On Model Identifiability of Latent Class Regressions

Abstract

Many statisticians are skeptical of latent class models despite a long tradition of application in the social sciences. A predominant concern is that potential non-identifiability of latent class models is a well-known problem. Without identifiability, standard inferences are meaningless. In this talk, I will focus on providing an identifiable regression extension of latent class model for studying relationships between measured multiple categorical outcomes and covariates of interest. An analysis of how measured health impairments affect older persons' functioning is used for illustration through the talk.