Latent class models have been used for many years to identify subgroups within a larger population which have similar characteristics. In medicine, these grouping have been related to syndromes, or clinical phenotypes. This work examines the fitting of latent class models in a Bayesian framework, with a particular examination of the problems encountered when fitting mixture models using MCMC. Specifically, the switching problem is encountered and resolved using two different methods. The question of the number of classes to choose is examined and some graphical diagnostics are proposed.
17
Oct 08