Introduction to Structural Equation Modeling
For students of the Doctoral School of Social
and Behavioural Sciences at Ghent University, a two-days seminar on Structural
Equation Modeling will be offered in April 2008.
The course will be given by Anne Boomsma at Ghent University (Practicumzaal A and PC room at Henri Dunantlaan 1).
Contents and Objectives of the Course
The purpose of this course is to provide a theoretical introduction to the analysis of covariance structures, or structural equation modeling as it is called, and to gain practical experience with this type of modeling using LISREL software.
For the description and analysis of theory-based directed relationships between several variables simultaneously, it is possible to construct models that can be visualized as path diagrams. The first aim of the researcher is to formulate an underlying structure of relations between variables (latent variables or hypothetical constructs) on the basis of theoretical considerations. By means of a random sample of covariances between observed variables, the next step is to examine the discrepancy between the postulated model and these sample data. The set of directed relations between the variables in a structural equation model, also known as a covariance structure, may be regarded as a complex `regression model'. The observed variables, being indicators of latent variables in such models, may be subject to measurement error, however. Social scientists use these types of models more and more frequently, whether justified from a statistical point of view or not. The LISREL (LInear Structural RELations) program, together with its preprocessor PRELIS, supports these analyses.
The course offers an introduction to the theoretical backgrounds of the statistical analysis of structural equation models. By means of practical assignments participants may learn how to construct covariance structure models, how to work with the LISREL and PRELIS program, and how to interpret the results of such analyses.
Participants who want to prepare themselves might start to read English abstracts (sheets of the lectures) from a Dutch manuscript of Boomsma, A. (2006). Covariantiestructuuuranalyse. Ongepubliceerd manuscript, Rijksuniversiteit Groningen, Vakgroep Statistiek & Meettheorie. All abstracts will be sent to the participants in advance.
Section 16.1 (page 16-4f.) of that manuscript also provides an advice – in English – on preliminary reading (sent to participants): Chapter 2 of Bollen (1989) or the first chapters of Brown (2006), for example, would be a good start. In addition, one might take a look into the English version of Chapter 15 (Reporting of Structural Equation Analyses) of Boomsma's manuscript, that will also as a separate paper be dispatched to the participants.
Monday, April 14, 2008
09.00 - 11.00 Introduction to structural equation modeling
11.15 - 12.00 Some probability theory and other preliminaries
12.00 - 13.00 Lunch
13.00 - 13.45 The LISREL model
14.00 - 14.45 Model specification; issues of standardization
15.15 - 16.00 The program command language SIMPLIS
16.15 - 17.00 Exercises
Tuesday, April 15, 2008
09.00 - 09.45 Identification problems; estimation methods
10.00 - 10.45 Exercises
11.00 - 12.00 Discussion exercises
12.00 - 13.00 Lunch
13.00 - 13.45 Model evaluation and modification
14.00 - 14.45 How to deal with ordinal, categorical variables?
15.15 - 16.00 Exercises
16.15 - 17.00 Reporting analyses of covariance structures, or How to publish?
Update: March 1, 2008
Anne Boomsma, University of Groningen, The Netherlands.