A0647
Title: Bayesian adaptive lasso factor analysis models with pre- and post-test binary data
Authors: Junhao Pan - Sun Yat-sen University (China) [presenting]
Lijin Zhang - Sun Yat-sen University (China)
Abstract: Binary data is frequently encountered in behavioral, educational and medical research. We extend previous work on the Bayesian covariance Lasso confirmatory factor analysis (CFA) model on the following aspects: (1) take the binary data into account by assuming that they are coming from an underlying latent continuous distribution with a threshold specification; and (2) handle potentially local dependency in item clusters by assigned the adaptive covariance Lasso prior to blocked diagonal residual covariance structure, which achieves model parsimony and generally fits the data better, while keeping the factor structure intact. We develop the Bayesian inference method based on the parameter expansion and Markov Chain Monte Carlo procedures. The simulation studies showed that the Bayesian estimates of the unknown parameters of interest are reliable. Real data on Pre- and Post-test Genetic Knowledge Items were also analyzed to evaluate the validity and practical usefulness of the proposed procedure.
