In various biomedical studies, we may encounter an appreciable amount of missing values. It will result in serious bias and incorrect inference if we do not address these missing data appropriately. In the literature, the methods under Missing-At-Random (MAR) Assumption are well developed. However, in many situations, for example, longitudinal dropout, there is a suspicion that the missing data mechanism is nonignorable.
In this talk, I will present some semiparametric pseudo likelihood methods on handling nonignorable missing data. I will first provide some conditions for population Identification, which is crucial for statistical estimation, then, I will establish some procedures based on semiparametric pseudo likelihoods. Afterwards, the asymptotic theory and an algorithm to maximize the pseudo likelihood function will be developed. This work is mainly motivated by a longitudinal study on HIV/AIDS CD4 cell counts. The proposed methods can be applied to many other health related areas. I will also analyze a data set from cotton factory workers on dyspnea. To illustrate our proposed methods, both real data examples and some simulation results will be discussed.