ABSTRACT
This chapter presents two types of extensions of generalized linear models (GLMs) for longitudinal or clustered data: marginal (population-averaged) models that focus on estimation of the effects of predictor variables on the average response in the population and treat the association structure of the repeated measures as a nuisance, and random effects (subject-specific) models that focus on inference at the individual level and jointly estimate the parameters characterizing the mean and the association structure. generalized estimating equations (GEE) and generalized linear mixed models (GLMM) differ in robustness of their inferences in the presence of missing data with GLMM, in general, providing valid results under a wider range of missing data assumptions. The chapter provides a short introduction to GLM, which are the basis for both GEE and GLMM extensions. Because of the non-linearity in the relationship between the predictors and the response with non-normal data, the population-averaged effect of a predictor is not the same as the subject-specific effect.
