Chapter 10 - Mixed-effects regression model for binary longitudinal data

Abstract

In Chapter 10, I concentrate on the mixed-effects logit model applied to analyze binary longitudinal data. First, the classical logit and probit models are reviewed, followed by the specification and statistical inference of the mixed-effects binary logit model. The specification and inference consist of the derivation of parameter estimates, a brief discussion on the interpretability of the conventional odds ratio in the longitudinal setting, and the prediction of the probability and the corresponding standard error. It is displayed that without information on the averaging of the random effects for the two population subgroups, the regression coefficient and its antilog for the mixed-effects logit model are not interpretable. Correspondingly, computation of the conditional effect and the conditional odds ratio is essential to display the covariate’s effect on the binary response. An empirical illustration is provided to display how to apply the mixed-effects logit model in the analysis of binary longitudinal data.