Seminar by Aelita Huang

Marginalized Transition Models for Clustered Longitudinal Binary Data

A marginalized transition model consists of two components: a marginal model, where covariates influence the marginal response probability of each individual in a cluster at each follow-up assessment, and a conditional model, which accounts for the dependence between an individual’s response at a given time point and their response history. In this work, we develop a marginalized transition model for longitudinal binary responses in a clustered setting, incorporating two additional association components: a cross-sectional association component, capturing within-cluster dependencies in responses, and a pairwise association component, embedded in transition events to reflect how peers may influence each other’s tendency to change behaviour within a cluster. We formulate second-order generalized estimating equations (GEE) to jointly estimate the model parameters and will discuss our simulation results and future research directions.