See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/245559395

Structural Equation Model of Customer Satisfaction for the New York City
Subway System

Article  in  Transportation Research Record Journal of the Transportation Research Board · January 2000
DOI: 10.3141/1735-16

CITATIONS READS
75 1,643

3 authors, including:

Johanna Bockman
George Mason University
33 PUBLICATIONS   622 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Gentrification in Washington, DC. View project

All content following this page was uploaded by Johanna Bockman on 19 September 2016.

The user has requested enhancement of the downloaded file.

Transportation Research Record 1735 ■ 133
Paper No. 00-0988

Structural Equation Model of

Customer Satisfaction for the
New York City Subway System
Kenneth R. Stuart, Marc Mednick, and Johanna Bockman

A more complex model in content and design than previously applied to view of causality—an issue that statisticians have been grappling
the measurement of customer satisfaction within the transportation with since Galton (1) first published the concept of a single corre-
industry is used in this study. Drawing from the results of previous stud- lation coefficient. This approach, by definition, does not allow the
ies that had a narrower focus, a network of 10 potentially important fac- impact of more than one variable in the model to be considered at
tors that affect customer satisfaction within the New York City subway a time.
system was postulated and tested using data collected from a cross sec- Multiple-regression analysis provides the potential for improve-
tion of adult residents. Results indicate that several factors have a direct ment when the impact of different variables is assessed. This tech-
influence on satisfaction, whereas others have an effect through inter- nique extends simple correlation and was pioneered by Edgeworth
mediary variables. Path coefficients for the posited model are statisti- (2), Pearson (3), and Yule (4) to permit more than one variable to
cally significant, although several factors have notably more impact have direct association with another. Figure 2 shows an example of
than others. Using model diagnostics, minor revisions and improve- such a model.
ments to the initial model have been made while adhering closely to the This technique may be used either strictly for prediction purposes
principles of the original theory. Future developments are discussed, as or as an attempt to explain relationships between variables. One
is the model’s application for planning and resource allocation. aspect of multiple-regression analysis is that the relationships of
variables to the outcome measure can be assessed simultaneously.
This facilitates analysis of their relative strengths and often is used
As public-transit organizations become more customer oriented, to provide insight into the importance of drivers of satisfaction.
there is increasing focus on the measurement of customer satisfac- There is still a significant weakness in the approach, however. It
tion in addition to using more traditional internal-service measures forces each variable under consideration to be either dependent or
(e.g., on-time performance and mean distance between failures of
independent in nature; thus, no variables can be both dependent and
equipment). Organizations that incorporate external evaluation of
independent in nature. The real world of customer satisfaction is more
their performance often turn to quantitative surveys of current and
likely to be a complex set of relationships in which some variables
potential customers as part of the assessment. This provides the
influence others, which in turn can affect still other measures.
opportunity to collect data on customer satisfaction with overall per-
formance as well as with specific aspects of the public-transit experi-
ence (e.g., service performance and service environment). Moreover, STRUCTURAL EQUATION MODELING
this allows transit management to develop a better understanding of
what drives customer satisfaction to better meet customer needs. An extension of regression analysis is the path analytic model, pio-
neered by Wright (5), in which a causal ordering of variables is pos-
tulated. In contrast to regression, each relationship between variables
TRADITIONAL STATISTICAL ANALYSIS
in the path (or structural) model represents a causal link rather than
The methods employed for assessing customer satisfaction vary con- a mere empirical association. This method can be used to examine a
siderably in complexity and may be as simple as looking at a single network of interrelated variables. Direct and indirect influences may
measure to help evaluate performance. An organization can monitor be studied, with some variables being both dependent and indepen-
progress over time by tracking this measure over successive time dent in nature. The model can be represented by a system of inter-
periods. This univariate approach, however, does not permit the related regression equations. Figure 3 is an example of a structural
analysis of causal factors that drive customer satisfaction; for this, model in which some variables are exogenous (no prior causation)
multivariate techniques are necessary. To help explain relationships, and others are endogenous (variables have prior influence). Addi-
a bivariate model (i.e., a pair of variables, such as the impact of ser- tionally, one of the variables is postulated to be both dependent and
vice frequency upon customer satisfaction) is often used. This type independent. Finally, customer satisfaction is a function of the direct
of model is depicted in Figure 1. impact of certain variables and, at the same time, the indirect impact
A positive aspect of the bivariate approach is that some insight of others.
into the strength of the relationship may be obtained. Its weakness, There is an important extension of path analysis that incorporates
though, is that it can provide an overly simplistic, even erroneous, multiple indicators of key attributes. Path analysis necessitates the
statistical assumption that observed measures correlate with their
MTA New York City Transit, 130 Livingston Street, Ninth Floor, Brooklyn, respective underlying constructs without error. Although combining
NY 11201. several indicators into an index can ameliorate this issue, as would
134 Paper No. 00-0988 Transportation Research Record 1735

FIGURE 1 Hypothetical two-variable model of customer

satisfaction.

having a particularly good single indicator, “latent” constructs with

multiple observable variables can be a statistical improvement.
Joreskog (6, 7) helped pioneer this advance by developing a viable
computer algorithm to solve the resulting complex set of simultane-
ous equations. Stuart (8) first applied this technique in the field of
public transportation with a study of bus operators and their work
environment. Golob (9–11) has examined travel demand and activi- FIGURE 3 Hypothetical structural equation model of customer
satisfaction.
ties extensively, and Mokhtarian and Salomon (12) began to incor-
porate attitudinal data into demand models. These efforts, among
others, have expanded the understanding of travel demand and have
policy implications. However, they have not looked specifically at this modeling effort, the potential major factors of service, service
customer satisfaction with transit service. environment, safety, and personal security are included.

STUDY OBJECTIVES AND SCOPE STUDY METHODOLOGY

The objectives of this study are as follows: Sample

Data were compiled from the Transportation Panel of the Metro-

• To develop a more complete model than heretofore used to
politan Transit Authority’s (MTA’s) New York City Transit. This
describe causal factors leading to customer satisfaction within a key
panel is composed of 1,500 individuals who represent a cross sec-
element of the transportation industry;
tion of adult residents of the city. They are interviewed quarterly
• To establish quantitative values for the relationships that are
(500 per month) via telephone regarding their general travel behav-
hypothesized to exist for variables leading to customer satisfaction;
ior, detailed travel behavior for the most recent 2 days, and attitudes
and
toward various transportation modes, including subway, bus, taxi,
• To describe potential uses for the model in strategic planning
and automobile. Attitude questions include overall satisfaction as
and resource allocation.
well as specific aspects of the transportation experience.
Data used were those that were gathered during the first quarter
The study focuses on the subway system of New York City. There
of 1999. Those residents who answered all of the questions (vari-
are a myriad of possible influences on satisfaction. For purposes of
ables) that are in the model are included in the sample. A total of
1,075 people met this criterion.

Variables Under Consideration

The following variables from the Transportation Panel are used in

the model. Each is measured on a 0-to-10 scale, with 0 being worst
and 10 being best:

• Overall satisfaction—the overall satisfaction that one has with

the NYC subway system.
• Value for money—a customer’s perceived value of a ride on
the subway given the fare.
• Speed of service—the overall speed of the ride, including
waiting and travel time.
• Personal security—the sense of personal security one has in the
system after 8:00 p.m.
• Safety—perceived safety from accidents while in the system.
• Courtesy—the courtesy of NYCT employees.
FIGURE 2 Hypothetical regression model of customer • Cleanliness— cleanliness of the system (including subway
satisfaction. stations and train cars).
Stuart et al. Paper No. 00-0988 135

• Panhandlers—presence of panhandlers within the system. • Personal security to overall satisfaction, and
• Frequency of service—perceived frequency of overall subway • Value for the money to overall satisfaction.
service.
• Predictability of service—perceived reliability of subway
service. FINDINGS
• Crowding—crowding on station platforms and in subway cars.
The overall model fits the data set, and each of the individual
hypothesized paths between variables is statistically significant.
Technique and Postulated Model All variables were measured on the same metric (using a scale of
0 to 10), and Table 1 shows the unstandardized coefficient (beta
Structural equation modeling was chosen, without latent traits, for weight) for each path. The corresponding critical ratio (a measure
this analysis. This decision stems in part from a desire initially to of the level of statistical significance) for each path is also shown.
develop management tools that are considered more concrete, eas- This critical ratio provides a measure of statistical significance of
ily understood, and, in turn, actionable. The 11 variables described the path coefficient—1.96 indicates significance at the 95 percent
in the previous section all were incorporated and are theorized to level of confidence.
have a particular set of relationships—some are causal, whereas Several key findings emerge. One is that all of the hypothesized
others are not. A similar model was described by Stuart and Schaller paths are statistically significant, although the strength of the rela-
(13). The exogenous variables are not deemed to be causally related, tionships can vary dramatically. Another finding is that the impact
although it is recognized that they have a nonzero correlation with of predictability of service on perceived speed of the trip is as
each other. The statistical software AMOS, from the Statistical important (even more so) than is frequency of service. This was
Package for the Social Sciences (SPSS), was used for this analysis, thought to be a possibility from earlier research but was never
although other credible software programs exist, such as LISREL quantified before.
(Linear Structural Relationships). The proposed structural model is
based on theoretical underpinnings.
The causal links stipulated below generally stem from prior Model Revisions
research on the New York City subway system. They are depicted
in Figure 4, in which straight lines represent causal links, and curved One of the important pieces of output besides the coefficients and
lines represent the noncausal relationships. The e represents error their level of significance is model diagnostics. This tool can help
that exists in estimation. The causal links are as follows: refine a model, although caution should be given to summarily
changing the theory to fit the data.
• Safety to value for the money, Diagnostics indicated other paths to consider, and two were cho-
• Courtesy to value for the money, sen. One is a new link between predictability of service and over-
• Cleanliness to value for the money, all satisfaction. This makes intuitive sense because predictability
• Cleanliness to personal security, usually provides a greater sense of comfort to customers. Another
• Panhandlers to personal security, is the link between safety from accidents and speed of service.
• Frequency of service to personal security, There was some initial skepticism of this possible causal associa-
• Frequency of service to speed of service, tion, but after discussion, it was decided that a customer’s comfort
• Predictability of service to speed of service, level in this area could make that person less anxious. One who is
• Crowding to overall satisfaction, anxious often overestimates the passage of time, so this link was
• Speed of service to value for the money, introduced.
• Speed of service to overall satisfaction, The model was rerun with only these two changes and is shown
• Personal security to value for the money, in Figure 5 with the new path coefficients.

FIGURE 4 Initial model of customer satisfaction.

136 Paper No. 00-0988 Transportation Research Record 1735

TABLE 1 Path Coefficients and Critical Ratios for Initial Model

Findings for the Revised Model satisfaction are desired. Furthermore, by associating a cost with a
given improvement and estimating the “ripple” impact it has on cor-
The addition of the two paths improved the overall model fit. How- responding customer perception in the model, it is possible to produce
ever, the specific link between predictability and overall satisfaction cost estimates for improving customer satisfaction under several dif-
reduced the strength of the link between speed and satisfaction. This ferent scenarios. The same impact on customer satisfaction may be
is understandable, given that predictability can now exercise influence made in a more cost-effective manner.
directly, not only indirectly. The impact of crowding on satisfaction
dropped but is still significant. Other paths changed relatively little,
and all remained significant. Table 2 shows the coefficients and ratios. THE FUTURE

The year 2000 draws us not only to the start of a new millenium, but
DISCUSSION OF RESULTS also to a greater understanding of customers and the keys to increas-
ing their satisfaction with transportation alternatives. The future for
One overarching finding is that although customer satisfaction is a structural equation models in helping to study customer satisfaction
complex and sometimes elusive concept in which customer ex- has many possibilities:
pectations and evaluations change, the components of what drives
customer satisfaction can be isolated and examined. Moreover, the • Developing more complex and realistic models to assess the
strength of the relationships can be quantified and compared with one drivers of satisfaction. These could incorporate additional items of
another in terms of both direct and indirect effects. Clearly, there are importance (e.g., communication within and outside the system);
applications for planning and resource allocation. Because the rela- • Linking mode choice among transportation alternatives with
tive weights of the path coefficients provide an indication of impor- customer satisfaction; and
tance in their effect on customer satisfaction, this can assist planners • Expanding use of actionable structural models in cost-benefit
in determining what efforts should be placed if enhancements to analysis to help establish priorities for capital expenditures.

FIGURE 5 Revised model of customer satisfaction.

 Stuart et al. Paper No. 00-0988 137

TABLE 2 Path Coefficients and Critical Ratios for Revised Model

ACKNOWLEDGMENTS Variables. Research Report 73-5. Statistics Department, University of

Uppsala, Uppsala, Sweden, 1973.
The authors wish to thank peer reviewers of the Transportation 7. Joreskog, K. G. Structural Equation Models in the Social Sciences:
Specification, Estimation, and Testing. In Advances in Factor Analysis
Research Board for their comments, which were used to modify and Structural Equation Models (K. G. Joreskog and D. Sorbom, eds.),
and improve an earlier draft of this article. Similarly, the assistance ABT Books, Cambridge, Mass., 1979.
and support of colleagues from the Marketing Research and Analy- 8. Stuart, K. R. The Effects of Stress, Coping, and Personality upon
sis group, the Division of Marketing and Service Information at Employee Satisfaction and Job Performance: An Analysis of Covari-
ance Structures Using Latent Variable Modeling. Ph.D. dissertation.
MTA New York City Transit, and MTA Headquarters are much University of California, Los Angeles, 1984.
appreciated. 9. Golob, T. F. The Dynamics of Household Travel-Time Expenditures and
Car-Ownership Decisions. Transportation Research, Part A, Vol. 24,
No. 6, 1990, pp. 443– 463.
10. Golob, T. F., and H. Meurs. A Structural Model of Temporal Change in
REFERENCES Multimodal Travel Demand. Transportation Research, Part A, Vol. 21,
No. 6, 1987, pp. 391–400.
1. Galton, F. Regression Toward Mediocrity in Hereditary Stature. Journal 11. Golob, T. F., and M. G. McNally. A Model of Activity Participation and
of Anthropological Institute, Vol. 15, 1885, pp. 246–263. Travel Interactions Between Household Heads. Transportation Research,
2. Edgeworth, F. Y. Correlated Averages. Philosophical Magazine, Vol. 34, Part B, Vol. 31, No. 3, 1997, pp. 177–194.
Series 5, 1892, pp. 190–204. 12. Mokhtarian, P. L., and I. Salomon. Modeling the Desire to Telecom-
3. Pearson, K. Mathematical Contributions to the Theory of Evolution: mute: The Importance of Attitudinal Factors in Behavioral Models.
III—Regression, Heredity, and Panmixia. Philosophical Transactions, Transportation Research, Part A, Vol. 31, No. 1, 1997, pp. 35–50.
Vol. 187, Series A, 1896, pp. 253–318. 13. Stuart, K. R., and R. B. Schaller. Panel Survey Incorporating Customer
4. Yule, G. U. On the Theory of Correlation. Journal of the Royal Statistical Attitudinal and Travel Data: Methodological Issues and Applications to
Society, Vol. 60, 1897, pp. 812–854. Transit Policy and Planning. Proc., Transportation Research Board
5. Wright, S. Correlation and Causation. Journal of Agricultural Research, Planning Applications Conference, Boston, Mass., 1999.
Vol. 20, 1921, pp. 557–585.
6. Joreskog, K. G., and M. van Thillo. LISREL—A General Computer Publication of this paper sponsored by Committee on Public Transportation
Program for Estimating a Linear Structural Equation of Unmeasured Marketing and Fare Policy.

View publication stats