EcoSta 2017: Start Registration
View Submission - EcoSta2017
A0353
Title: Multilevel heterogeneous factor analysis models with Bayesian covariance lasso prior Authors:  Junhao Pan - Sun Yat-sen University (China) [presenting]
Abstract: Multilevel confirmatory factor analysis (CFA) models are widely used in the social sciences to analyze data with heterogeneity. We extend previous work on multilevel CFA model to account for the dependency among observed indicators. A Lasso covariance prior was proposed to model the entire inverse residual covariance matrix of the observed indicators as a sparse positive definite matrix that contains only a few off-diagonal elements bounded away from zero, and Markov Chain Monte Carlo (MCMC) procedures was also developed to perform Bayesian inference. The proposed multilevel heterogeneous CFA model achieves model parsimony and generally fits the data better, while the factor structure (that is, the number of factors and how the observed indicators are loaded on the factors) is kept intact. Both simulated and real data sets were analyzed to evaluate the validity and practical usefulness of the proposed procedure.