Structural Equation Models

Structural equation models (SEM) are loved and loathed. SEM combine Path Analysis with Confirmatory Factor Analysis. That is, they are directed dependencies between several variables, which may entail latent variables, that are estimated in one model. Particularly researchers who work on attitudes favour SEMs for their easy combination of confirmatory factor analysis and path analysis in one model. Many also fancy that the presentation of SEM estimates highly resembles diagrams of our theoretical models. Moreover SEMs can be applied to panel or multilevel data, and handle categorical dependant variables. Others loathe SEMs for they strongly imply causality, without tackling unobserved heterogeneity and other biasing culprits. According to critics, SEM encourage to model everything at once and therefore to end up with a large set of uninformative correlations. Instead, regression analysis should be used to identify a small part of the overall causal story, but get that aspect right. In this session Michael Windzio will give us a brief introduction to SEMs and explain why the supposedly different worlds are maybe not that far apart after all.

Resources:

Tomarken, A.J. & Waller, N.G. (2005). Structural Equation Modeling: Strengths, Limitations, and Misconceptions. Annual Review of Clinical Psychology, 1(1), 31-65 

William, T. (1977). Structural Equation Models. Annual Review of Sociology, 3, 137-161.