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

net/publication/226330406

The data documentation initiative: A preservation standard for research

Article  in  Archives and Museum Informatics · March 2007

DOI: 10.1007/s10502-006-9036-0

CITATIONS READS

13 426

2 authors:

Karsten Boye Rasmussen Grant Blank

University of Southern Denmark University of Oxford
102 PUBLICATIONS   183 CITATIONS    84 PUBLICATIONS   4,085 CITATIONS   

SEE PROFILE SEE PROFILE

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

Oxford Internet Survey (OxIS) View project

The Oxfordshire Digital Inclusion Project View project

All content following this page was uploaded by Grant Blank on 21 May 2014.

The user has requested enhancement of the downloaded file.

Arch Sci
DOI 10.1007/s10502-006-9036-0

ORIGINAL PAPER

The data documentation initiative: a preservation

standard for research

Karsten Boye Rasmussen Æ Grant Blank

Received: 10 March 2005 / Accepted: 15 November 2006

Springer Science+Business Media B.V. 2007

Introduction

When the answer is the number ‘‘42’’ (or actually ‘‘forty-two’’) some know instantly that
the question is ‘‘The Great Question’’ concerning ‘‘Life, the Universe and Everything ...’’
(Adams 1986, p. 128). This demonstrates that even when we know the answer to a question
its meaning and usefulness are not always obvious. Context is required. This is especially
true for quantitative data. More information is needed in order to understand the numbers
and transform data into useful knowledge. This further information is itself data, and thus
metadata or ‘‘data about data’’. The importance of context and metadata are widely rec-
ognized in the social science community. This paper discusses a project to provide stan-
dardized metadata to document social science datasets: the Data Documentation Initiative
(DDI). The two authors worked together on the DDI committee and are authors of a
previous article on the DDI (Blank and Rasmussen 2004).
We begin by discussing the fundamental problems of social science dataset docu-
mentation. We focus on the value that standardized documentation in the form of the DDI
creates for the social science community.1 To understand the virtues of the new, we look
briefly at some historical documentation standards. We describe the fundamental features
of the DDI standard by describing a current application. The DDI is a standard still being

1
Although we focus on social science researchers, other stakeholders have an interest in the DDI: archives,
funding agencies, the layman, society as a whole. Methodologically we describe the value for researchers
based on the value of high-quality documentation as well as the ability easily to find relevant data.

K. B. Rasmussen (&)
University of Southern Denmark, Campusvej 55, 5230 Odense M, Denmark
e-mail: [email protected]

G. Blank
American University, 4400 Massachusetts Ave NW, Washington, DC 20016-8072, USA
e-mail: [email protected]

123
 Arch Sci

developed and we close with a discussion of future development plans and a summary of
the value of the DDI for research.

The background

Quantitative research data consists of values and categories describing characteristics of

entities, or to use object-oriented systems terminology, the categories and values are
descriptions or measurements of the attributes of objects. The object could be an individual
(a person), an artifact (a phone, a car etc.) or an aggregation in the form of some social or
geographic gathering (an organization, a political party, a city, a region, a nation, etc).2
Furthermore the object could be a representation of an act like the transactions performed
by the former representations (a phone call made on a certain phone, a phone call made by
a certain person, a sales receipt from a cash register, a car trip, etc.). Typical social science
data contains information about the object (often an individual) using numeric representations
of the data attributes. The data are often collected via a questionnaire. The complexity and
need for contextual information is clear from the following glimpse of a numeric data file.
Figure 1 illustrates—just like the opening anecdote—that without appropriate context
data are meaningless.

209112422910000000400001000507061121210112411910100000300010000803050510212
112222910000000400010001010010820213112222910200000301000101010050721215112
211910000000310000001010040450101113233900002010200000100910010811109111111
910002000300010000506060610106111342200100000301000000301050720216112211910
000000100000101010020210111411322910001000201010000102050720108211422910000
000110000000409050820201112222910000000100010101010021220112111222910000000
101000001010030521203112242200100000200000100910030631.......
Fig. 1 Numeric data

Data—information—knowledge

Peter Checkland—the British founder of the soft systems development—and Sue Holwell
(Checkland and Holwell 1998) supply a useful starting point. ‘‘Data’’ are viewed as a
cloud of jumbled, unordered, formless facts. ‘‘Information’’ refers to ‘‘meaningful facts’’
where meaning is derived from context. The point is important: meaning requires context
and meaning emerges from context. ‘‘Knowledge’’ is the ‘‘larger, longer living structures
of meaningful facts’’. How do we get from data to knowledge? According to Checkland
and Holwell our first task is to select. Billions of data could be collected on a person; we
select the data relevant to us. Checkland and Holwell (1998, p. 90) propose the term
‘‘capta’’ for what is actually and actively selected. In order to make use of the data, the
analyst has to have information on the process of selecting this ‘‘capta’’. This information
is the metadata or data description.
Thus, when we have data, we require data description in order to do useful analysis. The
data description is the key that supplies meaning to data. The point is that because data are

2
The basic problems of qualitative data are quite similar. In order to present more easily understandable
description this article concentrates on quantitative data and exemplifies the problems associated with it [See
the article by Louise Corti on qualitative data elsewhere in this issue; the editors].

123
Arch Sci

useless without the meaning supplied by the information embedded in the description,
preservation of the data description is as important as preservation of the data itself.

The use of computers

Data, such as the stream of numbers shown in Fig. 1, are described as ‘‘machine readable’’
when stored in a computer and this implies that these data will be processed by computers.
Streams of numerical data cannot be used either by computers or by people without
additional information. We need to know what each of the numbers represents. For
example, assume that the data file in Fig. 1 contains information about individuals, that
every person’s data record consists of 36 bytes (characters), and that the 18–19th byte in
each record contains information about the age-category for that particular person. Further,
assume that this is the 12th variable describing an individual. The information docu-
menting what each number represents is metadata.
Metadata must be ‘‘person readable’’ because it carries meaning. Ever since data were
first stored on computers, person readable documentation has been required. For decades it
was stored in paper form. Now metadata is almost always stored on computers, so it is also
machine-readable. Whether data is readable in the sense that it conveys meaning is a
question of the abilities of the human observer, the data itself, and the software that
processes the data. Schematically the metadata for the file in Fig. 1 could be presented in a
formal way as shown in Fig. 2.

Fig. 2 The information for RECORD=36

variable extraction
V12="AGECAT",COL=18-19

Using a standard format for the metadata, software is able to extract the location of the
age category variable. Once it has the location information, software is able to extract age
category information for all individuals in this particular file. Extraction of the variables in
the demonstration dataset will result in a data matrix with the cases (entities/rows) and
variables (attributes/columns) presented in Fig. 3.

Fig. 3 The data matrix

123
 Arch Sci

The true beauty of the idea of a standard metadata format lies in the fact that the
software that can find the location of the age category variable will also be able to find the
location of any and every variable in a file. Further, it can read the location of variables in
any other file that has been documented using the same standard format. It is a general tool.
The same software would for example be able to extract information from a file containing
car trips between two destinations, given the metadata in Fig. 4.

Fig. 4 The information for RECORD=32

variable extraction for
another file V9="DEPARTURE",COL=16-19
V10="DESTINATION",COL=20-23
V11="TRIP",COL=24-26

A standardized format for the metadata allows data in many types of files to be read and
processed.

Data documentation initiative

Analysing undocumented data is impossible. But even with documentation the process of
analysis is often difficult (e.g., the user must be able to understand the jargon of the
documentation), error prone (e.g., the documentation might be imperfect, and/or the user
might misunderstand the documentation), and time-consuming (e.g., users have to famil-
iarize themselves with the documentation and the software). Providing a standard format
for machine-readable metadata can reduce errors and simplify analysis. From these rea-
sons, the DDI is intended to become the cornerstone of many scientific infrastructure
projects.
The standard data matrix—as presented in Fig. 3—is a matrix of entities (e.g., indi-
viduals stored as rows) having attributes (columns or variables). Quantitative social science
research relies on many similar data matrices: social surveys, psychological test mea-
surements, economic and financial series, government statistics etc.

Basic data documentation

The rationale for the DDI is based upon a multi-faceted view of data documentation that
goes beyond the limited variable information available in statistical software. The rationale
focuses on the actual information stored and the reasons for storing it. In addition the
rationale includes the additional value that good documentation adds to research. This
section examines the fundamental elements of data documentation and their accompanying
benefits.

The dictionary and codebook

Basic documentation simply describes each variable. Historically basic variable informa-
tion (like the location or number and short description of the variable—‘‘column 18–19’’
and ‘‘AGECAT’’) is often called a ‘‘dictionary’’. However, the short description or the
plain identification (as in ‘‘V12’’), does not supply sufficient information, so textual

123
Arch Sci

documentation of the variable often included both a name and a longer label carrying a
somewhat more comprehensive description ‘‘AGECAT 10 YRS 2003’’. A longer
description of the variable might be needed in order to obtain valid information from the
data; for instance age is dependent upon the time the data were collected.
Most variables are numerically coded; this means that even though fancy tabulations
can be run from datasets containing only limited variable descriptions, without value-level
information the tabulations will not carry information. The ‘‘codebook’’ contains further
descriptions below the variable level including an explanation of the codes stored in the
numeric dataset. For example: numeric code ‘‘1’’ stands for ‘‘Male’’; ‘‘2’’ for ‘‘Female’’,
another code stands for missing data, etc. Obviously without this information tabulations
will be impossible to understand.
The producers of social science statistical software are aware that documentation is
important for analysis. Most statistical software allows documentation but it is often
limited to the location and name of individual variables (as exemplified in the Figs. 2 and 4
above) plus labels for variables and categories; in short, a dictionary and a codebook. In
order to supply meaning, documentation has to specify more than that (Blank 1993;
Rasmussen 1989).
Both the dictionary and the codebook document individual variables. For this reason,
they are often referred to collectively as the ‘‘variable level’’ documentation. There is
another level of metadata that we describe next: the ‘‘study level’’ documentation.

Further elements of the structured study description

From an analytical viewpoint we are getting closer to meaningful information. A table

using the dictionary and codebook can show the distribution of entities on a particular
variable but information about variables is not useful if they cannot be precisely linked to
the selection of objects. In order to interpret our results we need to know the population,
and if this is a sample, we need information on sampling procedures, appropriate
weighting, and other information about the study. Futhermore, most data are time-
dependent and this requires knowing the dates when data were collected. Data might have
relationships with other data; this could be because they contained data about the same
objects, or because the same instruments had been used for measurement, etc. Long-
running or complex studies often store data in multiple files and the relations between files
must be documented. All these elements—and more—have to be described. This is
information about how the collector or researcher turn data into ‘‘capta’’. This holistic
approach can be called ‘‘structured’’ when referring to the DDI as ‘‘structured codebook
standard’’ (Green et al. 1999, p. 31).
The electronic transmission of copies of the data file creates a need for additional
documentation. During transmission or copying the data may be jeopardized by unintended
corruption. Consequently basic documentation also includes record counts and complete
frequencies or descriptive statistics for every variable. Such redundancy ensures that the
data received are the same as the data sent. Further requirements include a precise
description of the storage media and some basic checkpoints.

Scientific benefits of documentation and archiving

Archiving in the digital age always means archiving both data and metadata. The benefits
of documentation are linked to the benefits of archiving. The archiving and availability of

123
 Arch Sci

high quality documentation will benefit research in the social sciences for the following
reasons:
(1) The use of metadata simplifies understanding and reuse of data, thus facilitating
secondary analysis. For discussions of the value of sharing data, see Sieber (1991);
Hauser (1987) and Fienberg et al. (1985). A summary with some American cases are
found in Fienberg (1994); they concentrate on health data, but are built upon
experience with social science investigations. A special case of reuse is the case
where the person performing secondary analysis is the original researcher. It is
obviously formally easy to reuse your own data, but returning to the data long after
details have faded from human memory is extremely difficult. Metadata with easy
access and comprehensive coverage can be of great value to the original researcher.
(2) Empirical social science research is expensive, and much of the cost is due to the
expense of data collection. Original data collection is much more costly than
secondary analysis. Consequently, it is wise behaviour to leverage expensive data in
as many secondary ways as possible.
(3) Secondary analysis is often the only possible way to investigate a time that is now
past. The benefit of strong documentation is that this research would not be possible
at all, had the data not been properly documented and preserved in an archive.
(4) A standard for quality metadata provides in effect a checklist of key information. This
forces the original investigator to systematically and rigorously understand and
describe the data—which may improve both collection and analysis. This raises the
quality of the investigation itself.
(5) The social sciences benefit from firm support for systematic, cumulative building on
prior knowledge. The addition of secondary data to newly collected primary data is
often a source of new knowledge creation. Documentation is crucial in order to be
able to ‘‘match’’ multiple datasets.3
(6) Data acquisition can benefit from the data documentation. For instance the metadata
describing headings, questions, codes, and structure of a questionnaire can be used as
input to automatically generate the computer software that is used to capture the data;
e.g. setting up the progression, the logic, and the screen layouts. This will further
improve the rigor of the questionnaire.
In addition to the benefits of standard metadata there are also disadvantages and costs.
(1) The production of the metadata demands resources, primarily work time of
researchers and assistants. Time requirements are a fierce impediment. If someone
other than the original researcher produces the metadata then the costs will be still
greater.
(2) Metadata demands knowledge. Obviously knowledge about the data and the research
being carried out is needed, but the further cost is that it demands knowledge of
metadata description, the structure and content of metadata and some technical
insight regarding the practical arrangement of the metadata.
These are important disadvantages, particularly for small studies. However, individual
researchers gain substantial benefits from the checklist as well as from generating the data

3
We use the somewhat vague term ‘‘match’’ here. It does not imply a match-merge of the two datasets at a
record level nor is our intention to imply a match on similar variables. The term is used in the broadest sense
where similarities and dissimilarities between the two datasets can be noticed and acted upon in the process
of further research.

123
Arch Sci

acquisition software. The further benefits for the social sciences in general stem from the
improved quality of secondary research in general. These benefits far outweigh the costs.
The accumulation of knowledge in science is supported through peer-reviewed publi-
cations. Many of the strengths of empirical social science research derive from the ability
of other researchers to have access to data to replicate and validate the original scientific
findings, as well as use the data in new and different ways. Some journals now require that
data be publicly archived and available to other researchers. An example from the
American Journal of Political Science: ‘‘AJPS requires that all manuscripts containing
analysis of quantitative data contain an initial footnote that states how individuals can
obtain the data and the documentation of statistical analysis necessary to replicate the
paper’’ (quoted from Meier (1995) in Rasmussen (2000)). With the growth of the Internet,
journals now assume that an interested party can contact the author for access to the
original data.
Most countries have boards for judging scientific dishonesty. In the description of the
purpose and work of the Danish Committee on Scientific Dishonesty is stated:
Scientific dishonesty shall mean intentional or grossly negligent conduct in the form
of falsification, plagiarism, non-disclosure or any similar conduct involving undue
misrepresentation of a person’s own scientific work and/or scientific results’’
(Danish Committee on Scientific Dishonesty 2005).
Examples of possible dishonesty include construction of data, selective and hidden cas-
sation of unwanted results, substitution with fictitious data, or knowingly erroneous use of
statistical methods. Very few actual breaches are found, as mentioned in reports from a
Danish scientific committee (Rasmussen 2000, p. 171, Fig. 119). Data documentation
supports the cumulative and ethical standards of science by making validation possible and
by improving the quality of the science.

A short history of documentation

To understand the objectives of the DDI it is necessary to understand how data archives
and libraries previously dealt with documentation. This clarifies the goals of the DDI
standard.

Paper documentation

Long after data began to be stored in digital form, most archives continued to keep
documentation on paper. Having two separate media—one for data and another for doc-
umentation—required separate systems for processing, storage, retrieval, and dissemina-
tion. Paper documentation has a variety of problems: storage is expensive, it requires
careful controls to track it in inventory, and it deteriorates over time. In contrast, digital
copies can be produced on demand or made directly accessible. From the point of view of
researchers, another problem is even more important: searching paper documentation is
slow and monotonous. In practice, to use paper the analyst must know beforehand which
studies are relevant to the question of interest. Few social scientists have such detailed
knowledge of prior work. In practice, users have been dependent upon archive personnel
and their ability to find—or remember—the appropriate studies.

123
 Arch Sci

Increasingly over the past 10 years, documentation has been available in searchable
form over the Internet. In practice the value of searchable documentation has been limited
and has never reached its potential. The ability to search across studies for similar items or
similar studies has been inadequate because no standard formats existed and because an
adequate mapping between existing formats could not be constructed. The result has been
that access to archived datasets for secondary analysis has been limited. The promise of the
Internet—to make information easily and widely available—has remained unfulfilled.

Machine readable documentation

Data archives have been very much aware of these limitations. They have been system-
atically transferring paper documentation to digital form (as characters or images) since the
1960s.4 In the 1970s the ICPSR and other archives developed electronic documentation in
the form of the OSIRIS codebook. This was a capable documentation standard focusing on
the variable level, although OSIRIS allowed limited information about the study. During
the 1970s through the 1990s, the ICPSR, and the Danish and Swedish archives developed a
series of software preprocessors and filters to add further documentation capabilities to
OSIRIS files (Rasmussen 1996, 2000). Everything about computing in the 1970s was very
expensive, and electronic documentation was also expensive. Only major archives and a
few large government data producers could afford the cost and were aware of the potential
savings of thorough documentation. Smaller organizations usually provided no electronic
documentation at all.
The other documentation formats developed during that time, including the best known
SPSS and SAS system files, described only the content of data files like variables and
values. This had serious limitations, even for documenting statistical output. For a dis-
cussion of the many design weaknesses that limited SPSS’s and SAS’s ability to produce
adequate documentation, even at the variable level, see Blank (1993).
The 1980s introduced personal computers (PCs) and attitudes shifted. The new
emphasis was on quick dissemination of datasets and documentation, and cooperation
lagged. With PCs every data library could develop its own systems, and they did! By the
end of the 1980s, much electronic documentation was hidden in non-standard, heteroge-
neous, individualized, personalized crypts of data deposits. Archives were supporting
hardware and software on microcomputers, minis, and mainframes. Support for all these
platforms drove up costs. Furthermore, prior to the Internet, dissemination incurred long
delays as data and documentation were copied and delivered by surface mail. By the 1990s,
the Internet brought users and archives back into closer contact. Users realized they could
obtain direct, fast access to data without geographical restrictions. Furthermore, users
wanted access to complex search systems to help precisely identify relevant datasets. It
became obvious to data disseminators that these wishes could only be met by developing
widely accepted, international standards for data documentation.
In 1993 staff from data archives and members of the International Association for Social
Science Information Service and Technology (IASSIST)—the professional association of

4
Even where archives have completely converted their documentation to electronic form, like the largest
archive, the Inter-university Consortium for Political and Social Research (ICPSR), most documentation is
only available as graphic images of the documentation pages (stored in PDF files). Such images are
unsearchable. Even when the archive does optical character recognition (OCR) on the images and produces
searchable text, there remain major problems. The most significant weakness is that such documentation is a
stream of text that cannot support field-structured searches.

123
Arch Sci

data archivists—formed the ‘‘The IASSIST Codebook Action Group’’ to work on the
problems of electronic codebooks. Since 1995 the ICPSR has been leading an international
effort to create a worldwide documentation standard to improve access to data. This came
to be the DDI.

Data documentation initiative

The term ‘‘data documentation’’ includes relevant metadata at both the variable-level and
the study-level. The product of the DDI is standardized, highly structured, electronic
documentation. ‘‘Standards are nice, as there are so many to choose among’’! Even though
a standard can be officially certified only by a standardization body (like the International
Standards Organization or ISO), we view the DDI as a practical standard because it is
widely and increasingly used. The DDI is a de facto standard. The DDI draws upon many
prior standards including:
• The OSIRIS codebook: This was the most comprehensive documentation format for
codebooks and the format in use at the ICPSR as well as at some archives in Europe.
• The Standard Study Description: A very elaborate descriptive scheme at the study
level. Several different dialects were used in European archives as well as the ICPSR.
The work on the Standard Study Description was reported with the Scheme in 1974
(Nielsen 1974). The Standard Study Description Scheme underwent continuous
improvement and the latest version is found as a basis of decision in a report for a
conference (Rasmussen 1981).
The following standards mentioned are now collectively described in the publication on
metadata from NISO (2001).
• The Dublin Core: A core of descriptive elements (for Web resources) published in
1995 by OCLC (Online Computer Library Center) and NCSA (National Center for
Supercomputing Applications). Several elements including the Title, Creator, Subject,
Description, Publisher, Contributor, Date, Type, Format, Identifier, Source, Language,
Relation, Coverage, and Rights are part of the DDI.
• The Text Encoding Initiative: The ‘‘TEI’’ is a standard for marking up texts for
research in the Humanities. Like the DDI, the TEI has a strong research emphasis; also
the word ‘‘Initiative’’ came from the TEI.
• Machine Readable Cataloging: MARC is the library record of an object (e.g., a book).
Objects themselves can be machine readable and this created special problems that are
discussed by Dodd (1982).
Several other standards were influential on the development of the DDI. This list first of all
illustrates the point that there are many standards to choose from. Secondly, even though
some of the information in the DDI could be stored in existing formats, no other standard
attempts to comprehensively define the metadata needed to document quantitative files.

Structured documentation in XML

To support publishing Charles Goldfarb and others developed Standardized General

Markup Language (SGML). In 1986, SGML became an ISO standard (ISO-8879). SGML

123
 Arch Sci

is a broad, flexible markup language, specifically designed to allow more-specific types of

documents to be defined within it. Specific types of documents are defined by a Document
Type Definition, or DTD. The standard language used to describe web pages, HTML, is a
document type defined in SGML, but HTML is directed towards screen display of web
pages: it instructs a computer what text should be bold, what should be italics, what should
be a normal paragraph (normal size characters), what should be a header (larger size, bold)
etc. The DDI started with the intention of building a DTD for metadata using SGML.
However in 1996 the World Wide Web Consortium (W3C) announced XML (eXtensible
Markup Language), which was much simpler and almost as flexible as SMGL. XML is
designed to support complex documents on the Internet. The DDI soon adopted the XML
standard. Compared with HTML the strength of XML is that it separates content from
display of information. XML allows creation of a set of rules to identify content by using
highly structured data. For a more detailed discussion of the advantages of XML, see Blank
and Rasmussen (2004).
To illustrate how XML works, we use a simple question: What is the sex of a
respondent?
XML uses printable and person-readable ASCII characters for both content and for tags.
There are no non-printing binary codes as in a Microsoft Word document. Tags are identified
by pairs of angle brackets. A variable name tag could be ‘‘ <VARNAME> ’’. The actual
content—in this example variable name—is placed within pairs of those tags: one marks the
beginning of the variable name text and another marks the end: ‘‘ <VARNAME> V58B </
VARNAME> ’’, see Fig. 6. The fully tagged variable might appear in the file like this.
Notice first, that the tags in Fig. 6 carry no information about display, they only describe
the content. The focus on content frees XML from any restrictions on the number or type
of tags. In XML, it is possible to define whatever tags the content requires.
Secondly, XML is not concerned with display. There are no built-in instructions for the
display of XML content-based tags. Display instructions can be added to XML through
stylesheets, using XSL, the eXtensible Stylesheet Language. This approach yields addi-
tional flexibility; by using different stylesheets we can display the same tagged text in
different ways and the tagging can concentrate on content and delivering content. It is the
choice of the Nesstar software to display the metadata as you see in Fig. 5.
A tagged document would not use the new lines, we added them to Fig. 6 to improve
readability. The elements of the variable are represented by the tags, and the structure of the
document is represented by the sequence and nesting of tags. The available tags and the
structure for the document are defined in the Document Type Definition (DTD) for the DDI.
The DTD defines the rules that a particular document has to obey to be a DDI-complaint
document. All DDI documents use the same DTD, although any single study is unlikely to
make use of all the available elements. The DDI DTD can be accessed from the web http://
www.ddialliance.org/dtd/index.html. Both study level and variable level metadata is con-
tained in standard, structured tags, making it easy to identify and retrieve. The DDI DTD thus
delivers a flexible, versatile platform for documenting social science data.

The value of the DDI for research

From an archival viewpoint, secondary research is the focus. The process of secondary
research can be viewed as a series of stages: retrieval, identification, access, analysis, and
publication. The DDI can create value at each stage.

123
Arch Sci

Fig. 5 The Nesstar display of a variable

Fig. 6 An XML-tagged variable

Examples of creating value for research by the DDI

The tagged example in Fig. 6 above makes it clear that software could perform a simple
string search on DDI-documents; the tagged DDI structure makes more complex searches

123
 Arch Sci

possible. For example a researcher might be looking for a study with (A) more than 1000
respondents, (B) carried out after 1990, containing variables (C) age and (D) highest
education. This searches logically for ‘‘A and B’’ at the study level as both have to be
present. However, at the variable level the search is for the occurrence of ‘‘C or D’’, since
a single variable containing both—‘‘C and D’’—age and highest education—is unlikely.
The DDI makes such combinatory complex searches possible.
Not only can individual datasets be located, but searching a large archive may return
whole lists of relevant datasets. Refining searches by expanding or contracting the time
frame or the geographical area is relatively simple. This directly supports the develop-
ment of research using more than one dataset for cross-national research or research
across time.
Once a dataset is identified, software can read a tagged documentation file and extract
the elements needed for analysis. With the DDI description it is straightforward to
transform the tagged information into sets of commands or system files for major statistical
packages (e.g., SAS, SPSS, STATA, SYSTAT). The data are accessible in the preferred
format or statistical package. And future software formats can be supported as well.
The DDI dramatically improves access to datasets. Datasets become available every-
where as long as the researcher has access to the Internet and is authorized to access the
data. In the future it will be possible to hyperlink publications directly to the data docu-
mentation. This may encourage further use of the data. XML features much more intel-
ligent linking than that available in HTML.
Fundamentally, the DDI will improve potential access to and utilization of data. The
DDI will improve the efficiency of research, and make better use of scarce research money.
This is valuable not only for the researchers but also for the society. The hope of DDI
designers is that better access to datasets will improve research and thereby transfer greater
benefits to society at large.

Applications utilizing the DDI

The DDI is a free standard; there are no license fees and applications can be developed
freely. Many major projects are now underway using the DDI standard. Some examples:
The Virtual Data Center (Altman et al. 2001) describes itself as ‘‘An operational, open-
source, digital library to enable the sharing of quantitative research data’’ (Virtual Data
Center 2004–2005). The European Union originally funded Nesstar (Networked Social
Science Tools and Resources) (Ryssevik and Musgrave 2001). Nesstar has since become
a private commercial organization with strong bonds to archives in both Europe and
North America (Nesstar 2004). Among census projects are the National Historical
Geographical Information System (NHGIS) that intends to ‘‘create and freely dissemi-
nate a database incorporating all available aggregate census information for the United
States from 1790–2000’’ (NHGIS 2004). In the USA there is the Cultural Policy and the
Arts National Data Archive (CPANDA 2006). Several other European projects are also
of significance. The Council of European Social Science Data Archives (CESSDA 1997)
has been involved in projects integrating the catalogues of the European data archives
and the European Union funded Multilingual Access to Data Infrastructures of the
European Research Area (MADIERA 2005) project is based on the DDI. A more
complete and updated list of DDI involved projects can be found at the DDI website
(DDI 2005).

123
Arch Sci

Nesstar as example

To give readers a better feel for how the DDI works in practice, we return to Figs. 5 and 6
above. They show how the XML DDI content for one variable (Fig. 6) is presented by
Nesstar software (Fig. 5). The variable is a basic element of a social science quantitative
dataset, but before a variable can be displayed the dataset itself has to be identified. Nesstar
has the ability to search for possible datasets.
Structured searches on either study level or variable level information, or both can be
combined. The example shown in Fig. 7 below is the Nesstar-server that supports indi-
vidual-level data as well as aggregated data and time series. The demonstration shows how
to specify a search for selected elements of documentation. The first screen contains a 2-
pane window: a list of datasets appears on the left while the right side displays the contents
of individual items for the dataset that is currently highlighted. The hierarchical structure
of the documentation is duplicated on the screen. The interface allows users to drill down

Fig. 7 The Nesstar server

123
 Arch Sci

Fig. 8 The Nesstar tabulation

to obtain detailed information about specific topics of the study as well as selecting
individual variables.
The ability to retrieve datasets via the Internet has been available from single archives
(e.g., ICPSR) as well as at the level of archive associations e.g. CESSDA (Council of
European Social Science Data Archives) developed a web-page (‘‘The integrated data
catalogue’’) for searching European archives at the study level. However, only recently has
it been possible to carry out serious analysis over the Internet. Figure 8 shows a tabulation
created by Nesstar over the Internet.
The importance of this demonstration is to show that Nesstar or similar software can be
used not only for searching datasets, but also for statistical analysis. The benefits of DDI
documented datasets are not just a promise for the future, they have already been imple-
mented in pioneering applications.

Applications producing the DDI

One of the major advantages of a standard is that software tools developed for specific
projects by one archive can be reused for other projects in other locations. These network
effects are among the most compelling features of the DDI project. Several software tools
to assist in the creation of XML documents are listed on the DDI Alliance website http://
www.ddialliance.org/DDI/related/tools.html. Among the tools is a program to convert data
definition information stored in SAS, SPSS, and STATA files into tagged DDI XML
elements. Nesstar Publisher software can, for instance, import an SPSS file and automat-
ically create DDI documentation. Researchers can use Publisher to add DDI-complaint
metadata without knowledge of XML. The Nesstar Publisher handles the technical details.
Instead of producing documentation by typing information at the end of a research
project, it is also possible to use other systems to generate the XML elements of the DDI.

123
Arch Sci

Data-creating software systems—like CATI, CAPI, and CASI systems (Computer Assisted
Telephone/Personal Interviewing/Self-Interviewing)—often have their own specification
language. Such specifications can be directly converted to DDI XML tags. For example, a
Blaise-to-DDI converter, supports the widely used Blaise software for interviewing and
survey processing. This capability means that large portions of DDI documentation can be
generated automatically. Obviously it is also possible to generate the Blaise specifications
from a DDI tagged file.

The impact of the DDI

We can summarize the preceding discussion of the impact of the DDI in the following
points:
• Data preservation: Without documentation the preservation of data alone is not
worthwhile.
• Access to the stored data: Elaborate metadata is necessary for an archive to deliver the
precise information that a user finds relevant.
• Data collection: The DDI can serve as a checklist to improve the design of surveys, and
also help with data collection through interfaces with special software.
• Data publishing: Governmental agencies may use the DDI as a means to publish data.
• Data analysis: Strong metadata improves the speed, accuracy, and reliability of
analysis.
• Research community: The DDI facilitates easy replication and extension of research
results, as well as simplifying secondary analysis of existing datasets.
• Development of tools: Standards facilitate the development of tools for creating,
storing, manipulating, and presenting metadata to users.

Further development of the DDI

The original DDI was designed primarily to document the final dataset. In this design the
documentation was largely a monolithic entity much like a paper codebook. It was not easy
to document parts of a dataset over its lifecycle. The next version of the DDI, version 3.0,
will introduce a new data model. A dataset can be thought of as having a lifecycle with
multiple stages. The lifecycle begins with the initial concept of the study, it moves to the
design stage, then the creation of instruments and a sampling frame, followed by data
collection, cleaning, analysis, and preparation of a file for public release. The new model
will facilitate documentation during each of the various stages of the lifecycle. The major
change introduced will be a new modular structure. Researchers could select particular
modules that document the study design, and only later document, say, the sampling frame,
and still later individual variables. Such a modular structure makes the DDI more useful
because it is easier to document each stage of a research project as it happens. Thus the
DDI can serve as the core repository of information even in the early stages of a project. It
will require reorganizing the DDI so that individual modules can stand alone but are easy
to combine and still remain compatible with the DDI as a whole.
The introduction of smaller computers and PDA’s combined with CATI and CAPI
systems has fostered the development of more complex forms of data. There is no reason to

123
 Arch Sci

believe that the trend toward more varied and more complex data has ended. Thus, a
documentation standard is not a goal, but rather a process. Standards like the DDI need to
be updated to support changing kinds of data, and changing needs of researchers and
teachers.
To continue development of the DDI standard, the ICPSR and the Roper Center for
Public Opinion Research have taken the lead in creating a self-supporting organization,
the Alliance for the DDI or DDI Alliance. The Alliance Steering Committee includes
officers of IASSIST and CESSDA, continuing the close association of the DDI to pro-
fessional data organizations. The DDI has come a long way since the start in 1995. The
DDI Alliance began operation on July 1, 2003 with a core of about 25 members.
Members of the Alliance are data archives, universities, government agencies, and other
institutions that would like to participate in further development of the DDI. Members
pay yearly dues, send representatives to expert committee meetings, and decide on
changes to the DDI.

Conclusion

The challenge of documentation is that it must serve multiple purposes. It must explain
the details of studies including the instruments used, the sample, the response rate, and
other relevant information. It should be friendly and accessible to novices, yet have the
depth to meet the needs of experts. Since the Internet is the medium of choice for
information search, documentation must integrate into the infrastructure on the Web,
including the ability to search across and within studies at the study- and variable-levels.
Finally, it should support automatic generation of system files for popular statistical
software. Some of these purposes work against each other; for example, the simplicity
that is important for novices often conflicts with the depth desired by experts. The beauty
of the DDI is that it is sufficiently flexible and adaptable to serve all these purposes and
more.
The use of the DDI will produce a striking improvement in access to a vast number of
archival datasets. Expanded use of these data has significant implications for the social
sciences. As data accumulate over many decades and societies, enhanced access makes
new studies possible and may lead to a significant improvement in our understanding of
changes across time as well as differences between societies: both longitudinal and
comparative studies become more feasible. Analysis of secondary data is a rich source of
social science knowledge. It is enhanced by increasing the availability of high-quality data.
Detailed, easily accessible documentation is necessary for this to become a reality. The
DDI promises exactly that: by facilitating flexible, user-friendly documentation, improving
its flow through networks, and enhancing our ability to link and display data in new ways,
the research processes mediated by that documentation could improve. Every social sci-
entist who uses secondary data can become more productive and create higher quality
research. This benefits all of us.

Acknowledgements The authors have participated in DDI development, but the DDI was developed by
people representing many academic institutes, governmental agencies, and other institutions. We are
thankful to these people and the many participating organizations. We are particularly grateful to Mary
Vardigan and Ann Green for conversations about the DDI and for their help in finding documents
describing the DDI. More information is available at the DDI website (http://www.ddialliance.org/).
Development of the DDI continues: initiatives can enjoy long lives when many people enthusiastically
support the effort.

123
Arch Sci

References

Adams D (1986) The Hitch Hiker’s guide to the galaxy. Heinemann, London
Altman M, Andreev L, Diggory M, King G, Sone A, Verba S, Kiskis DL, Krot, M (2001) A digital library
for the dissemination and replication of quantitative social science research. Soc Sci Comput Rev
19:458–470
Blank G (1993) Codebooks in the 1990s; or, aren’t you embarrassed to be running a multimedia-capable,
graphical environment like Windows, and still be limited to 40-byte variable labels? Soc Sci Comput
Rev 11:63–83
Blank G, Rasmussen KB (2004) The data documentation initiative: the value and significance of a
worldwide standard. Soc Sci Comput Rev 22(3):307–318
CESSDA (1997) Council of European Social Science Data http://www.nsd.uib.no/cessda/IDC. Cited 05 Nov
2006
Checkland P, Holwell S (1998) Information, systems and information systems: making sense of the field.
John Wiley & Sons
CPANDA (2006) Cultural Policy and the Arts National Data Archive http://www.cpanda.org. Cited 05 Nov
2006
Danish Committee on Scientific Dishonesty (2005) Executive Order No. 668 of 28 June 2005. (The official
translation to English)
DDI (2005) Data documentation initiative. http://www.ddialliance.org. Cited 05 Nov 2006
Dodd SA (1982) Cataloging machine-readable data files. Chicago, American Library Association
Fienberg SE (1994) Sharing statistical data in the biomedical and health sciences: ethical, institutional, legal
and professional dimensions. Annu Rev Public Health, vol 15 Annual Reviews, Inc., Palo Alto, CA
Fienberg SE, Martin ME, Straf ML (eds) (1985) Sharing research data. National Academy Press, Wash-
ington, DC
Green A, Dionne J, Dennis M (1999) Preserving the whole: a two-track approach to rescuing social science
data and metadata. Technical report 83. Washington, DC: Council on Library and Information Re-
sources. http://www.clir.org/pubs/reports/reports.html. Cited 05 Nov 2006
Hauser RM (1987) Sharing data: it’s time for ASA journals to follow the folkways of a scientific sociology.
Am Sociol Rev 52(6):vi–viii
MADIERA (2005) The MADIERA Project. http://www.madiera.net/. Cited 05 Nov 2006
Meier KJ (1995) Replication: a view from the streets. PS: Political Science and Politics, XXVIII(3):453–459
NESSTAR (2004) Publish your data on the web with Nesstar 3.0 http://www.nesstar.com. Cited 05 Nov
2006
NHGIS (2004) National Historical Geographical Information Center. http://www.nhgis.org. Cited 05 Nov
2006
Nielsen P (1974) Report on standardization of study description schemes and classification of indicators.
DDA, Copenhagen (intern)
NISO (2001) Understanding metadata. PDF at http://www.niso.org/standards/std_resources.html. Cited 05
Nov 2006
Rasmussen KB (2000) Datadokumentation. Metadata for samfundsvidenskabelige undersøgelser. [Data
documentation: metadata for social science research] Universitetsforlag, Odense. (In Danish)
Rasmussen KB (1996) Convergence of meta data. The development of standards for the description of social
science data. Paper presented at the 1996 Population Association of America Conference, New Or-
leans, LA. May
Rasmussen KB (1989) Data on data. In: Proceedings of the SAS European Users Group International
Conference 1989. SAS Institute, Cary, NC, pp 369–379
Rasmussen KB (1981) Proposed standard study description. The SD as a basis for on-line inventories of
social science data. (DOC00227) DDA, Odense (intern)
Ryssevik J, Musgrave S (2001) The social science dream machine. Soc Sci Comput Rev 19:163–174
Sieber JE (1991) Introduction: sharing social science data. In: Sieber JE (ed) Sharing social science data:
advantages and challenges. Sage Publications, Newberry Park, CA, pp 1–18
Virtual Data Center (2004–2005) http://thedata.org/. Cited 05 Nov 2006

123
View publication stats