ABSTRACT
In this chapter, we illustrate the basic formulation of the latent Markov (LM) model for categorical response variables; this formulation assumes the same measurement model, which is also time homogenous, and the same latent model for all observational units. Therefore, individual covariates are ruled out. At the same time, no constraints are posed either on the conditional distribution of each occasion-specific response variable, given the corresponding latent state, or on the initial and transition probabilities. As we will show, the resulting model is rather easy to estimate by the Expectation-Maximization (EM) algorithm.
