Participant observation, in H.R. Bernard (ed.),
Handbook of Methods in Cultural Anthropology, Walnut Creek, CA: AltaMira Press, pp. 259300.
Dumont, J. (1984), A matter of touristic indiffrance, American Ethnologist, 11 (1), 139151.
Durrenberger, E.P. (2003), Global processes, local systems, Urban Anthropology, 32 (34), 253279.
Dyer, P., L. Aberdeen and S. Schuler (2003), Tourism impacts on an Australian indigenous community: a
Djabugay case study, Tourism Management, 24, 8395.
Edensor, T. (1998), Tourists at the Taj: Performance and Meaning at a Symbolic Site, New York: Routledge.
Emerson, R.M., R.I. Fretz and L.L. Shaw (1995), Writing Ethnographic Fieldnotes, Chicago, IL: University
of Chicago Press.
Errington, F. and D. Gewertz (1989), Tourism and anthropology in a post-modern world, Oceania, 60 (1),
3754.
Ervin, J. (2003), Rapid assessment of protected area management effectiveness in four countries, BioScience,
53 (9), 833841.
Esman, M.R. (1984), Tourism as ethnic preservation: the Cajuns of Louisiana, Annals of Tourism Research,
11, 451467.
Foster, G.M. (1986), South Seas cruise: a case study of a short-lived society, Annals of Tourism Research, 13,
215238.

DWYER 9781781001288 Part II.indd 388 25/07/2012 11:07

 Participant observation 389

Freeman, D. (1999), The Fateful Hoaxing of Margaret Mead: A Historical Analysis of her Samoan Research,
Boulder, CO: Westview.
Geertz, C. (1973), The Interpretation of Cultures, New York: Basic Books.
Gmelch, S.B. (ed.) (2004 [2009]), Tourists and Tourism: A Reader, 2nd edition, Prospect Heights, IL: Waveland
Press.
Gold, R.L. (1969), Roles in sociological field observations, in G.J. McCall and J.L. Simmons (eds), Issues in
Participant Observation, Reading, MA: Addison-Wesley, pp. 3039.
Graburn, N.H. (1983), The anthropology of tourism, Annals of Tourism Research, 10, 933.
Graburn, N.H. (2002), The ethnographic tourist, in G.M. Dann (ed.), The Tourist as a Metaphor of the Social
World, Wallingford, UK: CABI Publishing, pp. 1940.
Greenwood, D.J. (1977 [1989]), Culture by the pound: an anthropological perspective on tourism as cul-
tural commoditization, in V.L. Smith (ed.), Hosts and Guests: The Anthropology of Tourism, 2nd edition,
Philadelphia: University of Pennsylvania Press, pp. 171185.
Greenwood, D.J. (2004), Culture by the pound: an anthropological perspective on tourism as cultural com-
moditization, in S. Gmelch (ed.), Tourists and Tourism: A Reader, Prospect Heights, IL: Waveland Press,
pp. 157169.
Handwerker, W.P. (2002), Quick Ethnography: A Guide to Rapid Multi-Method Research, Walnut Creek, CA:
AltaMira Press.
Hartmann, R. (1988), Combining field methods in tourism research, Annals of Tourism Research, 15, 88105.
Hoskins, J. (2002), Predatory voyeurs: tourists and tribal violence in remote Indonesia, American
Ethnologist, 29, 797808.
Ice, G.H. (2004), Technological advances in observational data collection: the advantages and limitations of
computer-assisted data collection, Field Methods, 16 (3), 352375.
Jafari, J. (2001), The scientification of tourism, in V.L. Smith and M. Brent (eds), Hosts and Guests Revisited:
Tourism Issues of the 21st Century, New York: Cognizant Communication Corporation, pp. 2841.
Jennings, G. (2001), Tourism Research. Milton, Australia: Wiley.
Johnson, J.C. and M.K. Orbach (1990), A fishery in transition: the impact of urbanization on Floridas spiny
lobster fishery, City & Society, 4 (1), 88104.
Johnson, R.B. and A.J. Onwuegbuzie (2004), Mixed methods research: a research paradigm whose time has
come, Educational Researcher, 33 (7), 1426.
Jorgensen, D.L. (1989), Participant Observation: A Methodology for Human Studies, Thousand Oaks, CA:
Sage Publications.
Kawulich, B.B. (2005), Participant observation as a data collection method, Forum: Qualitative Social
Research, 6 (2), art. 43.
Kemper, R.V., J.M. Roberts and R.D. Goodwin (1983), Tourism as a cultural domain: the case of Taos, New
Mexico, Annals of Tourism Research, 10, 149171.
Kikuchi, R.K., Z.A. Leo, V. Testa, L.X. Dutra and S. Spano (2003), Rapid assessment of the Abrolhos
Reefs, Eastern Brazil, Atoll Research Bulletin, 496, 172187.
Konstantinov, Y., G.M. Kressel and T. Thuen (1998), Outclassed by former outcasts: petty trading in Varna,
American Ethnologist, 25 (4), 729745.
Kottak, C.P. (2004), An anthropological take on sustainable development: a comparative study of change,
Human Organization, 63 (4), 501510.
Kottak, C.P. (2005), Assault on Paradise: The Globalization of a Little Community in Brazil, 4th edition, New
York: McGraw-Hill.
Lalone, M. (2005), An anthro-planning approach to local heritage tourism: case studies from Appalachia,
NAPA Bulletin, 23, 135150.
Lanfant, M.-F., J.B. Allcock and E.M. Bruner (eds) (1995), International Tourism: Identity and Change,
London: Sage.
LeCompte, M.D. and J.J. Schensul (1999), Analyzing and Interpreting Ethnographic Data, Walnut Creek, CA:
AltaMira Press.
Lemelin, R.H. (2006), The gawk, the glance, and the gaze: ocular consumption and polar bear tourism in
Churchill, Manitoba, Canada, Current Issues in Tourism, 9 (6), 516534.
Lepp, A. (2007), Residents attitudes towards tourism in Bigodi village, Uganda, Tourism Management, 28,
876885.
Lindeman, E.C. (1924), Social Discovery: An Introduction to the Study of Functional Groups, New York:
Republic Publishing Group.
McCall, G.J. and J.L. Simmons (eds) (1969), Issues in Participant Observation, Reading, MA: Addison-Wesley.
MacKay, K.J. and J.M. Campbell (2004), A mixed-methods approach for measuring environmental impacts
in nature-based tourism and outdoor recreation settings, Tourism Analysis, 9 (3), 141152.
Malinowski, B. (1922 [1984]), Argonauts of the Western Pacific: An Account of Native Enterprise and Adventure
in the Archipelagoes of Melanesian New Guinea, Prospect Heights, IL: Waveland Press.

DWYER 9781781001288 Part II.indd 389 25/07/2012 11:07

 390 Handbook of research methods in tourism

Marcus, G. (1995), Ethnography in/of the world system: the emergence of multi-sited ethnography, Annual
Review of Anthropology, 24, 95117.
Mason, A. (2005), Applied anthropology and heritage tourism: working for the Western Erie Canal Heritage
Corridor planning committee, NAPA Bulletin, 23, 159169.
Mead, M. (1928 [1961]), Coming of Age in Samoa, New York: Morrow Quill.
Ness, S.A. (2003), Where Asia Smiles: An Ethnography of Philippine Tourism. Philadelphia, PA: University of
Pennsylvania Press.
Nyaupane, G.P., D.B. Morais and L. Dowler (2006), The role of community involvement and number/type of
visitors on tourism impacts: a controlled comparison of Annapurna, Nepal and Northwest Yunnan, China,
Tourism Management, 27, 13731385.
ORourke, D. (director) (1988), Cannibal Tours [motion picture]. Australia: Institute of Papua New Guinea
Studios.
Oakes, M.P. (1998), Statistics for Corpus Linguistics, Edinburgh, UK: Edinburgh University Press.
Paget, E., F. Dimanche and J. Mounet (2010), A tourism innovation case: an actor-network approach, Annals
of Tourism Research, 37 (3), 828847.
Pansiri, J. (2005), Pragmatism: a methodological approach to researching strategic alliances in tourism,
Tourism Planning and Development, 2 (3), 191206.
Park, R.E., E.W. Burgess and R.D. McKenzie (1925), The City, Chicago, IL: The University of Chicago Press.
Peaslee, R.M. (2010), The man from New Line knocked on the door: tourism, media power, and Hobbiton/
Matamata as boundaried space, Tourist Studies, 10 (1), 5773.
Platt, J. (1983), The development of the participant observation method in sociology: origin, myth and
history, Journal of the History of the Behavioral Sciences, 19 (4), 379393.
Polkinghorne, D. (1983), Methodology for the Human Sciences, Albany, NY: State University of New York
Press.
Rabinow, P. (1977), Reflections on Fieldwork in Morocco, Berkeley, CA: University of California Press.
Ribeiro, N.F., E.W. Foemmel (2012, June), A map of the tourists mind? Using Graphic Layout Algorithm
(GLA) to visualize on-site destination image perception, Proceedings of the 43rd Travel and Tourism
Research Association (TTRA) Annual Conference, Virginia Beach, VA.
Ribeiro, N.F., E.W. Foemmel and T. Liechty (2011), Using a variety of methods to compare related free
lists and investigate the relationship between perceptions of meaning and behavior of a leisure experi-
ence, Proceedings of the 13th Canadian Congress of Leisure Research (CCLR) Annual Conference, St.
Catharines, ON.
Riley, R.W. and L.L. Love (2000), The state of qualitative tourism research, Annals of Tourism Research, 27
(1), 164187.
Roland, L.K. (2010), Tourism and the commodification of Cubanidad, Tourist Studies, 10 (1), 318.
Rozenberg, D. (1990), Tourism et utopie aux Balares, Paris: LHarmattan.
Salazar, N.B. (2010), Envisioning Eden: Mobilizing Imaginaries in Tourism and Beyond, New York: Berghahn
Books.
Saldanha, A. (2002), Music tourism and factions of bodies in Goa, Tourist Studies, 2 (1), 4362.
Sandiford, P.J. and J. Ap (1998), The role of ethnographic techniques in tourism planning, Journal of Travel
Research, 37, 311.
Schensul, S.L., J.J. Schensul and M.D. LeCompte (1999), Essential ethnographic methods: Observations, inter-
views, and questionnaires, Walnut Creek, CA: AltaMira Press.
Smith, V.L. (ed.) (1989), Hosts and Guests: The Anthropology of Tourism 2nd edition, Philadelphia, PA:
University of Pennsylvania Press.
Smith, V.L. and M. Brent (eds) (2001), Hosts and Guests Revisited: Tourism Issues of the 21st Century, New
York: Cognizant Communication.
Spradley, J. (1980), Participant Observation, New York: Holt, Rinehart, & Winston.
Stocker, K. (2007), Identity as work: changing job opportunities and indigenous identity in the transition to a
tourist economy, Anthropology of Work Review, 28 (2), 1822.
Stonich, S.C. (2005), Enhancing community-based tourism development and conservation in the Western
Caribbean, NAPA Bulletin, 23, 7786.
Strange, C. and M. Kempa (2003), Shades of dark tourism: Alcatraz and Robben Island, Annals of Tourism
Research, 30 (2), 386405.
Tashakkori, A. and C. Teddlie (1998), Mixed Methodology: Combining Qualitative and Quantitative Approaches,
Thousand Oaks, CA: Sage.
Tcherkzoff, S. (2001), Le mythe occidental de la sexualit polynsienne: 19281999 Margaret Mead, Derek
Freeman et Samoa, Paris: PUF.
Tedlock, B. (1991), From participant observation to the observation of participation: the emergence of narra-
tive ethnography, Journal of Anthropological Research, 47 (1), 6994.
Urry, J. (1990), The Tourist Gaze, London: Sage.

DWYER 9781781001288 Part II.indd 390 25/07/2012 11:07

 Participant observation 391

Vandegriff, D. (2008), This isnt paradise I work here: Global restructuring, the tourism industry, and
women workers in Caribbean Costa Rica, Gender & Society, 22 (6), 778798.
Van Maanen, J. (ed.) (1983), Qualitative Methodology, Beverly Hills, CA: Sage.
Veal, A.J. (1992), Research Methods for Leisure and Tourism: A Practical Guide, London: Longman.
Wallace, T. (2005), Tourism, tourists, and anthropologists at work, NAPA Bulletin, 23, 126.
Wallace, T. and D.N. Diamente (2005), Keeping the people in the parks: a case study from Guatemala,
NAPA Bulletin, 23, 191218.
Webb, B. (1926), My Apprenticeship, London: Longmans.
Weil, J. (1997), An Ecomuseo for San Vicente: ceramic artisans and cultural tourism in Costa Rica, Museum
Anthropology, 21 (2), 2338.
Whyte, W.F. (1943), Street Corner Society, Chicago, IL: University of Chicago Press.
Wilson, D. (1997), Paradoxes of tourism in Goa, Annals of Tourism Research, 24 (1), 5275.
Yamashita, S. (1999 [2003]), Bali and Beyond: Explorations in the Anthropology of Tourism, J.S. Eades (transl.),
New York: Berghahn Books.
Yarnal, C.M. and D.K. Kerstetter (2005), Casting off: an exploration of cruise ship space, group tour behav-
ior, and social interaction, Journal of Travel Research, 43, 368379.

DWYER 9781781001288 Part II.indd 391 25/07/2012 11:07

 20 Cross-cultural approaches
Carla Almeida Santos and Changsup Shim

NATURE OF THE CROSS-CULTURAL APPROACH AND ITS

EVOLUTION

Traditionally utilized in psychology, the emphasis of cross-cultural research is on the com-

parative examination of individuals from different countries or ethnic groups. Largely,
such research approaches culture as an underlying influence that collectively directs indi-
viduals beliefs, norms, values, customs, and ideological perspectives; all of which serve
to govern individuals behavior. Accordingly, such a view of culture assumes a collective
framework of experiences and beliefs that serve to construct a common consensus of not
only what to expect, but also how to behave in any given setting (Hofstede, 2001; Kim et
al., 2002). Culture, therefore, has largely been thought of as a standardized, causal, and
constant system of meanings shared by and amongst its group members. The result has
been a considerable body of cross-cultural research that serves to categorize national cul-
tures along a set of standard cultural dimensions (Hofstede, 2001). Such work has signifi-
cantly contributed and furthered our knowledge and understanding of consumer behavior
and marketing. Indeed, the study of cross-cultural difference reveals its importance and
centrality largely for its ability to account for consumer behavior, as well as suggest effec-
tive and efficient marketing strategies and tools aimed at a variety of targeted markets
(Hsieh and Tsai, 2009). In particular, and as it relates to the focus of this book, such
understanding has proven central in travel and tourism research given the proposed role
of cultural and national uniqueness in shaping tourist demand and behavior (Crotts and
Erdmann, 2000; Graburn, 1995; Pizam and Sussman, 1995; Reisinger and Turner, 2003).
From this perspective, it naturally follows that the vast majority of published cross-
cultural research conceptualizes culture as a variable, with statistical comparisons that
allow for description of the range and distribution of cross-cultural variation, as well as
the testing of hypotheses proposed to explain the variation recorded (Jackson and Niblo,
2003; van de Vijver and Leung, 1997). Consequently, extant cross-cultural research has
largely been premised on the assumption of universal principals governing the relation-
ship between culture and behavior. In other words, a defining assumption of traditional
cross-cultural research designs is that members of a given cultural group share specific
characteristic behaviors and patterns of traits. In so doing, the perspective inherent in
such research designs is that individuals who share the same culture always and neces-
sarily share an automatic tendency toward one or another form of behavior and pat-
terns of traits. Such an approach naturally favors universal laws and assumes culturally
comparable constructs. However, criticism has increasingly mounted for this approach
to culture and its lack of sensitivity toward understanding behavior as universal versus
cultural specific. In particular, there is increasing recognition that scientific knowledge
does not necessarily reflect and describe cultural reality, nor that culture can be studied,
recognized and understood as existing in a vacuum.

392

DWYER 9781781001288 Part II.indd 392 25/07/2012 11:07

 Cross-cultural approaches 393

Consequently, guided by a distrust of shared, cohesive cultural systems, social

science and humanities scholars began problematizing dominant cross-cultural research
approaches. They proposed that the assumption of universal principles governing the
relationship between culture and behavior failed to recognize cultural intricacies, com-
plexities and paradoxes, as well as differential identities and dynamics among members
of a given cultural group (Stanfield, 1993). To address such limitations, they proposed
that cross-cultural research embrace a variety of epistemological and methodologi-
cal approaches that consider how cultural perspectives and social situations are con-
tinuously produced, negotiated and modified (Gupta and Ferguson, 1997; Jackson and
Niblo, 2003). Qualitative research approaches were deemed particularly appropriate
in answering this call given that their methods often generate descriptive data that can
assist in investigating and identifying local, social, cultural and political processes as they
unfold and shift (Denzin and Lincoln, 2008; Hall and Kulig, 2004; Liamputtong, 2008,
2009; Papadopoulos and Lees, 2002; Smith, 2008; Tillman, 2006). As Morris suggests,
qualitative research is at the sociological vanguard for investigating cross-cultural
issues (2007, p. 410).
Seeking to answer the abovementioned call, cross-cultural qualitative research
approaches have developed around specific theory and application (e.g., ethnography,
grounded theory). Central to these approaches is the notion that culture is not something
that individuals possess; instead, a variety of cultures serve to form an inherent part
of each individual and are articulated in a variety of ways across multiple social settings.
Largely, these approaches, when put into practice, do share a significant overlap in data
collection methods. Cross-cultural qualitative research approaches are, therefore, less
reflective of a particular technique and instead more reflective of a particular epistemol-
ogy that serves to guide data collection and analysis. With this in mind, this chapter
discusses a particular technique used extensively in cross-cultural qualitative research
data collection: cross-cultural interviewing. The decision to focus on cross-cultural
interviewing is attributable to the fact that interviewing is one of the most fundamental
techniques used in qualitative research; and as a result, the large majority of qualitative
research makes use of individual or group interviews as one of its principal data collec-
tion methods. However, at the onset it should also be noted that it is not the intent of
this chapter to offer a prescriptive orthodoxy with regard to cross-cultural interviewing.
To be sure, the specifics of interviewing in cross-cultural research are identical to other
kinds of interviews (specifics such as building strong rapport with research participants,
thoughtful probing of research participants, maintaining visibility in the studied com-
munity, and so on). Instead, this chapter introduces a practical discussion regarding
cross-cultural qualitative research, as well as some of the distinctive issues and prac-
tices associated with cross-cultural interviewing that are often missing from published
research.

BACKGROUND AND TYPES OF PROBLEMS THAT A CROSS-

CULTURAL APPROACH IS DESIGNED TO HANDLE

In any given investigation methodological appropriateness is largely contingent upon

a variety of factors such as the issue being investigated, the knowledge available to the

DWYER 9781781001288 Part II.indd 393 25/07/2012 11:07

 394 Handbook of research methods in tourism

researchers, and the participants degree of recognition and acceptance of the techniques
used in the study. In general, when the objective is to generalize results or facilitate com-
parisons mixed or exclusively quantitative approaches in cross-cultural research are best
suited (Bhawuk and Triandis, 1996). However, when the objective is to investigate and
understand culturally unknown scenarios either from an in-depth or unique perspective,
qualitative approaches in cross-cultural research are best suited (Bhawuk and Triandis,
1996). Qualitative research is, therefore, traditionally proposed as essential when investi-
gating questions of subjective experience and situational meaning because it provides
a better opportunity for conveying sensitivity (Davies et al., 2009, p. 6).
Seeking to investigate and convey cultural sensitivities, a variety of principles serve
to guide cross-cultural qualitative research. Some of those principles are: (1) that
research participants generally know more about the topic, its complexities and para-
doxes than the researcher; (2) that those doing the interpreting be they the researcher
or participant ultimately affect the interpretation; and (3) that problematizing and
understanding more significantly contributes to advancing knowledge than does predic-
tion and control. In particular, cross-cultural qualitative research investigates subjective
human experiences from the perspective that understanding of reality can change over
time and in different social contexts (Dew, 2007, p. 434). Seeking to problematize and
understand meanings and interpretations, as well as the resulting experiences, qualitative
research emphasizes flexibility, fluidity, and interpretation in its approaches (Denzin and
Lincoln, 2008; Liamputtong, 2009).
A particular data collection technique frequently employed in cross-cultural qualita-
tive research is cross-cultural interviewing. This technique has proven particularly helpful
and central for investigating cross-cultural values, as well as perceptions and interactions
between individuals from different cultural backgrounds, information gathering proc-
esses, and processes of meaning making amongst others. One research area in which
cross-cultural interviewing is particularly useful is the study of cross-cultural values
(e.g., the basis for why individuals hold certain values). As Chan proposes, whereas
theory posits that values influence consumer behavior, empirical research fails to estab-
lish a strong link between values and consumer behavior, particularly in cross-cultural
contexts (2009, p. 322); to address this issue, he advocates for a qualitative approach
to the measurement of cross-cultural values. Specifically, Chan (2009) proposes that the
investigation of cross-cultural values is best addressed by cross-cultural interviewing.
Moreover, he proposes that cross-cultural interviews are better positioned to avoid the
problems of conceptual, functional, and translation equivalence that permeate much of
the existing cross-cultural research. Indeed, interviews play a vital role in the investiga-
tion of cross-cultural values because amongst others, they permit further elaboration and
probing on both the interviewers questions and the participants answers; in turn allow-
ing the researcher to capture and grasp the manifold aspects of the complex construct of
cross-cultural values.
Another focus area in cross-cultural qualitative research where interviews are par-
ticularly useful is the study of perception and interaction between individuals from dif-
ferent cultural backgrounds. Looking to engage with recent criticisms of the dominant
approach to the study of cross-cultural perception, Heijes (2011) investigates the role of
power dynamics on cross-cultural perception. Specifically, he examines the role of power
and power differences between European Dutch and African Curaaoans. He proposes

DWYER 9781781001288 Part II.indd 394 25/07/2012 11:07

 Cross-cultural approaches 395

that changes in cross-cultural perception between the two groups is largely based on
the amount of power imbalance. In particular, he explains that perceptions between
European Dutch and African Curaaoans change depending on the amount of power
asymmetry between the groups, whereas the diminution of power asymmetry led to the
possibility of effectively bridging the cultural differences by Curaaoans (Heijes, 2011,
p. 670). As the study sought to describe perception and interaction between respond-
ents, Heijes (2011) employed cross-cultural interviews which started with open-ended
questions and gradually moved toward semi-structured interviews so as to provide con-
tinuous structure and focus to data collection and interpretation. The findings further
promote the notion that research on cross-cultural perception must take into account
power dynamics between relevant groups; and that such dynamics are best studied, rec-
ognized and understood when examined via the use of interviews.

APPLICATIONS OF THE CROSS-CULTURAL APPROACH TO

TOURISM

Culture is often proposed as playing a significant role in tourism-related decisions,

including where to travel and how to behave, as well as perceptions regarding quality of
tourism services and experiences (Crotts and Erdmann, 2000; Crotts and Litvin, 2003;
Pizam and Sussman, 1995; Reisinger and Turner, 2003). To be sure, extant studies affirm
that tourist decision making, behavior, and expectations are affected by cultural factors
(Chen, 2000; Gursoy and Chen, 2000; Hsieh and Tsai, 2009; Kim and Prideaux, 2005;
Lee and Lee, 2009; Lee and Sparks, 2007; MacKay and Fesenmaier, 2000; Richardson
and Crompton, 1988). For example, Graburn (1995) suggests that the Japanese are more
likely to travel in groups due to their low sense of cultural confidence, while Europeans
and North Americans are more likely to travel on their own due to their high sense of
cultural confidence. Pizam and Sussman (1995) also propose that the Japanese, French
and Italians, unlike North Americans, are less tolerant of uncertainty. While Tsang and
Ap (2007) argue that, whereas Asian tourists place more emphasis on the quality of
service encounters as a key factor in their travel experience, Western tourists place more
emphasis on goal completion, efficiency, and time saving as key factors in their travel
experience. Therefore, given the proposed role of cultural and national uniqueness in
shaping tourist decision making, behavior, and expectations, cross-cultural research has
proven central in travel and tourism literature. Such research, however, is not without
its criticism both in terms of its adoption of universal and deterministic values, as well
as methodological approaches (Dimanche, 1994; Peabody, 1985). Indeed, while cross-
cultural tourism research continues to support the investigation of the role of culture
on tourist decision making, behavior, and expectations, it increasingly recognizes that
culture is a fluid construct with a limited lifespan for marketing purposes (Laing and
Crouch, 2005, p. 211).
Although extant cross-cultural tourism research continues to adopt universal and
deterministic values in regard to culture as its starting point, an increasing number of
scholars are calling for tourism studies to go beyond these dimensions (Chang et al.,
2010; Laing and Crouch, 2005; Maoz, 2007; Santos and Buzinde, 2007; Uriely et al.,
2009). Such research calls for an understanding of the social and cultural contexts

DWYER 9781781001288 Part II.indd 395 25/07/2012 11:07

 396 Handbook of research methods in tourism

and lived experiences of both the tourists and the toured, and largely falls into one of
three approaches. The first, and most traditional approach to cross-cultural qualitative
tourism research, is comparative studies that investigate cultural differences between dif-
ferent groups of individuals. Such studies focus mostly on issues that are strongly influ-
enced by a tourists particular ethnicity or nationality. For example, from a grounded
theory approach, Laing and Crouch (2005) conducted a cross-cultural qualitative
analysis of frontier travel experiences (i.e., travel to the most remote and amazing places
on earth, such as the poles, the harshest deserts, or the peaks of the highest mountains).
In order to compare the cultural differences in these travelers motivations as well as
understand the role of cultural background, in-depth interviews were conducted with
frontier travelers from Great Britain, America, and Australia. The analysis identified
several major themes: challenge/goal setting, self-actualization, novelty, and childhood
influences. Although they mainly sought to identify cross-cultural differences between
the three nationalities, they also found that, across all themes, participants from all three
countries shared similar motivations largely based on a common tradition of exploration
in British and Australian history, as well as the frontier myth in North American history.
Some comparative cross-cultural tourism studies do not seek to identify differences
between groups of tourists. They instead focus on the relationship between international
tourists and host community members. For example, Uriely et al. (2009) explored the
northern region of Sinai, which is largely frequented by Jewish-Israeli tourists. They
focused on what they called the bubble of serenity which resulted from the encounter
between Israeli tourists and their Egyptian hosts. The authors used a variety of qualita-
tive data collection methods including participant observation, semi structured in-depth
interviews, and informal conversations with Israeli tourists and their Egyptian hosts.
The questions addressing Israeli tourists included their motivation to travel to Sinai,
their feelings during the stay, and their interactions with the local hosts. At the same
time, Egyptian hosts were asked about the nature of their work, their feelings as service
providers, and their interaction with Israeli tourists. The results propose that Israeli
tourists desire an inexpensive and calm vacation, while Egyptians are willing to supply
it in exchange for economic benefits. Moreover, both parties were aware of their mutual
interests and readily served in line with them. The authors identify five main practices
for avoiding possible impediments to peaceful encounters: perceiving Sinai as an ex-
territory; avoiding politics; making friends; stressing similarities; and, finally, distin-
guishing between Good Guys and Bad Guys.
The second approach to cross-cultural qualitative tourism research investigates groups
of tourists from a particular national or ethnic background. The focus of such studies is
largely on how cultural background influences tourist motivation, behavior, and satis-
faction. The main purpose of these studies is to gain a deeper understanding of a given
group of individuals, as well as that groups differences from others. For example, Chang
et al. (2010) examined Chinese tourists food preferences by employing focus-group
interviews and on-site observations. The study focused on international tourists from
Mainland China, Hong Kong, and Taiwan; all of whom are thought to share a similar
food culture. Rather than large group discussions, small group interviews with three to
five tourists were decided upon to prevent the discussion from getting too broad. Based
on the interpretive analysis of the interviews, the authors presented their four main find-
ings: food preferences and underlying motivation; a typology of the Chinese tourists

DWYER 9781781001288 Part II.indd 396 25/07/2012 11:07

 Cross-cultural approaches 397

tourism dining behavior; the influence of Chinese food culture; and, finally, intergroup
disparities. The authors proposed that Chinese food culture strongly influences tourism
dining behavior; while at the same time there are some disparities in terms of patterns
between the participants from Mainland China, Hong Kong, and Taiwan. Another
example is Maozs (2007) study, in which she examined Israeli backpackers motivations
and travel patterns, and compared them to backpackers from several other countries.
As a participant observer, the author, an avid backpacker, conducted interviews as well
as stayed in guesthouses, spent time with backpackers in their favorite restaurants, tea
houses, and gathering places. After collecting data on Israeli backpackers, and then
comparing them to other backpackers, the results indicated that Israeli backpackers
have a highly collective orientation, tend to be patriotic, and are inclined to disregard
local norms.
The third approach to cross-cultural qualitative tourism research investigates the cul-
tural characteristics and dynamics of a particular tourist destination and its community.
Hence, most of these studies focus on the host community rather than on the tourists
who visit the community. More specifically, these studies examine the interrelationship
between tourism phenomena and the community influenced by the phenomena. For
example, from the perspective of colonialism, Lujan (1993) explored Taos Pueblo, one
of the most traditional Indian Pueblos in New Mexico, to understand the impact of the
increasing number of tourists on the ethnic community where 2000 tribal members reside.
The sample consisted of lifetime residents of the pueblo who were enrolled as members
in the pueblo. Several questions were asked during interviews, including questions about
the tribal members general feelings about tourists, their experiences with tourists, social
changes in the pueblo due to tourism, and personal changes due to tourism. In addi-
tion, informal interactions and observations were included in the analysis. The results
showed that although there were some differences between age groups, tourism largely
led the Native Indians to become more aware of their own culture and to maintain their
traditional lifestyle due to economic benefits, their strong cultural and religious founda-
tion, and a strong sense of community and responsibility. Similarly, from a symbolic
interactionist perspective, Santos and Buzinde (2007) examined the Puerto Rican neigh-
borhood of Humboldt Park, Chicago. Their study investigated how tourism influenced
Humboldt Parks representation of its Puerto Rican cultural identity. Interviews were
conducted with individuals residing in the ethnic neighborhood. Their approach focused
on how participants defined their own situation to create the reality they perceived and
from which they acted. The authors found that there were three main ways in which
Puerto Rican residents represented their cultural identity for tourism purposes: invoking
historical details; calling on elaborate architectural features; and focusing on the neigh-
borhoods rich legacy, presence, and persistence in Humboldt Park.

ADVANTAGES AND LIMITATIONS OF THE CROSS-

CULTURAL APPROACH

The growing influence of globalizing tendencies and competing research cultures call
for the development of culturally sensitive research methodologies. And while, to date,
the large majority of cross-cultural studies employ quantitative research methodologies,

DWYER 9781781001288 Part II.indd 397 25/07/2012 11:07

 398 Handbook of research methods in tourism

increasingly cross-cultural qualitative studies can be found in tourism (e.g., Laing and
Crouch, 2005; Lujan, 1993; Uriely et al., 2009). Accompanying this growth in cross-
cultural qualitative research, there are some distinctive issues and practices associated
with cross-cultural interviewing that must be considered. To begin, there is the issue
of understanding that culture serves to influence interviewing techniques, strategies,
and question interpretation; and, therefore, must be taken into account in research
design. Indeed, to increase their value, interviewing practices such as providing partici-
pants the opportunity to choose their preferred languages for the interviews are often
implemented. Also, seeking to avoid interviewer bias and recognizing that the ethnic
background of the interviewer can at times affect the respondents answers particularly
when the topic and focus of the questions is culture (Weeks and Moore, 1981) cross-
cultural qualitative research often employs interviewers who vary in gender and ethnic
background. In addition, another common practice is the use of local interpreters.
However, while such practices are of great benefit to advancing cross-cultural tourism
research, there are certainly issues that arise; one them being the concern that interview-
ers, research participants, and interpreters often interpret questions and answers in very
different ways.
We should also note that while cross-cultural interviewing is often thought of as inter-
viewing that is conducted by a national or cultural outsider with a national or cultural
insider, such a narrow approach is proving increasingly problematic. Indeed, there are
countless fieldwork accounts that illustrate the insideroutsider challenges faced by
researchers conducting interviews within their own societies (Bhopal, 2000; Phellas,
2000; Warren, 1977). To be sure, the insideroutsider challenge is generic to all forms
of interviewing conducted under the auspices of cultural difference, whether ethnicity
or culture writ large mediates the relationship between interviewer and interviewee
(Ryen, 2003, p. 430). Nonetheless, there are those who propose that within the context
of cross-cultural differences, the insideroutsider challenge is likely to grow uneven given
the homogenizing influence of globalization forces. We must, however, consider the pos-
sibility that increasing transnational flows of culture seem to be producing, not global
homogenization, but the growing assertion of heterogeneity and local distinctiveness
(Sibley, 1995, pp. 1834).
Another distinctive issue associated with cross-cultural interviewing is the necessity of
establishing good field relationships. As Kvale suggests, for valid interview data collec-
tion, it is up to the interviewer to create in a short period of time a contact that allows
the interaction to get beyond merely a polite conversation or exchange of ideas (1996,
p. 125). And while establishing good field relationships is central in any study, the issue
is greatly exacerbated in cross-cultural interviewing given that rapport is necessary for
valid cultural understanding. Rapport, however, is the outcome of communication and
is not established once and forever. Rather, and especially in cross-cultural contexts, it is
mediated by the complex external and internal ingredients of day-to-day involvement
(Ryen, 2003, p. 431). Therefore, cross-cultural interviewers must continuously work
out their roles and relationships with the community they are studying. Moreover, this
process is further complicated by the researchers own search for insider status (Rutten,
1995; Tadria, 1991). However, as Panini notes, the search for insider status is like
chasing a mirage (1991, p. 8). As such, in cross-cultural interviewing, the interviewer
must come to terms with the fact that their place in society ... is negotiated from the

DWYER 9781781001288 Part II.indd 398 25/07/2012 11:07

 Cross-cultural approaches 399

existing cultural shock of knowledge and action available to define and cope with stran-
gers (Warren, 1988, p. 189).
Other distinctive issues associated with cross-cultural interviewing are the potential
problems caused by the assumption that in sharing a common language one also shares
an agreement and understanding of the nonverbal elements of communication. As von
Raffler-Engel (1988) suggests, one of the most significant challenges in cross-cultural
research is understanding that there are a number of communicative practices that
lie below consciousness and as such cannot easily be understood by simply sharing a
common language. Such sensitivity toward, and understanding of the dynamics and
complexities of cross-cultural communicative practices can significantly diminish the
usual tendency by cross-cultural researchers to focus on verbal communication while
disregarding the central role of nonverbal elements of communication (Kvale, 1996).
Moreover, those conducting cross-cultural interviews are best served by remembering
that even within cultures who share the same language there are still vast differences in
linguistic, interpretive and signalling behavior. Indeed, as Gumperz suggests, Only by
looking at the whole range of linguistic phenomena that enter into conversational man-
agement can we understand what goes on in an interaction (1982, p. 186).
Finally, it is imperative that cross-cultural qualitative researchers take into account
that life narratives themselves are rarely straightforward. Cross-cultural interviews can
certainly generate a significant amount of data. Such interview data is often in the form
of a gamut of life narratives with their very own beginning and ending points, and rarely
do fit together into a harmonious whole. One approach to unpacking such complex
narratives is to allow interviews in their earlier stages to be as unstructured as possible,
while progressively moving toward more structured lines of questioning that allow the
researcher to more purposely explore themes of interest. In addition, it is essential that
in the process of gathering qualitative data, cross-cultural researchers extend beyond the
very limitations of cross-cultural interviewing by incorporating into their research design
both oral and written materials, as different genres of communication allow for different
interpretive advantages.

FURTHER DEVELOPMENTS AND APPLICATIONS OF THE

CROSS-CULTURAL APPROACH IN TOURISM RESEARCH

A major challenge for contemporary cross-cultural tourism research will continue to be

the fact that the world is becoming more and more interconnected; at times blurring cul-
tural differences and threatening the existence of a distinctive national culture. As Kasfir
and Yai (2004) suggest, global diasporas have transformed contemporary society into
a global community where distinct national cultures continue to diminish while global
governance continues to increase. To address such shifts and challenge traditional,
universal, and deterministic approaches to culture, cross-cultural tourism researchers
have begun to engage with a variety of qualitative methodological approaches that
allow for the study of culture without disconnecting it from its context. One particular
approach we believe holds promise for tourism research is the multiple case-study project
approach with its respective joint analysis strategies, originally developed by Miles and
Huberman (1994). This detailed and systematic approach calls for a team of researchers,

DWYER 9781781001288 Part II.indd 399 25/07/2012 11:07

 400 Handbook of research methods in tourism

each one working in a different site. Projects begin with a set of shared research ques-
tions and data-coding schemes. What follows are frequent case analysis meetings where
the researchers come together to jointly share and discuss their ongoing experiences
and emerging themes. During these meetings possible interpretations, explanations and
hypotheses are recorded and debated by the team. Based on the discussion, the team then
decides on the next steps regarding data collection, follow-up questions, coding schemes,
and any other required actions. This process continues until the project is completed.
Thus, analysis feeds into further data collection, leading to further analysis (Troman
and Jeffrey, 2007, p. 513). However, we should note that given the considerable amount
of time required to have the type of debates and discussions essential for sorting out con-
flicting interpretations across different languages, the Miles and Huberman model can
also prove difficult to implement in cross-cultural projects (Somekh and Pearson, 2002).
Finally, contemporary cross-cultural tourism research must continue to advance
notions of culture as a dynamic, open, and shifting system. Indeed, given the increasing
recognition that the diversity of human experience in social groups and communities,
with languages and epistemologies, is undergoing profound cultural and political shifts
(Smith, 2008, p. 137), cross-cultural tourism research can no longer simply rely on uni-
versal value-based approaches to culture. Instead, cross-cultural tourism research has
to take into account the power dynamics between tourists and the toured, both within
contemporary society as a whole, as well as within the specific communities where tour-
ists and the toured engage as a result of the tourism phenomena. Although the focus
of cross-cultural qualitative tourism studies varies significantly, the majority employ
individual or group interviews as the primary data collection method so as to clarify dif-
ferential identities and dynamics between cultural groups. Indeed, data generated from
interviews has the ability to interrupt past assumptions regarding tourism phenomena,
as well as contribute to the development of culturally relevant theories of tourism phe-
nomena. Specifically, to date, the use of interviews in cross-cultural tourism research has
furthered our ability to investigate and convey cultural sensitivities as it relates to the
social and cultural contexts, as well as lived experiences, of both tourists and the toured.

REFERENCES

Bhawuk, D. and H. Triandis (1996), The role of culture theory in the study of culture and intercultural train-
ing, in D. Landis and R.W. Brislin (eds), Handbook of Intercultural Training, Thousand Oaks, CA: Sage,
pp. 1734.
Bhopal, K. (2000), Gender, race and power in the research process: South Asian women in East London,
in C. Truman, D.M. Mertens and B. Humphries (eds), Research and Inequality, London: UCL, pp. 6779.
Chan, A. (2009), Measuring cross-cultural values: a qualitative approach, International Review of Business
Research Papers, 5 (6), 322337.
Chang, R.C.Y., J. Kivela and A.H.N. Mak (2010), Food preferences of Chinese tourists, Annals of Tourism
Research, 37 (4), 9891011.
Chen, J.S. (2000), Cross-cultural differences in travel information acquisition among tourists from three
Pacific-rim countries, Journal of Hospitality Tourism Research, 24, 239251.
Crotts, J.C. and R. Erdmann (2000), Does national culture influence consumers evaluation of travel services?
A test of Hofstedes Model of cross-cultural differences, Managing Service Quality, 10 (6), 410419.
Crotts, J.C. and S.W. Litvin (2003), Cross-cultural research: are researchers better served by knowing
respondents country of birth, residence or citizenship?, Journal of Travel Research, 42 (2), 186190.
Davies, B., J. Larson, N. Contro, C. Reyes-Hailey, A.R. Ablin, C.A. Chesla, B. Sourkes and H. Cohen (2009),
Conducting a qualitative culture study of pediatric palliative care, Qualitative Health Research, 19 (1), 516.

DWYER 9781781001288 Part II.indd 400 25/07/2012 11:07

 Cross-cultural approaches 401

Denzin, N.K. and Y.S. Lincoln (2008), Introduction: the discipline and practice of qualitative research, in
N.K. Denzin and Y.S. Lincoln (eds), Strategies of Qualitative Inquiry, 3rd edition, Thousand Oaks, CA:
Sage Publications, pp. 144.
Dew, K. (2007), A health researchers guide to qualitative methodologies, Australian and New Zealand
Journal of Public Health, 31 (5), 433437.
Dimanche, F. (1994), Cross-cultural tourism marketing research: an assessment and recommendations for
future studies, in M. Uysal (ed.), Global Tourist Behavior, Binghamton, NY: Haworth Press.
Graburn, N. (1995), Tourism, modernity and nostalgia, in A.S. Ahmed and C. Shore (eds), The Future of
Anthropology: Its Relevance to the Contemporary World, London: Athlone.
Gumperz, J.J. (1982), Discourse Strategies, Cambridge: Cambridge University Press.
Gupta, A. and J. Ferguson (1997), Beyond culture: space, identity, and the politics of difference, in A.
Gupta and J. Ferguson (eds), Culture, Power, Place, Durham, NC: Duke University Press, pp. 3351.
Gursoy, D. and J.S. Chen (2000), Competitive analysis of cross-cultural information search behavior,
Tourism Management, 21 (6), 583590.
Hall, B.L. and J.C. Kulig (2004), Kanadier Mennonites: a case study examining research challenges among
religious groups, Qualitative Health Research, 14 (3), 359368.
Heijes, C. (2011), Cross-cultural perception and power dynamics across changing organizational and national
contexts: Curaao and the Netherlands, Human Relations, 64 (5), 653674.
Hofstede, G. (2001), Cultural Consequences, 2nd edition, Thousand Oaks, CA: Sage.
Hsieh, A. and C. Tsai (2009), Does national culture really matter? Hotel service perceptions by Taiwan and
American tourists, International Journal of Culture, Tourism and Hospitality Research, 3 (1), 54.
Jackson, M.S. and D.M. Niblo (2003), The role of qualitative methodology in cross-cultural research,
Qualitative Research Journal, 3 (1), 1827.
Kasfir, S.L. and O.B.J. Yai (2004), Authenticity and diaspora, Museum International, 56 (12), 190197.
Kim, S.S. and B. Prideaux (2005), Marketing implications arising from a comparative study of inter-
national pleasure tourist motivations and other travel-related characteristics of visitors to Korea, Tourism
Management, 26 (3), 347357.
Kim, S.S., B. Prideaux and S.H. Kim (2002), A cross-cultural study on casino guests as perceived by casino
employees, Tourism Management, 23 (5), 511520.
Kvale, S. (1996), Interviews: An Introduction to Qualitative Research Interviewing, Thousand Oaks, CA: Sage.
Laing, J.H. and G.I. Crouch (2005), Extraordinary journeys: an exploratory cross-cultural study of tourists on
the frontier, Journal of Vacation Marketing, 11 (3), 209223.
Lee, G. and C. Lee (2009), Cross-cultural comparison of the image of Guam perceived by Korean and
Japanese leisure travelers: importanceperformance analysis, Tourism Management, 30(6), 922931.
Lee, S. and B. Sparks (2007), Cultural influences on travel lifestyle: a comparison of Korean Australians and
Koreans in Korea, Tourism Management, 28 (2), 505518.
Liamputtong, P. (2008), Doing research in a cross-cultural context: methodological and ethical challenges, in
P. Liamputtong (ed.), Doing Cross-Cultural Research: Ethical and Methodological Perspectives, Dordrecht,
the Netherlands: Springer, pp. 320.
Liamputtong, P. (2009), Qualitative Research Methods, 3rd edition, Melbourne, Australia: Oxford University
Press.
Lujan, C.C. (1993), A sociological view of tourism in an American Indian community: maintaining cultural
integrity at Taos Pueblo, American Indian Culture and Research Journal, 17(3), 101120.
MacKay, K.J. and D.R. Fesenmaier (2000), An exploration of cross-cultural destination image assessment,
Journal of Travel Research, 38 (4), 417423.
Maoz, D. (2007), Backpackers motivations: roles of culture and nationality, Annals of Tourism Research, 34
(1), 122140.
Miles, M.B. and A.M. Huberman (1994), Qualitative Data Analysis, 2nd edition, Thousand Oaks, CA: Sage.
Morris, E.W. (2007), Researching race: identifying a social construction through qualitative methods and an
interactionist perspective, Symbolic Interaction, 30 (3), 409425.
Panini, M.N. (1991), Introduction: reflections in feminism and fieldwork, in M.N. Panini (ed.), From
the Female Eye: Accounts of Women Fieldworkers Studying Their Own Communities, Delhi: Hindustan
Publishing, pp. 110.
Papadopoulos, I. and S. Lees (2002), Developing culturally competent researchers, Journal of Advanced
Nursing, 37 (3), 258264.
Peabody, D. (1985), National Characteristics, Cambridge: Cambridge University Press.
Phellas, C.N. (2000), Cultural and sexual identities in in-depth interviewing, in C. Truman and E.M. Bruner
(eds), Research and Inequality, Urbana, IL: University of Illinois Press, pp. 5264.
Pizam, A. and S. Sussman (1995), Does nationality affect tourist behaviour?, Annals of Tourism Research, 22
(2), 901917.
Rabinow, P. (1977), Reflections on Fieldwork in Morocco, Berkeley, CA: University of California Press.

DWYER 9781781001288 Part II.indd 401 25/07/2012 11:07

 402 Handbook of research methods in tourism

Reisinger, Y. and L. Turner (2003), Cross-Cultural Behaviour in Tourism: Concepts and Analysis, Oxford:
Butterworth Heinemann.
Richardson, S.L. and J. Crompton (1988), Vacation patterns of French and English Canadians, Annals of
Tourism Research, 15 (3), 430450.
Rutten, M. (1995), Farms and Factories: Social Profile of Large Farmers and Rural Industrialists in West India,
Delhi: Oxford University Press.
Ryen, A. (2003), Cross-cultural interviewing, in J.A. Holstein and J.F. Gubrium (eds), Inside Interviewing:
New Lenses, New Concerns, Thousand Oaks, CA: Sage Publications, pp. 429448.
Santos, C.A. and C. Buzinde (2007), Politics of identity and space: representational dynamics, Journal of
Travel Research, 45, 322332.
Sibley, D. (1995), Geographies of Exclusion: Society and Difference in the West, London: Routledge.
Smith, L.T. (2008), On tricky ground: researching the native in the age of uncertainty, in N.K. Denzin
and Y.S. Lincoln (eds), The Landscape Of Qualitative Research, 3rd edition, Thousand Oaks, CA: Sage
Publications, pp. 113143.
Somekh, B. and M. Pearson (2002), Intercultural learning arising from Pan-European collaboration: a com-
munity practice with a hole in the middle, British Educational Research Journal, 28 (4), 483484.
Stanfield, J. (1993), Epistemological considerations, in J. Stanfield (ed.), Race and Ethnicity in Research
Methods, London: Sage, pp. 1636.
Tadria, H.M. (1991), Challenges of participation and observation: fieldwork experience among some peasants
in Uganda, in M.N. Panini (ed.), From the Female Eye: Accounts of Women Fieldworkers Studying Their
Own Communities, Delhi: Hindustan Publishing, pp. 8898.
Tillman, L.C. (2006), Researching and writing from an African-American perspective: reflective notes on three
research studies, International Journal of Qualitative Studies in Education, 19(3), 265287.
Troman, G. and B. Jeffrey (2007), Qualitative data analysis in cross-cultural projects, Comparative Education,
43 (4), 511525.
Tsang, N.K. and J. Ap (2007), Tourists perceptions of relational quality service attributes: a cross-cultural
study, Journal of Travel Research, 45 (3), 355363.
Uriely, N., D. Maoz and A. Reichel (2009), Israeli guests and Egyptian hosts in Sinai: a bubble of serenity,
Journal of Travel Research, 47 (4), 508522.
van de Vijver, F.J.R. and K. Leung (1997), Methods and Data Analysis for Cross-Cultural Research, Newbury
Park, CA: Sage.
von Raffler-Engel, W. (1988), The impact of covert factors in cross-cultural communication, in F. Poyatos
(ed.), Cross-cultural Perspectives in Nonverbal Communication, Toronto: C.J. Hogrefe, pp. 71104.
Warren, C.A. (1977), Fieldwork in the gay world: issues and phenomenological research, Journal of Social
Issues, 33 (4), 93107.
Warren, C.A. (1988), Gender Issues in Field Research, Newbury Park, CA: Sage.
Weeks, M.F. and R.P. Moore (1981), Ethnicity-of-interviewer effects on ethnic respondents, The Public
Opinion Quarterly, 45 (2), 245249.

DWYER 9781781001288 Part II.indd 402 25/07/2012 11:07

 21 Archival research
Dallen J. Timothy

INTRODUCTION

Archives are one of the oldest and most utilized sources of data known to research-
ers. For centuries scholars have used written texts and visual images to piece together
histories of salient events in a nations development, lives of prominent citizens, or
occurrences of interest to the general population. Today, archived data are best known
for their value in historical research, but archives-based studies are becoming more
mainstream in many academic fields, including tourism, as postpositivist methods have
gained more traction in the realm of empirical research. In particular, archival data help
develop understandings of how tourist destinations grow and decline, the long-term
impacts of tourism in destinations, how certain populations have dealt with travel in
the distant and recent past, and how policies and legislative actions affect the industrys
growth throughout the world.
Archival research itself is not an analytical method but rather a set of approaches
to understanding physical data and their meanings. Archives are data sources, and the
methods that can be used to interpret archival data are manifold. Content analysis,
regression analysis, factor analysis, semiotics, and historical documentation are a few of
the most prevalent analytical methods for studying archived information. Since several
of these are discussed elsewhere in the book, this chapter does not delve too deeply into
any one analytical or interpretive method. It does, however, describe the nature of archi-
val research and its various forms, and it provides a background for the technique and
its application in tourism. It highlights several studies that have utilized archival sources
and outlines the main advantages and limitations associated with undertaking archives-
based scholarship.

USE OF ARCHIVAL DATA AND RESEARCH

In very general terms, archival research entails the use of data which the investigators
did not collect themselves; the data already existed when the study was started (Jackson,
2009). The researchers role is to select which information to use and how to analyse it
(McBurney and White, 2010). In the words of Dane (2011, p. 274), archival researchers
deal with peoples products rather than with the people themselves. It is an unobtrusive
research endeavor and involves the use of physical evidence (Babbie, 2010) wherein the
unit of analysis is any public record or artifact.
There are two general types of archive data sources. The first is numerical data sets
that were collected by someone else. These are usually created to fulfill the needs of a
public or private agency, such as the Census Bureau or the Ministry of Tourism. For the
agency that collected it, the primary purpose of this information is to gauge trends and

403

DWYER 9781781001288 Part II.indd 403 25/07/2012 11:07

 404 Handbook of research methods in tourism

patterns in certain populations. In the tourism context these data sets are compiled to
help understand global or regional travel trends, or demand for destinations or a par-
ticular tourism product.
The second type of archival data derives from a variety of records, images, and artifacts
that have remained in collections from bygone eras or from more recent times. These
are records of the activities of individuals, organizations, or groups, which can be used
to study people and events by examining their recorded history (Hill, 1993). Unlike the
numerical data sets mentioned earlier, this second type of archived data is less interested
in causal relationships and more interested in exploration, description, and explanation.
Like numerical data sets, these artifacts are also produced and held in reserve by someone
other than the researcher. A few key examples of these sources include newspapers, web-
sites, historic and contemporary maps, cultural artifacts and souvenirs, photographs and
postcards, gravestones, brochures and books, posters, personal journals, ship manifests,
oral histories, transcripts, planning documents, recorded music and speeches, tax and
land records, letters, guest registries, poetry, birth certificates, marriage licenses, legal
documents, and death records (Danto, 2008; Galgano et al., 2008; Williams, 2007). Of all
research approaches utilized by tourism scholars, this is one of the most creative and flex-
ible, as almost anything remaining from the past can be mined as a data source, depending
upon the research questions being investigated and the availability of the source.
Archival sources can be direct or indirect (Howell and Prevenier, 2001). Direct sources
are usable in their original form or derived directly from their creators. Letters written
by an immigrant, poems penned by an aspiring poet, or photographs taken during a
late nineteenth-century vacation are examples of direct sources. Indirect sources include
things such as inventories of someones collection, which then can help identify the per-
sons interests and help understand his or her history.
Archival data have been used for centuries. Within the first three centuries of the
Christian era, ecclesiastical leaders and historians compiled the New Testament from
various gospel sources and letters from the apostles (Riches, 2000). This is probably
one of the more famous uses of archived data. As well, there are many accounts of par-
liaments and congresses around the world basing laws and constitutions on historical
records. The United States, the United Kingdom, Canada, Australia, and New Zealand
have based many elements of their respective constitutions and laws on one of the most
famous historic documents in the world: the Magna Carta, which limited royal powers
and promoted human rights. Even during the medieval period, scholars were already
using books, letters, and other records to compile histories and document colonial suc-
cesses (de Hamel, 1986).
Historians and geographers were heavily engaged in archival research in the nine-
teenth century as they compiled histories and historical geographies of places that were
still under colonial rule and in their initial stages of industrialization (Milne, 1898;
Ratzel, 1896). Today, nearly all historians and many social scientists from a wide range
of disciplines engage in archival research, including that associated with tourism. It is
seen as a valuable approach to understanding historical problems, telling stories, unfold-
ing important events, understanding peoples lives and travel experiences, explaining
tourisms impacts, and identifying past and present patterns of supply and demand.
Although archives are essentially collections of documents held by a person or
organization, the term also refers to the place where they are kept. There are thousands

DWYER 9781781001288 Part II.indd 404 25/07/2012 11:07

 Archival research 405

of archives throughout the world under the ownership or stewardship of public agencies,
private companies and organizations, churches and other religious societies, cultural
clubs, libraries, non-profit organizations, and universities.
Until quite recently, archival work was done by hand, searching through card cata-
logs, folders, and boxes of stored items to find facts that helped tell stories or to identify
key words and symbols. Later, however, a significant breakthrough occurred as many
organizations began photographing and recording their collections on microfilm or
microfiche, which people could access by visiting the archives and utilizing microform
readers. The research process evolved even further when many organizations began
lending, or selling copies of, their microforms to other institutions and even to private
individuals. The use of microforms to store visual and textual data was an important
breakthrough, and in many cases researchers no longer had to travel away from their
home base to glean data from historical documents and artifacts.
The process has been improved even more in recent years with the digitization of
archival records and data sets that can be accessed from anywhere in the world through
the internet. Organizations have scanned millions of old birth certificates, census records,
newspapers, and the like to be digitized for mass consumption. Today, direct and indi-
rect secondary sources no longer have to be old and dusty but can be utilized by a wider
range of users. This digital revolution has brought about new ways for people to relate to
history and place (Meethan, 2008, p. 101), thus affecting both academic researchers and
the general public. This democratization of public and private records and the concomi-
tant increase in information accessibility has spread the use of archival research beyond
the domain of historians and historical geographers to include many other disciplines,
such as mainstream tourism studies.

ARCHIVAL METHODS

As noted already, archival research is not a method. Archives are sources of data that
can be assessed using a variety of analytical and interpretive methods, which can be clas-
sified into several broad types. Quantitative analysis of existing data, content analysis,
semiotic analysis, and historical fact finding are a few examples.

Existing Data Sets

Existing information, such as border-crossing data provided by a government agency,

can be analysed in many ways. Descriptive statistics, such as measures of central ten-
dency or frequencies, are often used to understand basic distribution patterns. Different
variables within existing data sets can also be analysed with bivariate or multivariate
tests to understand causal relationships between selected variables, as long as adequate
information was collected to represent the desired variables.

Content Analysis

Content analysis is a well-established research method that helps make inferences about
theoretically pertinent problems (Dane, 2011; Neuendorf, 2002). It is used in studying

DWYER 9781781001288 Part II.indd 405 25/07/2012 11:07

 406 Handbook of research methods in tourism

archived materials to record the occurrence and frequency of certain words, images,
names, events, or ideas. Broadly speaking, all forms of archival research involve content
analysis, because signs, symbols, patterns, and meanings are sought from archived mate-
rials regardless of the methodological approach utilized. Content analysis is an obser-
vational technique used to analyse symbolic or actual content of all forms of recorded
material (Hall and Valentin, 2005). This method in tourism studies has tended to con-
centrate on visual images and advertising, such as brochures and websites, although
more scholars today are beginning to examine recurring concepts in novels, guidebooks,
movies, and other textual sources.
Content analysis can be done quantitatively or qualitatively. From a quantitative per-
spective, frequency counts can provide simple descriptive statistics, which help research-
ers understand the occurrences and distributions of key words and concepts. Bivariate
and multivariate tests may also be used in analysing the frequency of variables in texts,
documents, and visual images, and checking their relationships with other variables.
There are several qualitative methods used to interpret content data. Template analysis
is often used to interpret archived data. This technique requires the researcher to develop
a coding template, which encapsulates various themes identified by the researcher as
important for the research questions, and allows these themes to be organized in a
meaningful way (King, 1998). Broad themes envelop narrower and more specific themes,
which are often identified (e.g., based upon a literature review) ahead of time as being
relevant to the study. When data bits appear in the archived texts or images, they are
coded according to the most relevant a priori theme. To check for reliability, this process
is typically undertaken by at least two independent assessors to assure consistency in
the coding and theming. Once all transcripts have been coded the template of themes
becomes the foundation for interpreting the data and summarizing the findings (King,
1998).

Semiotic Analysis

Qualitative content analysis may also include semiotics, discourse analysis, or textual
analysis. Semiotic analysis refers to the study of signs, and in the context of tourism it has
focused overwhelmingly on understanding the symbolisms and communicative elements
of visual images. Semiotics researchers are particularly interested in the meanings of
signs and symbols, their real and symbolic social contexts and their effects on the people
who produce and/or consume them. The meanings of signs and how these meanings are
transmitted are key in understanding archived objects such as photographs or souvenirs.
Researchers must decode the messages of signs found in physical objects and texts to
identify meanings that will help make sense of a medium and reveal the significance of
its production and social relations (Chandler, 2007; Danesi, 2007; Metro-Roland, 2009).
Physical objects may be read as texts just as written documents are considered texts
(Geertz, 1973).
Overlapping concepts include discourse and textual analysis, which focus on under-
standing written or spoken language and have been useful in understanding archived
written texts, such as speeches, planning documents, books, diaries, and newspapers.
Like other forms of semiotic analysis, discourse analysis is interested in the underlying
meanings of texts and their contexts. This approach views language and text as very

DWYER 9781781001288 Part II.indd 406 25/07/2012 11:07

 Archival research 407

important and emphasizes the social frameworks within which various discourses are
embedded (Gee, 2005). Textual analysis describes how the existence or absence of infor-
mation creates meaning (Buzinde, 2010).

Fact Finding/Historical Approach

Historical fact finding is another approach to archives-based research. For centuries

historians have documented the lives of individuals, important events, and the spatial
and temporal development of places via historical documentation. Historians construct
meanings and contextualize information from archival sources. They uncover facts to
help tell a story or reveal the bigger picture of an event or place. This kind of work can
be highly subjective, as each historian creates his or her own depiction of the past based
upon the bits and pieces of history they are able to verify (Howell and Prevenier, 2001).
History researchers identify a historical problem or need. They then gather as much
pertinent information as possible about the issue, organize the evidence and verify its
authenticity inasmuch as possible. Once the information is organized, conclusions are
drawn and written into an expressive narrative (Busha and Harter, 1980).

THE MESSAGE IN THE MEDIUM

As already noted, there are multiple data sources and units of analysis involved in archi-
val research. According to Dane (2011, p. 274), any record of information is fair game
for archival research. This section examines several of these resources and provides
examples of their use in the context of tourism.

Secondary Data Sets

Analysis of existing data sets usually means using the data for a different purpose than
that for which they were collected (Dane, 2011). Among the most commonly used sec-
ondary data sets in tourism research are visitor satisfaction and perception data, arrival
numbers, tourist demographics, transport-related data, and other such information that
is deemed important and useful by various public and private agencies. As already noted,
any number of analytical approaches can be used on secondary data to test hypotheses
or reveal descriptive and causal relationships between variables.
Based on a secondary data set from Canada, Murphy et al. (2000) assessed the effects
of visitors experiences and trip quality on trip value and intension to return. The data
were collected by Tourism Victoria, the destination tourism agency of Victoria, British
Columbia, based on a convenience sample of 3088 tourists as they were leaving the city.
For Tourism Victoria, these data were collected to provide an understanding of tourists
perceptions of the citys environment, service quality, and value. Just over 600 of these
surveys were used by Murphy and his colleagues in the study. They employed a Partial
Least Squares regression analysis and found that destination quality and infrastructure
were important factors in determining trip value. They also found that destination
quality and experience quality were important predictors of ones intention to revisit
Victoria, although trip value was not.

DWYER 9781781001288 Part II.indd 407 25/07/2012 11:07

 408 Handbook of research methods in tourism

Newspapers

Newspapers and other popular print material (e.g., magazines) are rich sources of data.
They reveal much about public thinking, or at least the medias views of public thinking,
regarding a variety of topics. There has been a long and rich history of scholars analysing
newspaper content to understand the publics view of tourism-related issues (Frew, 2009;
West, 2007; Wise, 2011).
Morgan et al. (2011) analysed the effects of newspaper coverage of red tide (algae
bloom) at a beach and park on Floridas coast. They searched for the key words red tide
in the daily newspapers (online versions) of Fort Myers and Naples, Florida, between
the years 2002 and 2004, and compared these findings to secondary data related to
visitor use of Lovers Key State Park. Because of its usefulness in interpreting time series
data, a multiple regression analysis was employed by Morgan and her colleagues. They
concluded that there was a significant negative impact on park attendance when red tide
was mentioned in the local newspapers in the range of between 398 and 416 visitors, and
financial losses for the county of some 4.5 million dollars.
From a qualitative perspective, Andriotis (2010) used a template analysis to appraise
public debates about the behavior of British travelers in the Greek Islands. His use of
a template resulted in four categories of bad behavior and community perceptions of it
as noted in newspapers. The categories included binge drinking, sexual behavior, risk
taking, and destination community reactions. After identifying these types of behaviors
and reactions, he highlighted examples and manifestations of each type based on quotes
and descriptions from the newspapers.

Journals, Letters and Other Written Sources

Old, hand-written letters, journals, and other primary sources have long been a favored
information well for historians, genealogists, and social scientists. These documents
provide a glimpse of what life was like in the past and how people experienced life events
and the world around them (Richards, 2008). Life histories and event histories have long
relied on these important resources.
One study in Ghana revealed some very interesting patterns associated with the experi-
ences of African-Americans at a slave heritage site (Timothy and Teye, 2004). The data
source for the study was the guest registry at Elmina Castle, one of the most notorious
slave trading forts in West Africa built in the fifteenth century by the Portuguese. At the
conclusion of their visits, tourists are invited to record their thoughts and feelings in a
guest book. The books contents were provided to the studys authors for the decade
of the 1990s and included some 14 120 entries, which were sorted and analysed qualita-
tively into nationalities. American visitors were further divided into whites and African-
Americans, and a textual analysis was undertaken to assess the slave site experience of
each group of US tourists. Textual and template analysis resulted in seven themes, which
epitomized the experiences of African-Americans: grief and pain, good versus evil (black
versus white), revenge, forgiveness and healing, coming home, in memory of our ances-
tors, and God and holy places.
Historical analysis of travelers during various periods of history is also popular in
tourism studies. In particular, with the rapid growth in feminist scholarship since the

DWYER 9781781001288 Part II.indd 408 25/07/2012 11:07

 Archival research 409

1980s, more attention has been paid to women travelers and their struggles to go beyond
socially dictated gender roles, to be taken seriously as recognized explorers and adven-
turers. Feminist histories of travel are prominent illustrations of how historical research-
ers have utilized the travel journals and diaries of allocentric women exploring parts of
the world that had historically been the domain of men (Kolocotroni and Mitsi, 2008;
Nittel, 2001; Squire, 1995; Towner, 1988, 1995).
Information in travel diaries and letters has also been widely used to illustrate the
history of the Grand Tour, one of the most salient eras in tourism history one that has
received a great deal of academic attention (Black, 1992; Brodsky-Porges, 1981; Towner,
1984, 1985, 1988, 1996) and is often highlighted as a precursor of todays phenomena of
cultural tourism, educational travel, independent travel, and backpacker tourism.

Books and Guidebooks

There has been growing interest in recent years about how places and travel are depicted
in various types of best-selling novels. Through content and semiotic analyses, research-
ers have attempted to understand how tourist destinations are portrayed in popular
books (Beaulieu and Lvy, 2003; Cate-Arries, 2000). Likewise, we have seen a growth in
the number of studies that provide discourse and other semiotic analyses of the content
of specialized tourist guide books (Laderman, 2002; Lew, 1992).
Using a semiotic textual analysis, Buzinde (2010, p. 224) analysed the discursive
construction of slavery at historic plantations in the American south as depicted in a
Lonely Planet guide book. She first prepared the raw data in a way that would more
easily accommodate cross comparisons. Second, she conducted an in-depth review of
the text to understand the narrative for the period in question. The third step entailed a
closer reading of the text and extrapolation of recurring themes and language that spoke
about plantation life. Finally, Buzinde interpreted the emergent themes within the larger
context of the study. The process resulted in the author identifying and supporting with
quotes and examples three main themes: deflection from slavery, presencing slavery, and
critical reflections on slavery. The findings suggest that the United States still bears the
burden of racial inequalities in its tourism narratives and that more dialogue is needed to
reach more balanced race relations in a country that prides itself on being a post-racial
society.

Photographic Images

Tourism and photographs go hand-in-hand. Tourists take photographs, and photos

have long been used by governments and the private sector to promote places and
businesses (Garrod, 2009). Postcards, brochures, and websites are all popular users of
place-based pictures that promote destinations and create images abroad. Photographs,
postcards, and tourist brochures stimulate strong interest among tourism scholars,
primarily because of the messages and image representations they portray (Burns and
Lester, 2005; Cohen, 2001; Echtner, 2002; Jokela and Raento, 2011). Semiotic analyses
and other forms of content analysis have been key in past research that has analysed
tourism visual images. Feighery (2009) argues that tourism organization stock photos,
which are utilized in promotional materials, can be helpful in understanding how places,

DWYER 9781781001288 Part II.indd 409 25/07/2012 11:07

 410 Handbook of research methods in tourism

bodies, and spaces are scripted through tourism. In this regard there are clear political
implications in terms of image creation and power relations between a places heritage
identity and what tourism and public officials want to portray to visitors. Similarly,
semiotic analyses often highlight political undertones as readings of postcards and other
photographic images help create imaginative geographies and contrived, idealized
tourist spaces (Larsen, 2006; Timothy, 2011).
Vespestads (2010) content analysis of brochures used to market Norway abroad,
found that tangible assets and snapshots of nature tend to dominate most of the adver-
tizing media Norway produces. Despite a rich built and living cultural heritage, this
illustrates how the country sells itself to British and Russian tourists as a primarily
nature-based destination. Istanbul, Turkey, highlights cultural heritage in its brochures.
One study of Istanbul brochures found that local tourism officials and the Turkish
Ministry of Tourism prefer to depict the city as a historically dynamic place where
visitors can travel through time via depictions of many of the civilizations that have
flourished there through the centuries (zdemir, 2010).
Since the 1800s, postcards have been an important part of the tourist experience.
Postcard images traditionally have represented what destinations and postcard produc-
ers have wanted the outside world to accept as being part of the destination (Arreola,
2001). In an interesting study of Christmas postcards, Cohen (2007) argued that they
have the potential to influence travelers decision making. He found that Thai Christmas
postcards depict Santa Claus enjoying the tropics and concluded that they contribute to
tourism promotion by encouraging northerners to get out of the cold and spend their
winter holidays in balmy Thailand.
Stokowski (2011) used both semiotic analysis and quantitative content analysis to
assess photographs published in local newspapers in two small Colorado communities
over a 22-year period while gambling tourism was the towns development focus. Her
aim was to understand the number of photos published on the front page of the news-
paper, the number of pictures that included scenes with people, and the number of pic-
tures that depicted people smiling in relation to key moments in gaming development.
Her content analysis entailed tabulating raw numbers and percentages of each type of
picture, compared across years. She found that the number of photos published on the
papers front pages declined over the years and that the proportion of photographs with
people in them went up and down in relation to the establishment and growth of gam-
bling tourism. Similarly, the depiction of people smiling in the photos declined during
the 22-year period. Stokowskis semiotic analysis suggests that as gaming tourism grew,
the communities became more serious as discontent and antagonism set in. In the initial
stages of tourism development, smiles were more prominent on the faces of community
members and decreased as gaming tourism matured.

Souvenirs

Like the photographs noted above, souvenirs are an important part of material culture
that reveal considerable meaning about places and can influence visitors experiences
(Collins-Kreiner and Zins, 2011; Timothy, 2005). In this sense, souvenirs and handicrafts
can also be viewed as an archival medium. Souvenirs research has typically involved
tabulating the range of gift items for sale in a given destination, shop, or museum. Much

DWYER 9781781001288 Part II.indd 410 25/07/2012 11:07

 Archival research 411

of the research associated with souvenirs and handicrafts has centered on the materials,
styles, designs, places, and crafters and how these reinforce or undermine the authentic-
ity of cultural artifacts for sale to tourists (Asplet and Cooper, 2000; Littrell et al., 1993).
Most of this line of inquiry, however, derives from tourists perceptions of authenticity
rather than the researchers own content analysis of the objects.
In terms of the cultural artifacts themselves, content analyses can be done to examine
the frequency of certain characteristics of a single type of souvenir or across a range of
souvenirs available in a destination. Semiotic approaches have been utilized in the past,
although not enough, in souvenir studies. Gordons (1986) analysis resulted in a typology
of souvenirs based on their forms, functions, and meanings for tourists: pictorial images,
piece-of-the-rock items, symbolic shorthand, markers, and local products. Shenhav-
Keller (1993) undertook a semiotic textual analysis of archived Israeli souvenirs at a
craft and souvenir market and concluded that the countrys attitude toward its recent
and ancient heritage, views of religion and culture, and relations with the Palestinians
were all expressed in the souvenirs on display. The analysis suggests that souvenirs in
Israel emphasize the boundary between Jews and non-Jews. They also contribute to the
anonymity of Palestinian crafters by utilizing Arab-made souvenirs as representative of
the state of Israel and in so doing ignore the cultural identity of the original Palestinian
crafters.

Electronic Archives and Modern Technology

Since the 1990s there has been growing interest in the use of web-based media for archi-
val research. The internet itself is likely the worlds largest archive, as so much textual
and visual data is stored within it. Online archives nowadays allow for easier content
analysis because they enable scholars to search for key words in titles or textual bodies,
as well as sources and date ranges (Hall and Valentin, 2005). Many traditional archives
have been scanned and logged into digital formats that can be accessed via the internet.
Likewise, new web-based media themselves may also be seen as archives that provide
abundant data for qualitative or quantitative inquiry. Website images, tourism mar-
keting web pages, social media, online photo banks, and webcams are all examples of
internet archives that are now receiving widespread research coverage (Alderman and
Modlin, 2008; Choi et al., 2007; Tang et al., 2011; Timothy and Groves, 2001).
Social media, such as Facebook, Twitter, and blogs have become a significant outlet
for people logging their travel behaviors and experiences. Some tourism service provid-
ers have also jumped on the bandwagon and are using these as inexpensive marketing
tools. These social media, together with web-based services such as Google Earth, Flickr,
WikiMapia, OpenStreetMap, and Mapapedia have created a new type of electronic
archive that allows members of the general public to create and share their own maps,
photographs, stories, and daily travel experiences. These new abilities and archival
spaces have been dubbed variously neogeography (Turner, 2006) or VGI (volunteered
geographic information) (Goodchild, 2007). These virtual venues now allow tourists to
make available an astonishing amount of geographic and destination attribute data to
other potential tourists (Sui, 2008).
There is an emerging literature that uses various textual analyses, content analysis,
and semiotic assessments to understand aspects of travel blogs and other social media

DWYER 9781781001288 Part II.indd 411 25/07/2012 11:07

 412 Handbook of research methods in tourism

networks (Enoch and Grossman, 2010; Leung et al., 2011; Pudliner, 2007). Paris (2011)
followed eight backpacker tourist bloggers and used a content analysis to examine their
online behavior. His aim was to understand the role of social media in the travel experi-
ence; the content analysis enabled him to formulate social media maps that defined travel
bloggers virtual spaces and meanings. Using a template approach, Timothy and Groves
(2001) analysed the image content of 300 webcams and illustrated the value of webcam
pictures for tourism research from the perspective of context/setting (site-specific attrac-
tions, natural scenery, resorts, cruise ships and ports, shopping centers, infrastructure,
border crossings, tourist services, and special events), data types (weather, crowd density,
changes in facilities, changes in infrastructure, and car number plates) and influential
factors that affect the researchers ability to glean reliable visual data (update frequency,
image resolution, range/distance of coverage, and quality of coverage/obscured images).

ADVANTAGES AND DISADVANTAGES OF ARCHIVAL

RESEARCH

As with any research approach or data-collection method, the use of archives has both
benefits and drawbacks. One of the most commonly cited advantages is that researchers
do not have to interact with study participants (Jackson, 2009). This, in effect, diminishes
the reactivity between the researcher and the researched. Thus it is unobtrusive and non-
reactive (Keppel et al., 1992), so that answers are recorded and not influenced by what
the subject feels the researcher might want to hear.
A second advantage is that archives-based investigations are less time consuming and
usually less expensive. When considering secondary data sets, the data are already col-
lected, saving time, effort, and cost (Jackson, 2009; Keppel et al., 1992). In most cases, to
address research questions archived documents can be used without large budgets, thus
allowing researchers to identify trends and patterns that exist in physical evidence or
historical documents without having to collect primary data.
Another benefit is that secondary data sets are sometimes based on entire populations
rather than just samples at a single point in time (Keppel et al., 1992). Many research-
ers see this as being more reliable, or at least more desirable in undertaking empirical
research. Total visitor arrivals at a museum or across an international border are good
tourism examples that allow researchers to work with aggregate data that reflect a fairly
accurate total accounting of the test population.
Despite the multitude of advantages of archival research there are also some notable
drawbacks. One of the most glaring limitations of archival data is the source, and in the
case of data sets, the methods with which the data were collected. Most archived data
sets were collected for the use of whatever agency commissioned their collection. Thus,
the data were typically acquired for non-scientific purposes and might not have been
done as randomly or systematically as a researcher would normally desire (McBurney
and White, 2010). Secondary data sets are also less flexible. The variety of questions a
researcher wants to ask might not have been included. Studies are therefore constrained
by the questions that were asked in the original data collection exercise. It is sometimes
difficult to derive conceptually rich information from data that are only produced for
marketing or general informational purposes. Data in a study commissioned by the US

DWYER 9781781001288 Part II.indd 412 25/07/2012 11:07

 Archival research 413

National Park Service, for instance, which aims to gauge visitors use of a heritage trail
would be difficult to use for understanding attachment to place concepts if related ques-
tions were not included on the original questionnaire. Because all archive-based research
derives conclusions based upon information collected by another person or agency,
there is no control in terms of who is studied or how they were studied (Jackson, 2009,
p.86).
Another problem is that many private archives are closed to scholars and can only be
accessed by permission from the owners, some of whom are unwilling to provide entry
except in exceptional circumstances. The vast archives in the Vatican City are home
to large collections of pilgrimage-related accounts, travel diaries, and ancient books
that would be of interest to many tourism historians, but the microstate is reticent to
allow researchers into its documents rooms. Although the vaults are opening up to
more researchers with special permission, access to the Vatican archives is still heavily
restricted (Boyle, 2001; Coombs, 1989).
Another glaring drawback is subjectivity. Information published in some newspapers
and on the internet is filtered and biased by reporters, editors, and web designers, who do
not always write occurrences verbatim or in their precise chronological order (Keppel et
al., 1992). Thus, depictions of past events, places, and people are often filtered through
selective memory mechanisms in favor of more positive twists or in ways that make the
event or person more heroic than they really were (Timothy, 2011). In archive-speak
this is known as selective deposit. Likewise, selective survival refers to the limited
number of archived items available for analysis (Keppel et al., 1992), which can result in
incomplete pictures of the past being revealed in historical narrative and other analyses.

CONCLUSION

Archives have been used as sources of data for centuries and will likely continue to be an
important element of social science and historical research for centuries to come. Their
primary value is the provision of information that does not have to be collected first-
hand by the researcher; the information already exists in numerical, textual, visual, or
physical form. In them are volumes of data about peoples lived experiences and travel
accounts, the successes and failures of service providers, images of place, and destina-
tions touristification processes, to name but a few.
These extant data can be analysed quantitatively via statistical tests and content
analysis, or qualitatively through various semiotic interpretations and deconstruction-
ist frameworks or historical storytelling. There is a growing literature elucidating the
meanings and contexts of many archived documents and images using many different
types of analysis. Some of those described here have focused on brochures, travel diaries,
newspapers, photographs, online blogs, guide books, websites, souvenirs, and secondary
numerical data sets.
Despite this emerging interest in archival research among tourism scholars, there
is a need to delve more deeply into the meanings of physically archived objects, texts
and images. How are destinations trying to depict themselves to the world by way
of their website images, postcards, and brochures? What underlying power relations
exist between key players, souvenir producers, and marketers? How can archived

DWYER 9781781001288 Part II.indd 413 25/07/2012 11:07

 414 Handbook of research methods in tourism

sources provide better understandings of tourism development in different parts of the

world? These are only a few of the multitudinous examples of questions that need to
be addressed with more analyses of archival data. There is an urgent need in academic
research to understand place identity and sense of place in tourism settings, power, and
politics, transformational tourism, and many other pressing issues of today. Archival
research can help address these and many more important scholarly debates. As long as
humans inhabit the earth and tourism continues to grow and develop, there will be texts,
pictures, or objects to archive. As the world continues to transform through technologi-
cal changes and globalization, adaptations will need to be made to traditional archival
approaches, but archives will undoubtedly continue to be important sources of data to
understand the tourism phenomenon.

REFERENCES

Alderman, D.H. and E.A. Modlin (2008), (In)visibility of the enslaved within online plantation tourism
marketing: a textual analysis of North Carolina websites, Journal of Travel and Tourism Marketing, 25 (3),
265281.
Andriotis, K. (2010), Brits behaving badly: template analysis of newspaper content, International Journal of
Tourism Anthropology, 1 (1), 1534.
Arreola, D.D. (2001), La Cerca y las Garitas de Ambos Nogales: a postcard landscape exploration, Journal of
the Southwest, 43, 505541.
Asplet, M. and M. Cooper (2000), Cultural designs on New Zealand souvenir clothing: the question of authen-
ticity, Tourism Management, 21, 307312.
Babbie, E. (2010), The Practice of Social Research, 12th edition, Belmont, CA: Cengage.
Beaulieu, I. and J.J. Lvy (2003), Tourism, sexuality and eroticism in a few contemporary novels, Toros,
Revue de Recherche en Tourisme, 22 (1), 4450.
Black, J. (1992), The British Abroad: The Grand Tour in the Eighteenth Century, New York: St Martins
Press.
Boyle, L.E. (2001), A Survey of the Vatican Archives and of its Medieval Holdings, Toronto: Pontifical Institute
of Mediaeval Studies.
Brodsky-Porges, E. (1981), The grand tour travel as an educational device 16001800, Annals of Tourism
Research, 8 (2), 171186.
Burns, P. and J. Lester (2005), Using visual evidence: the case of Cannibal Tours, in B.W. Ritchie, P. Burns
and C. Palmer (eds), Tourism Research Methods: Integrating Theory with Practice, Wallingford, UK: CABI,
pp. 4961.
Busha, C. and S.P. Harter (1980), Research Methods in Librarianship: Techniques and Interpretations, New
York: Academic Press.
Buzinde, C.N. (2010), Discursive constructions of the plantation past within a travel guidebook, Journal of
Heritage Tourism, 5 (3), 219235.
Cate-Arries, F. (2000), Changing places: travel and tourism in Manuel Vzquez Montalbns detective
novels, in H.M. Fraser and B.D. Willis (eds), Essays on Hispanic and Luso-Brazilian Literature and Film
in memory of Dr. Howard M. Fraser, pp. 8390. Mobile, AL: University of South Alabama Publications.
Chandler, D. (2007), Semiotics: The Basics, London: Routledge.
Choi, S.J., X.Y. Lehto and A.M. Morrison (2007), Destination image representation on the web: content
analysis of Macau travel related websites, Tourism Management, 28 (1), 118129.
Cohen, C.B. (2001), Island is a woman: women as producers and products in British Virgin Islands Tourism,
in Y. Apostolopoulos, S. Sonmez and D.J. Timothy (eds), Women as Producers and Consumers of Tourism in
Developing Regions, Westport, CT: Praeger, pp. 4772.
Cohen, E. (2007), From benefactor to tourist: Santa on cards from Thailand, Annals of Tourism Research,
34 (3), 690708.
Collins-Kreiner, N. and Y. Zins (2011), Tourists and souvenirs: changes through time, space and meaning,
Journal of Heritage Tourism, 6 (1), 1727.
Coombs, L.A. (1989), A new access system for the Vatican Archives, American Archivist, 52 (4), 538546.
Dane, F.C. (2011), Evaluating Research: Methodology for People Who Need to Read Research, Thousand Oaks,
CA: Sage.

DWYER 9781781001288 Part II.indd 414 25/07/2012 11:07

 Archival research 415

Danesi, M. (2007), The Quest for Meaning: A Guide to Semiotic Theory and Practice, Toronto: University of
Toronto Press.
Danto, E.A. (2008), Historical Research, New York: Oxford University Press.
de Hamel, C. (1986), A History of Illuminated Manuscripts, Oxford: Phaidon Press.
Echtner, C. (2002), The content of third world tourism marketing: a 4A approach, International Journal of
Tourism Research, 4 (6), 413434.
Enoch, Y. and R. Grossman (2010), Blogs if Israeli and Danish backpackers to India, Annals of Tourism
Research, 37 (2), 520536.
Feighery, W. (2009), Tourism, stock photography and surveillance: a Foucauldian interpretation, Journal of
Tourism and Cultural Change, 7 (3), 161178.
Frew, E. (2009), Festival image creation: the role of local and interstate newspapers, Tourism Analysis, 14
(2), 221230.
Galgano, M.J., J.C. Arndt and R.M. Hyser (2008), Doing History: Research and Writing in the Digital Age,
Boston, MA: Thomson.
Garrod, B. (2009), Understanding the relationship between tourism destination imagery and tourist photog-
raphy, Journal of Travel Research, 47 (3), 346358.
Gee, J.P. (2005), An Introduction to Discourse Analysis: Theory and Method, London: Routledge.
Geertz, C. (1973), The Interpretation of Cultures, New York: Basic Books.
Goodchild, M. (2007), Citizens as sensors: the world of volunteered geography, GeoJournal, 69, 211221.
Gordon, B. (1986), The souvenir: messenger of the extraordinary, Journal of Popular Culture, 20 (3),
135146.
Hall, C.M. and A. Valentin (2005), Content analysis, in B.W. Ritchie, P. Burns and C. Palmer (eds), Tourism
Research Methods: Integrating Theory with Practice, Wallingford, UK: CABI, pp. 191209.
Hill, M.R. (1993), Archival Strategies and Techniques, Newbury Park, CA: Sage.
Howell, M.C. and W. Prevenier (2001), From Reliable Sources: An Introduction to Historical Methods, Ithaca,
NY: Cornell University Press.
Jackson, S.L. (2009), Research Methods and Statistics: A Critical Thinking Approach, 3rd edition, Belmont,
CA: Cengage.
Jokela, S. and P. Raento (2011), Collecting visual materials from secondary sources, in T. Raki and D.
Chambers (eds), An Introduction to Visual Research Methods in Tourism, London: Routledge, pp. 5369.
Keppel, G., W.H. Saufley and H. Tokunaga (1992), Introduction to Design and Analysis: A Students Handbook,
New York: W.H. Freeman and Company.
King, N. (1998), Template analysis, in G. Symon and C. Cassell (eds), Qualitative Methods and Analysis in
Organizational Research, London: Sage, pp. 118134.
Kolocotroni, V. and E. Mitsi (eds) (2008), Women Writing Greece: Essays on Hellenism, Orientalism and
Travel, New York: Editions Rodopi.
Laderman, S. (2002), Shaping memory of the past: discourse in travel guidebooks for Vietnam, Mass
Communication and Society, 5 (1), 87110.
Larsen, J. (2006), Picturing Bornholm: producing and consuming a tourist place through picturing practices,
Scandinavian Journal of Hospitality and Tourism, 6 (2), 7594.
Leung, D., R. Law and H. Lee (2011), The perceived destination image of Hong Kong on Ctrip.com,
International Journal of Tourism Research, 13 (2), 124140.
Lew, A.A. (1992), Place representation in tourist guidebooks: an example from Singapore, Singapore Journal
of Tropical Geography, 12 (2), 124137.
Littrell, M.A., L.F. Anderson and P.J. Brown (1993), What makes a craft souvenir authentic?, Annals of
Tourism Research, 20, 197215.
McBurney, D.H. and T.L. White (2010), Research Methods, Belmont, CA: Cengage.
Meethan, K. (2008), Remaking time and space: the internet, digital archives and genealogy, in D.J. Timothy
and J.K. Guelke (eds), Geography and Genealogy: Locating Personal Pasts, Aldershot, UK: Ashgate, pp.
99112.
Metro-Roland, M. (2009), Interpreting meaning: an application of Peircean semiotics to tourism, Tourism
Geographies, 11 (2), 270279.
Milne, J.G. (1898), A History of Egypt under Roman Rule, London: Methuen.
Morgan, K.L., S.L. Larkin and C.M. Adams (2011), Empirical analysis of media versus environmental
impacts on park attendance, Tourism Management, 32 (4), 852859.
Murphy, P., M.P. Pritchard and B. Smith (2000), The destination product and its impact on traveller percep-
tions, Tourism Management, 21, 4352.
Neuendorf, K.A. (2002), The Content Analysis Guidebook, Thousand Oaks, CA: Sage.
Nittel, J. (2001), Wondrous Magic: Images of the Orient in the 18th and 19th Centuries British Women Travel
Writing, Berlin: Galda Wilch Verlag.
zdemir, G. (2010), Photographs in brochures as the representations of induced image in the marketing of

DWYER 9781781001288 Part II.indd 415 25/07/2012 11:07

 416 Handbook of research methods in tourism

destinations: a case study of Istanbul, in P.M. Burns, J.M. Lester and L. Bibbings (eds), Tourism and Visual
Culture, Volume 2: Methods and Cases, Wallingford, UK: CAB International, pp. 169180.
Paris, C.M. (2011), Understanding the statusphere and blogosphere: an analysis of virtual backpacker spaces,
in R. Law, M. Fuchs and F. Ricci (eds), Information and Communication Technologies in Tourism, 2011,
Vienna: Springer, pp. 443456.
Pudliner, B.A. (2007), Alternative literature and tourist experience: travel and tourist weblogs, Journal of
Tourism and Cultural Change, 5 (1), 4659.
Ratzel, F. (1896), The territorial growth of states, Scottish Geographical Magazine, 12, 351361.
Richards, P.L. (2008), Knitting the Transatlantic bond: one womans letters to America, 18601910, in
D.J. Timothy and J.K. Guelke (eds), Geography and Genealogy: Locating Personal Pasts, Aldershot, UK:
Ashgate, pp. 8197.
Riches, J.K. (2000), The Bible: A Very Short Introduction, Oxford: Oxford University Press.
Shenhav-Keller, S. (1993), The Israeli souvenir: its text and context, Annals of Tourism Research, 20, 182196.
Squire, S.J. (1995), In the steps of genteel ladies: women tourists in the Canadian Rockies, 18851939,
Canadian Geographer, 39 (1), 215.
Stokowski, P.A. (2011), The smile index: symbolizing people and place in Colorados casino gaming towns,
Tourism Geographies, 13 (1), 2144.
Sui, D.Z. (2008), The Wikification of GIS and its consequences: or Angelina Jolies new tattoo and the future
of GIS, Computers, Environment and Urban Systems, 32, 15.
Tang, L., R. Scherer and A.M. Morrison (2011), Web site-based destination images: a comparison of Macau
and Hong Kong, Journal of China Tourism Research, 7 (1), 219.
Timothy, D.J. (2005), Shopping Tourism, Retailing and Leisure, Bristol, UK: Channel View.
Timothy, D.J. (2011), Cultural Heritage and Tourism: An Introduction, Bristol, UK: Channel View.
Timothy, D.J. and D.L. Groves (2001), Webcam images as potential data sources for tourism research,
Tourism Geographies, 3 (4), 394404.
Timothy, D.J. and V.B. Teye (2004), American children of the African diaspora: journeys to the motherland,
in T. Coles and D.J. Timothy (eds), Tourism, Diasporas and Space, London: Routledge, pp. 111123.
Towner, J. (1984), The Grand Tour: sources and a methodology for an historical study of tourism, Tourism
Management, 5 (3), 215222.
Towner, J. (1985), The grand tour: a key phase in the history of tourism, Annals of Tourism Research, 12 (3),
297333.
Towner, J. (1988), Approaches to tourism history, Annals of Tourism Research, 15 (1), 4762.
Towner, J. (1995), What is tourisms history?, Tourism Management, 16 (5), 339343.
Towner, J. (1996), An Historical Geography of Recreation and Tourism in the Western World, 15401940,
Chichester, UK: Wiley.
Turner, A. (2006), Introduction to Neogeography, New York: OReilly.
Vespestad, M.K. (2010), Promoting Norway abroad: a content analysis of photographic messages of nature-
based tourism experiences, Tourism, Culture and Communication, 10 (2), 159174.
West, T. (2007), Dark tourism: a content analysis of popular media regarding motivations to travel to Ground
Zero, Master of Science Thesis, Arizona State University.
Williams, R.C. (2007), The Historians Toolbox: A Students Guide to the Theory and Craft of History, Armonk,
NY: M.E. Sharpe.
Wise, N.A. (2011), Post-war tourism and the imaginative geographies of Bosnia and Herzegovina and
Croatia, European Journal of Tourism Research, 4 (1), 524.

DWYER 9781781001288 Part II.indd 416 25/07/2012 11:07

 22 Community case study research
Dianne Dredge and Rob Hales

INTRODUCTION

Case studies are increasingly recognized as a valuable research strategy for the study
of tourism (Xiao and Smith, 2006). Defined as an empirical inquiry that investigates a
contemporary phenomenon within its real-life context, case study research is adopted
by researchers seeking to describe, explore and/or explain complex and dynamic social
systems (Yin, 2009, p. 13). Community case studies represent a particular subset of case
study research. For the purpose of this chapter, a community case study in tourism focuses
research attention on the way social processes within a community or multiple communi-
ties produce meanings about tourism and shape individual, organizational and institu-
tional responses. This chapter explores and critically discusses community case studies
in tourism and their application and value to tourism research and practice. The chapter
concludes by identifying future directions and the need to address a number of issues if
community case study research in tourism is to maximize its change-making potential.

THE NATURE OF TOURISM COMMUNITY CASE STUDIES

Defining Community

Community is a very difficult term to define and even the best attempts can be ambigu-
ous. It is often taken for granted that community denotes a group of people who share
some common features, i.e., they share the same ethnic background, are domiciled in
proximity to each other, or they possess the same values. However, communities are far
from homogeneous, which for many community case study researchers, is both a source
of fascination and frustration. To illustrate, community may be defined in a variety of
ways including:

A spatially coherent group of people. However, ambiguity emerges when

thenotionof scale is introduced, e.g., the local community and the global com-
munity.
A group of people sharing ethnic, cultural, professional or other characteristics.
However, that people share one or a number of common values and beliefs does
not mean they are a homogenous community. People can share some common
characteristics in one domain (e.g., ethnic background) and at the same time
exhibit opposing values in another domain (e.g., adherence to cultural traditions,
educational attainment or voting behaviour).
A group of people sharing of common beliefs, attitudes, interests, identities or other
types of connections. As in the above example, ambiguity emerges where some

417

DWYER 9781781001288 Part II.indd 417 25/07/2012 11:07

 418 Handbook of research methods in tourism

people share common beliefs in one domain (e.g., attitudes to tourism) but vary in
other ways (e.g., level of involvement in tourism).

In essence then, the term community is inherently vague and dependent upon the
context and the purpose for which it is defined. Individuals can belong to more than
one community and communities can overlap and intersect depending upon the issues
under investigation. However, good community case study research does not allow
the problem of definition to get in the way. It accepts that community is socially con-
structed; the focus of the research is on the interactions between community members
and this is often dynamic. Moreover, the concept of community is unbounded in that
the researcher tends to determine the number of participants in the case by making
value judgments about the minimum threshold needed to understand the social proc-
esses within that community. In this way community case studies are distinguished
from other types of case studies that focus on objects such as particular policies or
decisions.

CHARACTERISTICS OF COMMUNITY CASE STUDIES

Community case studies are empirical and context dependent, and cannot be abstracted
from the situated interactions and practices of community members. They are also
dependent upon the way in which community is defined and by the particular theoretical
explanations and frameworks that a researcher may choose to adopt. Notwithstanding
this diversity, community case studies share a number of common characteristics:

Tourism community case study research commonly seeks to describe, explore and/
or explain tourism or an aspect of tourism within the social context of a commu-
nity or set of communities.
Tourism community case study research focuses on the social interactions around
an episode or event of some kind in a singe case or a multiple comparative case
study setting.
Tourism community case study research is undertaken in situ, exploring the phe-
nomenon or a particular dimension of tourism in its natural setting.
Tourism community case study research is undertaken over a period of time (e.g.,
a longitudinal case study).
Tourism community case study research focuses on relationships between partici-
pants directly or indirectly involved in tourism or affected by it.
Tourism community case studies draw from several data sources and employ
multiple methods. Processes of triangulation and crystallization are often used in
generating understandings.
Tourism community study research commonly (but not always) includes learn-
ing and reflection on the part of the researcher, and these understandings may be
shared with community members (Swanborn, 2010; Yin, 2009, 2011).

These characteristics find their origins in rich historical debates about the character,
role and value of social sciences research in general. This background is now discussed.

DWYER 9781781001288 Part II.indd 418 25/07/2012 11:07

 Community case study research 419

BACKGROUND TO COMMUNITY CASE STUDY RESEARCH

Situating Social Research

The epistemological, ontological and methodological shifts occurring within the social
sciences from the mid-twentieth century have had significant implications for case study
research (Flyvbjerg, 2001; Gerring, 2007). At the risk of oversimplifying the discursive
twists and turns taking place over the course of the twentieth century, the dominance of
positivism in the early part of the century provided overwhelming support for rational
scientific methods characterized by systematic data collection, analysis and the produc-
tion of universal truths. In this environment, case studies were often criticized as being
less scientific, less reliable and therefore less useful, than scientific approaches (Flyvbjerg,
2006). However, as questions started to emerge about the rational and objective assump-
tions on which the scientific method was based, criticisms of positivism began to emerge.
By the middle of the last century, the idea that there were overarching universal truths
and cause-and-effect relationships that could explain the social world was increasingly
questioned. Understanding the dynamics of social processes, as opposed to hypothesis
testing and prediction, became an important imperative in social research. As a result,
case study research, as an approach to uncover complex social phenomenon in situ,
became increasingly accepted.
Underpinning this shift was the increasing recognition of the social situatedness of
research. In the Postmodern Condition: A Report on Knowledge, Lyotard (1979) argued
that all science was a form of story-telling, an interpretation based on the particular
choices and preferences of the researcher who is situated within a particular research
context. No single truth existed. Multiple realities could co-exist, which in turn opened
up opportunities for researchers to take different but nevertheless legitimate approaches
in researching the social world. This led to significant fault lines between natural and
social science research communities across the world. Even within the social sciences,
there were divisions between those comfortable devising hypotheses and emulating the
scientific method, and those seeking to explore and investigate social phenomenon from
a range of emerging postmodern perspectives.
These divisions played out over the last decades of the twentieth century, some-
times in bitter and subversive ways as evidenced by the Science Wars (Flyvbjerg, 2001;
Segerstrale, 2000). By the end of the twentieth century however, the battlelines had sof-
tened and researchers had become more tolerant of the contributions of various research
approaches and methods. An intellectualization of research methods was taking place.
Researchers began to focus greater attention on the way the philosophical positioning
of the researcher intersected with and influenced the choice of research approach and
methods. These choices would inevitably influence findings, so the emphasis shifted
from choosing the right methods to explaining the values, position and influence of
the researcher in the research process. For example, research commissioned by an event
organizer with the aim of establishing the impacts of an event on a local community
will likely yield very different results when compared to an ethnographer who also
examines the community impacts of the same event. Being clear about these influences
allows readers to better assess the value of the research and its findings in context and to
compare against their own experiences and knowledge (May and Perry, 2011).

DWYER 9781781001288 Part II.indd 419 25/07/2012 11:07

 420 Handbook of research methods in tourism

Tourism, as a complex multidisciplinary field of research, has benefited greatly from

a breaking down of these methodological divisions. As Jamal and Hollinshead (2001)
note, the very nature of tourism, and the inter- and multi-disciplinary influences upon it,
implies that it is not a distinct social practice; research questions cannot be easily isolated
nor simply aligned with quantitative or qualitative methodologies. Further, they argue,
the drive for scientific research that seeks out the truth only limits fuller understand-
ings of the situatedness of tourism as a dynamic phenomenon. These authors call for a
dialogue that includes and values multiple approaches, theories, practices, methods,
techniques that can assist those of us in tourism studies to do justice to the research topic
and research questions we formulate and pursue (p. 78). Against this background, com-
munity case studies in tourism have evolved considerably.

COMMUNITY CASE STUDIES AS MESSY RESEARCH

Previously, researchers under pressure to publish were bounded by the expectations and
protocols of their disciplinary-based research communities and the academic tribes to
which they belonged (Becher and Trowler, 2001; Tribe, 2010; Trowler, 1988). In those
disciplines where case studies were less recognized, researchers often found themselves
trying to create the illusion of a rational scientific approach, when in fact their research
was much more organic and creative, drawing fluidly and simultaneously from a variety
of theoretical and practical understandings. It was not until the latter part of the twenti-
eth century that this began to change and social research came to be viewed as complex,
messy and dynamic (Law, 2004).
In the case of tourism research, scientific approaches were inadequate in understand-
ing complex problems, and calls were made for an extended range of approaches and
methods. Law (2004) argued that, in trying to describe and analyse things in a coher-
ent way, the social sciences had made a mess of trying to make sense of things that are
inevitably messy, diffuse and complex. The methods, rules and frameworks derived from
different social science disciplines have helped to produce the reality that we understand,
but the weaknesses and flaws in these frameworks may also have produced black spots
in our understanding (Dredge et al., 2011). Law called upon researchers to explore other
realities and understandings by drawing upon the rich hinterland of pre-existing social
and material realities to bundle and reassemble it in different ways.
These views have been increasingly supported by those seeking a post-disciplinary
tourism research agenda; an approach that adopts a creative and flexible approach to
investigating problems and that embodies greater scholarly tolerance for different views
and methods (Coles et al., 2006; Etchner and Jamal, 1997; Hollinshead and Jamal, 2007).
These ideological shifts have stimulated an up-shifting in empirical, contextualized
research such as community case studies (Seidman and Alexander, 2001).

AIMS, VALUES AND PRIVILEGING PERSPECTIVES

In this context, there has been significant debate, refinement and development of
the case study as a legitimate research approach (Gerring, 2007; Swanborn, 2010;

DWYER 9781781001288 Part II.indd 420 25/07/2012 11:07

 Community case study research 421

VanWynsberghe and Khan, 2007; Yin, 2009, 2011). Yin (2009) defines various types of
case studies based on a 2 3 matrix wherein case studies can either be single or multiple
case studies and can be focused on description, exploration or explanation. Yins matrix
is useful in thinking about case study design, and can help a researcher make decisions
about the end goal of the research in terms of whether they are seeking to describe,
explore or explain their case. However, as discussed above, the values, identity and posi-
tioning of researchers can privilege certain perspectives and influence findings (Mills et
al., 2010).
Within the research process researchers also communicate, share information and
influence the community and other audiences so Yins idea that a case study describes,
explores or explains needs to be revisited. Noted anthropologist Clifford Geertz captures
this conundrum, arguing that finding somewhere to stand in a text that is supposed to
be at one and the same time an intimate view and a cool assessment is almost as much
of a challenge as gaining the view and making the assessment in the first place (Geertz,
1988, p. 10). This in turn raises questions about whether case studies can be categorized
as simply descriptive, exploratory or explanatory where the researcher is separate to
and independent of research problem, the end goal of the research and the community.
In other words, Yins categorizations do not adequately capture the relationships and
change that both the community and the researcher experience in many community case
study research processes. Inter-subjective communication between the researcher and
community members can be transformative (Reis, 2011). This then becomes a point of
departure from Yins typology because transformative community case studies go well
beyond description, exploration and explanation to incorporate values-driven action.

RESEARCHER REFLEXIVITY AND POLITICAL ACTION

Inter-subjective communication refers to the process of communication and exchange

between researchers and community members wherein all participants are reflecting and
building intercultural understandings, and in the process, they are being transformed
(Prus, 1997). There is a blurring of boundaries between researcher and community,
between insider and outsider, and between theory and practice (Alvesson and Skldberg,
2009; Chiseri-Strater, 1996). This dynamic process generates understandings that are
not inert facts but can become triggers for action and further reflection. In an evoca-
tive example, Noy (2004) explores how the identity of the backpacking community is
constituted through narrative performance of adventure stories. Encouraging a conver-
sational story-telling approach, Noy interviews Israeli backpackers. The backpackers
tell stories of their adventure travels interwoven with expressions of self-change, personal
growth and reflection. In the process, it was noted that the narratives were often told
in an excited tone and there was an expectation by the performer of the narrative that
the researcher shared mutual involvement in the story. Indeed, Noy notes that his own
response to one of the narratives was that he not only felt witness to the tourist attrac-
tion being described but the experience of the story evoked his own touristic self-change.
This example draws attention to the culture of research production and the way
research is produced, knowledge is exchanged and actions are identified and imple-
mented (May and Perry, 2011). In the past, the traditional approach of producing

DWYER 9781781001288 Part II.indd 421 25/07/2012 11:07

 422 Handbook of research methods in tourism

research that was presented as an end product (i.e., a research paper, a consulting report,
etc.) perpetuated a divide between research and practice. Researchers sought not to
intervene or take purposive action, merely to present facts. More recently, growing
emphasis on researcher reflexivity, or the extent to which researchers recognize their
roles and take purposive action to shape the outcomes, has led to optimism that research
can make a difference to the problems that communities are experiencing.
We argue that researchers undertaking community case study research can adopt
a position along a continuum of political action. At one end of the continuum is a
phronetic approach (as described by Flyvbjerg, 2001) which places the researcher in
a value-neutral position with respect to the community, but who is still politically
embedded in the case study. In this approach the mere choice of the community/case/
phenomenon under study is a political act because engagement by the researcher helps
in achieving the (political) aspirations of the community. Moreover, the researcher still
has personal perspectives and knowledge that will infiltrate their communications and
behaviours making a truly apolitical stance impossible. Nevertheless, in this process the
researcher seeks to understand all perspectives, promotes multi-perspective understand-
ings of the issues within the community, and resists attempts by certain community
members to legitimise particular stakeholder interests (Dredge, 2011). This approach to
the community case study has been variously described by social scientists as phronetic
case study research (Flyvbjerg, 2001; Maguire, 1997) and spiral case study research
(Mills et al., 2010) wherein the researcher is reflexively constructing meanings about the
case, sharing those meanings with community case study participants.
At the other end of the continuum, the researchers position is more direct and overt in
its political intent. The researcher chooses a case and actively engages in research openly
sharing and pursuing the aspirations of the community they engage with. The approach
sees the researcher embedding themselves within the case and siding with the com-
munity along with the sometimes disparate pursuit of their aspirations. This approach
also entails the researcher either communicating the findings of case study research in
the public sphere or iteratively feeding their findings back into the community. In both
instances the values that help align the action of the researcher are similar to the com-
munity. This approach can be described as the academic activist approach (Higgins-
Desboilles, 2010; Maxey, 1999).
The position of the researcher somewhere along this continuum will influence the way
the case study research evolves, the manner in which information exchange and knowl-
edge production takes place, and the role and influence of the researcher in the process.
In both approaches community case study research incorporates elements of participa-
tory action research, ethnography and quasi-experimental research designs. They require
that the researcher reflect upon and make decisions about such things as:

The type and level of political action desirable and appropriate from both the
researchers and the communitys perspective.
The desired level of involvement by the researcher (e.g., from low involvement to
highly overt political action) in the community case study.
The level and characteristics of power that the researcher has and to what advan-
tage this power can be used (or abused) to legitimize certain interests and out-
comes.

DWYER 9781781001288 Part II.indd 422 25/07/2012 11:07

 Community case study research 423

The level of personal and professional alignment the researcher has to the com-
munity over short, medium and long terms.

COMMUNITY CASE STUDIES IN TOURISM

Research, Practice and Ethical Action

Through application of theoretical frameworks, concepts and models, communication

and reflection, community case studies in tourism bring together theory and practice.
Moreover, this dialogic conversation between theory and practice can induce actions
by both researchers and community members that unleash tourisms world-making
capacity. Both Tribe (2004) and Jamal (2004) have argued for closer alignment between
tourism knowledge generated by research and practical action. Tribe (2002) builds
upon Schons (1983) reflective practitioner, arguing for a knowingin-tourism-ethical-
action wherein reflection and action are integrated where people act for the good of
tourism societies (p. 322). Jamal (2004) argues that Tribes argumentation does not go
far enough, and that, while action is good, further examination of the ethics underpin-
ning research is needed. Building upon Tribes work, and drawing from classical phi-
losophy of Aristotle, Jamal argues that researchers must address what Aristotle called
the good life a research life characterized by acting virtuously for the good of the
community.
These authors provide important justifications for valuing practical, engaged commu-
nity research that facilitates ethical action. Adding to this, we advocate that the dialogic
sense-making that takes place in community case studies when researchers and com-
munity members engage in communication and reflection results in Mode 2 knowledge
production (i.e., context and problem driven inquiry wherein both the researcher and
the researched contribute to the production of knowledge) (Nowotny, 2003). This sense-
making and ethical action approach can result in the researcher defining the problem
and researching in consultation with the community (rather than independently of it).
Unanticipated insights, derived through the co-production of local knowledge, can help
to generate solutions and joint actions.
Bodorks and Pataki (2009), for example, adopted a participatory action research
approach to learn about the possibilities for bottom-up regional sustainability
planning in a planning culture dominated by top-down institutional arrangements
in economically disadvantaged regions in Hungary. Ecotourism emerged as a
possibleaction identified by stakeholders. Reflecting on the value of this community-
based research the authors observed that it involved a different problem-structuring
process:

[It] takes a lot of time and effort; indeed, the mutual learning process is time-consuming.
Researchers are necessarily required to spend a great deal of time listening to the local inhab-
itants, striving to become familiar with their priorities, problems, and visions. As compared
to one-shot research processes typically applied in mainstream research, community-based
research can create a better learning environment through providing more flexibility and being
more responsive to local ideas. (p. 1130)

DWYER 9781781001288 Part II.indd 423 25/07/2012 11:07

 424 Handbook of research methods in tourism

TOURISM COMMUNITY CASE STUDY DESIGN

The post-disciplinary tourism research agenda with its attendant creative and flexible
approach to investigating problems has meant that methodology and techniques in
research have become more diverse (Tribe, 2004). Not surprisingly, community case
study design, data collection and analysis can appear messy. In what follows, three
frames of philosophical traditions in social research are used to distil the diversity found
in case study design, data collection and analysis in tourism studies. The data collection
methods and analysis used in community case studies are invariably guided by these tra-
ditions, and will be discussed later in the chapter.
The first is the postpositivist approach. In this approach it is possible to generate
hypotheses and test them within the case because within the case there are particular
units which can be discovered and delineated by the researcher (Ragin, 1993). Reality
is out there and it is the role of the researcher to discover these things and test for their
existence. The researcher is thus positioned as separate from data and the community
under study. A problem with this approach is that the aspect of tourism being examined
and tested is quite often conflated with the entirety of the case (VanWynsberghe and
Khan, 2007). To illustrate, noting that networks have become a fashionable topic in
the tourism literature, Costa et al. (2008) undertake a study exploring the benefits of
networks and partnerhips characteristics in sport and adventure tourism companies in
Portugal. In this example, the research problem is extracted from literature and defined
independently of the community under study. The findings relate to one aspect of the
community (i.e., the contribution of networks and partnerships to innovation) and
support the view that small-and medium-sized enterprises contribute to innovation and
economic development. However, and not diminishing the contributions of the study,
the process of identifying the research problem and approach were undertaken within
an academic context and whether or not the networks contribution to innovation is
important to the community is not relevant. There is also a tendency to generalize the
characteristics of the networks with the community of sport and adventure tourism com-
panies more generally.
In essence then, postpositivist approaches focus on understanding cause and effect,
and there is an emphasis on establishing a high level of internal validity (i.e., the extent
to which cause and effect can be confidently explained). The degree of emphasis on exter-
nal validity (i.e., the extent to which the case study is generalizable across other cases) in
the postpositivist approach is dependent on the purpose of the case study. For example,
in research examining residents attitudes towards cruise tourism in Canadas Arctic,
Stewart et al. (2011) develop a resident attitude typology from the literature that is then
tested in two communities. Local knowledge obtained from semi-structured interviews
provided input into a grounded approach whereby results and theory were generated
inductively and where data collection and analysis took place concurrently. This itera-
tive process was useful in improving the studys internal validity, i.e., the extent to which
the researchers were confident their results were representative of local community atti-
tudes. External validity was addressed in part by comparing the results obtained in the
two communities studied. It was also addressed through theoretical generalization (see
Hyde, 2008) whereby the cases were compared with the findings of other cases reported
in the literature. Whilst some similarities in community attitudes were noted, geographic

DWYER 9781781001288 Part II.indd 424 25/07/2012 11:07

 Community case study research 425

factors influence the type, duration and regularity of encounters thus making generaliza-
tions across communities difficult.
The second approach that can be adopted in community case studies is a critical
theory approach. It is not the purpose here to describe critical theory in detail, but to
note that the underpinning tenet is that there is a history and social ordering that has
led to contradictions in power, culture, race, ethnicity and gender. Transformation
towards a better or more just world is the aim, although critical research is often chided
for falling short of bringing about change and action (Bianchi, 2009). Critical theory
approaches to case studies tend to define the unit of analysis as the phenomenon under
study, although issues of power, influence and public interest are often central to such
studies (VanWynsberghe and Khan, 2007). For example, Dredge (2010) shows how con-
trasting notions of public interest emerge and are given meaning by various stakehold-
ers in an episode of community resistance over a proposed cruise ship terminal project.
Here, the unit of analysis is public interest. She illustrates how the states historically
embedded, top-down attitudes about public interest clash with locally driven bottom-up
ideas of civil democratic society. Similarly, in another community case study, Dredge
and Whitford (2011) examine how governance arrangements serve to (dis)empower dif-
ferent public spheres that emerge as a result of community resistance in the staging of
a major event. So, whilst the unit of analysis can be diverse, these critical community
case studies usually seek to highlight contradictions that lead to injustice, inequity and
marginalization.
Researchers adopting a critical theory approach also strive for high levels of
internal validity because the truthfulness of the research process is important in
any attempt to change the world. In tourism studies for example, this has been
achieved through approaches that embed the researcher in the community, where
barriers between the researched and the researcher dissolve and genuine community
interaction is adopted. To illustrate, in a study by Matarrita-Cascante (2010) in Costa
Rica 67 informants were interviewed and participant observation was undertaken
to better understand tourisms impact on two communities quality of life and well-
being. Community interaction was the unit of analysis, with the research finding
communication, participation, communion and tolerance were important factors in
developing community-based tourism. External validity was relatively less impor-
tant because the emphasis was not on generalizability across different communities
but rather on the power of example. In this case study, Matarrita-Cascante (2010)
illustrates the importance of interaction in purposive community-led tourism and
additional case studies were recommended as a way of extending the external validity
of these claims.
The third approach is the interpretivist approach. In this approach the focus of the
community case study is to offer a story like rendering of a problem or issue through
an in-depth description of everyday life and the real world. The interpretivist tradition
acknowledges multiple entry points into any given reality and that social constructs can
play a powerful role in tourism and travel. The unit of analysis is usually separated from
the case through rich description and attempt to understand the conditions under which
the concept, relationship, or event got the way it is (VanWynsberghe and Khan, 2007, p.
90). For example, Tucker (2007) undertakes a study of womens increasing involvement
in tourism work in Greme in central Turkey. She explores the separation of gender in

DWYER 9781781001288 Part II.indd 425 25/07/2012 11:07

 426 Handbook of research methods in tourism

Turkish village society and changing notions of power, control, honour and shame in
an ethnographic study spanning 10 years. In doing so, she seeks to explore the entangle-
ments that women are subject to in taking work outside the home, and in the process she
highlights changing attitudes towards and impacts of tourism.
Explorations of different communities of travellers are well represented in the lit-
erature, also demonstrating this interpretivist approach (e.g., Jennings, 2005; Noy,
2004; Reis, 2011). Wilson and Little (2008) explore the characteristics of solo female
travellers, focusing on how fear is perceived and experienced. Internal validity is not
as important in this approach as the other two approaches because rich description
and the power of example are emphasized, and the existence of multiple realities
means that no two stories are the same. As a result, external validity (generalizability)
is also not as important with the authors observing further research agendas which
encompass such perspectives would broaden and continue to problematize our
understanding of the complexities of womens geography of fear (Wilson and
Little, 2008, p. 183).

DATA COLLECTION AND ANALYSIS

A range of techniques can be employed in the data collection and analysis of community
case studies. The choice of philosophical tradition is an important part of the research
decision-making process and guides the type of data collection techniques employed.
Data collection can include contextual data, primary data from the community itself and
secondary data that is related to the community. A range of methods can be employed
and may include historical archival analysis, grounded theory, ethnography, textual
analysis, content analysis, media analysis and hermeneutics research to name a few. The
how to of these methods can be found in many standard texts on research methods
(e.g., Jennings, 2010).
The nature of community case studies and the need for researchers to gain an in-depth
knowledge of the phenomenon invariably means that the researcher becomes, to some
extent, part of the case. In this sense, the choices made about processes of data collection
and analysis are also dependent upon researchers positionality, identity and belong-
ing, since these aspects both open up or constrain opportunities for data gathering.
Therefore, how data gathering is undertaken will differ depending on a range of factors
including the:

Characteristics of researchers engagement with the community under study (e.g.,

number and diversity of participants, length of time spent in the community,
etc.);
Decisions about who defines the research problems and how they are defined (e.g.,
by the researcher, by the community, or in collaboration);
Construction and application of data collection tools (e.g., surveys and inter-
views);
Role and availability of secondary sources and archival materials to assist with
crystallization;
Power relationships between researchers and community case study participants;

DWYER 9781781001288 Part II.indd 426 25/07/2012 11:07

 Community case study research 427

Power relationships between researchers and institutions that may influence the
research and/or community (e.g., researchers current and future institutional
affiliations, direct and indirect clients of the research, etc.).

Ellis et al. (2011) offer a range of techniques for how the researcher can be reflexively
embedded in the data gathering exercise and they draw attention to how these choices
shape analysis. Reflexive techniques relevant to tourism community case study research
include: narrative ethnography; reflexive didactic interviews; co-constructed narrative
interviews; layered accounts using traditional data collection and analysis techniques
with reflection and indigenous ethnography. A detailed exploration of these techniques
is outside the scope of this chapter so, for further information on the how to of these
methods, see Ellis et al. (2011).
Adding a further challenge, discussions of triangulation and crystallization prompt
researchers to consider more closely how data is analysed and findings are presented.
Triangulation is much discussed in the literature as a strategy that combines multiple
methods of data collection and analysis to improve internal validity, to produce more
rigorous research outcomes and improve confidence in the findings (e.g., Bryman and
Burgess, 1994; Oppermann, 2000). More recently, others argue that reaching a single
point of understanding, metaphorically captured at the centre of a rigid two dimensional
triangle, is not consistent with the postmodern social constructionist turn (Taylor and
Trujillo, 2009). Rather, a crystal is a more appropriate metaphor from which multiple
points of view are reflected and refracted:

Crystallization combines multiple forms if analysis and multiple genres of representation into
a coherent text or series of related texts, building a rich and openly partial account of a phe-
nomenon that problematizes its own construction, highlights researchers vulnerabilities and
positionality, makes claims about socially constructed meanings, and reveals the indeterminacy
of knowledge claims even as it makes them. (Ellingson, 2009, p. 4)

Yet despite the increasing discussion of crystallization in the broader methods litera-
ture, in the tourism literature the term is employed with little explanation, and its appli-
cation remains locked within the researchers black box. Further examination of the
concept and its application would be useful.

ADVANTAGES AND LIMITATIONS OF COMMUNITY CASE

STUDIES IN TOURISM

Advantages of Community Case Study Research

A key value of community case studies lies in their capacity to contribute to human
learning and the development of intellectual, social, cultural and political capital.
Bourdieu (1977) and Dreyfus and Dreyfus (1986) explain that there is a developmental
leap between rule-bound knowledge and the fluid and dynamic performance of knowl-
edge demonstrated by experts. Reaching the world that world-making tourism profes-
sionals inhabit requires the development of, and fluency in, a range of knowledges that

DWYER 9781781001288 Part II.indd 427 25/07/2012 11:07

 428 Handbook of research methods in tourism

can be called upon dynamically to describe, explore, analyse, communicate and formu-
late actions. Flyvbjerg (2006, p. 222) argues that embedded case studies empower readers
to develop the reflective and fluid knowledge of experts:

Common to all experts ... is that they operate on the basis of intimate knowledge of several
thousand concrete cases in their areas of expertise. Context-dependent knowledge and expertise
and knowledge are at the very heart of expert activity. Such knowledge and expertise also lie
at the center of the case study as a research and teaching method ... If people were exclusively
trained in context-independent knowledge and rules ... they would remain at the beginners
level in the learning process.

The development of expert knowledge that directly contributes to the world-making

potential of tourism is an important advantage of community case study research.
Tourism has been described as world-making1 in the sense that it has a creative and
transformative role in the making of people and places and in the production of mean-
ings, values and understandings about the past, present and future (Hollinshead et al.,
2009). Where well managed within an integrated and sustainable approach, tourism can
give meaning to places and people, add value to cultural and environmental resources,
and promote peace and understanding. It can increase cultural awareness and tolerance;
address poverty, empower communities and contribute to improving economic and
social well-being (e.g., Higgins-Desboilles, 2006, 2008; Sharpley, 2009; United Nations
World Tourism Organization, 2010). Tourism community case studies can facilitate this
world-making activity in a number of ways.
First, community case study research builds deeper understandings of the complex
social world in which tourism takes place. Theerapappisit (2011) demonstrates this
world-making capacity in a community case study examining the participation of ethnic
communities in tourism planning in parts of the Mekong Region. The author exam-
ines stories told by local residents about the problems and benefits of participating in
tourism, observing that these different ways of storytelling can bring about a much
greater understanding of local contexts and traditional ways of life (p. 220). The case
study contributes to world-making by highlighting factors that inhibit and disempower
ethnic minorities that have been marginalized within tourism development dialogues,
and suggests that Buddhist ethics may serve as a developmental code of conduct to bring
disparate groups together.
Second, tourism community case studies can contribute to world-making by inter-
rogating the local implications of the grand narratives such as sustainable development,
globalization, public management and participative democracy. In a seminal study in the
late 1990s, Jamal and Getz (2000) describe the challenges of defining sustainable tourism
in the context of community roundtables where individuals and agencies had very dif-
ferent ideas and understandings about the purpose of such an exercise. In the process
the authors were also sensitized to the particular issues and interpretations held by local
actors, highlighting that the global narrative of sustainable development is interpreted
differently through the prism of ones own values and that addressing the challenge of
collaboration is paramount in moving sustainability debates forward.
Powerful examples of how community case studies contribute to knowledge produc-
tion are also found in studies of indigeneity and tourism. Scherrer and Doonan (2011)
set out to integrate Aboriginal and western knowledge in a system to capture, translate

DWYER 9781781001288 Part II.indd 428 25/07/2012 11:07

 Community case study research 429

and represent the intangible cultural impacts of tourism. The objective was to develop
a holistic and ontologically appropriate evaluation toolkit to assess cultural change
(p. 275). However, it soon became clear that the domination of Western thought within
the research process systematically marginalized Indigenous concepts and values, such
as access to land, making it difficult to develop standards and indicators. Fundamental
questions of power, control and the right and meaning of access to Indigenous lands
were at stake which may not have been uncovered without the deeply engaged commu-
nity case study approach that was adopted. The world-making contribution of this com-
munity case study was to highlight a fundamental disconnect between Indigenous values
and institutionalized management approaches.
Third, community case studies can contribute to world-making by challenging hidden,
and often unquestioned, assumptions; they can also prompt people and institutions
to think and reflect on the broader conditions and social constructs that shape their
engagement in and responses to tourism (Considine, 2005; Swan, 2008). Grybovych et
al. (2011), for example, explore the how planners and policy makers helped to address
sustainable tourism through actively engaging the community in a rezoning application
in Ucluelet, British Columbia. One of the authors (Mazzoni) had strong connections
with the community as a local resident and as the municipal planner. In the late 1990s,
Ucluelet was a community with a stagnant economy, a decreasing tax base and at the
mercy of fly-by-night and high risk developers (p. 81). The chapter tells the story of the
rezoning process of 800 acres of rainforest to permit a range of urban, recreation and
tourism uses and details the struggle to manage community resistance, and to engage
and negotiate acceptable outcomes. The long-term legacy and world-making contribu-
tion of this community case study was to develop a stronger consultation framework that
embraced the multi-vocality of the community.
Fourth, community case studies can contribute to world-making by interrogating con-
troversies and conundrums, inconsistencies and contradictions, i.e., local versus global;
residents versus tourists; top-down versus bottom-up; and corporate versus community.
Deeply engaged community case studies can contribute to the reframing of problems that
can in turn generate alternative solutions. In this way, community case studies align with
a movement that has become known as the Academy of Hope, which seeks to disrupt
the dominating discourses of neoliberalism and globalization (Pritchard et al., 2011; Ren
et al., 2010; Tribe, 2010). The Academy of Hope advocates for hopeful tourism inquiry,
and positions research as a conduit to generate transformative models for human devel-
opment. It strongly advocates humanist values including social justice, empowerment,
equality and tolerance; it is value-driven, highly contextualized and change-oriented.
To illustrate, Richards et al. (2010) explore impediments experienced by travellers with
a disability to participate in tourism. In this study the authors co-create knowledge about
the tourist experience using accounts of both sighted and visually impaired tourists.
In doing so they challenge an unsympathetic, unaccommodating sighted world that,
through their decision making, inhibits equity and enjoyment of the tourism experience.
In this paper, the authors clearly identify they are engaging in world-making through
advocacy scholarship and they seek to address these deeply embedded inequities.
Fifth, community case study research can also contribute to world-making by raising
questions about the strengths and weaknesses of literature and by promoting engaged,
critical reflection on theory. Pegas and Stronza (2010) employed an embedded ethno-

DWYER 9781781001288 Part II.indd 429 25/07/2012 11:07

 430 Handbook of research methods in tourism

graphic approach to investigate the use of ecotourism as a tool to protect sea turtles in
the coastal community of Praia Do Forte, Brazil. They found that sea turtle harvesting
that had put the conservation of the species at risk was the result of a complex interplay
of social, economic and cultural factors. The use of ecotourism to generate employ-
ment and produce economic diversity had contributed to the community changing its
attitudes towards sea turtles. However, incidents of illegal harvesting still occurred.
In this research the universalizing discourse that positions ecotourism as contributing
to environmental conservation was interrogated and a fuller understanding of social,
economic, cultural, demographic influences of ecotourism relevant to this community
were presented. The link between tourism, community development and environmental
conservation was found to be not as simple as it is often portrayed in the literature.

LIMITATIONS OF CASE STUDIES

Case studies, as a form of empirical, context-dependent research, have received strong

criticism over the years (Xiao and Smith, 2006). Community case studies, which are
strongly value-driven and characterized by deep research-community entanglements,
have been responsible for much of this criticism because they challenge long established
research standards inspired by positivism and scientific rationality. As a result, commu-
nity case studies have been criticized as being less scientific because they employ multiple
methods; they can be quite organic in their approach to data collection and they rely of
concepts such as triangulation and crystallization which cannot be explained independ-
ently of the researcher or outside the co-production of knowledge that takes place with
the community.
Reis (2011) grapples with these potential limitations in a case study of hunters and
trampers on Steward Island, New Zealand. Adopting a critical interpretivist perspective
and reflexive and creative methodologies, Reis reflects that she could not separate herself
from the fieldwork: there is no field out there to be entered and then left (p. 16);
nor was she able to isolate herself and her life experiences from the research in order to
discover facts. The evolution of her understanding was non-linear: my project changed
dramatically, and negotiating that understanding with others became a major process
in my research (p. 16). In this example, Reis acknowledges these tensions but instead of
framing them as limitations, they became liberating opportunities to explore alternate
ways of understanding herself and the tramping community she identified with.
A second criticism is that case studies cannot be easily generalized and their findings
applied to other cases. However, in a world where we recognize the multiple interests
within communities and the multi-vocality of individuals themselves, external validity
and reliability are less important and developing fuller understandings about how the
world works at the community level is becoming more important (Xiao and Smith,
2006).
A third criticism of community case studies is that they are not useful in explaining
cause and effect relationships and are therefore less useful in identifying solutions and
actions (see Flyvbjerg, 2006). This criticism is again derived from a positivist worldview
and does not take into account the complexity, multiplicity and constant change that
characterizes tourism in a social world. It is simply not possible to isolate all variables

DWYER 9781781001288 Part II.indd 430 25/07/2012 11:07

 Community case study research 431

and test the strength of relationships between them. Rather, the aim should be to provide
plausible context-dependent explanations of phenomenon that in turn may be used to
grow critical, creative thinking, deeper understandings of complex communities and
contribute to Mode 2 knowledge production.
DeBerry-Spence (2010) illustrates this in an examination of the MASAZI Visitor and
Welcome Centre, Ghana, a not-for-profit social venture, the aim of which is to support
micro-businesses that preserve and promote Ghanaian arts and culture. Everything in
this market space is negotiated and there exist complex dynamic relationships between
administrators, vendors, customers and tour operators. In this case study the author
explores the third space, a space where scholarship and practice blend in the making of
both theory and practice (p. 608). Working over time in the market, DeBerry-Spence
observes that nothing is clear cut; processes of negotiation and acts of resistance abound
and not everyone experiences these interactions in the same way. Binary positions of
cause and consequence are not able to be isolated.
Fourth, case studies have also been criticized because they contain bias and tend to
confirm researchers preconceived ideas. This criticism is underpinned by an apparent
expectation that researchers are objective, a myth we have previously acknowledged
as misleading. All research, including that framed by scientific rationality, is subject to
researcher and institutional entanglements that make it impossible to be objective. In
the case studies discussed above, Reis (2011) and DeBerry-Spence (2010) celebrate these
entanglements and argue that these research experiences have brought transcendental
change in the way these researchers engage with and position themselves in the research.
DeBerry-Spence argues that scholarship and advocacy do not have to be mutually exclu-
sive and that awareness and change can happen from researchers actively engaging in
embedded case studies of communities.
The above criticisms of case study research are derived from tensions between
researchers that adhere to notions of objectivity and rationality and those that embrace
value-laden postmodern, multicultural, transdisciplinary research. The former tend to be
rule bound and seek truth through research which interrogates cause and effect, while
the latter seek a shift towards responsible, ethical, value-based social science research,
to pursue knowledge that directly contributes to creating a more just and sustainable
world (Pritchard et al., 2011). Xiao and Smith (2006) conclude that the perception of
case studies as atheoretical, area-specific, one time and not following methodological
procedures was unfair and not justified:
... the majority of these instances [i.e. case studies] have followed scientific research procedures
with sound analytic techniques. A substantial proportion of work relied on longitudinal and/
or triangulated observations for the published report. Some have come up with moderate to
extensive discussions in the literature or theoretical contexts. (Xiao and Smith, 2006, p. 747)

FURTHER DEVELOPMENT OF THE COMMUNITY CASE

STUDY APPROACH

In the development of this chapter it has become increasingly clear that, while case study
research in tourism is relatively well developed, community case study research has not

DWYER 9781781001288 Part II.indd 431 25/07/2012 11:07

 432 Handbook of research methods in tourism

Table 22.1 Factors to consider in tourism community case studies

Factor Explanation
Definition of Dynamically defined dependent upon the context, the problem
community and the way the research unfolds in situ (i.e., the researcher will
make decisions about the definition of the community in the field)
Researchers position Researcher has a relationship with the community;
in relation to the embeddedness of the researcher is located on a continuum from
community casual to deeply connected
Approach Includes postpositivist, constructivist and interpretivist approaches
Methods Multiple methods including ethnography, interviews, participant
observation, focus groups, archival research, grounded theory,
media analysis, hermeneutics
Knowledge production On a continuum from researcher generated knowledge to co-
creating knowledge derived from reflexive engagement with
communities
Reflexivity On a continuum of political activism from phronetic (value
neutral) to political activism depending upon values-based and
change-making action of the researcher

received sufficient critical discussion. It is useful to make a distinction between case

studies focused on objects such as policies or decisions, and case studies that focus on the
social processes in communities that shape and give meaning to tourism. More impor-
tantly however, this distinction and further development of the community case study
research method is warranted if researchers are to enhance the world-making capacity
of tourism. By virtue of the entanglements that researchers develop before, during and
after the community case study research process, there are opportunities for change and
innovation to occur. Little attention has been placed on the role and ethical responsi-
bilities of researchers in undertaking community case study research, which needs to be
addressed in future work.
It has also become apparent that researchers seeking to undertake community case
studies are motivated by a set of personal and professional values, are influenced by
institutional factors, and are required to make a number of choices in framing their
research (Thomas, 2010). Drawing from the discussion contained in this chapter, we
identify in Table 22.1 those factors that need to be considered and justified by researchers
in framing their community case studies in tourism.
The consideration of the factors in Table 22.1, and the choices and trade-offs required,
are made through the prism of a researchers ethical position, their work conditions,
institutional values and other influences. While the idea of undertaking community case
studies and acting virtuously for the good of society is certainly a position to strive for,
it remains a difficult proposition in the current research environments of many western
countries (Peck, 2001). Emphasis on performance indicators, such as publications and
citations, rather than on the world-making contributions of research means that the
value of community case study research needs to be vigorously and continuously argued.
Furthermore, as the development of the ideas in this chapter took hold, it became
increasingly evident that we were touching upon deep tensions in ontological politics and

DWYER 9781781001288 Part II.indd 432 25/07/2012 11:07

 Community case study research 433

that, as researchers and practitioners, we were deeply entangled in these tensions. Our
two perspectives are given below in the closing of this chapter.

REFLECTIONS

I (Dianne) remain convinced that a deep critical approach to studying tourism in

situ reveals understandings and insights into globalization, neoliberalism, extra-local
rule systems and new public management that are not revealed by adopting scientific
methods. Through this approach I have becoming increasingly aware that the problems
and gaps in tourism literature are not necessarily of interest nor are the findings directly
or immediately useful to industry, policy makers and communities. Moreover the time-
lines required for deeply engaged community case study research sit uneasily against
the needs of industry and government for quick results to specific predetermined (and
usually narrow) questions. As a result the long-term benefits of community case studies,
co-production of knowledge and reflexive action remain, to some extent, unproven ben-
efits.
Specific case studies I have undertaken include community resistance to a proposed
development of a cruise ship terminal and an international car rally, and the develop-
ment and operation of local interest networks in a number of destinations in Australia.
These embedded case studies have allowed me to develop both context specific exper-
tise and theoretical understandings about a range of interconnected themes including
intergovernmental relations, networks, participative democracy, knowledge production
and public interest in tourism. However, my capacity to take or facilitate good actions
based on reflection and communication is often limited by the temporal nature of the
research task, which is often subject to ethical approval timelines, contractual obliga-
tions (if it is consulting-based research) and institutional demands for timely academic
outputs.
There is also an uneasy tension in community case study research that seeks to locate
itself at that end of the continuum that is embedded, reflexive and change-oriented, and
the scientific rationality expected by many industry and government clients. Not only is
the community case study research more organic and less able to be stage-managed by
industry or government clients (one may argue that the community is the real client),
but the action and change orientation of community case study research does not
provide industry and government decision makers with the opportunity to implement
recommendations and move through a rational policy cycle.
Moving beyond these ontological tensions in the way consulting-based research is
framed as a rational scientific exercise, and the organic process of a community case
study approach, requires long-term commitment from researchers, their universi-
ties, government and other clients (Dredge and Hales, 2012). Embedded research
takes time, long-term commitment and confidence in researchers, in the institutions
they work for, in communities and government. Not surprisingly, the idea of a
researcher-in-residence embedded within and working with communities seems a long
way off.
When I (Rob) engage in community case studies in tourism, I face problems in decid-
ing how to act as a researcher and the meaning and consequences of those actions. As

DWYER 9781781001288 Part II.indd 433 25/07/2012 11:07

 434 Handbook of research methods in tourism

previously described in this chapter the positionality of the researcher can be placed on
a continuum between value free to academic activist and, whilst there can be some guid-
ance on the factors to consider in community case study research, what to do is not laid
down in a formula to guide research action in each context.
How to position myself in any case is inevitably determined by my actions within a
case and quite often I am drawn to a particular side (or not) through considering the
factors in Table 22.1; by considering who in the community I engage with, the approach
that best suits the purpose of the research and for whom the knowledge is produced and
for what effect. How I am drawn can be best understood as follows.
Previously in this chapter the concept of virtue ethics (Jamal, 2004) was offered as a
guide to a research life characterized by acting virtuously for the good of the commu-
nity. It was noted that there was a lack of adoption of this ethical approach in commu-
nity case study research. With the risk of oversimplification, this may be attributed to
the inherent problems of what is good and who is community and how the researcher
decides such notions. A way I have overcome this problem was to use a normative
guide similar to the moral philosophy of Kants categorical imperative (Jamal and
Menzel, 2009) as this allows personal decisions of the academic to be guided and justi-
fied by universal rights such as international human rights or sustainability principles.
In my experience such guides help with how to act virtuously in tourism community
case studies.
For example, I was recently engaged in a consultancy which was to provide
advice to government on the Indigenous consent to a potential World Heritage
nomination. Because of the political issues in engagement with the community
(i.e., key representative groups did not want to participate), this meant that the
recommendations were partially completed. Additionally, after the consultancy was
completed the research publication based on the project targeted this issue, among
others, and used the United Nations Declaration on the Rights of Indigenous Peoples
(United Nations, 2007) to justify the argument for improvements in community
engagement. At the same time, I knew that the project was based on a government
consultancy and had issues of confidentiality. In order to deal with issues of confiden-
tiality the publication of the paper was delayed to respect the political processes that
the consultancy had addressed. Thus, this type of research is very time consuming
and does not sit easily under the publish or perish paradigm in current university
performance regimes.
In sum, we both agree that the positionality of the researcher can be envisaged along
a continuum and that the choice of action by the researcher is a highly contextual and
complex decision. We both agree that community case studies using progressive pur-
poseful approaches have the potential for significant emancipatory impact on the com-
munity irrespective of the philosophical tradition and researcher reflexivity adopted in
the community case study approach. However, community case study research is time
consuming, physically exhausting and presents various challenges for project manage-
ment and negotiating academic workload. Despite these challenges the potential for
community learning is enormous and further reflection on the value of community case
study research offers the possibility of undertaking research that can make a difference
to people and communities.

DWYER 9781781001288 Part II.indd 434 25/07/2012 11:07

 Community case study research 435

NOTE

1. Tourism is not the only profession that can make a claim to world-making. A range of other professions
have world-making potential, including nursing, medicine and engineering.

REFERENCES

Alvesson, M. and K. Skldberg (2009), Reflexive Methodology: New Vistas for Qualitative Research, 2nd
edition, London and Thousand Oaks, CA: Sage.
Arendt, H. (1958), The Human Condition, Chicago, IL: University of Chicago Press (reprinted 1998).
Becher, T. and P.R. Trowler (2001), Academic Tribes and Territories: Intellectual Enquiry and the Culture of
Disciplines, Buckingham, UK and Philadelphia: Open University Press and The Society for Research into
Higher Education.
Bianchi, R.V. (2009), The critical turn in tourism studies: a radical critique, Tourism Geographies, 11 (4),
484504.
Bodorks, B. and G. Pataki (2009), Linking academic and local knowledge: community-based research and
service learning for sustainable rural development in Hungary, Journal of Cleaner Production, 17 (12),
11231131.
Bourdieu, P. (1977), Outline of a Theory of Practice, Cambridge: Cambridge University Press.
Bryman, A. and R.G. Burgess (1994), Analysing Qualitative Data, New York: Routledge.
Chiseri-Strater, E. (1996), Turning in upon ourselves: positionality, subjectivity and reflexivity in case study
and ethnographic research, in P. Mortensen and G. Kirsch (eds), Ethics and Representation in Qualitative
Studies of Literacy, Urbana, IL: National Council of Teachers of English, pp. 115131.
Coles, T., C.M. Hall and D.T. Duval (2006), Tourism and post-discplinary enquiry, Current Issues in
Tourism, 9 (4/5), 293319.
Considine, M. (2005), Making Public Policy, Cambridge: Polity Press.
Costa, C., Z. Breda, R. Costa and J. Miguns (2008), The benefits of networks for small and medium sized
tourism enterprises, in N. Scott, R. Baggio and C. Cooper (eds), Networks Analysis and Tourism: From
Theory to Practice, Clevedon, Bristol: Channel View Publications, pp. 96130.
DeBerry-Spence, B. (2010), Making theory and practice in subsistence markets: an analytic autoethnography
of MASAZI in Accra, Ghana, Journal of Business Research, 63 (6), 608616.
Dredge, D. (2010), Place change and tourism development conflict: evaluating public interest, Tourism
Management, 31 (1), 104112.
Dredge, D. (2011), Phronetic Tourism Planning Research: reflections on critically engaged tourism planning
research and practice, IV Critical Tourism Studies Conference: Tourism Futures: Creative and Critical
Action, Cardiff: Welsh Centre for Tourism Research.
Dredge, D. and R. Hales (2012), Implications of the New Age of public management in higher education
tourism research, paper presented at the Council of Australasian Tourism and Hospitality Educators
Conference, Melbourne, 69 February.
Dredge, D. and M. Whitford (2011), Event tourism governance and the public sphere, Journal of Sustainable
Tourism, 19 (4/5), 479499.
Dredge, D., J. Jenkins and M. Whitford (2011), Stories of practice, in D. Dredge, J. Jenkins and M. Whitford
(eds), Stories of Practice: Tourism Planning and Policy, Surrey, UK: Ashgate.
Dreyfus, H. and S. Dreyfus (1986), Mind Over Machine: The Power of Human Intuition and Expertise in the
Area of the Computer, New York: Free Press.
Ellingson, L.L. (2009), Engaging Crystallization in Qualitative Research: An Introduction, Thousand Oaks, CA:
Sage.
Ellis, C., T. Adams and A. Bochner (2011), Autoethnography: an overview, Forum: Qualitative Social
Research, 12 (1), article 10.
Etchner, C. and T.B. Jamal (1997), The disciplinary dilemma of tourism studies, Annals of Tourism Research,
24, 868883.
Flyvbjerg, B. (2001), Making Social Science Matter: Why Social Inquiry Fails and How it Can Succeed Again,
Cambridge: Cambridge University Press.
Flyvbjerg, B. (2004), Phronetic plannning research: theoretical and methodological reflections, Planning
Theory and Practice, 5 (3), 283306.
Flyvbjerg, B. (2006), Five misunderstandings about case study research, Qualitative Inquiry, 12 (2),
219245.

DWYER 9781781001288 Part II.indd 435 25/07/2012 11:07

 436 Handbook of research methods in tourism

Geertz, C. (1988), Works and Lives: The Anthrolologist as Author, Standford, CA: Standford University
Press.
Gerring, J. (2007), Case Study Research: Principles and Practices, Cambridge: Cambridge University Press.
Grybovych, O., D. Hafermann and F. Mazzoni (2011), Tourism planning, community engagement and policy
innovation in Ucluelet, British Columbia, in D. Dredge and J. Jenkins (eds) Stories of Practice: Tourism
Policy and Planning, Farnham & Burlington: Ashgate, pp. 79104.
Higgins-Desboilles, F. (2006), More than an industry: the forgotten power of tourism as a social force,
Tourism Management, 27, 11921208.
Higgins-Desboilles, F. (2008), Justice tourism and alternative globalisation, Journal of Sustainable Tourism,
16 (3), 345364.
Higgins-Desboilles, F. (2010), In the eye of the beholder? Tourism and the activist academic, in P.M.
Burns, C.A. Palmer and J.M. Lester (eds), Tourism and Visual Culture Volume 1: Theories and Concepts,
Wallingford: CABI, pp. 98106.
Hollinshead, K. and T. Jamal (2007), Tourism and the third ear: further propsects for qualtitative inquiry,
Tourism Analysis, 12, 85129.
Hollinshead, K., I. Ateljevic and N. Ali (2009), Worldmaking agency: worldmaking authority. The sovereign
constitutive role of tourism. Tourism Geographies, 11 (4), 427443.
Hyde, K. (2008), Independent traveler decision-making, in A. Woodside (ed.), Advances in Culture, Tourism
and Hospitality Research, Volume 2, Bingley, UK: Emerald, pp. 43152.
Jamal, T. (2004), Virtue ethics and sustainable tourism pedagogy: phronesis, principles and practice, Journal
of Sustainable Tourism, 12 (6), 530545.
Jamal, T. and D. Getz (2000), Community roundtables for tourism related conflicts: the dialectics of consensus
and process structures, in B. Bramwell and B. Lane (eds), Tourism Collaboration and Parternships: Politics,
Practice and Sustainability, Clevedon, UK: Channel View Press, pp. 159182S.
Jamal, T. and K. Hollinshead (2001), Tourism and the forbidden zone: the underserved power of qualitative
inquiry, Tourism Management, 22, 6682.
Jamal, T. and C. Menzel (2009), Good actions in tourism, in J. Tribe (ed.), Philosophical Issues in Tourism,
Bristol, UK: Channel View Publications, pp. 227243.
Jennings, G. (2005), Caught in the irons: one of the lived experiences of long-term ocean cruising women,
Tourism Review International, 9 (2), 177193.
Jennings, G. (2010), Tourism Research, Brisbane, Australia: John Wiley & Sons.
Law, J. (2004), After Method: Mess in Social Science Research, Abingdon, UK: Routledge.
Lyotard, J.F. (1979), The Postmodern Condition: A Report on Knowledge, Minneapolis, MN: University of
Minnesota.
Maguire, S. (1997), Business ethics: a compromise between politics and virtue, Journal of Business Ethics, 16
(12/13), 14111418.
Matarrita-Cascante, D. (2010), Beyond growth: reaching tourism-led development, Annals of Tourism
Research, 37 (4), 11411163.
Maxey, I. (1999), Beyond boundaries: activism, academic, reflexivity and research, Area, 31 (3), 199208.
May, T. and B. Perry (2011), Social Research and Reflexivity: Content, Consequence and Context, Thousand
Oaks, CA: Sage.
Mills, A., G. Durepos and E. Wiebe (2010), Encyclopedia of Case Study Research Volume 2, Thousand Oaks,
CA: Sage.
Nowotny, H. (2003), The potential of transdisciplinarity, Interdiscipline, available at http://www.helga-
nowotny.eu/downloads/helga_nowotny_b59.pdf.
Noy, C. (2004), Performing identity: touristic narratives of self-change, Text and Performance Quarterly, 24
(2), 115138.
Oppermann, M. (2000), Triangulation: a methodological discussion, International Journal of Tourism
Research, 2, 141146.
Peck, J. (2001), Neoliberalizing states: thin policies/hard outcomes, Progress in Human Geography, 25 (3),
445455.
Pegas, F. and A. Stronza (2010), Ecotourism and sea turtle harvesting in a fishing village of Bahia, Brazil,
Conservation and Society, 8 (1), 15.
Pritchard, A., N. Morgan and I. Ateljevic (2011), Hopeful tourism: a new transformative perspective, Annals
of Tourism Research, 38 (3), 941963.
Prus, R. (1997), Subcultural Mosaics and Intersubjective Realities: An Ethnographic Research Agenda for
Pragmatising the Social Sciences, Albany, NY: State University of New York University Press.
Ragin, C. (1993), Introduction to qualitative comparative analysis, in T. Janoski and A. Hicks (eds), The
Comparative Political Economy of the Welfare State, New York: Cambridge University Press, pp. 299319.
Reis, A. (2011), Bringing my creative self to the fore: accounts of a reflexive research endeavour, Creative
Approaches to Research, 1 (1), 218.

DWYER 9781781001288 Part II.indd 436 25/07/2012 11:07

 Community case study research 437

Ren, C., A. Pritchard and N. Morgan (2010), Constructing tourism research: a critical inquiry, Annals of
Tourism Research, 37 (4), 885904.
Richards, V., A. Pritchard and N. Morgan (2010), (Re)Envisioning tourism and visual impairment, Annals of
Tourism Research, 37 (4), 10971116.
Sandercock, L. (2003), Out of the closet: the importance of stories and storytelling in planning practice,
Planning Theory and Practice, 4 (1), 1128.
Scherrer, P. and K. Doonan (2011), Capturing intangible culural impacts of tourism on Aboriginal land in
Australias Kimberley Region, Tourism Recreation Research, 36 (3), 271280.
Schon, D. (1983), The Reflective Practitioner, New York: Basic Books.
Segerstrale, U. (2000), Beyond the Science Wars: The Missing Discourse about Science and Society, New York:
State University of New York Press.
Seidman, S. and J.A. Alexander (2001), The New Social Theory Reader, New York and London: Routledge.
Sharpley, R. (2009), Tourism Development and the Environment: Beyond Sustainability, London: Earthscan.
Stewart, E.J., J. Dawson and D. Draper (2011), Cruise tourism and residents in Arctic Canada: development
of a resident attitude typology, Journal of Hospitality and Tourism Management, 18 (1), 95106.
Swan, E. (2008), Lets not get too personal: critical reflection, reflexivity and the confessional turn, Journal of
European Industrial Training, 32 (5), 385399.
Swanborn, J. (2010), Case Study Research: What, Why and How? Thousand Oaks, CA: Sage.
Taylor, B.C. and N. Trujillo (2009), Qualitative methods, in F.M. Jablin & L.L. Putnam (eds.), The New
Handbook of Organizational Communication: Advances in Theory, Research and Methods, Thousand Oaks,
CA: Sage, pp. 161193.
Theerapappisit, P. (2011), The Mekong tourism dilemma: converging forces, contesting values, Stories of
Practice: Tourism Planning and Policy, Surrey, UK: Ashgate, pp. 199226.
Thomas, R. (2010), Academics as policy-makers: (not) researching tourism and events policy formation from
the inside, Current Issues in Tourism, 14 (6), 493506.
Tribe, J. (2002), The philosophic practitioner, Annals of Tourism Research, 29 (2), 228257.
Tribe, J. (2004), Knowing about tourism: epistemological issues, in J. Phillimore and L. Goodson (eds),
Qualitative Research in Tourism: Ontologies, Epistemologies and Methodologies, London and New York:
Routledge, pp. 4662.
Tribe, J. (2010), Tribes, territories and networks in the tourism academy, Annals of Tourism Research, 37 (1),
733.
Trowler, P.R. (1988), Academics Responding to Change: New Higher Education Frameworks and Academic
Cultures, Buckingham, UK: The Society for Research into Higher Education & Open University Press.
Tucker, H. (2007), Undoing shame: tourism and womens work in Turkey, Journal of Tourism & Cultural
Change, 5 (2), 87105.
United Nations (2007), The UN Declaration on the Rights of Indigenous Peoples, New York: United Nations,
available at www.un.org/esa/socdev/unpfii/en/drip.html.
United Nations World Tourism Organization (2010), Tourism and the Millennium Development Goals, Madrid:
World Tourism Organization, available at http://www.unwto.org/tourism&mdgsezine/.
VanWynsberghe, R. and S. Khan (2007), Redefining case study, International Journal of Qualitative Methods,
6 (2), Article 6, available at http://www.ualberta.ca/~iiqm/backissues/6_2/vanwynsberghe.htm.
Wilson, E. and D.E. Little (2008), The solo female travel experience: exploring the geography of womens
fear, Current Issues in Tourism, 11 (2), 167186.
Xiao, H. and S.L.J. Smith (2006), Case studies in tourism research: a state-of-the-art analysis, Tourism
Management, 27 (5), 738749.
Yin, R. (2009), Case Study Research: Design and Methods, 4th edition, Thousand Oaks, CA: Sage Publications.
Yin, R.K. (2011), Applications of Case Study Research, Thousand Oaks, CA: Sage.

DWYER 9781781001288 Part II.indd 437 25/07/2012 11:07

DWYER 9781781001288 Part II.indd 438 25/07/2012 11:07
 PART III

MIXED METHODS

In Chapter 23, Svetlana Stepchenkova discusses Content analysis, a research tool

focused on the actual content and internal features of media. Content analysis is used
to determine the presence of certain words, concepts, themes, phrases, characters,
or sentences within texts or sets of texts and to quantify this presence in an objective
manner. Researchers quantify and analyse the presence, meanings and relationships of
such words and concepts, then make inferences about the messages within the texts, the
writer(s), the audience, and even the culture and time of which these are a part.
Although initially limited to studies that examined texts for the frequency of the
occurrence of identified terms (word counts), by the mid-1950s researchers were already
starting to consider the need for more sophisticated methods of analysis, focusing on
concepts rather than simply words, and on semantic relationships rather than just pres-
ence (de Sola Pool 1959). While both traditions still continue today, content analysis
now is also utilized to explore mental models, and their linguistic, affective, cognitive,
social, cultural and historical significance. Due to the fact that it can be applied to
examine any piece of writing or occurrence of recorded communication, content analysis
is currently used in a large range of fields, ranging from marketing and media studies, to
literature and rhetoric, ethnography and cultural studies, gender and age issues, sociol-
ogy and political science, psychology and cognitive science, and many other fields of
inquiry. Additionally, content analysis reflects a close relationship with socio- and psy-
cholinguistics, and is playing an integral role in the development of artificial intelligence.
Uses include identifying international differences in communication content, detecting
the existence of propaganda, Identifying the intentions, focus or communication trends
of an individual, group or institution, describing attitudinal and behavioural responses
to communications; and revealing psychological or emotional state of persons or groups.
In Chapter 24, Ulrike Gretzel and Heather Kennedy-Eden discuss Meta-analyses.
The main benefits of meta-analyses are the more objective appraisal of evidence than
in narrative reviews, the greater transparency which makes meta-analyses replicable,
the potential to resolve uncertainty and disagreement in existing research, the ability to
analyse the effects of research designs/study conceptualizations, as well as the opportu-
nity to generate important research questions for future research.
Quantitative meta-analysis is a methodology that involves the statistical integration of
research findings from separate studies It is a form of systematic review that uses statisti-
cal techniques to derive quantitative estimates of the magnitude of statistical effects and
their associated precision. For quantitative meta-analysis to work, the findings in the
individual studies need to be conceptually comparable (i.e., they must deal with the same

439

DWYER 9781781001288 Part III.indd 439 25/07/2012 11:08

 440 Handbook of research methods in tourism

constructs) and configured in similar statistical form. The advantages of quantitative

meta-analysis in comparison to narrative reviews lies in the explicit structure imposed
on the summary of research findings, the ability to code both magnitude and direction
of effects, the opportunity to relate study characteristics to effect sizes, and the ability to
synthesize a large number of studies resulting in estimates with greater statistical power.
Qualitative meta-analysis follows the same basic principles established for quantita-
tive meta-analysis in terms of comprehensiveness and systematic approaches but applies
qualitative research methods to the analysis of the studies included in the research syn-
thesis exercise. Consequently, it is interpretive rather than aggregative. Qualitative meta-
analysis does not aim at hypothesis testing. Rather, its main contributions lie in building
of more abstract theories (meta-theory), theory explication and development of new
theory. It requires assessing the qualitative research methodology as well as the influence
of paradigmatic parameters and theoretical assumptions that guided the researchers.
At the same time, it also needs to critically reflect on these issues in terms of how they
affected the meta-analysis itself. Challenges faced by qualitative meta-analysis involve
the loss of contextual information that is seen as essential to qualitative interpretation,
which might lead to more descriptive rather than interpretive approaches implemented
in the meta-analysis.
The authors note that the growing maturity of tourism as a field has spurred some
meta-analyses. The technique has been successfully applied to a variety of research
themes in tourism for which significant bodies of research exist and for which a synthesis
and/or critical evaluation of approaches was necessary to achieve theoretical or meth-
odological advancement. Quantitative meta-analyses have addressed income multipli-
ers, tourism demand, service quality, service recovery paradox, tourism advertising and
website evaluation. Examples of qualitative meta-analyses that aim specifically at theory
explication and theory building include analysis of the different theoretical models used
to explain travel decision-making processes, the ways in which the problem of peak oil
has been conceptualized in tourism research as well as the literature on tourism and the
environment.
The robustness of properly derived meta-analysis results makes them especially suit-
able for informing policy decisions and establishing practical guidelines. Like any kind
of research, meta-analysis depends on the quality of the data and the rigor applied in the
coding and analysis stages, and a thorough understanding of the statistical assumptions
made.
In Chapter 25, Ehsan Ahmed discusses Network analysis. A social network is a social
structure made up of individuals (or organizations) called nodes, which are tied (con-
nected) by one or more specific types of interdependency, such as friendship, kinship,
common interest, financial exchange, dislike, sexual relationships, or relationships of
beliefs, knowledge or prestige. Nodes are the individual actors within the networks, and
ties are the relationships between the actors. In its simplest form, a social network is a
map of specified ties between the nodes being studied. The notion of connectedness or
dependency therefore implies existence of an aggregated structure of relationships that
extends beyond the conventional view of relationships between two objects in isola-
tion. Studies on networks aim to capture these aggregated views of relationships and
offer analytical insights to understand the role of relationships on the objects embedded
within these relationships. Network analysis can generally be thought of as coming from

DWYER 9781781001288 Part III.indd 440 25/07/2012 11:08

 Part III Mixed methods 441

two different but complementary perspectives: the view from the individual element
embedded in a set of relationships (actor level) and the view from an aggregated form of
analysis (network level).
The origins of network theory and practice may be found in research conducted in the
late 1950s and 1960s on social behaviour and exchange, the social psychology of groups,
exchange and power, operational research in local government and inter-organizational
and exchange analysis.
The tourism industry may be regarded as an amalgamation of a wide range of sup-
pliers clustering together to provide experiences of value to the tourist. By using the
concepts of network and corresponding analytical tools, a network analyst can concep-
tualize, visualize and analyse these linkages and relationships between several tourism
elements. A tourism network can be understood as a set of formal and informal (social)
relationships that shape collaborative actions between government, the tourism industry
and the general public. Tourism networks are often loosely articulated groups of inde-
pendent suppliers linked to deliver an overall tourism experience. The multifaceted con-
nections of the tourism firm with its suppliers, customers, surrounding community, and/
or environment can be of various levels (e.g., local, regional, global), forms (e.g., infor-
mal, formal) and interest (e.g., economic, philanthropic). A tourism firm has connections
not only with its suppliers and customers but also with its surrounding community and
extended environment.
Ahmed distinguishes four different types of network research to provide the context
for his discussion of its application to tourism. These are: the impact of actor(s) on
other actor(s), the impact of individual actors on a network, the impact of a network
on individual actors, and whole network or network-level interaction. The more that is
known about the relationships between the various stakeholders in the tourism industry
collaboration, alliance, partnership, business clusters and similar other concepts
the better will be our understanding of the required strategies to maintain competitive
advantage at the various levels from the firm to the destination.

REFERENCE

Pool, I de S. (1959), Trends in Content Analysis, Oxford: Illinois Press.

DWYER 9781781001288 Part III.indd 441 25/07/2012 11:08

DWYER 9781781001288 Part III.indd 442 25/07/2012 11:08
 23 Content analysis
Svetlana Stepchenkova

NATURE OF CONTENT ANALYSIS AND ITS EVOLUTION

Content analysis was famously defined by Harold Lasswell as the technique that aims
at describing, with optimum objectivity, precision, and generality, what is said on a given
subject in a given place at a given time (Lasswell et al., 1952, p. 34). More recently,
Weber (1990, p. 9) defined content analysis as a research method that uses a set of pro-
cedures to make valid inferences from text. While these definitions emphasize textual
materials as objects of content analysis, other scholars consider the method applicable to
a broader range of content. According to Cartwright (1953, p. 424), the term refers to
the objective, systematic, and quantitative description of any symbolic behavior. In a
similar definition Berelson (1952, p. 18) states that content analysis is a research tech-
nique for the objective, systematic, and quantitative description of the manifest content
of communication. Shapiro and Markoff (1997, p. 14) comparatively analysed various
definitions of the term along several dimensions and suggested their own minimal defi-
nition: content analysis is any methodological measurement applied to text (or other
symbolic material) for social science purposes. Content analysis examines data for
patterns and structures, singles out the key features, develops categories, and aggregates
them into perceptible constructs in order to seize meaning of communications (Gray and
Densten, 1998); thus, the process involves the systematic reduction of the content flow,
whether textual or otherwise symbolic. Content analysis is capable of capturing a richer
sense of concepts within the data due to its qualitative basis and, at the same time, can be
subjected to quantitative data analysis techniques (Insch and Moore, 1997).
Arguably, the first content analysis study was conducted in the eighteenth century in
Sweden. It compared a popular set of 90 religious hymns of unknown authorship (The
Songs of Zion) with an orthodox set from established songbooks by counting occurrences
of selected religious symbols. No difference between the two sets of hymns was found,
and the claim that The Songs of Zion were contagious was dismissed (Krippendorff,
2004). At the beginning of the twentieth century, with the growth in newspapers circula-
tion numbers, the technique termed quantitative newspaper analysis emerged. Early
studies employed mostly subject-matter categories such as domestic affairs, politics,
crime, and sports to classify news content; for example, Mathews (1910) was concerned
with the question of whether newspapers were devoted to coverage of religious, scientific,
and literary affairs of the day, i.e., real news, or had shifted their attention in favor of
sports, gossip, and scandals. The interest in studying public opinion, spurred by various
social and political problems of the Great Depression era, the emergence of radio as
a new communication channel, and increasing public acceptance of methodological
advancements from social and behavioral sciences, led to formulation of the theoretical
and methodological foundations of the content analysis technique in the 19301940s
(Krippendorff, 2004). The topics of inquiry in those days ranged from presentation of

443

DWYER 9781781001288 Part III.indd 443 25/07/2012 11:08

 444 Handbook of research methods in tourism

African Americans in news media (Simpson, 1934) to comparative analysis of US and

US former enemies history textbooks (Walworth, 1938) to studies of nationalism in
childrens books (Martin, 1936) to investigation of values expressed in Boy Scouts and
Hitler Youth literature (Lewin, 1947).
Between the 1930s and 1950s, content analysis was propelled by the work of Harold
Lasswell and his associates through studies of mass communication and, especially, one
of its aspects propaganda. During World War II, the US government organized groups
of researchers charged with the responsibilities to content analyse enemy newspapers,
radio broadcasts, leaders speeches, propaganda films, etc. The Lasswell group adopted
a view of reality as a symbolic environment of words, images, and their meanings, which
surrounds human beings from their early existence. These symbols, though frozen at
any given moment, exhibit changes if studied over longer periods of time. These changes
are indicative of changes in reality: predominant ideologies in society, preoccupation
with certain issues, and other political and social matters (Lasswell et al., 1952). During
this period, the group contributed to development of content analysis methodology by
addressing the issues of sampling, reliability and validity of content categories, as well as
measurement problems (Krippendorff, 2004).
From the early to mid-twentieth century, the number of content analysis studies
progressed very quickly. During the first two decades the average annual number of
content analysis studies was 2.5, in the 1920s, 1930s, and 1940s increasing to 13.3,
22.8, and 43.3, respectively (Holsti, 1969, p. 20). During the 1950s, the average annual
number almost reached 100. Survey of 1719 titles published between 190058 revealed
that four out of five papers reported empirical studies while only 14.2 percent of all
titles were concerned with methodological issues (Barcus, 1959). By discipline, sociol-
ogyanthropology, general communication, and political science accounted for 27.7
percent, 25.9 percent, and 21.5 percent of all titles, respectively. The most popular
areas of inquiry were the studies of social values, propaganda analyses, journalistic
studies, media inventories, and psychologicalpsychoanalytic research; collectively
they accounted for 60 percent of all empirical content analysis studies (Holsti, 1969, p.
21). In 1955, the Allerton House conference at the University of Illinois was called to
discuss the state of the method, share the developments across various fields of social
science, and, at the same time, stipulate and address the methodological problems
of the day. The scholarly legacy of the conference is the volume Trends in Content
Analysis (Pool, 1959).
Another significant multidisciplinary gathering occurred at the University of
Pennsylvania in 1967, and is known as the Annenberg conference. Within a decade after
the Allerton House, research in linguistics and cognitive processes produced new theories
relevant to content analysis. Furthermore, Stone and his colleagues (Stone et al., 1966)
created The General Inquirer, the first language processing software which operated
on mainframe IBM computers that supported the PL/1 programming language. The
computer-based techniques allowed for the first time processing of huge amounts of
complex data, routinization of time-consuming operations, as well as transformation
of the data using programing languages. The aforementioned advances in theory and
technology required rigorous explication of some previously intuitive techniques and
procedures (Gerbner et al., 1969). The resulting volume, The Analysis of Communication
Content (Gerbner et al., 1969) was mostly concerned with text analysis; however, two

DWYER 9781781001288 Part III.indd 444 25/07/2012 11:08

 Content analysis 445

notable exceptions dealt with analysis of music (Brook, 1969) and visual records (Ekman
et al., 1969).
The period of 197095 has been characterized as the routinization of the technique
(Stone, 1997). Content analysis projects, which previously were contained in academic
spheres, extended to the business environment where private companies made content
analysis their specialization (e.g., the Gallop Organization). However, it was felt that
by the late 1990s, the number of studies employing content analysis had changed from a
flood (following the Annenberg conference) to a trickle (Shapiro, 1997). Computers were
still expensive and not user-friendly. Some scholars continued with manual approaches,
feeling that perfect reliability obtained in computer-assisted analyses often came at the
expense of validity. During this time, social sciences shifted away from grand theories
towards more practical, grounded approaches (Stone, 1997). However, emergence of
Web 2.0 applications in the early 2000s brought new steam to the field, as technical capa-
bilities, theoretical foundations, empirical knowledge, and practical need converged. A
whole new research area of online behavior opened up, not only generating interest from
social scientists but making content analysis relevant to a multitude of businesses all
wanting to harness the power of social networks and user-generated content.

BACKGROUND AND TYPES OF PROBLEMS THAT CONTENT

ANALYSIS IS DESIGNED TO HANDLE

In a number of situations, content analysis is a suitable research method. The technique

should be considered when the analyst cannot obtain the data directly from human
subjects due to time, space, or financial constraints and is limited to the documentary
evidence only. Establishing authorship of unsigned works (Miles and Selvin, 1966) is an
example of such analysis. Content analysis can also be used on free-elicitation survey
responses for enrichment and triangulation purposes, in parallel with other techniques
(Echtner and Ritchie, 1993). The technique is preferable in situations most notably in
psychology and psychiatry when the subjects own language, with all its complexities
and nuances, is crucial to the investigation (Mahl, 1959). Finally, the technique is helpful
in studies of general media, movies, literature, and online forms of communication,
where due to massive volumes all relevant data can rarely be examined by the researcher.
With the proper methodological steps taken to ensure representativeness and suitability
of the sample for the analysis, findings obtained from the sample can be inferred to the
larger universe of the textual materials (Holsti, 1969).
There are two general classes of epistemologies employed for content analysis in social
sciences: qualitative and quantitative. The former term refers to non-statistical and
exploratory methods, which involve inductive reasoning, while the latter term refers to
methods that are capable of providing statistical inferences from text populations (Berg,
1995). A central idea of quantitative content analysis is that many words of text can
be classified into much fewer content categories (Weber, 1990, p. 7). George (1959)
and later Shoemaker and Reese (1996, p. 32) argued that the process of reducing large
volumes of text to quantitative data does not provide a complete picture of meaning
and contextual codes, since texts may contain many other forms of emphasis besides
sheer repetition. Newbold et al. (2002, p. 80) agreed that there is no simple relationship

DWYER 9781781001288 Part III.indd 445 25/07/2012 11:08

 446 Handbook of research methods in tourism

between media texts and their impact, and it would be too simplistic to base decisions
in this regard on mere figures obtained from a statistical content analysis. Quantitative
content analysis does not always account for source credibility, political or social context
of messages being examined, and audience characteristics (Macnamara, 2003). However,
despite its limitations, quantitative content analysis has long been employed in social
studies due to its clear methodological reasoning based on the assumption that the most
frequent theme in the text is the most important, as well as the ability to incorporate such
scientific methods as a priori design, reliability, validity, generalizability, replicability,
and hypothesis testing (Neuendorf, 2002, p. 10).
From a philosophical perspective, the quantitative tradition is based on the positivist
premise that there is something like an objective reality (social facts) out there that
can be observed, measured, analyzed and thus understood (Newbold et al., 2002, p.
59); thus, decontextualization of the textual material and selection of the outsider vari-
ables for analysis of social phenomena are the main issues in quantitative paradigm. In
contrast, the qualitative epistemologies share the view that reality is a social and cultural
creation, which can only be interpreted, approximated but not fully apprehended; thus in
qualitative tradition the focus is on complexity, context, and detail (Denzin and Lincoln,
1994). Qualitative tradition heavily relies on researchers reading of content and includes
such approaches as rhetorical, narrative, semiotic, and discourse analyses to textual data
that cannot easily be summarized (Neuendorf, 2002).
Because it must consider multiple interpretational perspectives, the qualitative
approach is time consuming and rarely involves large samples. It has been also pointed
out that in qualitative data analysis causality cannot be established without high levels of
subjectivity (Mehmetoglu and Dann, 2003), and qualitative studies have also been criti-
cized as impossible to do with scientific reliability (Macnamara, 2003, p. 6). The com-
plete separation of the two traditions, however, is not always possible, given the diversity
of approaches and wide range of content analysis applications. Modern media scholars
tend to view qualitative and quantitative content analysis as complementary and parts
of a continuum of methods that can be applied to capture meaning and impacts of texts
(Curran, 2002; Newbold et al., 2002).
Two dimensions of quantitative content analysis, interpretational and structural,
became a basis for a 2 3 taxonomy suggested by Roberts (2000). On the structural
dimension of Roberts taxonomy there are thematic, semantic, and network text analy-
ses. The thematic approach is rooted in contingency analysis and involves counting
themes belonging to a certain theoretical construct within the text blocks. In the semantic
text analysis, textual data are broken into semantic units, for example, subjectaction
object triplets. Every semantic unit is associated with a certain numerical sequence,
which reflects the a priori established codes of themes (Franzosi, 1997). Lastly, in the
network analysis text blocks are represented as networks of interrelated themes, and
theme linkages are measured by specially generated variables (Kleinnijenhuis et al.,
1997). The interpretational dimension reflects the perspective from which the data is
interpreted, i.e., that of the speakers or the researchers. Perspective is considered repre-
sentational if texts are used to identify the speakers intended meaning. If the researchers
perspective is dominant, texts are interpreted in terms of researchers theory and the
analysis becomes instrumental. However, Roberts (2000) acknowledged that in many
instances text analysis involves both perspectives. It is representational when words for

DWYER 9781781001288 Part III.indd 446 25/07/2012 11:08

 Content analysis 447

thematic categories are coded based on their face value, yet the researcher might use the
data on these themes to interpret the text instrumentally. Any quantitative text analysis
produces a numerical matrix suitable for further statistical analysis. What is considered
a classic, textbook content analysis is an instrumentalthematic method applied to the
textual material; the following is a brief description of the method.

Textual Population and Sampling

A major research consideration in content analysis is defining a population of textual

materials suitable to answer a particular research question. Textual population is usually
a set of messages or documents, e.g., all newspaper editorials for a given year or all online
hotel reviews for a particular period of time. It may, however, be a population of people
who in the process of the study produce messages that are to be analysed. Sampling
procedure may involve as many as three steps: selecting sources of communication,
sampling documents, and sampling within documents (Holsti, 1969, p. 130). When
all sources in the population or all sampling units are considered equally informative,
non-probability samples should be avoided whenever possible (Neuendorf, 2002). When
sources are unequally informative or differ in influence, circulation, or format, content
analysts concern is that the selected texts are relevant to the research question and help
to answer it fairly. Texts must be sampled in view of what they mean, the interpretation
they enable, and the information they contain (Krippendorff, 2004, p. 113).
Relevance, or purposive, sampling often focuses on prestige press because it is felt
to represent most adequately the views of political elites, and, therefore, is considered
more influential than local papers (Lacy et al., 1991). Danielson and Lasorsa (1997)
used only two papers New York Times and Los Angeles Times in their study of politi-
cal and social changes in American society. Some studies selected sources on the basis of
the largest circulation: for example, Kayser (1953) made a comparative study of major
daily newspapers of 15 countries for a 7-day period. For sampling within the sources,
sampling guidelines are available from experimental studies: e.g., Mintz (1949) com-
pared four methods of sampling newspaper headlines in Pravda for the entire month.
The results revealed that every-fifth-day samples and odd-day samples did not differ
significantly from the figures for the entire month, but weekly samples were inferior.
Research by Riffe and associates (e.g., Riffe et al., 1993; Riffe et al., 1996a; Riffe et al.,
1996b) provide the most comprehensive guidelines with regard to optimal sampling pro-
cedure and sample size for news media.

Units of Content

Most researchers recognize three main unit types in content analysis: sampling units,
recording units, and context units (Neuendorf, 2002). Sampling units e.g., newspaper
articles, books, or websites provide a basis for identifying textual population and
selecting a sample for content analysis. The size of the sampling unit should be adequate
to represent the phenomenon under investigation.
To obtain a numerical matrix for statistical analysis, the researcher must designate
the textual unit for coding called the recording unit; this is one of the most fundamental
decisions in the content analysis project (Weber, 1990). Holsti (1969, p. 116) defines

DWYER 9781781001288 Part III.indd 447 25/07/2012 11:08

 448 Handbook of research methods in tourism

recording unit as the specific segment of content that is characterized by placing it in

a given category. In some cases, recording units can coincide with the sampling units
(e.g., Stepchenkova and Eales, 2011) but never exceed them. The smallest recording units
are words or symbols; they were used, for example, in studies on readability (Flesch,
1948) and authorship detection (Miles and Selvin, 1966). Theme, which is understood as
a single assertion about some subject matter, was widely used in research on propaganda
(Holsti, 1969), values (Namenwirth, 1973), and attitudes (Osgood et al., 1957). Coding
themes by hand is time consuming: the coder often needs to reduce the sentence into
its component themes, i.e., recording units, before these recording units can be placed
into established categories. This process is a source of error and can seriously reduce
reliability in content analysis. Sentences, paragraphs, and entire documents are usually
difficult to classify into a single category, which is a main reason why these recording
units are rarely utilized.
Often, the meaning of a recording unit is seen more clearly from a surrounding context;
therefore, a context unit has to be identified as well. For example, a researcher may be
required when he meets a word democracy to search an entire sentence or a paragraph
in order to find out what attitude toward democracy is exhibited by the source of content
(Geller et al., 1942). In addition, a reporting unit is the element for which findings are
reported (Neuendorf, 2002). In most social science investigations, the individual person
is both the unit of data collection and the reporting unit. In content analysis research,
however, units of data collection and reporting can differ: while sampling unit can be an
article or a story, the reporting unit can be a month, a quarter, or a year.

Categories

In the often-quoted words of Berelson (1952, p. 147), [c]ontent analysis stands or falls
by its categories. Particular studies have been productive to the extent that the categories
were clearly formulated and well adapted to the problem and to the content. There
are common principles that researchers should follow while developing categories: they
should be exhaustive (i.e., there should be a category for every relevant item in the text),
mutually exclusive (i.e., no recording unit should be placed in more than a single cate-
gory), and independent (i.e., assignment of any recording unit into a single category does
not affect classification of other data units) (Holsti, 1969, p. 95). Weber (1990) particu-
larly stresses the requirement of mutual exclusiveness and the statistical consequences of
its disregard. The number of categories and the complexity of the coding scheme vary
significantly among the studies, depending mostly on the level of detail that the analyst
aims to provide. The classification system involves painstaking description and detailing
of developed categories, as well as explication of coding rules for assigning numerical
values to recording units.
There are two main traditions in the quantitative content analysis research: substitu-
tion model and correlational model. In the substitution tradition, text is analysed with
a priori established categories, which are understood as a group of words with similar
meaning and/or connotations (Weber, 1983, p. 140). For example, words ice, frost,
snowflakes, and winter all represent the same idea of cold, thus, can be united under
one category, that of cold. Various categories are organized into dictionaries used
for making necessary substitutions in the text and obtaining category frequency counts.

DWYER 9781781001288 Part III.indd 448 25/07/2012 11:08

 Content analysis 449

When frequencies are organized in a matrix, correlations between categories can be

calculated. In contrast, the correlational model discerns categories from the text itself.
In this tradition categories are groups of words with different meaning or connotations
that taken together refer to some theme or issue (Weber, 1983, p. 140). These themes, or
categories, are extracted by a means of factor analysis or other multidimensional scaling
technique from the matrix of word co-variations (Iker, 1974). One can follow the debate
on comparative merits and disadvantages of the two approaches in Muskens (1985),
Weber (1983, 1986).

Reliability and Validity

Among three types of reliability stability, reproducibility, and accuracy (Krippendorff,

2004) the issue of reproducibility, or inter-coder reliability, traditionally gets more
attention than the other two reliability types (Shapiro, 1997). Unless an objective
instrument such as a computer is used for coding, the analysts must rely upon pooled
judgments of coders. Inter-coder reliability is a function of coders skill, insight, and
experience, clarity of categories and coding rules, as well as the degree of ambiguity in
the data. Inter-coder reliability reflects the degree of shared understanding of the two
coders. Since the nature of the data is usually beyond the investigators control, opportu-
nities for enhancing reliability are most often limited to improving coders, categories, or
both. Experimental studies have demonstrated that training of coders conducted prior to
the coding procedure can significantly increase inter-coder agreement (e.g., Kaplan and
Goldsen, 1965). Defining an acceptable level of reliability in content analysis is a problem
for which there is no single solution. Useful guidelines are given in Krippendorff (2004).
In contrast to reliability, validity cannot be asserted or declared through repeated
trials. Validity is the extent to which research design on the whole and measurement pro-
cedure in particular lead to inferences about the concept under investigation. Adequate
sampling and reliability of coding are necessary but not sufficient conditions for validity.
Computer-assisted content analysis, though perfectly reliable, can be invalid if concep-
tually inappropriate words, symbols, or categories are specified for counting. Content
analysts are always confronted with the choice what to prefer: more interesting and less
reliable text interpretations or highly reliable or somewhat superficial text analyses.

APPLICATION OF CONTENT ANALYSIS TO TOURISM

Tourism marketing and management involves communication of ideas and images; thus,
studies on destination image, tourism promotion, and, most recently, customer experi-
ences and satisfaction with tourism and hospitality services are areas of applicability of
the content analysis technique in tourism-related contexts. Tourism researchers employ
sorting and categorization techniques to identify the frequencies of certain concepts,
words, or people in textual and pictorial materials and treat the most frequent ones
as variables, or dimensions, of the theoretical construct under study (Echtner, 2002).
Scholars often utilize the technique at preliminary stages of the research, for question-
naire development (e.g., Baloglu and McCleary, 1999; Poria et al., 2006) or within
Echtner and Ritchies (1993) destination image measurement framework that involves

DWYER 9781781001288 Part III.indd 449 25/07/2012 11:08

 450 Handbook of research methods in tourism

analysis of both survey-based and free-elicitation responses (e.g., Murphy, 1999;

OLeary and Deegan, 2003; Rezende-Parker et al., 2003).
In a number of tourism studies, however, the technique has stood on its own. Within
the pushpull framework, Dann (1977) used tourists open-ended responses to inves-
tigate tourist motivations. Tapachai and Waryszak (2000) examined destination images
of Thailand and the United States as a response to stimuli descriptions. Echtner (2002)
analysed 115 brochures from 12 developing world countries and showed how myths
around the destinations are created by stereotyping and emphasizing certain attractions,
actors, and situations. Bandyopadhyay and Morais (2005) compared the representations
of India in the brochures produced by the Indian government and American tourism
media; the approach revealed the differences in two representations as they dealt with the
colonial nature of international tourism and the postcolonial stage of Indias national-
ism. Stepchenkova and Eales (2011) analysed media messages about Russia in the pres-
tige UK press and quantified them based on volume of the material, frequency of certain
topics and themes, as well as favorability of coverage. The developed quantitative series
were put in an econometric model of UK arrivals to Russia as variables accounting for
the qualitative factor of Russias destination image.
Advances in computer-assisted text analysis (CATA) software has made data han-
dling easier for analysts, allowing for larger samples, and providing ways for data visu-
alization. Andsager and Drzewiecka (2002) examined responses of college students to
guidebook image stimuli with VBPro (Miller, 1993); the obtained matrix was cluster ana-
lysed to determine relationships between most frequent words to describe destinations.
To investigate images of Australian cities and construct the perceptual maps, Ryan and
Cave (2005) used data from qualitative interviews; the analysis was aided by CATPAC
(Woelfel, 1998) and TextSmart software. Xiao and Mair (2006) analysed articles about
China published in major English newspapers with the Nudist Nvivo program; the
authors concluded that contrasting perceptions emerged in destination representational
dynamics. Govers et al. (2007a) looked at images of seven destinations using online
survey responses; the results produced with CATPAC were a three-dimensional visuali-
zation of image similarities and differences. Stepchenkova and Morrison (2008) derived
stereotypical and affective images of Russia from open-ended survey responses and
compared favorability of affective images for the US visitors and non-visitors combin-
ing two CATA packages. With the internet becoming a prominent means of destination
marketing, several studies focused on destination websites as a promotional tool (Govers
and Go, 2005; Stepchenkova and Morrison, 2006; Choi et al., 2007), while Govers et al.
(2007b) used the data from online communities at three prominent travel websites to
measured images of Middle Eastern destinations.
In the internet era, travelers themselves can serve as both image formation agents and
information sources. Travel blogs and virtual communities generate accessible, credible,
and readily available information able to influence travel behavior and purchase deci-
sions (Litvin et al., 2008). A comprehensive survey of the still emerging field of travel
blogs has been given recently by Banyai and Glover (2011). Besides interest in demo-
graphic profile of the bloggers, researchers have mostly focused on projected destination
image and identity creation (Banyai, 2010; Enoch and Grossman, 2010; Karlsson, 2006;
Law and Cheung, 2010; Schmallegger and Carson, 2009), monitoring visitors attitudes
and experiences (Carson, 2008; Pan et al., 2007; Wenger, 2008), as well as visited loca-

DWYER 9781781001288 Part III.indd 450 25/07/2012 11:08

 Content analysis 451

tions, activities, and tourism products (Carson, 2008). In the context of online com-
munities, Arsal et al. (2010) studied the influence of comments and advice provided by
residents about a destination on travel-related decisions of potential visitors.
Customer experiences and satisfaction with tourism and hospitality services related
online is an emerging topic of research because of its high practical relevance. Content
analysis has been applied to comments from hotel distribution websites to investigate
helpfulness of online hotel reviews (Black and Kelley, 2009), as well as commendations
and concerns that travelers express when rating hotels (Stringam and Gerdes, 2010).
Studies of hotel complaints with user-generated data has extended to examining sources
of satisfaction and dissatisfaction with hotels (Lu and Stepchenkova, 2012; OConnor,
2010), customer motives for complaining (Lee and Hu, 2004; Sparks and Browning,
2010), consistency of complaints with the expected level of service and room rates (Jeong
and Jeon, 2008), and cultural aspects of the complaints (Au et al., 2010). Using com-
ments from online restaurant guides, Pantelidis (2010) compared customer comments
made in times of economic plenty and times of recession.
Studies that employed content analysis may not necessarily cite the technique as their
research method; however, by applying principles of categorization and systematic data
reduction to the textual data they adhered to Shapiros and Markoff s (1997) minimal
definition. Crotts et al. (2009) studied guest satisfaction and competitive position of the
hotels by drawing inferences on rationale, opinion, details and feelings, as well as repeat
purchase intent of the visitors; the authors identified their technique as stance-shift anal-
ysis, which is a form of quantitative content analysis (Mason and Davis, 2007). Inversini
et al. (2010) studied how the online reputation of London, UK, transpired through travel
blogs; the authors classified reputation drivers into seven categorical dimensions (e.g.,
products and services, society, or environment) citing the Destination Online Reputation
analysis as their methodological approach. Dwivedi (2009) used consumer queries
posted on travel message boards to investigate the image of India, defining the technique
as netnography (Kozinets, 2006). The technique included a quantitative component of
data reduction, summarizing the wealth of textual data into six most highly discussed
attributes such as natural resources, culture, or history and art.
All of the abovementioned studies use textual data as the primary source. Imagery
analysis is less developed in comparison to the analysis of texts (for examples in tourism
research, see MacKay and Couldwell, 2004; MacKay and Fesenmaier, 1997), and
content analysis techniques used on texts are not automatically transferable to studying
imagery. To illustrate, Tussyadiah and Fesenmaier (2009) collected 120 video clips about
New York City from YouTube to investigate the roles that new media play as media-
tors of tourist experiences. Following a method proposed by Rose (2001), the research-
ers iteratively decomposed the visual material into sets of sequences, scenes, shots, and
frames to quantitatively code visual data.

ADVANTAGES AND LIMITATIONS OF CONTENT ANALYSIS

CONCEPTUALLY AND FOR POLICY FORMULATION

Content analysis is an unobtrusive research methodology that typically does not involve
human subjects and can effectively utilize vast amounts of material already available

DWYER 9781781001288 Part III.indd 451 25/07/2012 11:08

 452 Handbook of research methods in tourism

on the information superhighway. The technique has a rich history and very broad
application area. From the perspective of usefulness of the technique to the tourism
industry for example, destination marketing organizations (DMOs) content analysis
is suitable for comparing communications about a destination and examining inter-
national differences in content. Auditing communication content in the target markets
against the stated promotional objectives could help DMOs evaluate the effectiveness of
the effort. Content analysis would allow maintaining the consistent brand and staying
up to date on how the destination transpires in both general media and user-generated
textual universe. As a research technique, content analysis is also suitable for description
of attitudinal and behavioral responses of individual and organizational customers and
stakeholders towards destination initiatives. Monitoring the responses of the potential
travelers following destination promotional campaigns would allow adjustment of mes-
sages and correction of the communication flows. Since content analysis can reflect
cultural patterns of groups, institutions, or societies, it can also be used for examining
competition. The technique is relevant for making known the focus of individual or soci-
etal attention, as well as shifts in consumer preferences.
Obviously, the technique cannot be everything for everyone; it has its limitations.
Readers are reminded that content analysis where the speakers intentions and motives
are the purpose of investigation, i.e., representational content analysis, requires higher
interpretation levels and can be a source of increased error. Coding representational
communication into content categories assumes, in the words of Pool (1959, p. 4)
that what an author says is what he means, which, as critics point out, overempha-
sizes manifest content of communication and leaves out latent message. The issue of
manifest versus latent content has drawn a great deal of attention in the course of the
technique development; however, not all documents are rich in latent meaning, and the
importance of the latent content may be overestimated sometimes (Duriau and Reger,
2004). Moreover, manifest content can yield valid inferences despite the strategies of
the source if what is said is beyond the voluntary control of the source (Osgood, 1959;
Roberts, 2000). Echtners (2002) investigation of the brochures produced by 12 develop-
ing world countries DMOs is a good example; the analysis revealed that the brochures
content promoted some unflattering stereotypes towards the countries, which was
unlikely to be the intended DMOs strategy. From a more technical perspective, content
analysis projects are large in scope and can be very time consuming which to a certain
point can be remedied by using CATA packages; however, computer-assisted analysis
clearly requires some amount of training.

FURTHER DEVELOPMENTS AND APPLICATIONS OF

CONTENT ANALYSIS IN TOURISM RESEARCH

Stone (1997) made several predictions about the development of content analysis, which
so far have proved remarkably accurate. We have been witnessing increased interest in
content analysis studies, primarily due to nonparallel availability of texts on the infor-
mation superhighway combined with the rising processing power of computers and
developments in CATA programs. Social scientists have been making significant contri-
butions to explaining personal and organizational behavior in an ever-changing social

DWYER 9781781001288 Part III.indd 452 25/07/2012 11:08

 Content analysis 453

environment that now increasingly involves humancomputer interaction. Analysts have

shifted their interest from the top-bottom grand theories approaches towards smaller
scale bottom-up projects; well-founded valid inferences are produced with computers
helping the investigators better situate thematic content analyses. Practical convenience
and cost/benefit considerations have been guiding the division of labor between human
coders and machines. Finally, over the past decade much of the thematic text analysis
has been very practically oriented, as evidenced by the growing number of business, hos-
pitality, and tourism content analysis applications.
CATA programs have routinely been used in content analysis research. Overall,
quantitatively oriented software generally supports automatic coding of text using
dictionaries, while qualitatively oriented programs better carry out operations associ-
ated with theme identification and category coding (Alexa and Zuell, 2000; Lowe, n.d.).
Programs from the quantitative group typically include word frequency analysis, cat-
egory frequency counts, key-word-in-context (KWIC) function, cluster analysis, and
visualization, which Lowe (n.d) called basic handful. Popping (1997) further divided
quantitative software into six groups by its orientations towards thematic, semantic,
or network content analysis, as well as being more suitable for representational or
instrumental interpretation of texts. Programs with a qualitative focus are more of an
electronic version of the set of marginal notes, cross-references and notepad jottings
that a researcher will generate when analyzing a set of texts by hand (Lowe, n.d., para
1). There is also a group of programs the development environments that assists in
construction of dictionaries and grammars and is oriented towards a specific application
area, e.g., linguistics.
However, despite all developments in computer-assisted text processing, Mehmetoglu
and Dann (2003), as well as Stepchenkova and Mills (2010), point out a relatively low
acceptance of CATA in tourism research, with selection being confined to just a few
CATA packages. The adoption of CATA is slowed down by insufficient functionality
of any given CATA product in certain areas; at the same time, functional possibilities
offered by various software programs are not used to their full advantage (Alexa and
Zuell, 2000). This is due, at least partially, to an enormously wide range of possible
content analysis applications, which makes it impractical to create a program which
can support all conceivable operations for all types of content analysis. Studies that use
CATA in tourism research often give very cursory justification for selection of a CATA
package, and descriptions of the technicalities of the analyses are largely missing. To
assist researchers, articles that would recommend a particular CATA package for certain
types of content analysis, like the one by Mehmetologlu and Dann (2003), are highly
welcomed.
Automated content analysis promises unparalleled opportunities for social scientists;
however, many developments are still needed. Comparability of results from various
content analysis projects and, thus, accumulation of comparable records are highly
dependent on the usage of standardized categories. The two best-known dictionaries
Harvard IV Psychological Dictionaries (Dunphy et al., 1974) and the Lasswell Value
Dictionary (Lasswell and Namenwirth, 1968) were created specifically for that purpose.
However, with the focus shifted towards more grounded, bottom-top approaches, these
dictionaries have been used much less often. Take, for example, the issues of attitudinal
analysis and standardized categories in studies of customer opinions, experiences, and

DWYER 9781781001288 Part III.indd 453 25/07/2012 11:08

 454 Handbook of research methods in tourism

complaints. Content analysis of attitudes is a long-standing tradition; in studies of atti-

tudes, the number of favorable and unfavorable characteristics attributed to a symbol
or idea is often viewed as an indicator of the attitudes held by the writers, the readers, or
their common culture toward that symbol or idea. Attitude objects are valued, liked
or disliked in degrees ranging from positive to neutral to negative (Osgood et al., 1957,
p. 189). Thus, in studies involving customers opinions or complaints, classification of
text on the positivenegative dimension, with subsequent hypothesis testing, would
enrich the analysis beyond traditional category counting (for an example, see Lu and
Stepchenkova, 2012). However, it is more difficult for individual researchers to create
dictionaries for measuring positivenegative sentiments than for counting categories of
themes.
Presently, several areas of tourism and hospitality inquiry are well-defined, with certain
topics being recurrent. Coupled with the fact that the one-time procedure of dictionary
development often significantly offsets time gains obtained as a result of CATA process-
ing, it would make sense to create standardized dictionaries for certain content analysis
applications e.g., hotel reviews on various consumer websites which would be easily
expandable if needed. An efficient and user-friendly CATA package with a built-in atti-
tudinal measure will further foster the tourism and hospitality research. Integration of
the two types of dictionaries with the standardized categories and favorability analysis
would bring CATA analysis even closer to being automated. However, research-
ers are warned against push-button content analysis (Deacon, 2007). As Neuendorf
(2002) rightfully noted, no content analysis can be completely automated, the role of
researcher on any stage of the project, whether it is design, data collection, analysis, or
interpretation is the decisive factor of success.

REFERENCES

Alexa, M. and C. Zuell (2000), Text analysis software. Commonalities, differences and limitations: the results
of a review, Quality and Quantity, 34, 299321.
Andsager, J.L. and J.A. Drzewiecka (2002), Desirability of differences in destinations, Annals of Tourism
Research, 29 (2), 401421.
Arsal, I., K.M. Woosnam, E.D. Baldwin and S.J. Backman (2010), Residents as travel destination informa-
tion providers: an online community perspective, Journal of Travel Research, 49 (4), 400413.
Au, N., R. Law and D. Buhalis (2010), The impact of culture on e-complaints: evidence from Chinese
consumers in hospitality organizations, Information and Communication Technologies in Tourism 2010
Proceedings of the International Conference. Lugano, Switzerland, pp. 285296.
Baloglu, S. and K.W. McCleary (1999), A model of destination image formation, Annals of Tourism Research,
26 (4), 868897.
Bandyopadhyay, R. and D. Morais (2005), Representative dissonance: Indias Self and Western Image,
Annals of Tourism Research, 32 (4), 10061021.
Banyai, M. (2010), Draculas image in tourism: Western bloggers versus tour guides, European Journal of
Tourism Research, 3 (1), 522.
Banyai, M. and T.D. Glover (2011), Evaluating research methods on travel blogs, Journal of Travel Research,
DOI:101177/0047287511410323.
Barcus, F.E. (1959), Communications content: analysis of research, 19001958, PhD thesis, University of
Illinois.
Berelson, B. (1952), Content Analysis in Communication Research, Glencoe, IL: Free Press.
Berg, B.L. (1995), Qualitative Research Methods for the Social Sciences, Boston, MA: Bacon and Allyn.
Black, H. and S.W. Kelley (2009), A storytelling perspective on online customer reviews reporting service
failure and recovery, Journal of Travel and Tourism Marketing, 26 (2), 169179.

DWYER 9781781001288 Part III.indd 454 25/07/2012 11:08

 Content analysis 455

Brook, B.S. (1969), Style and content analysis in music: the simplified plaine and easie code, in G. Gerbner,
O.R. Holsti, K. Krippendorf, W.J. Paisley and P.J. Stone (eds), The Analysis of Communication Content,
New York: Wiley, pp. 287296.
Carson, D. (2008), The blogosphere as a market research tool for tourism destinations: a case study of
Australias Northern Territory, Journal of Vacation Marketing, 14, 111119.
Cartwright, D. (1953), Analysis of qualitative material, in L.F. and D. Katz (eds), Research Methods in the
Behavioral Sciences, Niles, IL: Dryden, pp. 421470.
Choi, S., X.Y. Lehto and A.M. Morrison (2007), Destination image representation on the web: content analy-
sis of Macau travel related websites, Tourism Management, 28 (1), 118129.
Crotts, J.C., P.R. Mason and B. Davis (2009), Measuring guest satisfaction and competitive position in the
hospitality and tourism industry, Journal of Travel Research, 48 (2), 139151.
Curran, J. (2002), Media and Power, London: Routledge.
Danielson, W.A. and D.L. Lasorsa (1997), Perceptions of social change: 100 years of front-page content
on The New York Times and The Los Angeles Times, in C.W. Roberts (ed.), Text Analysis for the Social
Sciences: Methods For Drawing Statistical Inferences from Texts and Transcripts, Mahwah, NJ: Lawrence
Erlbaum Associates, pp. 103116.
Dann, G. (1977), Anomie, ego-enhancement and tourism, Annals of Tourism Research, 4 (4) 184194.
Deacon, D. (2007), Yesterdays papers and todays technology digital newspaper archives and push button
content analysis, European Journal of Communication, 22, 525.
Denzin, N.K. and Y.S. Lincoln (1994), Introduction: entering the field of qualitative research, in N.K. Denzin
and Y.S. Lincoln (eds), Handbook of Qualitative Research, Thousand Oaks, CA: Sage, pp. 117.
Dunphy, D.C., C.G. Bullard and E.E.M. Crossing (1974), Validation of the General Inquirer Harvard IV
Dictionary, Pisa Conference on Content Analysis, Pisa, Italy.
Duriau, V.J. and R.K. Reger (2004), Choice of text analysis software in organization research: insight from
a multi-dimensional scaling (MDS) analysis, 7th International Conference on the Textual Data Statistical
Analysis, Louvain-la-Neuve, Belgium, pp. 382389.
Dwivedi, M. (2009), Online destination image of India, International Journal of Contemporary Hospitality
Management, 21 (2), 226232.
Echtner, C.M. (2002), The content of Third World tourism marketing: a 4A approach, International Journal
of Tourism Research, 4, 413434.
Echtner, C.M. and J.R.B. Ritchie (1993), The measurement of destination image: an empirical assessment,
Journal of Travel Research, 31 (4), 313.
Ekman, E.M., W.V. Friesen and T.G. Taussig (1969), VID-R and SCAN: tools and methods for the auto-
matic analysis of visual records, in G. Gerbner, O.R. Holsti, K.Krippendorf, W.J. Paisley and P.J. Stone
(eds), The Analysis of Communication Content, New York: Wiley, pp. 297312.
Enoch, Y. and M.L. Grossman (2010), Blogs of Israeli and Danish backpackers to India, Annals of Tourism
Research, 37 (2), 520536.
Flesch, R. (1948), A new readability yardstick, Journal of Applied Psychology, 32, 221233.
Franzosi, R. (1997), Labor unrest in the Italian service sector: an application of semantic grammars, in C.W.
Roberts (eds), Text Analysis for Social Sciences: Methods for Drawing Statistical Inferences from Texts and
Transcripts, Mahwah, NJ: Lawrence Earlbaum Associates, pp. 131145.
Geller, A., D. Kaplan and H.D. Lasswell (1942), An experimental comparison of four ways of coding editorial
content, Journalism Quarterly, 19, 362370.
George, A. (1959), Quantitative and qualitative approaches to content analysis, in A. Raftery (ed.),
Sociological Methodology, Oxford: Basil Blackwell, pp. 135144.
Gerbner, G., O.R. Holsti, K. Krippendorf, W.J. Paisley and P.J. Stone (1969), The Analysis of Communication
Content: Developments in Scientific Theories and Techniques, New York: Wiley.
Govers, R. and F.M. Go (2005), Projected destination image online: website content analysis of pictures and
text, Information Technology & Tourism, 7 (2), 7389.
Govers, R., F.M. Go and K. Kumar (2007a), Promoting tourism destination image, Journal of Travel
Research, 46 (1), 1523.
Govers, R., F.M. Go and K. Kumar (2007b), Virtual destination image a new measurement approach, Annals
of Tourism Research, 34 (4), 977997.
Gray, J.H. and I.L. Densten (1998), Integrating quantitative and qualitative analysis using latent and manifest
variables, Quality & Quantity, 32, 419431.
Holsti, O.R. (1969), Content Analysis for the Social Sciences and Humanities, Reading, MA: Addison-Wesley.
Iker, H.P. (1974), An historical note on the use of word-frequency contiguities in content analysis, Computers
and the Humanities, 8, 9398.
Insch, G.S. and J.E. Moore (1997), Content analysis in leadership research: examples, procedures, and
suggestions for future use, Leadership Quarterly, 8 (1), 125.
Inversini, A., E. Marchiori, C. Dedekind and L. Cantoni (2010), Applying a conceptual framework to analyze

DWYER 9781781001288 Part III.indd 455 25/07/2012 11:08

 456 Handbook of research methods in tourism

online reputation of tourism destinations, in Information and Communication Technologies in Tourism

2010 Proceedings of the International Conference, Lugano, Switzerland, pp. 321332.
Jeong, M. and M. Jeon (2008), Customer reviews of hotel experiences through consumer generated media
(CGM), Journal of Hospitality Marketing & Management, 17 (1), 121138.
Kaplan, A. and J.M. Goldsen (1965), The reliability of content analysis categories, in H.D. Lasswell, N.
Leites and associates (eds), Language of Politics: Studies in Quantitative Semantics, Cambridge, MA: MIT
Press, pp. 83112.
Karlsson, L. (2006), The diary weblog and the travelling tales of diasporic tourists, Journal of Intercultural
Studies, 27 (3), 299312.
Kayser, J. (1953), One Weeks News: Comparative Study of Seventeen Major Dailies for a Seven-day Period,
Paris: Paul Dupont, for UNESCO.
Kleinnijenhuis, J., J.A. de Ridder and E.M. Rietberg (1997), Reasoning in economic discourse: an application
of the network approach to the Dutch press, in C.W. Roberts (ed.), Text Analysis for the Social Sciences:
Methods for Drawing Statistical Inferences from Texts and Transcripts, Mahwah, NJ: Lawrence Erlbaum
Associates, pp. 191207.
Kozinets, R.V. (2006), Netnography 2.0., in R.W. Belk (ed.), Handbook of Qualitative Methods in Marketing,
Cheltenham, UK and Northampton, MA, USA: Edward Elgar, pp. 129143.
Krippendorff, K. (2004), Content Analysis: An Introduction to its Methodology, Thousand Oaks, CA: Sage
Publications.
Lacy, S.R., F.G. Fico and A.F. Simon (1991), Fairness and balance in the prestige press, Journalism
Quarterly, 68, 363370.
Lasswell, H.D., D. Lerner and I. de S. Pool (1952), The Comparative Study of Symbols, Stanford, CA: Stanford
University Press.
Lasswell, H.D. and J.Z. Namenwirth (1968), The Lasswell Value Dictionary. Vol. 13, New Haven, CT: Yale
University.
Law, R. and S. Cheung (2010), The perceived destination image of Hong Kong as revealed in the travel blogs
of Mainland Chinese Tourists, International Journal of Hospitality and Tourism Administration, 11 (4),
303327.
Lee, C.C. and C. Hu (2004), Analyzing hotel customers e-complaints from an Internet complaint forum,
Journal of Travel & Tourism Marketing, 17 (2), 167181.
Lewin, H.S. (1947), Hitler youth and the Boys Scounts of America: a comparison of aims, Human Relations,
1, 206227.
Litvin, S.W., R.E. Goldsmith and B. Pan (2008), Electronic word-of-mouth in hospitality and tourism man-
agement, Tourism Management, 29, 458468.
Lowe, W. (n.d.), Software for content analysis: a review, vol. 1, 2008, available at http://www.ou.edu/cls/
online/lstd5913/pdf/rev.pdf.
Lu, W. and S. Stepchenkova (2012), Ecotourism experiences reported online: classification of satisfaction
attributes, Tourism Management, 33 (3), 702712.
MacKay, K.J. and C.M. Couldwell (2004), Using visitor-employed photography to investigate destination
image, Journal of Travel Research, 42 (4), 390396.
MacKay, K.J. and D.R. Fesenmaier (1997), Pictorial element of destination in image formation, Annals of
Tourism Research, 24, 537565.
Macnamara, J.R. (2003), Media content analysis: its uses, benefits and best practice methodology, vol. 2008.
Mahl, G.F. (1959), Exploring emotional states by content analysis, in I.d.S. Pool (ed.), Trends in Content
Analysis, Urbana, IL: University of Illinois Press, pp. 89130.
Martin, H. (1936), Nationalism and childrens literature, Library Quarterly, 6, 405418.
Mason, P.R. and B. Davis (2007), More than the words: using stance-shift analysis to identify critical opinions
and attitudes in online focus groups, Journal of Advertising Research, 47 (4), 496506.
Mathews, B.C. (1910), A study of a New York daily, in Independent, 68, 8286.
Mehmetoglu, M. and G.M.S. Dann (2003), ATLAS/ti and content/semiotic analysis in tourism research,
Tourism Analysis, 8, 113.
Miles, J. and H.C. Selvin (1966), A factor analysis of the vocabulary of poetry in the seventeenth century, in
J. Leed (ed.), The Computer and Literary Style, Kent, OH: Kent State University Press.
Miller, M.M. (1993), Users Guide to VBPro: A Program for Qualitative and Quantitative Analysis of Verbatim
Text, Knoxville TN: University of Tennessee.
Mintz, A. (1949), The feasibility of the use of samples in content analysis, in H.D. Lasswell, N. Leites, R.
Fadner, J.M. Goldsen, A. Gray, I.L. Janis, A. Kaplan, D. Kaplan, A. Mintz, I.d.S. Pool and S. Yakobson
(eds), The Language of Politics: Studies in Quantitative Semantics, New York: George Stewart, pp. 127152.
Murphy, L. (1999), Australias image as a holiday destination: perceptions of back-packers visitors, Journal
of Travel and Tourism Marketing, 8 (3), 2145.
Muskens, G. (1985), Mathematical analysis of content, Quality & Quantity, 19, 99103.

DWYER 9781781001288 Part III.indd 456 25/07/2012 11:08

 Content analysis 457

Namenwirth, J.Z. (1973), The wheels of time and the interdependence of value change, Journal of
Interdisciplinary History, 3, 649683.
Neuendorf, K.A. (2002), The Content Analysis Guidebook, Thousand Oak, CA: Sage.
Newbold, C., O. Boyd-Barrett and H. Van Den Bulck (2002), The Media Book, London: Arnold (Hodder
Headline).
OConnor, P. (2010), Managing a hotel image on TripAdvisor, Journal of Hospitality Marketing &
Management, 19 (7), 754772.
OLeary, S. and J. Deegan (2005), Irelands image as a tourism destination in France: attribute importance
and performance, Journal of Travel Research, 43 (3), 247256.
Osgood, C.E. (1959), The representational model and relevant research methods, in I.d.S. Pool (ed.), Trends
in Content Analysis, Champaign, IL: University of Illinois Press, pp. 3388.
Osgood, C.E., G.J. Suci and P.H. Tannenbaum (1957), The Measurement of Meaning, Champaign, IL:
University of Illinois Press.
Pan, B., T. MacLaurin and J.C. Crotts (2007), Travel blogs and the implications for destination marketing,
Journal of Travel Research, 46 (1), 3545.
Pantelidis, I.S. (2010), Electronic meal experience: a content analysis of online restaurant comments, Cornell
Hospitality Quarterly, 51 (4), 483491.
Pool, I. de S. (ed.) (1959), Trends in Content Analysis, Champaign, IL: University of Illinois Press.
Popping, R. (1997), Computer programs for the analysis of texts and transcripts, in C.W. Roberts (ed.),
Text Analysis for the Social Sciences: Methods for Drawing Statistical Inferences from Texts and Transcripts
Mahwah, NJ: Lawrence Erlbaum, pp. 79100.
Poria, Y., A. Reichel and A. Biran (2006), Heritage site perceptions and motivations to visit, Journal of Travel
Research, 44 (3), 318326.
Rezende-Parker, A.M., A.M. Morrison and J.A. Ismail (2003), Dazed or confused? An exploratory study of
the image of Brazil as a travel destination, Journal of Vacation Marketing, 9 (3), 243259.
Riffe, D., C.F. Aust and S.R. Lacy (1993), The effectiveness of random, consecutive day and constructed week
samplings in newspaper content analysis, Journalism Quarterly, 70, 133139.
Riffe, D., S.R. Lacy and M.W. Drager (1996a), Sample size in content analysis of weekly news magazines,
Journalism and Mass Communication Quarterly, 73, 635644.
Riffe, D., S.R. Lacy, J. Nagovan and L. Burkum (1996b), The effectiveness of simple and stratified
random sampling in broadcast news content analysis, Journalism and Mass Communication Quarterly,
73, 159168.
Roberts, C.W. (2000), A conceptual framework for quantitative text analysis, Quality and Quantity, 34,
259274.
Rose, G. (2001), Visual Methodologies, London: Sage.
Ryan, C. and J. Cave (2005), Structuring destination image: a qualitative approach, Journal of Travel
Research, 44 (2), 143150.
Schmallegger, D. and D. Carson (2009), Destination image projection on consumer-generated content web-
sites: a case study of the Flinders Ranges, Information Technology & Tourism, 11, 111127.
Shapiro, G. (1997), The future of coders: human judgments in a world of sophisticated software, in C.W.
Roberts (ed.), Text Analysis for the Social Sciences: Methods of Drawing Statistical Inferences from Texts and
Transcripts, Mahwah, NJ: Lawrence Erlbaum, pp. 225238.
Shapiro, G. and J. Markoff (1997), A matter of definition, in C.W. Roberts (ed.), Text Analysis for the Social
Sciences: Methods for Drawing Statistical Inferences from Texts and Transcripts, Mahwah, NJ: Lawrence
Erlbaum Associates, pp. 931.
Shoemaker, P. and S. Reese (1996), Mediating the Message: Theories of Influences on Mass Media Content,
White Plains, NY: Longman.
Simpson, G.E. (1934), The Negro in the Philadelphia Press, Philadelphia, PA: University of Pennsylvania.
Sparks, B. and V. Browning (2010), Complaining in cyberspace: the motives and forms of hotel guests com-
plaints online, Journal of Hospitality Marketing & Management, 19 (7), 797818.
Stepchenkova, S. and J. Eales (2011), Destination image as quantified media messages: the effect of news on
tourism demand, Journal of Travel Research, 50 (2), 198212.
Stepchenkova, S. and J.E. Mills (2010), Destination image: a meta-analysis of 20002007 research, Journal of
Hospitality Marketing and Management, 19 (6), 575609.
Stepchenkova, S. and A.M. Morrison (2006), The destination image of Russia: from the online induced per-
spective, Tourism Management, 27 (5), 943956.
Stepchenkova, S. and A.M. Morrison (2008), Revisiting Echtner and Ritchie: Russias destination image
among American pleasure travelers, Tourism Management, 29 (3), 548560.
Stone, P.J. (1997), Thematic text analysis, in C.W. Roberts (ed.), Text Analysis for the Social Sciences:
Methods for Drawing Statistical Inferences from Texts and Transcripts, Mahwah, NJ: Lawrence Erlbaum
Associates, pp. 3554.

DWYER 9781781001288 Part III.indd 457 25/07/2012 11:08

 458 Handbook of research methods in tourism

Stone, P.J., D.C. Dunphy, M.S. Smith and D.M. Ogilvie (1966), The General Inquirer: A Computer Approach
to Content Analysis, Cambridge, MA: M.I.T. Press.
Stringam, B.B. and Jr. J. Gerdes (2010), An analysis of word-of-mouth ratings and guest comments of online
hotel distribution sites, Journal of Hospitality Marketing & Management, 19 (7), 773796.
Tapachai, N. and R. Waryszak (2000), An examination of the role of beneficial image in tourist destination
selection, Journal of Travel Research, 39 (1), 3744.
Tussyadiah, I.P. and D.R. Fesenmaier (2009), Mediating tourist experiences: access to places via shared
videos, Annals of Tourism Research, 36 (1), 2440.
Walworth, A. (1938), School Histories at War: A Study of the Treatment of Our Wars in the Secondary School
History Books of the United States and in Those of its Former Enemies, Cambridge, MA: Harvard University
Press.
Weber, R.P. (1983), Measurement models of content analysis, Quality and Quantity, 17, 127149.
Weber, R.P. (1986), Correlational models of content: reply to Muskens, Quality & Quantity, 20 (2/3), 273275.
Weber, R.P. (1990), Basic Content Analysis, Newbury Park, CA: Sage.
Wenger, A. (2008), Analysis of travel bloggers characteristics and their communication about Austria as a
tourism destination, Journal of Vacation Marketing, 14, 169176.
Woelfel, J. (1998), CATPAC: Users Guide, New York: RAH Press.
Xiao, H. and H.L. Mair (2006), A Paradox of Images: representation of China as a tourist destination,
Journal of Travel & Tourism Marketing, 20 (2), 114.

DWYER 9781781001288 Part III.indd 458 25/07/2012 11:08

 24 Meta-analyses of tourism research
Ulrike Gretzel and Heather Kennedy-Eden

INTRODUCTION

As the name suggests, meta-analysis is a research method that involves the analysis of
analyses. It aims at assessing a field of study beyond one particular study (Timulak,
2009, p. 591) by aggregating research findings from a number of existing empirical
research studies. It is a relatively new methodology, especially in the social sciences, that
is still under development and also controversial in some respect because of the potential
biases it can introduce and the risk of comparing things that are not comparable. With
the growing maturity of tourism as a field of inquiry one can expect that the integration
of research findings provided through meta-analyses will grow in importance. A chapter
dealing with meta-analysis in a tourism research methods book is therefore warranted.
However, rather than replicating what others have already written, this chapter only
provides brief summaries of the general issues and procedures and focuses attention
on the applicability of meta-analyses to solving research problems in tourism and the
specific challenges faced when conducting a meta-analysis related to tourism research.
Examples of published meta-analytic research in tourism are presented to illustrate
the processes involved and to demonstrate the benefits for knowledge advancement in
tourism.

IMPORTANCE OF META-ANALYSIS AS A FORM OF

RESEARCH SYNTHESIS

Science is a cumulative endeavour that requires the integration of research related to the
same phenomena (Cooper et al., 2009). This ensures that a constant re-invention of the
wheel can be avoided and that knowledge is actually advanced. Research integration is
typically achieved through reviews of existing research. Based on their goal, approach,
coverage and perspective used one can distinguish broadly between two types of reviews:
narrative reviews and systematic reviews (Figure 24.1). Narrative reviews are summaries
or assessments that can be written from a neutral perspective or from an argumentative
point of view. They are inherently subjective as the selection of what research is sum-
marized is not systematic and the criteria that led to the inclusion of a piece of research
in the summary are typically not made explicit. These reviews also do not have to be
comprehensive but rather identify the most important/relevant research. In summary,
they are selective and not guided by formal rules (Egger and Smith, 1997). Looking at
the goal and format of narrative reviews, two types emerge: the typical literature review
included in a publication or grant proposal and the review article. They differ mainly in
scope, with the literature review being more selective and more focused, but also in their
role: A review article critically evaluates more broadly what is known in a particular field

459

DWYER 9781781001288 Part III.indd 459 25/07/2012 11:08

 RESEARCH INTEGRATION

DWYER 9781781001288 Part III.indd 460

Summary of existing research

Subjective, selective, using Objective, comprehensive,

implicit criteria using explicit criteria

Narrative Review Systematic Review

Summary/assessment of existing research findings Synthesis of published and unpublished primary

and theoretical arguments regarding a limited research findings to answer research questions
topic as found in academic publications with the same rigor applied to primary research

460
Literature Review Review Article Quantitative Meta-Analysis Qualitative Meta-Analysis

Presentation of findings, Critical evaluation of the Statistical analysis of a Systematic analysis of

arguments or definitions state of knowledge in a collection of analysis primary qualitative
from academic particular area of research results from individual research findings for the
publications to explicate studies for the purpose of purpose of theory
the need for research in an testing specific research explication, building and
area and/or justify the hypotheses development
conceptualization of a
particular research study

Figure 24.1 Research review typology

25/07/2012 11:08
 Meta-analysis 461

of research so that knowledge/research gaps can be identified. A literature review posi-

tions an empirical research study within an existing body of literature.
In contrast, systematic reviews focus on the synthesis of primary research rather
than descriptive summaries or critiques, ideally including both published and unpub-
lished studies and with the goal of being as exhaustive as possible in the inclusion and
meticulous in the execution (Clarke, 2011). Systematic reviews are needed when there is
a substantive question, several primary studies perhaps with disparate findings and
substantial uncertainty (Hemingway and Brereton, 2009, p. 3). Egger and Smith (1997)
describe systematic reviews as types of reviews that have been prepared with strategies
in place to avoid bias. Consequently, systematic means that clear criteria for the inclu-
sion or exclusion of research in the systematic review are established and made explicit.
Meta-analysis is a possible component of a systematic review aiming at a comprehensive
description of a phenomenon beyond one particular study, usually also assessing the
influence of the method of investigation on the research findings. As such it allows for
cumulating knowledge across studies by aggregating outcome measures and correcting
errors and biases in research findings as well as overcoming research artefacts that occur
in all individual empirical studies (Woodside and Dubelaar, 2003). Meta-analysis is
therefore only applicable to empirical research (Lipsey and Wilson, 2001). Of course, a
meta-analysis is only as good as the quality of the primary research it includes, making it
important to carefully judge studies before they are included. The typical steps involved
in a meta-analysis are:

1. Definition of a research question

2. Identification of existing research studies
3. Assessment of the studies
4. Combining of the results of the primary research studies through coding and analy-
sis
5. Interpretation of the findings.

According to Egger and Smith (1997), the main benefits of meta-analyses are the more
objective appraisal of evidence than in narrative reviews, the greater transparency which
makes meta-analyses replicable, the potential to resolve uncertainty and disagreement in
existing research, the ability to analyse the effects of research designs/study conceptuali-
zations, as well as the opportunity to generate important research questions for future
research. Imminent challenges for meta-analyses are the issue of comparability of studies
(the apples and orange problem) and the bias that can be introduced by focusing only on
published studies as certain types of studies with certain types of findings have a higher
likelihood of being published (file drawer problem) (Lipsey and Wilson, 2001).
Meta-analysis has been applied to various research questions in a variety of disciplines
and research areas. This chapter provides an overview from the perspective of tourism
research and cannot possibly discuss all aspects relevant to actually conducting meta-
analyses. Thus, it is not a systematic review of meta-analysis articles and books but
rather a selective narrative review to provide interested readers with an introduction.
Detailed information about the specific steps and challenges related to meta-analysis
can be found in the literature (for instance, Berman and Parker, 2002; Borenstein et al.,
2009; Briggs, 2005; Charlton, 1996; Chatzisarantis and Stoica, 2009; Cooper et al., 2009;

DWYER 9781781001288 Part III.indd 461 25/07/2012 11:08

 462 Handbook of research methods in tourism

Delahaye et al., 1991; Dickersin and Berlin, 1992; Lipsey, 2007; Lipsey and Wilson, 2001;
Montori et al., 2003; Paterson et al., 2001; Rosenfeld, 2004; Rosenthal and DiMatteo,
2001; Rothstein et al., 2006; Schmidt, 2008; Shelby and Vaske, 2008; Uttl et al., 2008;
Wheatley, 2001). Meta-analysis as a method of research synthesis can be either qualita-
tive or quantitative in nature. Although the main principles are the same across quantita-
tive and qualitative meta-analyses, there are specific steps and also challenges unique to
the chosen research approach. Therefore, the following sections will discuss them sepa-
rately. However, qualitative and quantitative meta-analyses can be combined to result in
mixed-methods systematic reviews (Harden, 2010).

QUANTITATIVE META-ANALYSIS

Quantitative meta-analysis is a methodology that involves the statistical integration

of research findings from separate studies (Egger and Smith, 1997). Rosenfeld (2004)
defines it as a form of systematic review that uses statistical techniques to derive quan-
titative estimates of the magnitude of statistical effects and their associated precision.
For quantitative meta-analysis to work, the findings in the individual studies need to
be conceptually comparable (i.e., they must deal with the same constructs) and config-
ured in similar statistical form. To facilitate comparisons across studies, quantitative
meta-analysis uses a statistic that describes the strength and magnitude of an effect, the
so-called effect size. This effect size is usually standardized to account for differences in
sample sizes across the studies. The standardized effect sizes are treated as data and sub-
jected to various forms of analysis.
The practical steps involved in conducting quantitative meta-analyses are explained in
detail in Higgins and Green (2011), a free handbook available online, as well as in Lipsey
and Wilson (2001). In general, the quantitative meta-analysis starts with the problem
identification and formulation of research hypotheses. These hypotheses cover the main
effects as well as moderating effects of research design characteristics. Then, studies to be
included in the meta-analysis are gathered based on criteria determined by the research
question(s), the so-called inclusion criteria (Berkeljon and Baldwin, 2009). These studies
are then subjected to a review and eliminated if they have characteristics that exclude
them from further analysis, the so-called exclusion criteria. Inclusion/exclusion criteria
can relate to characteristics of study participants, a specific study design, the presence
of certain outcome variables, publication language, etc. It is common practice to report
not only the databases searched but also the search terms and the number of studies
included/excluded at each stage of the process.
Once the sample of studies is determined, information from the studies is extracted
in the form of codes, thus requiring a coding manual to explicitly describe the coding
procedures and facilitate inter-rater reliability. The most important variable coded is
the effect size. There are many types of effect sizes that can be selected and the choice
usually depends on the kind of data that is reported in the published primary studies and
whether one wants to investigate one-variable relationships (e.g., the proportion of trav-
ellers using social media to plan trips) or two-variable relationships (e.g., pre-post test
contrasts, group contrasts, association between variables). Even more complex is the for-
mulation of multi-variate problems (Jackson et al., 2011). Common effect size measures

DWYER 9781781001288 Part III.indd 462 25/07/2012 11:08

 Meta-analysis 463

are proportions, arithmetic means, the Pearson product moment correlation coefficient,
and odds ratios. Sampling error is smaller for effect sizes estimated from large samples
than for those estimated from small samples; thus, effect sizes are usually weighted by
their precision, with the weight typically being the inverse of the squared standard error
(inverse variance weight). Formulas for calculating various effect size types as well as a
very useful effect size decision tree to choose the correct measure are included in Lipsey
and Wilson (2001). As mentioned above, multivariate effects are difficult to deal with
but recent advances in meta-analysis statistics offer ways in which multivariate meta-
analyses can be accomplished (see Jackson et al., 2011, for details). Variables related to
the research methodology are also encoded.
The process continues with the analysis of the data in a way similar to primary research.
Analysis often starts with the elimination of outliers that could influence the results and
typically continues with the transformation of effect size measures into more convenient
forms for analysis. However, the interpretation of results might require transformations
back to more interpretable statistics. Further, specific standards against which results
can be gauged are used to guide the interpretation. Again, Lipsey and Wilsons (2001)
practical guide has specific information for transforming and interpreting the various
effect size options. The regular analysis is typically followed by a sensitivity analysis that
explores whether decisions made in the process affect the results substantially (Berkeljon
and Baldwin, 2009). The findings can then be interpreted and questions for future
research as well as practical implications can be derived.
Lipsey and Wilson (2001) describe the advantages of quantitative meta-analysis
in comparison to narrative reviews as lying in the explicit structure imposed on the
summary of research findings, the ability to code both magnitude and direction of effects,
the opportunity to relate study characteristics to effect sizes, and the ability to synthesize
a large number of studies resulting in estimates with greater statistical power. Berkeljon
and Baldwin (2009) emphasize the unique suitability of meta-analysis for answering
questions of generalizability due to its ability to aggregate findings of studies conducted
in different settings and on different samples. The robustness of properly derived meta-
analysis results also makes them especially suitable for informing policy decisions and
establishing practical guidelines.
While the comparability issue and file drawer problem apply to all meta-analyses,
they are of course especially problematic for quantitative meta-analyses that statistically
integrate findings. If known, differences in research design and measurement need to be
coded so that their effects can be analysed. To deal with the file drawer problem, which
causes studies with greater and statistically significant effects to be more likely included
because they have a greater likelihood of being published, efforts have to be undertaken
to uncover unpublished studies. One can also calculate the number of non-significant
or unavailable studies that would be necessary to change the cumulative effect size to
a nonsignificant value (fail-safe N, see Lipsey and Wilson, 2001). Selection biases can
also result from studies having to be excluded because of incomplete reporting. Another
issue can be the inclusion of a large portion of studies produced by or in collaboration
with a single author (Banas and Rains, 2010). In this case, comparisons of effect sizes of
studies involving this author and of those which do not can help with identifying such
bias. Multivariate meta-analysis has a number of limitations mostly due to estimation
difficulties because of a lack of information, additional assumptions being required, the

DWYER 9781781001288 Part III.indd 463 25/07/2012 11:08

 464 Handbook of research methods in tourism

complexity of its interpretation, and the potential exaggeration of publication biases

(Jackson et al., 2010).
Like any kind of research, meta-analysis depends on the quality of the data and the
rigour applied in the coding and analysis stages, and a thorough understanding of the
statistical assumptions made. However, it involves aggregation of data and, thus, also
aggregation of potential biases, which can be very problematic. Consequently, quanti-
tative meta-analysis require utmost care and what Lipsey and Wilson (2001) call a fish
bowl approach that renders decision processes, criteria, data and analysis and interpreta-
tion procedures explicit.

QUALITATIVE META-ANALYSIS

Timulak (2009) defines qualitative meta-analysis as rigorous secondary qualitative

analysis of primary qualitative findings (p. 591). It therefore follows the same basic
principles established for quantitative meta-analysis in terms of comprehensiveness and
systematic approaches but applies qualitative research methods to the analysis of the
studies included in the research synthesis exercise. Consequently, it is interpretive rather
than aggregative (Paterson et al., 2001). In accordance with the general goals of qualita-
tive research, qualitative meta-analysis does not aim at hypothesis testing. Rather, its
main contributions lie in (1) building of more abstract theories (meta-theory); (2) theory
explication; and (3) development of new theory (Schreiber et al., 1997). To reach these
aims requires assessing the qualitative research methodology as well as the influence
of paradigmatic parameters and theoretical assumptions that guided the researchers
(Becken, 2011). At the same time, it also needs to critically reflect on these issues in terms
of how they affected the meta-analysis itself. Therefore, it is important that the episte-
mology of a given qualitative meta-analysis and its embedding in a particular theoretical
framework is reflected by the researchers throughout the analysis and commented on in
a published account of it (Timulak, 2009, p. 593).
Challenges faced by qualitative meta-analysis involve the loss of contextual infor-
mation that is seen as essential to qualitative interpretation, which might lead to more
descriptive rather than interpretive approaches implemented in the meta-analysis.
Timulak (2009) also stresses that building on an existing body of knowledge can limit the
openness to new discoveries in meta-analysis as it increases the tension between formu-
lating a theory and assessing an existing theory.
While the overall steps involved in qualitative meta-analysis are similar to those for
quantitative analysis, there are some important differences that require highlighting.
First, while qualitative research also starts with the formulation of research questions,
these are more global than the specific hypotheses to be tested in quantitative efforts and
can actually change in the process if discoveries warrant reformulation or adding of new
questions. Second, the collection of studies is also critical but comparability is an even
greater issue for qualitative studies. In addition, it has to be decided whether quantitative
studies should be considered or not. Further, while quantitative methods can easily deal
with and actually often require large data sets, there are limits to the number of studies
that can be qualitatively analysed by a researcher. Theoretical sampling and saturation
can be used as guiding principles to deal with study selection issues. Evaluations of the

DWYER 9781781001288 Part III.indd 464 25/07/2012 11:08

 Meta-analysis 465

quality of studies to be included in the analysis are again more difficult in the qualitative
realm and need to focus on appraising studies based on their relevance to the research
question as well as the credibility of the research procedure, sampling, data analysis and
interpretation.
While quantitative meta-analysts can often calculate measures based on descriptive
data included in the publications (e.g., means and standard deviations) or can even get
access to raw data sets, it is very difficult for qualitative meta-analysts to obtain raw data.
Therefore, special care needs to be given to making sure that original meanings and con-
textual information are understood and appropriately coded. As far as the analysis of the
data is concerned, various qualitative methods approaches can be applied (Thorne et al.,
2004). An important step unique to qualitative meta-analysis involves conducting cred-
ibility checks. According to Timulak (2009), these can involve (1) auditing (an auditor
checks all steps in the meta-analysis to identify if alternative approaches and concep-
tualizations would have been possible); (2) independent analysts (researchers conduct
the entire analysis or parts of it independently and later try to reach a consensus); (3)
triangulation (use of additional, including quantitative data to complement the original
analysis); (4) validation by the primary researchers (similar to validation by participants
in primary studies); and (5) representativeness to the sample (checking how representa-
tive the findings are of the overall sample of studies included in the analysis).
The growing number of qualitative studies in tourism will likely encourage a greater
number of qualitative meta-analyses. It is important to note that rigor and making
assumptions and decisions explicit are criteria as important for qualitative meta-analyses
as they are for quantitative ones but are probably even harder to communicate.

ROLE OF META-ANALYSES IN TOURISM

Tourism is a relatively young and small field of scientific inquiry. It is also a field that
encompasses a large array of topics studied from various disciplinary backgrounds
and with a multitude of methodologies. Experimental laboratory studies, which hold
external conditions constant and often involve standardized procedures, are common
practice in many disciplines but are rare in tourism research. In addition, tourism is an
applied field with an extensive number of studies being conducted by the industry or by
research consultants. Further, there are many institutions around the world who grant
research-based master degrees in tourism that involve small individual studies rather
than the work of research groups and are usually not published. All this leads to rather
small numbers of scientific publications focusing on one topic using comparable meth-
odologies. At the same time, it also results in a large number of publications in the form
of master theses, industry reports, etc. This grey literature is often only promoted and
available within institutional boundaries and therefore usually not open to meta-analytic
inspection.
Challenges to meta-analyses in tourism are further rooted in the way scientific journals
related to tourism operate. There is a very small number of top journals in tourism and a
very large number (a longtail) of lower level journals. The top journal editors have only
recently started to demand rigorous reporting of measurement items, descriptive statis-
tics, correlation matrices and other information needed to judge and compare studies.

DWYER 9781781001288 Part III.indd 465 25/07/2012 11:08

 466 Handbook of research methods in tourism

For older publications, reporting procedures were not standardized and, thus, critical
measures necessary for quantitative meta-analyses are often not readily available. This is
still very much the case for lower level journals.
In addition, although replication is a pillar of scientific research, replication is not
rewarded in scientific journals. Journal editors and reviewers look for innovation and
often do not recognize the merit of replication to challenge or further develop existing
measures and methodologies. While this is true for many areas, it is particularly true for
tourism. Rigorous testing and re-testing of measures as is common in psychology, for
instance, does not occur in the tourism field. The file drawer problem is also a relevant
one for tourism, with studies that resulted in insignificant findings not being published
in tourism journals. Moreover, there is a general emphasis on citing only the most recent
publications in an area, often neglecting original works and the historical development
of theoretical positions.
Despite all this, the growing maturity of tourism as a field has spurred interest in
reviews of the literature and of existing research related to specific topic areas. While
most of these reviews have to date been narrative reviews, there are also some attempts
of systematic reviews, at least as far as the identification and collection of materials to
be reviewed is concerned. Ever better search technologies promise progress as far as the
identification of grey literature is concerned and the above-mentioned efforts to make
reporting in tourism-related scientific publications more complete will hopefully pave the
way for opportunities to conduct meta-analyses in the future. Better search opportuni-
ties and greater access to publications also raises the awareness of research streams and
of contradictory findings. This is very important for tourism as tourism-related research
is published in a wide array of journals spanning many disciplines.
The importance of meta-analyses for the advancement of the field is especially high
in tourism, where generalizability of results is a challenge because of the complexity
and diversity of tourism settings. Due to the lesser degree of standardization in research
methodologies, there is also a great need to inspect the effects of certain research designs
and even seemingly unimportant details such as identifying the sponsor of the research
(see Woodside and Dubelaar, 2003) on research outcomes. Weed (2006) presents a dis-
cussion of research synthesis for tourism research highlighting various application areas.
That meta-analysis in tourism is not only possible but can indeed drive theory develop-
ment, can shed light on contradictory findings and can lead to advancements in method-
ologies is explicated in the studies presented in the next section.

APPLICATIONS OF META-ANALYSES IN TOURISM

One of the research topics that has received great attention in tourism by academic and
industry researchers is the one of income multipliers. It therefore lends itself well to
meta-analysis. Baaijens et al. (1998), for example, looked at the question of why different
researchers find different multiplier values for different regions and specifically wanted
to understand the impact of using different estimation methods (standard inputoutput
model versus Archers ad hoc method). Their sample consisted of 11 multiplier values
from a set of nine studies selected based on (1) source of study; (2) year of the data collec-
tion; (3) method of research; (4) reported multiplier values; (5) features of the economy;

DWYER 9781781001288 Part III.indd 466 25/07/2012 11:08

 Meta-analysis 467

(6) features of tourism; (7) features of the environment; and (8) tourism policy. The small
sample size limited the kinds of analyses that could be performed but, nevertheless, the
results suggest that regional tourism policies and the regional external environment of
the destination influence multiplier values. As such, the study provides important direc-
tions for future research but also policy making in tourism.
Meta-analyses have also been conducted in the area of tourism demand. Brons et al.
(2002) meta-analysed 37 studies on air transport passengers. They used transfer distance,
fare class and geographic location to explain the variance among price elasticity esti-
mates and employed research method, time horizon used and period of data collection as
moderator variables. They found that long-run elasticities are higher in absolute value,
that business class passengers are less sensitive to price, and that European passengers
are not more price sensitive than their US and Australian counterparts. Crouch has
published several meta-analytic studies on tourism demand (Crouch, 1992, 1994, 1995,
1996). Crouch (1995), for instance, meta-analysed 80 studies of international tourism
demand and found that demand elasticities vary regionally in terms of both origin and
destination, suggesting that international tourism demand coefficients depend on the
pair of countries (origin and destination) of interest. Lim (1999) meta-analysed 70 studies
of international tourism demand to test the hypothesis that international travel is posi-
tively related to income in the country of origin and negatively related to both transpor-
tation costs and relative tourism prices. The hypothesized relationships were confirmed
for income and tourism prices but were inconclusive for the transportation cost variable.
Service quality is also a topic that has been studied widely in tourism and hospitality
research. Lynn and McCall (2000) conducted a meta-analysis that linked service quality
to tipping behaviour. They included both published (seven) and unpublished (six) studies
in their sample and were thus able to look at the effect based on 2547 dining parties at 20
different restaurants. They selected the restaurant as the unit of analysis rather than the
study and tip as a percentage of bill size was used as the dependent variable, which was
linked to service quality evaluations. Coefficient r was selected as the effect size measure.
The findings suggest that there is a small but reliable and positive relationship between
service evaluations and tip sizes while patronage frequency is neither a strong nor con-
sistent confound of the service-tipping relationship. De Matos et al. (2007) conducted a
meta-analysis of the service recovery paradox (a situation in which a customers post-
failure satisfaction is higher than the pre-failure satisfaction) to resolve the mixed results
that appeared in existing studies. Their sample included 21 studies with 24 observations
in hotels, restaurants and other settings. The meta-analysis used the correlation coef-
ficient as the effect size and found that there is a significant and positive service recovery
effect on satisfaction but not on repurchase intention, word of mouth and corporate
image. The research design (cross-sectional versus longitudinal), subjects (student versus
nonstudent) and service category (hotel, restaurant, other) significantly moderated the
effect on satisfaction.
There is also an abundance of conversion studies in tourism research, especially
outside the academic literature. Woodside and Dubelaar (2003) conducted a quantitative
meta-analysis of 32 published tourism-advertising studies to test whether the identifica-
tion of the sponsor influenced the response rate (H1) and the estimated conversion rate
(H2). They found support for both hypotheses, concluding that response rates are lower
and conversion rate estimates are higher for studies that identify the sponsoring brand.

DWYER 9781781001288 Part III.indd 467 25/07/2012 11:08

 468 Handbook of research methods in tourism

Website evaluation research has also been subjected to meta-analysis by tourism

researchers. Park and Gretzel (2007) meta-analysed 153 academic papers related to
e-satisfaction, e-loyalty and e-service quality. The theoretical dimensions of website
quality mentioned in these papers were extracted from the text of these papers and then
sorted into categories until nine categories that could no longer be consolidated emerged.
More recently, Law et al. (2010) and Ip et al. (2011) conducted systematic reviews of the
website evaluation literature relevant to tourism and hospitality extracting and catego-
rizing the methodologies employed in tourism website evaluation efforts.
Bloom et al.s (2009) study of vacation effects on health and well-being is a qualitative
meta-analysis that illustrates very well the importance of reporting details of the inclu-
sion and exclusion processes that lead up to the final sample selection. The research ques-
tions investigated were whether empirical evidence could be found for an improvement
of health and well-being due to a vacation, how long such an effect lasts and whether
specific vacation activities/experiences influence it. A systematic literature search was
carried out in two bibliographical databases: PsychInfo and Medline. No publication
year limits were set and the final search date was 15 June 2008. The following search
terms were used within the fields title or keywords:

Vacation or holiday (1702 hits)

Well-being or health or quality of life or satisfaction OR stress or burnout or
recovery or sleep or mood or affect (829 536 hits)

The combination of these two searches resulted in 125 hits. The first selection round
eliminated non-English papers (minus 22), and dissertations, short communications,
letters, non-empirical and/or non-peer reviewed papers (minus 38), leading to 65 remain-
ing papers. In the second round, papers not dealing with healthy, working samples
(minus 14), studies not dealing with vacation effects (minus 35), and studies not includ-
ing both a pre-test and a post-test (minus 5) were excluded, resulting in 11 remaining
papers. Finally, a third round led to the exclusion of 4 more papers. In two cases, papers
were written by the same authors, based on the same sample with the second paper not
offering extra information. A third paper was excluded because it investigated outcome
variables in spouses instead of vacationers themselves. The fourth paper was excluded
because it did not fit the definition of vacation as a voluntary period off work. The entire
process resulted in a final selection of seven studies. These studies indicate that vacations
have a weak positive effect on well-being that unfortunately does not seem to last long.
Examples of qualitative meta-analyses that aim specifically at theory explication and
theory building can also be found in the tourism literature. Sirakaya and Woodside
(2005), for instance, analyse different theoretical models used to explain travel decision-
making processes. Similarly, Becken (2011) looks at the ways in which the problem of
peak oil has been conceptualized in tourism research while Jennings (2006) critically
analyses the literature on tourism and the environment. Weed (2009), on the other hand,
conducts a meta-review of reviews on sports tourism to illustrate how unsystematic
narrative reviews are. Due to their comprehensiveness and systematic identification of
research studies, the reviews by Li et al. (2005) and Song and Li (2008) of tourism fore-
casting research can also be seen as examples of systematic reviews although they mostly
list and describe rather than synthesize existing approaches.

DWYER 9781781001288 Part III.indd 468 25/07/2012 11:08

 Meta-analysis 469

These examples show that despite the specific challenges that meta-analysis faces in
tourism, it has been successfully applied to a variety of research themes in tourism for
which significant bodies of research exist and for which a synthesis and/or critical evalua-
tion of approaches was necessary to achieve theoretical or methodological advancement.
These meta-analyses also illustrate the importance of research synthesis in tourism, not
only from a theoretical perspective but also from a practical point of view, which is very
important for an applied field like tourism.

CONCLUSION

Meta-analyses are rigourous research endeavours that seek to overcome the limitations
of individual studies by producing robust and generalizable findings or higher level/
better conceptualized/more comprehensive theories. Like all research, they have limita-
tions that need to be carefully considered. Potential biases need to be understood and if
possible eliminated or explicitly taken into account as part of the analysis. If done well,
they can make significant contributions to a field of scientific inquiry, including tourism.
Meta-analyses have been applied to tourism-specific research problems but the oppor-
tunities are far from being exhausted. Further, tourism has so far not contributed to the
development of meta-analysis as a research methodology, which is unfortunate. This
is probably partly due to a lack of awareness of the benefits of meta-analyses and their
applicability beyond standardized experimental studies such as clinical trials. Further,
some of the reviews found in the tourism realm are systematic and explicit in the data
collection process but not in other areas of the research. There is certainly some con-
fusion as to what a meta-analysis is and what it is not. The tourism-related examples
presented in this chapter clearly illustrate the need for systematic reviews in tourism
and will hopefully inspire the readers to formulate meta-analytic research questions and
collect and analyse the necessary data following the recommended steps and reporting
procedures.

REFERENCES

Baaijens, S.R., P. Nijkamp and K. Van Montfort (1998), Explanatory meta-analysis for the comparison and
transfer of regional tourist income multipliers, Regional Studies, 32 (9), 839849.
Banas, J.A. and S.A. Rains (2010), A meta-analysis of research on inoculation theory, Communication
Monographs, 77 (3), 281311.
Becken, S. (2011), A critical review of tourism and oil, Annals of Tourism Research, 38 (2), 359379.
Berkeljon, A. and S.A. Baldwin (2009), An introduction to meta-analysis for psychotherapy outcome
research, Psychotherapy Research, 19 (45), 511518.
Berman, N.G. and R.A. Parker (2002), Meta-analysis: neither quick nor easy, BMC Medical Research
Methodology, 2 (10), available at http://www.biomedcentral.com/1471-2288/2/10.
Bloom, J.d., M. Kompier, S. Geurts, C.d. Weerth, T. Taris and S. Sonnentag (2009), Do we recover from
vacation? Meta-analysis of vacation effects on health and well-being, Journal of Occupational Health, 51
(1), 1325.
Borenstein, M., L.V. Hedges, J.P.T. Higgins and H.R. Rothstein (2009), Introduction to Meta-Analysis, West
Sussex, UK: John Wiley & Sons.
Briggs, D.C. (2005), Meta-analysis: a case study, Evaluation Review, 29 (2), 87127.
Brons, M., E. Pels, P. Nijkamp and P. Rietveld (2002), Price elasticities of demand for passenger air travel: a
meta-analysis, Journal of Air Transport Management, 8 (3), 165175.

DWYER 9781781001288 Part III.indd 469 25/07/2012 11:08

 470 Handbook of research methods in tourism

Charlton, B.G. (1996), The uses and abuses of meta-analysis, Family Practice, 13 (4), 397401.
Chatzisarantis, N.L.D. and A. Stoica (2009), A primer on the understanding of meta-analysis, Psychology of
Sport and Exercise, 10 (5), 498501.
Clarke, J. (2011), What is a systematic review?, Evidence-Based Nursing, 14 (3), 64.
Cooper, H., L.V. Hedges and J.C. Valentine (eds) (2009), The Handbook of Research Synthesis and Meta-
Analysis, 2nd edition, New York: Russell Sage Foundation.
Crouch, G.I. (1992), Effect of income and price on international tourism, Annals of Tourism Research, 19 (4),
643664.
Crouch, G.I. (1994), Demand elasticities for short-haul versus long-haul tourism, Journal of Travel Research,
33 (2), 27.
Crouch, G.I. (1995), A meta-analysis of tourism demand, Annals of Tourism Research, 22 (1), 103118.
Crouch, G.I. (1996), Demand elasticities in international marketing: a meta-analytical application to tourism,
Journal of Business Research, 36 (2), 117136.
Delahaye, F., G. Landrivon, R. Ecochard and C. Colin (1991), Meta-analysis, Health Policy, 19 (23),
185196.
De Matos, C.A., J.L. Henrique and C.A.V. Rossi (2007), Service recovery paradox: a meta-analysis, Journal
of Service Research, 10 (1), 6077.
Dickersin, K. and J.A. Berlin (1992), Meta-analysis: state-of-the-science, Epidemiologic Reviews, 14 (1),
154176.
Egger, M. and G.D. Smith (1997), Meta-analysis: potentials and promise, BMJ, 315, 13711374.
Harden, A. (2010), Mixed-methods systematic reviews: integrating quantitative and qualitative findings,
National Center for the Dissemination of Disability Research Technical Brief No. 25, available at http://
www.ncddr.org/kt/products/focus/focus25/Focus25.pdf.
Hemingway, P. and N. Brereton (2009),What is a systematic review?, available at http://www.medicine.ox.ac.
uk/bandolier/painres/download/whatis/Syst-review.pdf.
Higgins, J.P.T. and S. Green (eds) (2011), Cochrane Handbook for Systematic Reviews of Interventions Version
5.1.0 [updated March 2011], The Cochrane Collaboration, available at www.cochrane-handbook.org.
Ip, C., R. Law and H.A. Lee (2011), A review of website evaluation studies in the tourism and hospitality fields
from 1996 to 2009, International Journal of Tourism Research, 13 (3), 234265.
Jackson, D., R. Riley and I.R. White (2011), Multivariate meta-analysis: potential and promise, Statistics in
Medicine, doi: 10.1002/sim.4172.
Jennings, G. (2006), A meta-analysis of current research in tourism and environment, in G. Jennings and N.P.
Nickerson (eds), Quality Tourism Experiences, Oxford: Elsevier Butterworth-Heinemann Publications, pp.
174183.
Law, R., S. Qi and D. Buhalis (2010), Progress in tourism management: a review of website evaluation in
tourism research, Tourism Management, 31 (3), 297313.
Li, G., H. Song and S.F. Witt (2005), Recent developments in econometric modeling and forecasting, Journal
of Travel Research, 44, 8299.
Lim, C. (1999), A meta-analytic review of international tourism demand, Journal of Travel Research, 37 (3),
273284.
Lipsey, M.W. (2007), Unjustified inferences about meta-analysis, Journal of Experimental Criminology, 3 (3),
271279.
Lipsey, M.W. and D.B. Wilson (2001), Practical Meta-Analysis (Vol. 49), Thousand Oaks, CA: Sage
Publications.
Lynn, M. and M. McCall (2000), Gratitude and gratuity: a meta-analysis of research on the service-tipping
relationship, Journal of Socio-Economics, 29 (2), 203214.
Montori, V.M., Swiontkowski, M.F. and D.J. Cook (2003), Methodologic issues in systematic reviews and
meta-analyses, Clinical Orthopaedics and Related Research, 413, 4354.
Park, Y. and U. Gretzel (2007), Success factors for destination marketing web sites: a qualitative meta-
analysis, Journal of Travel Research, 46 (1), 4663.
Paterson, B.L., S.E. Thorne, C. Canam and C. Jillings (2001), Meta-Study of Qualitative Health Research,
Thousand Oaks, CA: Sage.
Rosenfeld, R.M. (2004), Meta-analysis, Outcomes Research in Otorhinolaryngology, 66 (4), 186195.
Rosenthal, R. and M.R. DiMatteo (2001), Meta-analysis: recent developments in quantitative methods for
literature reviews, Annual Review of Psychology, 52 (1), 5982.
Rothstein, H.R., A.J. Sutton and M. Borenstein (2006), Publication Bias in Meta-Analysis: Prevention,
Assessment and Adjustments, West Sussex, UK: John Wiley & Sons.
Schmidt, F. (2008), Meta-analysis: a constantly evolving research integration tool, Organizational Research
Methods, 11 (1), 96113.
Schreiber, R., D. Crooks and P.N. Stern (1997), Qualitative meta-analysis, in J.M. Morse (ed.), Completing a
Qualitative Project: Details and Dialogue, Thousand Oaks, CA: Sage, pp. 311326.

DWYER 9781781001288 Part III.indd 470 25/07/2012 11:08

 Meta-analysis 471

Shelby, L.B. and J. J. Vaske (2008), Understanding meta-analysis: a review of the methodological literature,
Leisure Sciences: An Interdisciplinary Journal, 30 (2), 96110.
Sirakaya, E. and A.G. Woodside (2005), Building and testing theories of decision making by travellers,
Tourism Management, 26 (6), 815832.
Song, H. and G. Li (2008), Tourism demand modelling and forecasting: a review of recent research, Tourism
Management, 29 (2), 203220.
Thorne, S., L. Jensen, M.H. Kearney, G. Noblit and M. Sandelowski (2004), Qualitative metasynthesis: reflec-
tion on methodological orientation and ideological agenda, Qualitative Health Research, 14, 13421365.
Timulak, L. (2009), Meta-analysis of qualitative studies: a tool for reviewing qualitative research findings in
psychotherapy, Psychotherapy Research, 19 (45), 591600.
Uttl, B., B. Odegard, K. Kisinger and M. Henry (2008), Meta-analysis of meta-analysis: transparency
matters, Canadian Journal of Experimental Psychology, 62 (4), 298.
Weed, M. (2006), Undiscovered public knowledge: the potential of research synthesis approaches in tourism
research, Current Issues in Tourism, 9 (3), 256268.
Weed, M. (2009), Progress in sports tourism research? A meta-review and exploration of futures, Tourism
Management, 30 (5), 615628.
Wheatley, K. (2001), Understanding meta-analysis, Evidence-based Oncology, 2 (3), 116122.
Woodside, A.G. and C. Dubelaar (2003), Increasing quality in measuring advertising effectiveness: a meta-
analysis of question framing in conversion studies, Journal of Advertising Research, 43 (1), 7885.

DWYER 9781781001288 Part III.indd 471 25/07/2012 11:08

 25 Network analysis
Ehsan Ahmed

INTRODUCTION

We live in a social world, where, to some extent, every social object (living and non-
living) is connected with, or dependent on others. In principle, the span of this general-
ized connectedness or dependency is without limit, can integrate multiple forms and
objects, and can span over several attributes (e.g., age, race, species) and contents such
as friendship, kinship, resource, information and so on. A tourism firm, for instance,
has connections not only with its suppliers and customers but also with its surrounding
community and extended environment. Moreover, the multifaceted connections of the
tourism firm with its suppliers, customers, surrounding community, and/or environment
can be of various levels (e.g., local, regional, global), forms (e.g., informal, formal) and
interest (e.g., economic, philanthropic) (Scott, 2000). The notion of connectedness or
dependency therefore implies existence of an aggregated structure of infinite relation-
ships that extends beyond the conventional view of relationships between two objects in
isolation. Studies on networks aim to capture these aggregated views of relationships and
offer analytical insights to understand the role of relationships on the objects embedded
within these relationships.
The importance of collaboration, alliance, partnership, business clusters and similar
other concepts to better understand aggregated tourism relationships have been fre-
quently noted by a number of writers in the tourism literature. Nevertheless, to date,
little has been undertaken to examine these widely discussed tourism topics through
the lens of network analysis. On the contrary, numerous business studies in the wider
field of discipline identify the importance of networks studies on strategic alliances and
collaborative relationships for businesses (Borgatti and Foster, 2003; Lemmetyinen,
2009). Arguably, there are plausible opportunities for tourism researchers to incorporate
network analytical approaches into network studies. This chapter informs these oppor-
tunities and aims to outline a prospective research avenue for the tourism researchers
and practitioners working or willing to work with tourism networks. It attempts to
present what work has been done to study tourism networks, pointing out some of the
differences and commonalities and then discussing where future research might best be
headed. In so doing, the subsequent sections of this chapter are arranged as follows:
First, the chapter discusses the fundamental nature of network analysis techniques
and its evaluation. Second, the chapter provides the background and types of problem
that network analysis is designed to handle. This explicates the fundamental concept of
tourism relationships within a network context. In this section the author also presents
a four-fold typology of network research to classify empirical approaches to study
tourism networks. Third, a review process of the modest literature on tourism networks
is provided. Utilizing the four-fold typology and literature review, the chapter discusses
the applications of network analysis to tourism, including the discussion of empirical

472

DWYER 9781781001288 Part III.indd 472 25/07/2012 11:08

 Network analysis 473

works that have used the techniques and findings. Fourth, advantages and limitations
of network analysis are pointed out. Finally, the chapter offers a discussion concern-
ing what future directions might be taken by tourism researchers hoping to expand this
important, but understudied, field of tourism research.

NATURE OF NETWORK ANALYSIS AND ITS EVALUATION

Fundamentally, a network can be described as a set of items, called nodes, with connec-
tions between them called edges (Newman, 2003). Using this concept, network analysis,
in general, identifies relevant nodes, focuses on the connections between the nodes and
seeks to model the nature, pattern and implications of the connections between the
nodes (Scott, 2000; Wasserman and Faust, 1994). Transferring to tourism theories and
practices, tourism elements (e.g., employees, tourism firms, stakeholders, etc.) can be
presented by nodes (or actors). Corresponding relationships (e.g., monetary transaction,
communication, membership, acquaintanceship, etc.) between the tourism elements can
be presented by edges (or links). In this way, tourism nodes and corresponding edges
can be aggregated to form a network of relationships (tourism network) and thus be
employed to analyse relational properties of the tourism elements embedded within
the tourism network. Figure 25.1 depicts a small example tourism network comprising
five tourism related organizations (A, B, C, D, and E) and seven corresponding links.
In the network, for example, actor A has direct links with actors B, C, and E but does
not have any direct link with actor D. Again, as we can see in Figure 25.1, actor A has
only outgoing ties with actor B, C, and E and actor D has both outgoing and incoming
ties with actor E. Built around this common or core concept, network analysis can be
utilized to investigate and analyse the nature and pattern of links between and/or among
the tourism elements (A, B, C, D, and E) involved in different types of relationships. In
a real-life tourism perspective, thus actor A may be a holiday resort; actor B is a tour
operator, and actor C has spa and retreat facilities receiving special package deals from
the resort (actor A) to attract potential visitors. The tour operator (actor B) uses the
information and attributes of the spa and retreat facilities in its promotional materials.
Actor E is the local tourism association where both A and C are members. Actor D is the
regional tourism organization (RTO) that looks after the overall tourism development
of the region, and in turn charges a membership fee from the local tourism association.

D B

E C

Figure 25.1 A small network with five tourism organizations (nodes) and seven
corresponding links (edges)

DWYER 9781781001288 Part III.indd 473 25/07/2012 11:08

 474 Handbook of research methods in tourism

In this way a network can represent an infinite number of tourism elements (actors) and
a wide range of relationships between the tourism elements embedded in the network.
The concept of a network, at a very rudimentary stage, was a metaphor for the complex
interactions among people in a social setting. The origins of network theory and practice
may be found in research conducted in the late 1950s and 1960s on social behaviour and
exchange (Homans, 1958, 1974), the social psychology of groups (Thibaut and Kelly,
1959), exchange and power (Blau, 1964), operational research in local government
(Ward, 1964) and inter-organizational and exchange analysis (Levin and White, 1961;
Miller, 1958; Reid, 1964). From the mid-1980s onwards network analysis and theory
began to attract the attention of growing numbers of economists, planners, sociologists,
geographers, psychologists, and politicians.
In sociology, the birthplace of network analysis, a network is defined as a specific
type of relation (ties) linking a defined sets of persons, objects or events, and the sets
of persons, objects or events on which a network is defined are called actors or nodes
(Mitchell, 1969). Thus, from a sociological perspective, a network consists of a set of
nodes (usually individuals), and ties representing some relationship between the nodes.
Typically, a social network analyst would seek to model these relationships to depict
the structure of a group or the individual members belong to the group (Wasserman
and Faust, 1994). Today, there are many definitions of a network but, as pointed out by
Jarillo (1988, p. 31), many have been developed by applying this basic definition to new
areas such as the study of business organizations, internet blogs, tourism destinations,
sexually transmitted diseases, terrorist attacks, academic publications, world wide web,
or even to some extent biological systems or ecosystems.
Network concepts and theories in tourism studies date back only few years, as it
became increasingly recognized that relationships between enterprises can stimulate
inter-organizational learning and knowledge exchange, resulting in qualitative and/or
quantitative benefits to businesses, communities and destinations (Morrison et al., 2004,
p. 2). A tourism network can be understood as a set of formal and informal (social) rela-
tionships that shape collaborative actions between government, the tourism industry and
the general public (Atkinson and Coleman, 1992; Dredge, 2006b). Scott et al. (2008) have
described tourism networks as loosely articulated groups of independent suppliers linked
to deliver an overall tourism experience.

BACKGROUND AND TYPES OF PROBLEMS THAT NETWORK

ANALYSIS IS DESIGNED TO HANDLE

To date, a wide range of concepts and theories are available to study networks, with
many more to be expected in the future. Typically, the majority of these approaches are
crystallized around three interrelated network contexts. These are: (1) actors of relation-
ships, (2) content/s of relationships, and (3) form/s (nature and pattern) of relationships.
Actors of relationships refers to any entity that forms or has the ability to form con-
nections or linkages with other actors. This addresses who such as individual, group,
organization, or to some extent non-human elements (e.g., natural environment, season-
ality, etc.) having linkages or relationship with others. Content of relationships refers to
the subjective type of relation represented in the connection between actors (e.g., friend-

DWYER 9781781001288 Part III.indd 474 25/07/2012 11:08

 Network analysis 475

Table 25.1 Typology of network research

Independent Variable Dependent Variable or Outcome Focus

or Input Focus
Actor Network
Actor variable Impact of actors on other Impact of individual actors on a
actors through dyadic network
(oneto one) interactions
Network variable Impact of a network on Whole network or network-level
individual actors interactions

Source: Adapted from Provan et al. (2007).

ship, information and knowledge, power and authority, etc.). This addresses what is the
purpose of connection between the actors? Finally, form (nature and pattern) of rela-
tionships refers to the relational properties (e.g., presence, strength, direction, density,
etc.) of the connections between actors. This addresses how the connections are taking
place between and/or among actors of a network?
Although, these three contextual grounds are customary and reflect simple mecha-
nisms, they are not easy to define empirically in a tourism context. A tourism destination
network, for example, comprises of a large number of tourism elements (actors) directly
and indirectly related to tourism. These large numbers of tourism elements are again
embedded in a myriad of relationships (links) with other elements within and outside the
destination and can include several types of content, attributes, and forms of relation-
ships (e.g., information exchange, economic dependency, friendship, memberships, etc.).
Therefore, in a practical sense, the actors and their relationships within a tourism context
can be anything from a list of infinite possibilities. This leads to the question: is there
any specific research approach to study tourism networks, or alternatively, what are the
empirical bases for studying tourism networks?
Most scholars who study the topic would agree that no single grand theory of net-
works exists (Faulkner and De Rond, 2000; Galaskiewicz, 2007; Kilduff and Tsai, 2003;
Monge and Contractor, 2003). However, theorizing about networks can generally be
thought of as coming from two different but complementary perspectives: the view from
the individual element embedded in a set of relationships (actor level) and the view from
an aggregated form of analysis (network level). Wasserman and Galaskiewicz (1994)
also make this distinction, referring to a micro-level versus a macro-level network focus.
Kilduff and Tsai (2003) and Marsden (1990) refer to the important distinction between
a focus on the egocentric network versus the whole network. Building on these perspec-
tives, Provan et al. (2007) categorized the research on networks along two dimensions:
the independent variable or input focus being utilized for the study (actor variables
or network variables) and the dependent variable or outcome focus adopted by the
researcher (actor or network outcome). Using these dimensions, they also demonstrated
the possibility of four different types of network research. These are: (1) impact of
actor/s on other actor/s, (2) impact of individual actors on a network, (3) impact of a
network on individual actors, and (4) whole network or network-level interaction (see
Table 25.1).

DWYER 9781781001288 Part III.indd 475 25/07/2012 11:08

 476 Handbook of research methods in tourism

Utilizing the aforementioned typology, this chapter reviews tourism research on net-
working.

APPLICATION OF NETWORK ANALYSIS TO TOURISM

A Review of the Empirical Literature on Tourism Networks

In an effort to identify recent empirical work done on tourism networks, the author
undertook an extensive review of the literature. First, a search for web content uti-
lizing snow-ball techniques was conducted using Google Scholar. Consistent with
the broad range of definitions of networks in the tourism literature, the search terms
included tourism networks, tourism collaboration, tourism relationship, and tourism
alliances. Web content found was reviewed and a corresponding list of references
and citations were used to identify subsequent empirical works focusing on tourism
networks. This process continued until most of the publications identified were
already included in the list. For the sake and ease of access, only journal articles were
recruited. Sources such as chapters or books were excluded due to accessibility and/
or length of treatment. The author also limited the search to the 10-year period from
2001 to 2011 because most of the empirical work on tourism networks has been done
in recent years and because the study of networks is still relatively new in the tourism
discipline.
After the initial search, the author culled through the abstracts of the selected arti-
cles. Often, the abstracts provided ample information about the methods and unit of
analysis. Based on a reading of these abstracts, it was possible to eliminate the articles
that were not based on empirical data. Initially, full-length journal articles typical of
published research in a tourism networks and relationships were identified for exami-
nation. Notably, some of the articles often talked of a tourism network, but did not
study the network itself, thus ignoring the basic network theoretical insight that actors
and actor-to-actor relationships are likely to be influenced by the aggregated set of
relationships (Mitchell, 1969). For example, many articles address computer networks,
system designs, websites, and technological applications, and many dealt with the neural
network paradigm. In similar perspective, many articles used the words alliance,
relationships, collaboration, partnerships, and networks but predominantly
focused on dyadic relationship limited to only two actors in isolation. Interestingly, in
many instances, the term network is used in everyday speech without precision as a
definition of a particular phenomenon. The concept of a network has blurry edges
such as in its usage as part of a network ideology that advocates egalitarian, open com-
munities (Wellman and Berkowitz, 1988, p. 81) or confusion with the term networking
in a tourism context. This problem of definition of the term network is not confined to
tourism research and applies also to network studies in the wider management literature.
This wide usage has led to a diverse literature where the term network analysis is used
as a metaphor, homology, paradigm or method (Wellman, 1988) in different contexts.
In this chapter the definition of network used is a complex set of inter-relationships in
a social system (Mitchell, 1969) and articles that did not focus on aggregated view of
actor/s, contents, and forms of relationships were eliminated from the consideration list.

DWYER 9781781001288 Part III.indd 476 25/07/2012 11:08

 Network analysis 477

Table 25.2 Search-summary of empirical tourism-network articles obtained from

academic journals, 2001 to July, 2011

Year No. of empirical Perceived Typology Type of analysis

articles reviewed
A B C D Mixed Descriptive Analytic
2001 4 1 2 1 4(110101211)
2002 1 1 1(010101110)
2003 1 1 1(010101110)
2004 2 1 1 2(010101111)
2005 4 1 3 3(110101210) 1(010101110)
2006 9 2 2 4 1 7(210121211) 2(010101210)
2007 7 1 1 4 1 6(111101311) 1(010101110)
2008 7 1 6 5(110101410) 2(010101210)
2009 16 4 2 1 8 1 12(412111411) 4(010101410)
2010 18 5 1 2 9 1 12(511111510) 6(010111411)
2011 4 1 1 2 1(011101010) 3(110101210)
(as of June)
Total 73 16 5 5 41 6 54 19

A major limitation of the search process is the language barrier of the author. Many
articles on tourism networks are published in languages other than English, but were
not included in the final list. Once relevant articles and potentially relevant articles were
identified, the author went on to read the complete articles to make sure that each fit the
requirements of the search.
Each article that fulfilled the requirement of focusing on tourism networks was then
indexed. A summary was produced for the search result which list year, number of
articles per year, typology, and types of analysis (Table 25.2). The final list of reviewed
journal articles includes 73 empirical tourism network studies. All studies included in
the review involve some type of data collection and analysis. Although the review was
thorough, I make no claims that it is exhaustive and, despite my best efforts, I may
well have unintentionally omitted a small number of tourism network studies from the
review.

Search Findings

The article summary in Table 25.2 provides identifiable markers for comparisons of the
research being conducted in tourism network studies. Several distinct themes emerged
in the empirical research articles reviewed from tourism network literature. For one
thing, most of these studies focus on the two extreme perspectives of network research
approaches: actor level (typology A) and network level (typology D) (57 out of 73).
There is a noticeable paucity of research approaches to examine the impact of indi-
vidual actor/s on the whole network (typology B) and vice versa (typology C). Another
clear finding is that there is an emerging popularity of network studies in recent times,
indicated by the number of article published in the years 2009 (16) and 2010 (18) in
comparison to the number of articles published in the previous years. Moreover, the sole

DWYER 9781781001288 Part III.indd 477 25/07/2012 11:08

 478 Handbook of research methods in tourism

dominance of the descriptive approach to study tourism networks in the early works is
diminishing and an emergence of analytical and quantitative research approach can be
noticed in the most recent articles.
What follows is an attempt to present the selected empirical works by utilizing the
network research typology depicted in Table 25.1. In so doing, all the selected articles are
organized in four types of network research approaches. There are 16 articles in category
A, 5 articles in category B, 5 articles in category C, and 41 articles in category D. It is also
important to note here that, many of the selected articles successfully focused on more
than one typology (e.g., A and D, or A to B to D). Moreover, some of the studies largely
fall under the grey area between the actor-level and network-level boundary line of the
four typologies. Articles of these types are classified under a new category and named as
mixed typology. There are six articles identified in the mixed typology.

Typology A: Impact of Actor/s on Other Actor/s

In this type of network research, researchers often utilize characteristics and attributes of
individual actors to explain their relationships with other actors. Actor-level attributes
such as trust, economic condition, nature of business, legal structure, etc. help to explain
the nature of an actors involvement with others. These views, often referred to as ego-
centric, are concerned with trying to explain how involvement of an individual actor in
a network affects its actions and outcomes. Network research of this category in tourism
is predominantly descriptive in nature where several possible dyadic (actor-to-actor)
relationships such as alliances, partnerships, social nominations are taken into account
and analysed. Information collected from the actors in general covers a wide range of
actor-attribute information along with relational data. The majority of the empirical
works, identified as typology A, utilize in-depth interview and thick description sup-
ported by theoretical review and secondary information. The data analysis approach
used highlights actor-level information and theoretical frameworks to explain how
complex web of relations and shared concern of the tourism actors contemplates specific
tourism phenomena of interest. The analysis helps the researcher to answer several rela-
tional questions that include: (1) the impact of dyadic or network ties on organizational
performance, (2) which types of links are most or least beneficial to individual network
members, (3) which network positions might be most or least influential, and (4) how
the position of organizations in a network might shift over time in response to changes
within and outside the network (Provan et al., 2007). Identified examples of this type in
the tourism network context are as follows.
Tyler and Dinan (2001) investigate the role of interested groups (umbrella groups,
professional groups, trade groups, decision makers, and public sector representatives) in
Englands emerging tourism policy network within the context of a theoretical discussion
on public policy analysis, policy styles, and policy networks. Saxena (2005) investigated
the patterns of interaction and communication of the actors to highlight how a complex
web of relations provides relational capital for different actors to enable greater learn-
ing and co-operation, accommodate shared concerns about the local environment, and
collectively sell of the destination. His empirical works rest on 45 semi-structured inter-
views comprising tourism managers in three case study areas Castleton, Bakewell, and
Tideswell in the Peak District National Park (PDNP). Nash (2006) implements network

DWYER 9781781001288 Part III.indd 478 25/07/2012 11:08

 Network analysis 479

analysis to identify tourism issues in peripheral locations (North Scotland) and develops
an empirical model representing the aggregated views of all respondents from each study
region. Sheehan et al. (2007) discuss urban tourism promotion and explored the relation-
ships between the DMO, charged with crafting and executing destination promotion,
and its two most powerful stakeholders the city (or urban government) and hotels
(or accommodation sector). Tinsley and Lynch (2007) studies small tourism business
relations within a rural tourism destination and their contribution to destination devel-
opment. Wilkinson and March (2009) discuss the managerial application of network
research in tourism. Based on interview data from the tourism managers they offer a
method for investigating and conceptualizing network relationships between firms in the
Australian wine region of the Hunter Valley. Lemmetyinen and Go (2009) propose that
local tourism businesses must develop new key capabilities in order to face future global
competition. Their study uses case methodology and in-depth interviews to examine
organizational realities as a product of the subjective enactments or social constructions
of individual actors through the perceptions of two coordinators. The case-analysis find-
ings identify the coordination of cooperative activities in tourism business networks as
a prerequisite for enhancing the value-creation process, and building the brand-identity
process across the network. Fyrberg and Jriado (2009) extended the understand-
ing of tourism networks within the service-dominant logic. Their work uses empirical
data from a travel industry network incorporating in-depth interviews and a survey
of approximately 100 meetings professionals. They found that the network approach
provides a deeper understanding of how actors integrate with one another and how this
interaction leads to co-created outcomes that can be translated into value. Based on the
narrative accounts of 22 actors in a strategic network of organizations promoting cruise
tourism in the Baltic Sea region, Lemmetyinen (2009) analyses the extent of coordination
of cooperation within the tourism industry. The study reveals different forms of coor-
dination which enhance network development. Furthermore, it pinpoints the need for
coordinated activities on the different levels and dimensions of cooperation. Tribe (2010)
extends network study in a different but interesting tourism network context. His article
critically analyses the territories and tribes of tourism studies focusing on the culture
and practices of the academics working in the tourism discipline. Here, in his article,
actornetwork theory is deployed to link relevant objects (living and non-living) and
reveal academic networks by interviewing tourism academics. Presenza and Cipollina
(2010) examine 354 hospitality firms acting in Molise Region (Italy). Each operator in
their study was asked through an interview to judge the importance of collaborating
with other stakeholders to enhance the effectiveness of their management and market-
ing activities. Their study suggests that public stakeholders are more important for both
management and marketing activities than those in the private sector. Wegner et al.
(2010) report on in-depth interviews that explores a number of different types of partner-
ships involved in tourism and protected areas with the aim of providing recommenda-
tions to policy makers regarding how successful partnerships operate. Seven key themes
for policy recommendations and potential implication emerge in their study. Based on a
semi-structured interviews among 21 small accommodation operations around the city
of Perth (Western Australia), including self-contained, bed and breakfast, and farm-stay
businesses, Alonso (2010) examined the importance of relationships among the small
hospitality operations. The study by Weidenfeld et al. (2010) in Cornwall, England, is

DWYER 9781781001288 Part III.indd 479 25/07/2012 11:08

 480 Handbook of research methods in tourism

based on in-depth interviews with tourist attraction managers and key informants, com-
plemented by a survey of 435 tourists. Their study provides insights into the relationship
between the nature of the tourism product, spatial clustering and tourism behavior.
Zach and Racherla (2011) explore the determinants of perceived value derived from
inter-organizational collaborations in a tourism destination. Examining the tourism
organizations operating within Elkhart County, Indiana, they propose a theoretical
model of perceived value, drawing upon the rich stream of literature related to strategic
collaborations and inter-organizational networks. Their result suggests that a significant
positive value of collaboration is achieved from dyadic relationships. Importantly, the
results suggest that the positive effect achieved from one-to-one partnerships decreases
once an organization collaborates with several other organizations.

Typology B: Impact of Individual Actor/s on a Network

In this type of research, analysts utilize actor-level phenomena to explain how individual
actors and their actions might affect outcomes at the network level, such as network
structures, stability, and effectiveness. Here, the key focus is the aggregated outcomes at
the network level rather than for the individual actors that comprise the network.
In tourism, this approach is most likely to be found in the studies of tourism networks
led by destination management organizations (DMO), larger organizations, entrepre-
neurs, or local tourism authorities such as government, local community, or policy
makers. Here, the driving tourism organizations actions are likely to affect the entire
network through policy development within a destination network. Examples identified
in the literature review include work by Henriksen and Halkier (2009), Lemmetyinen and
Go (2010b), Wray (2009), Ziene and Hazel (2007).
Ziene and Hazel (2007) investigate destination networks by focusing on multiple
ownership, or portfolio entrepreneurship, when more than one small or micro business
within a specific destination is owned by the same entrepreneur. By investigating supply
side and demand side of two destinations, the authors conclude that multiple ownership
create webs of power which significantly impact on the business structure and operation
of tourism destinations. Wray (2009) demonstrates how concepts derived from the policy
community, policy network and issues management theories can be used to understand
the roles, activities and interactions of government, corporate and pressure group stake-
holders engaged in tourism policy, planning, and management in destination contexts.
The application of this approach is demonstrated through a case analysis of the tourism
policy and planning system that underpins the destination system of Byron Bay, a signifi-
cant domestic and international destination on the East Coast of Australia. Henriksen
and Halkier (2009) examine the process of policy change from local promotions towards
regional tourism policies. Adopting an institution perspective, and based on in-depth
case study, the authors conclude that the issue of localism has been effectively addressed
by establishing and operating as a network-based body where individual stakeholders
are mutually dependent on the specific capacities of their partners, a consensual style of
decision making is prevailing, and a division of labor has been established that engages
local actors in destination-wide tasks while at the same time enabling them to maintain
close links with small tourism businesses in their area. Lemmetyinen and Go (2010b)
investigate the role of a regional DMO in coordinating cooperation and enhancing the

DWYER 9781781001288 Part III.indd 480 25/07/2012 11:08

 Network analysis 481

value creation process in three tourism regions and finds that by taking active roles in
the project the DMO is able to enhance the value creation process through integrated
marketing efficiency.

Typology C: Impact of a Network on Individual Actors

In this type of network research, researchers focus at the level of the network and try to
understand the impact of network-level structures and behavior on individual actors.
Here, the fundamental assumption is that actor/s are embedded in a myriad of social
relationships, and it is impossible to understand their behavior without understand-
ing the relational context in which they function (Granovetter, 1985). In this type of
network research, the outcome or behavior of an actor is explained through structural
attributes that incorporate the interaction among the different network members (Burt,
1987).
A common theme found in tourism literature is the impact of network involvement on
organizational development, learning, or innovation. Identified examples in the tourism
literature include work by Erku-ztrk (2010), Kelliher et al. (2009), Mackellar (2006),
Reinl and Kelliher (2010), Stokes (2006).
Stokes (2006) examines the inter-organizational network that influences the event
tourism strategy of public-sector development agencies in Australia. A qualitative
methodology of convergent interviews, followed by multiple case research across six
Australian states and territories, was employed where networks of events agencies that
impact on their strategy processes for events tourism are the core focus. In a single case
study of regional festival in New South Wales, Australia, Mackellar (2006) demonstrates
the links between interactive network and innovation. The research demonstrates inno-
vation for individual organizations or business clusters as an effect of the event networks.
Kelliher et al. (2009) examine the Tourism Learning Network (TLN), a small firm
learning network model within the Filte Ireland (Irish tourism development agency).
Their findings suggest that the approaches underlying TLN facilitate the development
of organizational capabilities and result in active and substantial TLN involvement
among participating tourism enterprises. Erku-ztrk (2010) explores the role of local,
global, and associational networking, as well as company size, in the contribution of
tourism companies to the level of their creativeness. The study reveals that the level of
creativeness of a company is influenced by its level of networking at both global and
local levels, as well as its institutional thickness. Based on the research findings carried
out on another TLN, Reinl and Kelliher (2010) proposes a framework of cooperative
learning. Their research findings offer insight into how network structures, support and
interrelationships leverages organizational resources and may facilitate learning process
completion in the micro-firm environment.

Typology D: Whole Network or Network-level Interaction

The majority of the works identified by the author falls under this category of tourism
network research. Researchers using this approach choose to study the impact of multi-
level actions and structures on network level outcomes. Typically, these structural issues
are aggregated across the entire network to understand the outcome at the network level.

DWYER 9781781001288 Part III.indd 481 25/07/2012 11:08

 482 Handbook of research methods in tourism

Several distinct themes emerge in the empirical tourism network articles. Apart
from few exceptions (e.g., Zach et al., 2008), the majority of the whole network studies
in tourism focus at tourism destination(s). Specific geographic location is defined as
the tourism context of the whole network and corresponding tourism elements within
the geographic boundary are considered as the actors of the network. Here, examina-
tion of the whole destination network(s) aims to facilitate an understanding of how
a multilateral web of relationships operates within and influences a tourism region
or destination. In concert, by using a whole network approach, several interesting
tourism themes are discussed and examined in the tourism literature. For example,
Baggio et al. (2010b), Beaumonta and Dredge (2009), Cooper et al. (2009), Gibson
et al. (2005), Nordin and Svensson (2007), Pavlovich (2002), Shih (2005), Tinsley and
Lynch (2007), Wesley and Pforr (2010) discuss destination characteristics and govern-
ance; Camprub et al. (2008) and Lemmetyinen and Go (2010a) talk about destination
image and branding; Baggio and Cooper (2010), Baggio et al. (2010b), McLeod et al.
(2010), Miguns and Corfu (2008), Pavlovich (2002), Scott et al. (2008) and Xiao et
al. (2011) explore transfer of information and knowledge management; Novelli et al.
(2006), Petrou, et al. (2007) and Romeiro and Costa (2010) discuss product develop-
ment and innovation; Baggio and Corigliano (2009), Bhat and Milne (2008), Braun
(2003) focus on ICT diffusion and use; Dredge (2006b) and Pforr (2006) highlight
tourism policy networks; Baggio (2006), Baggio et al. (2010a), da Fontoura Costa and
Baggio (2009) and Mariussen et al. (2010) discuss the complexity approach by utilizing
web network data in tourism context; Fadeeva (2004), Halme (2001), Hede and Stokes
(2009), Pavlovich (2001) and Timur and Getz (2008) examine sustainable tourism
development; Anderson (2009), Brs et al. (2010), Hede and Stokes (2009), Moscardo
et al. (2009) focus on special interest tourism and tourism events; and Baggio (2007,
2011), Beesley (2005) and Erku-ztrk (2009) examine tourism collaboration and
cooperation.
A clear finding is that the identified empirical works on the whole network can
be classified into two distinct contextual grounds: (1) single network study, and (2)
multiple network study. In a single network study the network analysis is restricted
to a single network. Network analyses of this nature are again found in the literature
examining a single network based on static and/or longitudinal relational data. The
research outcome of the former approach (i.e., single network and static relational
data) explores the existing relationships within a single network and can help our
understanding of the nature and pattern of network relationships at a specific point
of time. Prominent examples include the work by Shih (2005) who tested a sample
of drive tourists for 16 destinations in Nantou, Taiwan. Travelers visiting sequence
to these 16 destinations were the network links between destinations. A number of
recommendations were made regarding the location and type of tourist facilities to be
offered and promoted based on the concurrent network analysis. In a similar fashion
Zach et al. (2008) collected empirical data from travelers who visited a regional des-
tination in the United States. The traveling sequence of the sample visitors were used
as the link between several places in the regional destination and then analysed with
respect to its network structure. In another study of this type, Baggio (2006) exam-
ined and analysed the network structure of the community of websites belonging to
Italian travel agencies. Bhat and Milne (2008), examined the network effects on des-

DWYER 9781781001288 Part III.indd 482 25/07/2012 11:08

 Network analysis 483

tination website development in New Zealand. The major part of their study is based
on the semi-structured interview of a network consisting of 35 CEOs or manage-
ment of the organizations involved in the website development. Baggio and Cooper
(2010) used a case study on Elba, Italy to illustrate the effect of network typology
on information diffusion. A static structural characterization of the network formed
by destination stakeholders was derived from the stakeholder interviews and website
link analysis. Some other prominent examples of this type (single network and static
relational data) are the works by Miguns and Corfu (2008), Baggio (2011), Baggio
and Corigliano (2009), Baggio et al. (2010a, b), Brs et al. (2010), Camprub et al.
(2008), Cooper et al. (2009), da Fontoura Costa and Baggio (2009), Lemmetyinen
and Go (2010a), Mariussen et al. (2010), Romeiro and Costa (2010), Sainaghi and
Baggio (2011).
The latter approach (i.e., single network and longitudinal relational data) collects rela-
tional data over a time period and can help to understand such issues as how networks
evolve, how they are governed, and, ultimately, how collective outcomes might be gener-
ated (Fyrberg and Jriado, 2009). Prominent examples include the work by Pavlovich
(2001, 2003) who examined tourism destination evolution and transformation of the
Waitomo Caves, New Zealand. Lisa Beesley (2005) conducted a 3-year qualitative study
of a collaborative tourism research project in Hawaii, United States. Dredge (2006b)
investigated the relationships between local government and industry to critically discuss
the role of networks. Her study rests on a case study of the local tourism networks in
Lake Macquarie over the period of 19702000. Using the Smith Beach coastal tourism,
Western Australia, and residential development as a case study, Wesley and Pforr (2010)
examined the nature of coastal tourism governance. Some other examples of this type
(single network and longitudinal relational data) are the works by Gibson et al.(2005),
Hede and Stokes (2009), Nordin and Svensson (2007), Novelli et al. (2006) and Pforr
(2006).
In contrast to single network approaches, multiple network studies are comparative
in nature. Network properties of the whole network are aggregated across an entire
network and then compared with those of other similar networks. Prominent examples
include work by Halme (2001) who studied six tourism networks in four European
countries to investigate learning towards sustainable development in multi-stakeholder
publicprivate networks; Pavlovich (2002) who explored tourism network structure
in the UK, Italy, and Egypt; Petrou et al. (2007) who focused on the role of business
networks using illustrative examples and findings from the case-study regions in the
UK, Spain, and Greece; Scott et al. (2008) examined the structural properties of inter-
organizational networks within destinations. Their paper outlined four Australian case
studies that demonstrate the utility of network analysis. Erku-ztrk (2009) examined
the role of cluster types and firm size in designing the level of network relations among
12 geographic clusters of Antalya located at the Southern part of Turkey. Beaumonta
and Dredge (2009) investigated how different tourism governance networks operate and
the effects of this governance on local tourism policy. Three local tourism networks from
Queensland, Australia, were examined in their study. These are (1) council-led network
governance structure, (2) a participant-led community network governance structure,
and (3) a local tourism organization (LTO)-led industry network governance structure.
Some other examples of this type (multiple networks) are the works by Anderson (2009),

DWYER 9781781001288 Part III.indd 483 25/07/2012 11:08

 484 Handbook of research methods in tourism

Fadeeva (2004), McLeod et al. (2010), Moscardo et al. (2009), Timur and Getz (2008),
Tinsley and Lynch (2007).

Mixed Typology

At times the network researcher may be interested to examine several levels of

network input and outcome. For example, relational data can be collected at a dyadic
level (relationship between two actors) and analysis can proceed at the network level
through aggregating the relationships between the dyads. In this way it is possible to
utilize the same network data to examine various range of network outcomes (e.g.,
dyads, triads, clusters, whole networks). Network analysis is therefore not strictly
confined within the four typologies presented in Table 25.1. Rather, a network analyst
has the flexibility to amalgamate more than one typology, switch from one typology to
another, and/or explore a wide range of network input variables and outcomes. Many
of the works found in the literature review undertake permutations and combinations
of the actor-level and network level dimensions. Prominent examples include work
by Tinsley and Lynch (2001) who examined a small tourism business network and
its contribution to destination development on the West Coast of Scotland. In their
study, relational data were first collected from the small tourism businesses through
in-depth interview (typology A). Information obtained was used to analyse network-
ing between small tourism businesses and its contribution to destination development
(typology D). Beesley (2004) examined knowledge-based networks. She explored the
emergent pattern of knowledge not only within individual, group, organizational, and
inter-organizational levels of learning, but also among them. Dredge (2006a) focused
on the theoretical and practical implications of local tourism policy networks on
collaborative planning. She examined the structures and social interactions between
several clusters relevant to destination management policy and practice of Redland
Shire located to the southeast of Brisbane, Australia. Data were collected to under-
stand collaborative planning at the individual organizational level and sub-network
level of a tourism destination. Sorensen (2007) examined the local and non-local
social networks and innovation in tourism. Relational and attribute data were col-
lected from individual tourism firms by interviewing key personnel. The analysis,
conducted both at the organizational level and network level (local and non-local),
has enhanced our understanding of different types of social network geography and
their innovation benefit. Andersson and Ekman (2009) conducted a web-based survey
with present members of ambassador (tourist) networks of five Swedish cities and
towns. Their study indentified four main dimensions of networks and, on this basis,
outlined a typology of four main categories of tourist networks. How individual influ-
ences their networks and subsequently how the networks influences other networks as
a place branding tool is the major focus of this study. Pyo (2010) examined a tourism
chain performance comprises of four attractions in Icheon City, located an hours
drive southeast from Seoul, in the Republic of Korea. Network efficiency of the four
attraction areas is measured by collecting preference data from the tourist and stra-
tegic information from the attraction managers. Analysis was undertaken both at the
individual level and tourism chain level.

DWYER 9781781001288 Part III.indd 484 25/07/2012 11:08

 Network analysis 485

ADVANTAGES AND LIMITATIONS OF NETWORK ANALYSIS

CONCEPTUALLY AND FOR POLICY FORMULATION
The notion of networks seems to be relevant in industry analysis and is gaining popular-
ity in tourism research as it has the ability to mimic and analyse the aggregated nature
of relationships between various tourism stakeholders. In particular, from a structural
point of view, tourism industry can be seen as ensembles of elements of many different
sorts, many of which are connected by some kind of relationship (Leiper, 1989). As
many authors have noted (e.g., Leiper, 2000; Pavlovich, 2001, 2003; Scott et al., 2008),
these relationships within the tourism industry are numerous, diversified in nature, and
can be regarded as an assemblage of elements with many aggregated structures forming
a unitary whole. In a similar perspective, Wilkinson and March (2009), Crotts et al.
(1998), and Smith (1988) have described the tourism industry as an amalgamation of a
wide range of suppliers clustering together to provide experiences of value to the tourist.
By using the concepts of network and corresponding analytical tools, a network analyst
can conceptualize, visualize and analyse these linkages and relationships between several
tourism elements (Scott et al., 2008). For instance, relational mapping, one of the funda-
mental network analytical tools, can provide a graphical representation of multifaceted
relationships between several tourism elements. Theoretical frameworks of this nature
have the ability to inform a wide range of relational forms such as who is connected to
whom, number and direction of connections, most or least connected elements, exist-
ence of sub-group or cluster (Iacobucci, 1996). This information can be used to improve
connectivity by intervening to mend broken or indirect links, or reconfiguring the aggre-
gated relationships to be more efficient. Here, the diagnostics can aid several tourism
phenomena such as collaboration and innovation (Cooper and Hall, 2008), information
and knowledge management (Baggio and Cooper, 2010; Cooper 2006), policy and plan-
ning (Dredge, 2006b; Pforr, 2004, 2006), destination governance (Pavlovich, 2001, 2003),
sustainable development (Halme, 2001).
Although there is a wide range of possibilities for utilizing network analysis in tourism
studies, the network literature is far less extensive than many others in the wider litera-
ture on tourism research. Indeed, the broadest conclusion that can be drawn from the
literature review of this chapter is that there is simply not very much of it. Numerous
business studies discuss the importance of networks, strategic alliances and collabora-
tive relationships for businesses. Despite such a rich body of knowledge, however, such
discussion appears to have fallen short of incorporating network analysis in tourism
research in a serious way. In fact, tourism studies exploring these dimensions are few
and far between. What follows is a discussion of what the author believes as the primary
limitation and methodological issues that must be overcome if significant progress is to
be made on the topic.
As noted earlier, network analysis depends on relational rather than attribute data.
Here, the presence and nature of relationships between actors is the focus, rather than
the characteristics of each individual actor. Because of this focus on relationships, the
techniques used to analyse networks differ substantially from conventional survey
research methods and require supplementary knowledge and skills. Suppose we are
studying economic ties among several tourism stakeholders within a destination. For
example, an accommodation provider has been selected to be in our sample. When we

DWYER 9781781001288 Part III.indd 485 25/07/2012 11:08

 486 Handbook of research methods in tourism

ask the accommodation provider, he/she identifies strong economic ties with seven stake-
holders. We need to track down each of those seven stakeholders and ask them about
their economic ties, as well. The seven stakeholders are in our sample because the accom-
modation provider is (and vice versa), so the sample elements are no longer independ-
ent. This means that actors are usually not sampled independently in network studies,
as in many other kinds of studies (most typically, surveys). If one actor happens to be
selected, then we must also include all other actors to whom our sample has (or could
have) ties. As a result, network approaches tend to study whole populations by means of
census, rather than by sample.
Another limitation of the network analysis can be pointed out as the boundary speci-
fication issue, i.e., the managerial and theoretical perspectives to distinguish whom to
consider and whom not to consider within a network (Laumann et al. 1989). This is a
challenging issue for any network studies both in tourism and non-tourism contexts. For
example, a tourist attraction of any destination is connected to a wide range of support
organizations that are directly and indirectly related to tourism. Here, the relationship
between the attraction and the local tourism organization (LTO) is explicit and easily
identifiable. They may have a formal membership program, belong to the same industry,
work with similar types of customer product and value, share resources and knowledge,
or may even organize promotions and events together. On the contrary, the relation-
ship between the very same attraction and any other non-tourism organization such
as road transport authority or fire service of that destination is significant but hard to
relate in a tourism perspective. Similarly, two attractions within the same destination
may compete with one another and may not have direct or identifiable relationship
in between. However, they might influence one another through the invisible string of
competition. Therefore, a network study may involve any number of actors, elements
and type of relationships from a list of infinite possibilities. Nevertheless, in a practi-
cal sense, it is not possible to consider every element and all possible types of relational
content. The network researcher has to stop somewhere in the recruitment process and
must draw a boundary line to define the size of the network. The problem may also arise
if the researcher is too stringent on the boundary issue. Strict boundary specification
may cause omission of pertinent elements of the network. For example, membership
network of local tourism authority may exclude non-tourism organizations (e.g., local
chambers of commerce, SMEs, donor agencies, public hospitals, etc.) having significant
contributions to the overall tourism outcome of the destination. Such strict delineation
process may create confusion and can lead to misleading results (Albert & Barabsi,
2002; Newman, 2003). Fortunately, several boundary specification approaches have
been suggested in the wider literature of network studies. Every approach has its own set
of advantages and disadvantages. Interested readers may wish to review Laumann et al.
(1989), Marsden (1990), and Newman (2003) for further reading.
Another limitation of the network study is the time, cost and difficulties to access
network data. As discussed before, collecting network data is not a random process
as actors of the network members are not independent unit of analysis. Therefore, in
many instances, network researchers have to seek information from the respondents
that are difficult to access or not willing to participate in the research. For example, an
external event management firm may organize a single annual event in a destination and
works with a large number of local tourism stakeholders during the event. Realizing

DWYER 9781781001288 Part III.indd 486 25/07/2012 11:08

 Network analysis 487

the economic benefit of the event, a large number of actors (stakeholders) may indicate
a relationship with the event management firm. In this case, the network researcher
is required to include the event management firm within the destination network and
may even have to find the event management firm, or in extreme case, have to wait for
the event to take place in the next year! Again, depending on the characteristics of the
tourism context, the network researcher may have to include more than one network and
multiple tourism elements ranging from individual actor perspectives to a large number
of aggregated relationships. For example, a simple tourism network study may include
50 actors (accommodation, transportation, attractions, etc.), five relational contents
(acquaintanceship, money, information, etc.), and on an average ten relationships per
actor. Studying this network means a total of 2500 (50 5 10) data points for analysis.
Trying to conduct a longitudinal study with this simple network of 50, the researcher will
add 2500 additional data points (or more) for every cycle of data collection, a daunting
task. In addition to network data, if the network researcher is also interested in collecting
actors attribute data and if there are ten attributes per actor, this will add another 500 (50
10) data points for the analysis. Dealing with this large number of data is difficult and
at times network researchers are bound to compromise valuable actor to actor relational
data due to resource and time constraints.
Finally, networks are defined by their actors and the connections among them. Here,
the individual connections between two actors are aggregated to form the network. In
other words, connections between the actors are the fundamental building blocks of a
network. However, there may be many connections between two actors in a network.
For example, tourism stakeholders exchange personnel, money, information, and
form groups and alliances. Relations among two destinations may be characterized
by numerous forms of culture, economic, and political exchange. However, if we
consider how the relation between two actors may influence their behavior or provide
each with opportunity and constraint, it may well be the case that not all of these ties
matter. Among many of the possible relational contents, it is always difficult to identify
the most appropriate or a single element of connection that can provide reasonable
research validity and reliability for all the network actors. For instance, within a set of
tourism stakeholders, the monetary relationship between an accommodation provider
and transportation provider may be more significant than the information ties, and
exchange of employees may not come into consideration as these two types of organi-
zations usually require different type of human expertise. Whereas, the relationships
between an accommodation provider and destination organization within the same
network the information ties are more meaningful than the monetary ties as they
predominantly exchange information and knowledge. A reasonable solution to this
problem may be to construct several layers of network containing one type of relation-
ship in each layer (multilevel networking). Again, in such case, the problem is to decide
how many network layers are appropriate to measure relationships across a set of
actors and how these layers are going to overlap with one another. Moreover, a single
relational attribute (e.g., money, information) may contain several sub-level measures
such as strength of relations, direction of relations, frequency of relations, and so on.
Therefore, despite the core element of network analysis, it is always difficult to define
relationship(s) as it contains a wide range of hidden issues such as form, measures, type,
reliability, and validity.

DWYER 9781781001288 Part III.indd 487 25/07/2012 11:08

 488 Handbook of research methods in tourism

FUTURE DEVELOPMENT AND APPLICATION OF NETWORK

ANALYSIS IN TOURISM RESEARCH
It is useful to build on what we know and do not know to highlight those areas where
future researchers might most productively focus their efforts in tourism network analy-
sis. Consistent with the themes of the literature review, we can identify two broad areas
of tourism network research in the future: (1) network properties and processes and (2)
network development and evolution.

Network Properties and Process

Fundamentally, network analysis can be seen as a process that is used to collect and
analyse relational data. In network research, the presence and nature of relationships
among the actors is the focus (Knoke and Kuklinski, 1991) and the behavior of actors
is seen as a function of their relationship within a network (Mizruchi and Marquis,
2006). In essence and as noted earlier, any network research is based on three contextual
grounds: (1) actor(s) of the network, (2) content of relationship, and (3) form of relation-
ships. Even though these three contextual grounds are the ingredients of any fundamen-
tal network research setting, tourism network research, in many instances, has tended to
neglect these issues. Complete research design, defining and validating actors, contents
and forms of relationship, is sparse in the literature reviewed. As a consequence, there
is a relative dearth of work on empirically informed explanation of the network settings
used to examine tourism issues. It is therefore suggested that a systematic research design
to study tourism networks is essential and should be further explored, developed, and
utilized.
Network analysis seeks to model the structure of relationships, focusing in particular
on the nature and pattern of relationships between network elements (Wasserman and
Faust, 1994). Based on this fundamental concept, the utility and implication of network
structure is widely discussed in the tourism network literature. Indeed, it is clear from the
literature that structural analysis on tourism networks has the ability to inform a wide
range of tourism issues and research agenda such as policy development and implemen-
tation, knowledge management, destination governance (for a review, see Scott et al.,
2008). Nonetheless, what is less clear is how to capture the nature and pattern of relation-
ships in quantitative terms that mimics the effects of relationships in tourism. Using thick
description in network analysis is manageable in case of small networks (e.g., network of
less than 25 to 40 actors). However, with a larger network (e.g., more than 100 actors),
it becomes increasingly difficult to manage. For example, let us compare two networks
having 5 and 50 actors, respectively. The network with 5 actors can have a maximum of
20 (5 3 (5 2 1)) possible links. In contrast, the larger network with 50 members can have
a maximum of 2450 (50 3 (50 2 1)) links. Similarly, a network with 100 actors can have
a maximum of 9900 (100 3 (100 2 1)) links, and a network with 1000 actors can have a
maximum of 999 000 (1000 3 (1000 2 1)) links, and so on. Scary but very few of the
tourism networks in reality have less than 100 actors!
Many of the recent works on tourism networks have attempted to employ quan-
titative approaches and there is an emerging trend to utilize quantitative analysis
to understand tourism networks. For example, in the literature review, 14 of the 18

DWYER 9781781001288 Part III.indd 488 25/07/2012 11:08

 Network analysis 489

quantitative studies using network analysis in tourism have appeared since 2008 (see
Table 25.2). This is a welcome change. However, despite the strong methodological
foundation, the majority of the quantitative approaches reviewed have paid very little
or no attention to descriptive network analysis. Although there is hardly any serious
alternative to quantitative methods for studying large-scale networks, additional
insights into the structure and content of relationships, their development over time,
the initial conditions at the formation stage of the network, and changing contexts
could be gained by the additional use of qualitative methodologies such as narrative
interviews, case study and participant observation. Nevertheless, from the literature
review, it can be argued that there is a generalized tendency of the tourism network
researcher to utilize only one of the approaches rather than using both. Quantitative
and qualitative approaches in network studies might complement each other and can
deliver a number of important insights which can be used to analyse tourism networks
in the future.
Tourism outcomes can be conceptualized as aggregated activities of several individual
stakeholders directly and indirectly related to the tourism industry. On this perspective,
network study is an ideal research avenue for tourism research as it has the ability to
represent and analyse the aggregated view of links and connections. Indeed, the concepts
and techniques of network analysis can be used as a metaphor to represent the tourism
industry both theoretically and practically. For example, any tourism stakeholder can
be represented by a node and several forms and types of relationships of that stake-
holder with others can be represented by links. The total combination of the nodes
and links then can be represented by a network and analysed by using network tools
and techniques. However, compared to the wider network literature, to date, a small
portion of the network tools and techniques is discussed and utilized in the tourism
network research. Despite all we know at this point regarding certain aspects of tourism
network research, there are many perspectives that have not been adequately addressed
in tourism network studies. Some important but under addressed research avenues for
future tourism network research can be stated as:

1. Overlapping and cross examining actors relational data (e.g., friendship, monetary,
information, power, influence, etc.) and attribute data (e.g., occupation, age, size,
economic condition, etc.).
2. Integrating, overlapping and cross examining multiple content of relationships such
as friendship, money and resource exchange, information and knowledge sharing,
flow of tourists, etc.
3. Examining correlation between several content and form of relationship such as how
the presence of one relationship (e.g., friendship) influences the presence (or absence)
of another relationship (e.g., information exchange).
4. Utilization of valued and directional data. For example, relationships can be meas-
ured in ordinal scale starting from very strong relationship to very weak relationship
or no relationship; a relationship can also be directional by indicating who reports
to (or does not report) to whom, etc.
5. Examining affiliation network effects such as how the choices of individual tourism
actors to join a specific group, community, and/or activity may affect the choices of
the individual actors.

DWYER 9781781001288 Part III.indd 489 25/07/2012 11:08

 490 Handbook of research methods in tourism

Finally, the literature review of this chapter indicates that the influence of an indi-
vidual organization on the whole network (e.g., influence of the state tourism organiza-
tion on a specific tourism region) and vice versa (e.g., influence of the accommodation
association on a small bed and breakfast hotel) are rarely examined and discussed in the
tourism literature (see Table 25.2, 10 out of 73). The author believes that it is a potential
ground of future network research in tourism.

Network Development and Evolution

A major theme for future research direction is that of network development and evolu-
tion. Although there has been considerable discussion on tourism network development
(e.g., destination governance by Pavlovich (2003, 2008), evolution of tourism policy
networks by Dredge (2006b), Pforr (2004, 2006), there has been scant discussion on the
process of tourism network development, such as how network structures evolve over
time and how or if these multilateral relationships are managed. The studies of tourism
network development, evolution and governance, have suggested a number of directions
for the study of networks: How do networks evolve from early birth to maturity and
beyond? Does evolution occur in predictable ways, either in specific evolutionary stages
or based on environmental conditions and internal pressures and changes? Are there
critical pre-network activities (e.g., presence of a strong relationship between the early
stakeholders) and structures that predict successful network evolution? Do networks
continually shift and evolve in significant ways, or does network stability emerge at some
point as an important factor for explaining network success? How does geography and
climate change influence the evolution of network? To what extent can a whole network
be stable (or dynamic) despite significant changes at the cluster or actor level or adop-
tion of technology? What, then, is the impact of network structural characteristics on its
development?
One of the main ways in which all forms of network research have changed during
recent years is that longitudinal research has become much more prevalent, opening the
way to in-depth consideration of network development (Provan et al., 2007). However,
there is only a limited amount of such studies found in the literature review (Pavlovich,
2001, 2003, 2008; Dredge 2006a, b). Moreover, the majority of the longitudinal studies
in tourism networks are limited to a single network. Whereas, longitudinal studies with
multiple (more than one networks among the same group of actors) network approach is
yet an untouched research avenue in tourism network research.

CONCLUSION

The chapter has provided an overview of studies of tourism networks, reviewing those
empirical studies carried out during the past 10 years. In contrast to the abundant
tourism research, especially with a focus on tourism collaboration, cooperation, alliance,
and inter-organizational relationships, network research is still of a manageable size for
a review chapter like this. On the more critical side, research on tourism networks has,
so far, left many important questions unanswered. Apart from giving an overview of
empirical studies of tourism networks, another aim of this chapter has been to raise such

DWYER 9781781001288 Part III.indd 490 25/07/2012 11:08

 Network analysis 491

important questions, especially in regard to network properties and processes, network

development and evolution, and network methodology in tourism.
Despite all the efforts the author has made to present a complete review of tourism
network research, the study has clear limitations. Admittedly, the search procedure used
was somewhat subjective, so that some studies that might be considered to be impor-
tant in the topic area may not have been included. The search process was narrowly
focused but serves to point out the relative dearth of empirical literature on the topic.
Nevertheless, the limited set of empirical works were investigated in a way that should
provide useful insights for future researchers choosing to focus on tourism networks or to
include this important analytical approach in even more in-depth network implications.

REFERENCES

Albert, R. and A.L. Barabsi (2002), Statistical mechanics of complex networks, Review of Modern Physics,
74, 4797.
Alonso, A.D. (2010), Importance of relationships among small accommodation operations around the city of
Perth, Tourism and Hospitality Research, 10, 1424.
Anderson, W. (2009), Promoting ecotourism through networks: case studies in the Balearic Islands, Journal
of Ecotourism, 8, 5169.
Andersson, M. and P. Ekman (2009), Ambassador networks and place branding, Journal of Place
Management and Development, 2, 4151.
Atkinson, M.M. and W.D. Coleman (1992), Policy networks, policy communities and the problems of govern-
ance, Governance, 5, 154180.
Baggio, R. (2006), Complex systems, information technologies and tourism:a network point of view,
Information Technology and Tourism, 8, 1529.
Baggio, R. (2007), The web graph of a tourism system, Physica A, 379, 727734.
Baggio, R. (2011), Collaboration and cooperation in a tourism destination: a network science approach,
Current Issues in Tourism, 14, 183189.
Baggio, R. and M. Antonioli Corigliano (2009), On the importance of hyperlinks: a network science
approach, Information and Communication Technologies in Tourism, 2009, 309318.
Baggio, R. and C. Cooper (2010), Knowledge transfer in a tourims destination: the effect of a network struc-
ture, The Service Industries Journal, 30, 17571772.
Baggio, R., N. Scott and C. Cooper (2010a), Improving tourism destination governance: a complexity science
approach, Tourism Review, 65, 5160.
Baggio, R., N. Scott and C. Cooper (2010b), Network science: a review focused on tourism, Annals of Tourism
Research, 37, 802827.
Beaumonta, N. and D. Dredge (2009), Local tourism governance: a comparison of three network approaches,
Journal of Sustainable Tourism, 18, 728.
Beesley, L. (2004), Multi-level complexity in the management of knowledge networks, Journal of Knowledge
Management, 8, 71100.
Beesley, L. (2005), The management of emotion in collaborative tourism research settings, Tourism
Management, 26, 261275.
Bhat, S.S. and S. Milne (2008), Network effects on cooperation in destination website development, Tourism
Management, 11311140.
Blau, P.M. (1964), Exchange and Power in Social Life, New York: Wiley.
Borgatti, S.P. and P.C. Foster (2003), The network paradigm in organisational research: a review and typol-
ogy, Jounrnal of Management, 29, 9911013.
Brs, J.M., C. Costa, D. Buhalis, N. Scott and E. Laws (2010), Network analysis and wine routes: the case of
the Bairrada Wine Route, Service Industries Journal, 30, 16211641.
Braun, P. (2003), Regional tourism networks: the nexus between ICT diffusion and change in Australia,
Information Technology and Tourism, 6, 231243.
Burt, R.S. (1987), Social contagion and innovation: cohesion versus structural equivalence, American Journal
of Sociology, 92, 12871335.
Camprub, R., J. Guia and J. Comas (2008), Destination networks and induced tourism image, Tourism
Review, 63, 4758.

DWYER 9781781001288 Part III.indd 491 25/07/2012 11:08

 492 Handbook of research methods in tourism

Cooper, C. (2006), Knowledge management and tourism, Annals of Tourism Research, 33, 4764.
Cooper, C. and M. Hall (2008), Contemporary Tourism: An International Approach, Amsterdam: Elsevier.
Cooper, C., N. Scott and R. Baggio (2009), Network position and perception of destination stakeholder
importance, Anatolia, 20, 3345.
Crotts, J., A. Aziz and A. Rashid (1998), Antecedents of suppliers commitment to wholesale buyers in the
international travel trade, Tourism Management, 19, 127134.
da Fontoura Costa, L. and R. Baggio (2009), The web of connections between tourism companies: structure
and dynamics, Physica A: Statistical Mechanics and its Applications, 388, 42864296.
Dredge, D. (2006a), Networks, conflict and collaborative communities, Journal of Sustainable Tourism, 14,
562581.
Dredge, D. (2006b), Policy networks and the local organisation of tourism, Tourism Management, 27,
269280.
Erku-ztrk, H. (2009), The role of cluster types and firm size in designing the level of network relations: the
experience of the Antalya tourism region, Tourism Management, 30, 589597.
Erku-ztrk, H. (2010), The significance of networking and company size in the level of creativeness of
tourism companies: Antalya case, European Planning Studies, 18, 12471266.
Fadeeva, Z. (2004), Translation of sustainability ideas in tourism networks: some roles of cross-sectoral net-
works in change towards sustainable development, Journal of Cleaner Production, 13, 175189.
Faulkner, D.O. and M. De Rond (2000), Perspectives on cooperative strategy, Cooperative Strategy:
Economic, Business, and Organizational Issues, 1.
Fyrberg, A. and R. Jriado (2009), What about interaction? Networks and brands as integrators within
service-dominant logic, Journal of Service Management, 20, 420432.
Galaskiewicz, J. (2007), Has a network theory of organizational behaviour lived up to its promises?,
Management and Organization Review, 3, 118.
Gibson, L., P.A. Lynch and A. Morrison (2005), The local destination tourism network: development issues,
Tourism Planning & Development, 2, 8799.
Granovetter, M. (1985), Economic action and social structure: the problem of embeddedness, American
Journal of Sociology, 91, 481510.
Halme, M. (2001), Learning for sustainable development in tourism networks, Business Strategy and the
Environment, 10, 100114.
Hede, A.-M. and R. Stokes (2009), Network analysis of tourism events: an approach to improve marketing
practices for sustainable tourism, Journal of Travel & Tourism Marketing, 26, 656669.
Henriksen, P.F. and H. Halkier (2009), From local promotion towards regional tourism policies: knowledge
processes and actor networks in North Jutland, Denmark, European Planning Studies, 17, 14451462.
Homans, G.C. (1958), Social behaviour as exchange, American Journal of Sociology, 63, 597606.
Homans, G.C. (1974), Theories and Paradigms in Contemporary Sociology, Itasca, IL: F.E. Peacock Publishers.
Iacobucci, D. (1996), Networks in Marketing, Thousand Oaks, CA: Sage Publications.
Jarillo, J.C. (1988), On strategic networks, Strategic Management Journal, 9, 3141.
Kelliher, F., A. Foley and A. Frampton (2009), Facilitating small firm learning networks in the Irish tourism
sector, Tourism and Hospitality Research, 9, 8095.
Kilduff, M. and W. Tsai (2003), Social Networks and Organisations, London: Sage Publications.
Knoke, D. and J.H. Kuklinski (1991), Network analysis: basic concepts, in G. Thompson, J. Frances, R.
Levacic and J. Mitchel (eds), Markets, Hierarchies and Networks, Lodon: Sage Publications, pp. 173182.
Laumann, E.O., P.V. Marsden and D. Prensky (1989), The boundary specification problem in network analy-
sis, in L.C. Freeman, D.R. White and K.A. Romney (eds), Research Methods in Social Network Analysis,
Fairfax, VA: George Mason University Press.
Leiper, N. (1989), Tourism and tourism system, Vol. Occasional Paper No. 1, Palmerston North Massey
University.
Leiper, N. (2000), An emerging discipline, Annals of Tourism Research, 27, 805809.
Lemmetyinen, A. (2009), The coordination of cooperation in strategic business networks: the Cruise Baltic
Case, Scandinavian Journal of Hospitality and Tourism, 9, 366386.
Lemmetyinen, A. and F.M. Go (2009), The key capabilities required for managing tourism business net-
works, Tourism Management, 30, 3140.
Lemmetyinen, A. and F.M. Go (2010a), Building a brand identity in a network of Cruise Baltics destinations:
a multi-authoring approach, Journal of Brand Management, 17, 519531.
Lemmetyinen, A. and F.M. Go (2010b), The role of the DMO in creating value in EU-funded tourism
projects, Scandinavian Journal of Hospitality and Tourism, 10, 129152.
Levin, S. and P.E. White (1961), Exchange as a conceptual framework for the study of interorganisational
relationships, Administrative Science Quarterly, 5, 583601.
Mackellar, J. (2006), An integrated view of innovation emerging from a regional festival, International
Journal of Event Management Research, 2, 3748.

DWYER 9781781001288 Part III.indd 492 25/07/2012 11:08

 Network analysis 493

Mariussen, A., R. Daniele, D. Bowie, N. Scott and E. Laws (2010), Unintended consequences in the evolution
of affiliate marketing networks: a complexity approach, Service Industries Journal, 30, 17071722.
Marsden, P.V. (1990), Network data and measurement, Annual Review of Sociology, 16, 435463.
McLeod, M.T., D.R. Vaughan and J. Edwards (2010), Knowledge networks in the tourism sector of the
Bournemouth, Poole, and Christchurch conurbation: preliminary analysis, The Service Industries Journal,
30, 16511667.
Miguns, J. and A. Corfu (2008), e-Destination structure: a network analysis approach, Information and
Communication Technologies in Tourism 2008, 580591.
Miller, W.B. (1958), Inter-institutional conflict as a major impediment to delinquency prevention, Human
Organisation, 17, 2023.
Mitchell, J.C. (1969), The concept and use of social networks, in J.C. Mitchell (ed.), Social Network in Urban
Situations, Manchester: University of Manchester Press, pp. 150.
Mizruchi, M.S. and C. Marquis (2006), Egocentric, sociometric, or dyadic? Indentifying the appropriate level
of analysis in the study of organisational networks, Social Networks, 28, 187208.
Monge, P.R. and N.S. Contractor (2003), Theories of Communication Networks, Oxford: Oxford University
Press.
Morrison, A., P. Lynch and N. Johns (2004), International tourism network, International Journal of
Contemporary Hospitality Management, 16, 197202.
Moscardo, G., B. McCarthy, L. Murphy and P. Pearce (2009), The importance of networks in special interest
tourism: case studies of music tourism in Australia, International Journal of Tourism Policy, 2, 523.
Nash, R. (2006), Casual network methodology: tourism research application, Annals of Tourism Research,
33, 918938.
Newman, M.E.J. (2003), The structure and function of complex networks, SIAM Review, 45, 167256.
Nordin, S. and B. Svensson (2007), Innovative destination governance: the Swedish ski resort of Are, The
International Journal of Entrepreneurship and Innovation, 8, 5366.
Novelli, M., B. Schmitz and T. Spencer (2006), Networks, clusters and innovation in tourism: a UK experi-
ence, Tourism Management, 27, 11411152.
Pavlovich, K. (2001), The twin landscapes of Waitomo: tourism network and sustainability through the land-
scape group, Journal of Sustainable Tourism, 9, 491504.
Pavlovich, K. (2002), Pyramids, pubs, and pizzas: an interpretation of tourism network structures, Tourism
Culture and Communication, 4, 4148.
Pavlovich, K. (2003), The evoluation and transformation of a tourism destination network: the Waitomo
Cave, New Zealand, Toursim Management, 2, 203216.
Pavlovich, K. (2008), Network governance and connectivity: a case study, in N. Scott, R. Baggio and C.
Cooper (eds), Network Analysis and Tourism From Theory to Practice, England: Channel View Publication.
Petrou, A., E.F. Pantziou, E. Dimara and D. Skuras (2007), Resources and activities complementarities: the
role of business networks in the provision of integrated rural tourism, Tourism Geographies, 9, 421440.
Pforr, C. (2004), The makers and shapers of tourism policy in the Northern Territory of Australia: a policy
network analysis of actors and their relational constellations, Journal of Hospitality and Tourism Research,
9, 134151.
Pforr, C. (2006), Tourism policy in the making an Australian network study, Annals of Tourism Research,
33, 87108.
Presenza, A. and M. Cipollina (2010) , Analysing tourism stakeholders networks, Tourism Review, 65, 1730.
Provan, K.G., A. Fish and J. Sydow (2007), Interorganisational networks at the network level: a review of the
empirical literature on whole networks, Journal of Management, 33, 479516.
Pyo, S. (2010), Measuring tourism chain performance, The Service Industries Journal, 30, 16691682.
Reid, W. (1964), Interagency coordination in delinquency prevention and control, Social Service Review, 38,
418428.
Reinl, L. and F. Kelliher (2010), Cooperative micro-firm strategies leveraging resources through learning
networks, Entrepreneurship and Innovation, 11, 141150.
Romeiro, P. and C. Costa (2010), The potential of management networks in the innovation and competitive-
ness of rural tourism: a case study on the Valle del Jerte (Spain), Current Issues in Tourism, 13, 7591.
Sainaghi, R. and R. Baggio (2011), Complex and chaotic tourism systems: towards a quantitative approach,
International Journal of Contemporary Hospitality Management, 23, 77.
Saxena, G. (2005), Relationships, networks and the learning regions: case evidence from the Peak District
National Park, Tourism Management, 26 (2), 277289.
Scott, J. (2000), Social Network Analysis a Handbook (2nd edn), London: Sage Publications.
Scott, N., C. Cooper and R. Baggio (2008), Destination networks: four Australian cases, Annals of Tourism
Research, 35, 169188.
Sheehan, L., J.R.B. Ritchie and S. Hudson (2007), The destination promotion triad: understanding asymmet-
ric stakeholder interdependencies among the city, hotels, and DMO, Journal of Travel Research, 46, 6474.

DWYER 9781781001288 Part III.indd 493 25/07/2012 11:08

 494 Handbook of research methods in tourism

Shih, H.-Y. (2005), Network characteristics of drive tourism destinations: an application of network analysis
in tourism, Tourism Management, 27, 10291039.
Smith, S.L.J. (1988), Defining tourism a supply-side view, Annals of Tourism Research, 15, 179190.
Sorensen, F. (2007), The geographies of social networks and innovation in tourism, Tourism Geographies, 9,
2248.
Stokes, R. (2006), Network-based strategy making for events tourism, European Journal of Marketing, 40,
682695.
Thibaut, J. and H.H. Kelly (1959), The Social Psychology of Groups, New York: John Wiley.
Timur, S. and D. Getz (2008), A network perspective on managing stakeholders for sustainable urban
tourism, International Journal of Contemporary Hospitality Management, 20, 445461.
Tinsley, R. and P. Lynch (2001), Small tourism business networks and destination development, International
Journal of Hospitality Management, 20, 367378.
Tinsley, R. and P.A. Lynch (2007), Small business networking and tourism destination development: a com-
parative perspective, The International Journal of Entrepreneurship and Innovation, 8, 1527.
Tribe, J. (2010), Tribes, territories and networks in the tourism academy, Annals of Tourism Research, 37,
733.
Tyler, D. and C. Dinan (2001), The role of interested groups in Englands emerging tourism policy network,
Current Issues in Tourism, 4, 210252.
Ward, R.A. (1964), Operational Research in Local Government, London: Allen and Unwin.
Wasserman, S. and K. Faust (1994), Social Network Analysis, New York: Cambridge University Press.
Wasserman, S. and J. Galaskiewicz (1994), Advances in Social Network Analysis: Research in the Social and
Behavioral Sciences, Thousand Oaks, CA: Sage Publications, Inc.
Wegner, A., D. Lee and B. Weiler (2010), Important ingredients for successful tourism/protected area part-
nerships: partners policy recommendations, The Service Industries Journal, 30, 16431650.
Weidenfeld, A., R.W. Butler and A.M. Williams (2010), Clustering and compatibility between tourism attrac-
tions, International Journal of Tourism Research, 12, 116.
Wellman, B. (1988), Structural analysis: from method and metaphor to theory and substance, in B. Wellman
and S.D. Berkowitz (eds), Social Structures: A Network Approach, Cambridge: Cambridge University Press.
Wellman, B. and S.D. Berkowitz (1988), Social Structures: A Network Approach. Cambridge: Cambridge
University Press.
Wesley, A. and C. Pforr (2010), The governance of coastal tourism: unravelling the layers of complexity at
Smiths Beach, Western Australia, Journal of Sustainable Tourism, 18, 773792.
Wilkinson, I. and R. March (2009), Conceptual tools for evaluating tourism partnership, Tourism
Management, 30, 455462.
Wray, M. (2009), Policy communities, networks and issue cycles in tourism destination systems, Journal of
Sustainable Tourism, 17, 673690.
Xiao, H., M. Li and E.C.K. Lin (2011), Diffusion patterns and knowledge networks: an inductive analysis of
intellectual connections in multidisciplinary tourism studies, Journal of Travel & Tourism Marketing, 28,
405422.
Zach, F. and P. Racherla (2011), Assessing the value of collaborations in tourism networks: a case study of
Elkhart County, Indiana, Journal of Travel & Tourism Marketing, 28, 97110.
Zach, F., U. Gretzel and D.R. Fesenmaier (2008), Tourist activated networks: implications for dynamic pack-
aging systems in tourism, Information and Communication Technology in Tourism, 6, 198208.
Ziene, M. and T. Hazel (2007), Webs of power: multiple ownership in tourism destinations, Current Issues in
Tourism, 10, 279295.

DWYER 9781781001288 Part III.indd 494 25/07/2012 11:08