Transition Models for Multivariate Longitudinal Binary Data

• Activity Number: 82
• Type: Contributed
• Date/Time: Monday, August 9, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Leilei Zeng*+ and Richard J. Cook
• Companies: University of Waterloo and University of Waterloo
• Keywords: association parameters ; longitudinal data ; estimating functions ; Markov model ; multivariate process ; transition probability

Abstract

In many settings with longitudinal binary data interest lies in modeling covariate effects on transition probabilities. When interest lies in tracking how two processes change together, one may examine the degree to which changes in one process are correlated with changes in another process. In such settings, the associations between the transition occurrences for the two processes are the scientific focus. Under Markov assumptions, use of marginal transition models permits separate modeling of covariate effects on the transition probabilities for univariate longitudinal binary data, but no insight into the associations can be obtained. While time-dependent covariates may be constructed for one process based on the other, the two processes are then treated asymmetrically. We propose a method of estimation and inference for joint transitional models for bivariate longitudinal binary data based on GEE2 or alternating logistic regression. This approach enables one to model covariate effects on marginal transition probabilities as well as on the association parameters between the two processes.