A Simulation Approach to Locally Derived Criteria for Model Fit

• In Event: Roundtable Session 117
  In Roundtable Session: Issues in Model Specification and Fit in Factor Analysis

Abstract

In Structural Equation Modeling (SEM), researchers often rely on global fit indices, and associated fit criteria, to evaluate statistical models. However, little empirical evidence has revealed that the recommended fit criteria actually protect against incorrect conclusions about model fit. The purpose of this study is to propose and evaluate a novel procedure for choosing proper fit criteria. The procedure uses Monte Carlo simulation to generate a unique sampling distribution of fit index values; from which, “local” fit criteria are derived. Results from a simulation study indicate that local fit criteria can make correct conclusions about model fit with nearly 90% accuracy. This simulation approach to local fit criteria may provide a strong alternative to conventional evaluative methods in SEM.

Author

• Brent F. Olson, The University of British Columbia