Model Comparison and Evaluation in Structural Equation Models

• Activity Number: 91
• Type: Roundtables
• Date/Time: Monday, July 30, 2012 : 7:00 AM to 8:15 AM
• Sponsor: Health Policy Statistics Section
• Author(s): Douglas David Gunzler*+
• Companies: Case Western Reserve University
• Keywords: structural equation modeling ; model fit ; nested models ; nonhierarchical models ; equivalent models ; causal direction

Abstract

Is the causal path between alcohol abuse and depression important for a model describing suicidal ideation in a given study? Answering questions of causal directionality, and deciding whether to include or exclude causal paths, especially those involving more complex causal networks, are essential for health policy researchers. Structural equation modeling (SEM)-based approaches are well-suited for confirmatory analysis of a hypothesized causal relationship, with the ability to include both observed and latent variables in a single framework. However, justifying why a particular model was used over alternative models for a given study is often difficult. In addition, many researchers use multiple indices and numerical cutoff points as guidelines for evaluating model fit, without a comprehensive understanding of the strengths and weaknesses of the fit statistics. This roundtable discusses the model fit indices for SEM and approaches to choosing the most appropriate causal paths for a given data set and study.