Mixed Markov Latent Class Models

Abstract

The process of change in a discrete characteristic is seldom the same for every respondent in a panel survey. Mixed Markov latent class models describe cross-tables of consecutive measurements in terms of several Markov chains. Each of these chains may be a stayer chain or some kind of mover chain. Markov chains may or may not be latent; i.e., they may or may not be unreliably measured. In this paper, we extend the mixed Markov latent class model to several subpopulations. We give formulas for estimation under equality restrictions when some parameters are fixed, and we give estimation priorities for several types of restrictions. Equality restrictions on parameters across subpopulations can be tested by means of a likelihood ratio test. To illustrate, we reanalyze a data set from Wiggins (1973).