A Comparison of Exploratory Structural Equation Modeling and Bayesian Structural Equation Modeling for the Investigation of Factor Structure

• In Event: 61.084 - Criteria and Bias in Measuring Latent Constructs and Classes

Abstract

Conventional confirmatory factor analysis (CFA) or structural equation modeling (SEM) commonly removes cross-loadings smaller than .3 in order to reach a simple-structure model. This may cause varying degrees of model misspecifications that deteriorate the model fit and parameter estimation. Two modeling approaches with less restrictive assumptions have been developed to allow free estimation of the cross-loadings: exploratory SEM (ESEM) and Bayesian SEM (BSEM). This simulation study compares BSEM and ESEM for estimating SEM models containing measurement components with small cross-loadings. Design factors include factor structures, item distributions, and sample sizes. Results are evaluated based on overall model fit, parameter estimates, and standard errors of parameter estimates. We hope to provide knowledge about flexible modeling of complex measurement structures.

Authors

• Xinya Liang, University of Arkansas

• Yanyun Yang, Florida State University