Variable Selection via the Horseshoe Prior in Bayesian Factor Analysis

• In Event: Advanced Methods With Practical Implications

Abstract

Exploring the loading pattern in factor analysis can be considered as a variable selection problem, which incorporates both exploratory and confirmatory features of the factor analysis. Recent developments have used Bayesian shrinkage priors to perform variable selection, showing a balance between model fit and model complexity. Our study proposes a new method that extends the horseshoe prior [Carvalho et al., 2010] to Bayesian structural equation modeling (BSEM), regarded as BSEM-HS, for determining the factor-loading structure. The performance of BSEM-HS is also compared with a popular BSEM method using the ridge prior (BSEM-RP) [Muthén & Asparouhov, 2012] and the frequentist counterpart exploratory factor analysis (EFA).

Authors

• Xinya Liang, University of Arkansas

• Jyotishka Datta, University of Arkansas - Fayetteville