Structural equation modeling (SEM)

Docente: James Jaccard

Tipo di corso: Corso per dottorandi


Structural equation modeling (SEM) is a powerful approach to data analysis that is becoming increasingly popular throughout the social, organizational, and health sciences. It embraces theoretical frameworks that emphasize causal thinking. It permits formal testing of theories in the context of a multivariate framework that can take into account measurement error and wide range of scaling and methodological phenomena. Knowledge of SEM not only strengthens one´s ability to analyze data, but it also strengthens one´s conceptual skills for thinking about alternative causal models.

This course is an introductory treatment of structural equation modeling designed to provide students with an appreciation of the major concepts and issues in structural equation modeling and causal modeling. Participants will learn how to use the computer program AMOS, and the strengths and weakness of different computer packages for SEM analyses. It is assumed that students have only basic knowledge of correlation and regression and the statistical theory underlying these techniques. Thus, minimal statistical background is assumed. However, to accommodate more advanced students in the class, time will be set aside during the last half hour of each class to discuss topics at an advanced level and to entertain questions at any level of inquiry.

After completing the course, students will be able to formulate their own causal models, be better able to evaluate causal models that they encounter in the research literature, and will have the basic skills to design studies and analyze data using SEM. They will have working knowledge of the AMOS software and will be aware of other SEM software that can be used for more advanced applications of SEM.

Students will be given four assignments in which they will analyze data and write it up in accord with protocols distributed in class. Students also will prepare an in-class presentation summarizing and evaluating an article of their choice that uses SEM. 

The course is divided into four parts:

Part 1: The Basics of Structural Equation Modeling

  • Review of correlation, regression, and psychometric theories of measurement
  • Discussion of causality, causal models, and introduction to path/influence diagrams
  • Issues in formulating and testing single indicator causal models with continuous variables
  • Issues of statistical identification
  • Issues of model fit
  • Global indicesof model fit
  • Focused indices of model fit
  • Interpretation of parameter estimates
  • The notion of direct effects, indirect effects, and total effects
  • Statistical assumptions
  • Computer programming using AMOS

Part II: Measurement Models

  • Latent variable and multiple indicator models in SEM
  • The concepts of reliability and validity revisited in the context of SEM
  • Nested models and comparing different causal models
  • Correlated measurement error and its role in SEM
  • Common applications of SEM based measurement models
  • Confirmatory versus exploratory factor analysis
  • Higher order factor analysis

Part III: Introduction to Latent Variable Structural Models

  • Latent variables in structural equation models
  • Empirical illustrations of latent variable SEMs
  • The use of equality constraints in SEM and testing nested models (revisited)
  • Longitudinal designs
  • Limited information estimation versus full information estimation strategies
  • Categorical variables in SEM

Part IV: Advanced Issues in Latent Variable Modeling

  • Multiple group analyses: Comparing models across groups
  • Tests of statistical assumptions and analytic remedies
  • Small sample size scenarios
  • Missing data
  • Interaction effects, non-linear models and ordinal analysis
  • Reciprocal causality