HANDLING MISSING RESPONSES IN GENERALIZED LINEAR MIXED MODEL WITHOUT SPECIFYING MISSING MECHANISM

In longitudinal studies, missingness of data is often unavoidable. Valid estimators from the generalized linear mixed model usually rely on the correct specification of the missing data mechanism. An incorrectly specified missing mechanism may lead to a biased estimator. In this article, we propose a class of unbiased estimating equations using pairwise conditional technique to deal with the generalized linear mixed model under benign non-ignorable missingness where specification of the missing model is not needed. We show that the proposed estimator is consistent and asymptotically normal under certain conditions. Simulation results and an example using longitudinal course of neuropsychological data are also shown.