Bayesian Model Selection Methods for Testing Measurement Invariance

• In Event: 75.039 - Structural Equation Modeling Methods Medley: New Approaches for Modeling Growth, Multilevel Mediation, and Invariance Testing

Abstract

To examine measurement invariance in a multigroup confirmatory factor analysis (MCFA), one important issue is understanding the performance of various model selection methods to detect non-invariant models. The objective of this simulation study was to systematically compare the performance of various Bayesian model selection methods: deviance information criterion (DIC), widely available information criterion (WAIC), leave-one-out cross-validation (LOO), and posterior predictive p-value, with other more frequently used methods: likelihood ratio tests, AIC, small-sample corrected AIC (AICC), BIC, and sample-size adjusted BIC (SABIC), for testing measurement invariance in MCFA. Results showed that DIC, WAIC, LOO had the best balance between the Type I error rates and statistical power in MCFA.

Authors

• Xinya Liang, University of Arkansas

• Yong Luo, Educational Testing Service