FULL-SEMIPARAMETRIC-LIKELIHOOD-BASED INFERENCE FOR NON-IGNORABLE MISSING DATA

During the past few decades, missing-data problems have been studied extensively, with a focus on ignorable missing case, where probability depends only observable quantities. By contrast, research into non-ignorable data is quite limited. The main difficulty in solving such that and regression likelihood function are tangled together presentation, model parameters may not be identifiable even under strong parametric assumptions. In this paper we discuss semiparametric for propose maximum full estimation method, which an efficient combination of conditional marginal nonparametric biased sampling likelihood. extra contribution can produce efficiency gain but also identify underlying without additional We further show proposed estimators response mean semiparametrically efficient. Extensive simulations real analysis demonstrate advantage method over competing methods.