Markov models with random effects for binary panel data

In the analysis of time-dependent choice data, it is important to take into account that individuals differ both in their preferences at a particular point in time and in the way they change their preferences over time. As a solution to this problem, this paper adopts Humphreys (1998) random-effects approach and combines the family of discrete-time, latent Markov models (Langeheine & van de Pol, 1990) with Thurstone's (1927) random utility framework. It is shown that the synthesis of both model classes yields a flexible and parsimonious framework for the analysis of time-stable preferences, systematic preference shifts, and structural dependencies between current and past preference states. Results from a simulation study and an application illustrate the feasibility and usefulness of random-effects Markovian models.