Transition Models for Multivariate Longitudinal Binary Data

Abstract

In many settings with longitudinal binary data, interest lies in modeling covariate effects on transition probabilities of an underlying stochastic process. When data from two or more processes are available, the scientific focus may be on the degree to which changes in one process are associated with changes in another process. Analysis based on independent Markov models permits separate examination of covariate effects on the transition probabilities for each process, but no insight into between-process associations is obtained. We propose a method of estimation and inference based on joint transitional models for multivariate longitudinal binary data using GEE2 or alternating logistic regression that allows modeling of covariate effects on marginal transition probabilities as well as the association parameters. Consistent estimates of regression coefficients and association parameters are obtained, and efficiency gains for the parameters governing the marginal transition probabilities are realized when the association between processes is strong. Extensions to deal with multivariate longitudinal categorical data are indicated.