Prior Specification in Bayesian Factor Analysis With Ordinal Data: Variable Selection and Parameter Estimation

• In Event: Statistical Inference and Bayesian Estimation

Abstract

This study aims to examine Bayesian structural equation modeling (BSEM; Muthén & Asparouhov, 2012) for both variable selection and parameter estimation in sparse factor analysis models with ordered categorical data. A sparse factor loading matrix contains only a limited number of non-zero cross-loadings while the majority of cross-loadings being zero. A simulation study is conducted to examine to what extent BSEM is affected by the prior specifications and other factors including the sample size, cross-loading structure, and categorical distribution. Results will contribute to the understanding of ordinal BSEM in terms of its efficiency for detecting cross-loadings and accuracy of estimating model parameters.

Authors

• Xinya Liang, University of Arkansas

• Chunhua Cao, The University of Alabama