A Bayesian Approach to Latent Class Modeling with Ordinal Response Data

• Activity Number: 404 - Bayesian Clustering and Classification
• Type: Contributed
• Date/Time: Tuesday, August 1, 2017 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Padma Sharma*
• Companies: UC Irvine
• Keywords: MCMC ; latent class ; ordinal models ; Collapsed Gibbs ; treatment effects ; banking

Abstract

Latent class models have been applied in a wide range of modeling situations that entail a discrete distribution for unobserved heterogeneity. Bayesian algorithms for latent class models with ordinal outcomes have not been as extensively developed and implemented as those for their continuous and binary counterparts. The MCMC algorithms developed in this paper provide an efficient method to estimate the univariate and multivariate forms of a latent class ordinal probit model by introducing a novel marginalizing procedure within the Collapsed Gibbs sampler. The proposed method effectively addresses issues pertaining to identification, prior sensitivity and label switching. We apply this method to data from the Savings and Loans Crisis to uncover previously unexplored heterogeneous treatment effects of regulatory measures on thrift institutions.