Septième séminaire de statistique – A08
Le jeudi 06 novembre 2008, à 13 h 30,
à la salle 1240 du pavillon Alexandre-Vachon
A New Approach to Structural Equation Modeling:
Generalized Structured Component Analysis
Department of Psychology, McGill University
Structural equation models (SEMs) have become a remarkably popular tool for the specification and testing of hypothesized relationships among observed and latent variables in psychology and various fields. Two different approaches have been proposed for SEMs: Covariance structure analysis (CSA; Jöreskog, 1970) and partial least squares (PLS; Wold, 1973). Hwang and Takane (2004) recently proposed generalized structured component analysis (GSCA) as an alternative to the traditional approaches. In essence, the use of GSCA bypasses their major limitations (i.e., improper solutions in CSA and no global optimization criterion in PLS). The theoretical aspects of GSCA are discussed in brevity. An empirical application is then presented which demonstrates the usefulness of GSCA and how it compares to the traditional approaches. Recent extensions of GSCA that further strengthen its generality (e.g., dealing with longitudinal data and cluster-based respondent heterogeneity) are also discussed. Finally, the use of a free software program for GSCA, called VisualGSCA 1.0, will be demonstrated.