As a graduate student becoming acquainted with structural equation modeling (SEM), and ultimately using this technique in my dissertation, I now realize that I was just a "freshman." Presently I am using SEM to analyze longitudinal and cross-section data which qualifies me to be a "sophomore." The term sophomore is derived from the Greek words "wise" (sophos) and "foolish" (mohros). As a user of structural equation modeling I have been referred to as both of these. The purpose of this article is to briefly present a nontechnical perspective of SEM, and to list useful available resources.
Structural equation modeling, a statistical technique to test hypotheses about relations among latent and observed variables, is slowly gaining acceptance in behavioral accounting research. Articles using SEM have appeared in Accounting Horizons, Accounting, Organizations and Society, Behavioral Research in Accounting and The Journal of Accounting Literature. Presentations at the American Accounting Association annual, sectional, and regional meetings have included papers using SEM. Even more significant is the increased interest in and use of SEM by accounting doctoral students.
SEM is often referred to as LISREL (linear structural relations), one of the first names given to models which incorporated confirmatory factor analysis, multiple regression, path analysis, simultaneous equation models, and other techniques. A distinct strength of SEM is its simultaneous estimation of the measurement and structural models.1
Much of the early research on SEM occurred in the late 1960s and early 1970s by Karl Joreskog, Dag Sorbom, and others at the Educational Testing Services. Its use in the social sciences began appearing later in such disciplines as education, psychology, and political science. My first introduction to SEM in the business literature was a marketing study by Bagozzi (1980) examining the antecedents and simultaneity of performance and satisfaction. This technique gained popularity in marketing literature, as evidenced by a number of studies published in the 1980s using SEM, in such journals as the Journal of Marketing and the Journal of Marketing Research. In many business schools SEM is offered in graduate marketing research courses.
To gain wider acceptance in accounting behavioral research, SEM needs to be understood as a functional technique. While a thorough knowledge of mathematics is not a necessary prerequisite to using SEM, understanding SEM is best accomplished when, at a minimum (beyond basic statistics), comprehension of matrix algebra, factor analysis, and covariance structure models exist.
In its inception, SEM computer programs generated tremendous amounts of data which required spending much time analyzing the output, a fairly complex process. Top do so, knowledge of areas such as covariance structure models was necessary. Due to its newness, early published studies using SEM tended to be detailed as to the process. Published research using SEM has changed rather quickly, no longer are there detailed models, matrices, or explanations about using techniques such as LISREL. Rather, it appears that as its popularity is increasing expanded applications of SEM are appearing.
Two possible reasons why SEM has become more popular is the ease of using computer programs, especially PC based programs, and the amount of information available about SEM. As SEM became more popular (and accepted), computer programs became more "user friendly." Programs became easier to write and provided more detailed output, thus, requiring less analyses. Additionally, there have been more studies published outlining the procedures and interpretation of the computer-generated SEM output, thus, generating more reference material. These are auspicious (and possibly necessary) beginnings for SEM to be a preferred technique in behavioral accounting research.
However, since so much new material is being generated about SEM (for example, SAGE Publications has issued four new books on the subject in the past year), the ability to thoroughly understand the current state of SEM requires more than just writing the computer program and presenting the output. The ability to analyze beyond the information that is computer-generated is perhaps the most important aspect of using SEM. If skills are lacking in areas such as covariance structure models, we will be ill-prepared to advance beyond the current state of the available computer programs.
There are different SEM computer packages that are available for the mainframe and the PC. The PC based packages are more convenient to use than the mainframe packages. A potential limitation of PC SEM packages is the amount of memory available, but usually this is not a major problem unless the data sets become extremely large. In those cases many institutions have SEM programs (such as those found on SAS or SPSS) on the mainframe computer. I currently use three different Windows-based structural equation modeling packages; LISREL 8.03 for Windows and DOS (which includes PRELIS 2, a data screening program), EQS Windows 4.0, and Amos and Amos Draw for Windows. All are "user-friendly" and have very good documentation. The LISREL and EQS programs are better known than the Amos Programs. However, if you are considering in investing in a SEM computer program for personal or instructional use, you may want to consider starting with the Amos program, as it is available at a cost of only $50 with generous licensing terms for student use. Amos is available from James Arbuckle, Department of Psychology, Temple University, Philadelphia, PA, 19122. In addition to the different SEM software computer packages, there are many texts and articles, as well as support services available for SEM. The following list is not intended to be comprehensive. Rather, these are references which I have found useful in working with SEM. To obtain articles, a literature search using key words such as SEM, LISREL, covariance structure models, causal analysis, etc., will provide sources. A number of these articles appear in journals outside the business disciplines, many of which are in psychology journals.
Bollen, K. A. 1989. Structural Equations With Latent Variables. New York: Wiley.
___, and J. S. Long, eds. 1993. Testing Structural Equation Models. Thousand Oaks: SAGE Publications.
Byrne, B. M. 1989. A Primer of LISREL: Basic Applications and Programming for Confirmatory Factor Analytic Models. New York: Springer-Verlag.
___. 1994. Structural Equation Modelling With EQS and EQS Windows: Basic Concepts, Applications, and Programming. Thousand Oaks: SAGE Publications.
Finkel, S. E. 1995. Causal Analysis With Panel Data, Vol. 105. Thousand Oaks: SAGE Publications.
Hayduk, L. A. 1987. Structural Equation Modeling With LISREL: Essentials and Advances. Baltimore: John Hopkins University Press.
Hoyle, R. H., ed. 1995. Structural Equation Modeling: Concepts, Issues, and Applications. Thousand Oaks: SAGE Publications.
James, L. R., S. A. Mulaik and J. M. Brett. 1982. Causal Analysis: Assumptions, Models, and Data. Thousand Oaks: SAGE Publications.
Joreskog, K. G. and D. Sorbom. 1993. LISREL 8: Structural Equation Modeling With the SIMPLIS Command Language. Hillsdale: Lawrence Erlbaum Associates.
___, and___. 1993. LISREL 8 User's Reference Guide. Chicago: Scientific Software International, Inc.
___, and ___. 1993. PRELIS 2 User's Reference Guide. Chicago: Scientific Software International, Inc.
Long, J. S. 1983. Confirmatory Factor Analysis: A Preface to LISREL. Thousand Oaks: SAGE Publications.
___. 1983. Confirmatory Factor Analysis: An Introduction to LISREL. Thousand Oaks: SAGE Publications.
von Eye, Alexander & Clifford C. Clogg, eds. 1994. Latent Variables Analysis: Applications for Developmental Research. Thousand Oaks: SAGE Publications.