Bayesian variable selection for latent class analysis using a collapsed Gibbs sampler

Latent class analysis is used to perform model based clustering for multivariate categorical responses. Selection of the variables most relevant for clustering is an important task which can affect the quality of clustering considerably. This work considers a Bayesian approach for selecting the number of clusters and the best clustering variables. The main idea is to reformulate the problem of group and variable selection as a probabilistically driven search over a large discrete space using Markov chain Monte Carlo (MCMC) methods. Both selection tasks are carried out simultaneously using an MCMC approach based on a collapsed Gibbs sampling method, whereby several model parameters are integrated from the model, substantially improving computational performance. Post-hoc procedures for parameter and uncertainty estimation are outlined. The approach is tested on simulated and real data.