Identification And Inference For Partial Linear Model With Nonignorable Missing Data

Missing data,one type of complicated data,is widely exist in current research areas such as biological medicine,education and economics.Recent researches about missing data problems are mostly focus on ignorable missing case.However,some reality problems,especially sensitive questions,are more complicated.For example,in investigating faculties’ salary situation,the low and high salary faculties tend to reject to fill out the questionnaire,which leads to a nonignorable missing data case.The exist approaches used in ignorable missing data problem would lead to bias in estimate results when dealing with nonignorable problems.Therefore,researching effective approaches to dealing with nonignorable missing data is very meaningful.Partial linear model,which is a very useful tool and one of the most classic semi-parameter model,which has great advantages comparing with common paramatric models,has not been well developed in dealing with nonignorable missing data problems.In this thesis we use the empirical likelihood methods for parameter estimation in missing mechanism and focus on the statistical inference for the Partial Linear model with the nonignorable missing data.The main works include:(1)We built a parameterized partial linear model and consider the identification problem for this model when missing mechanism is logistic model with data missing nonignorable according to identification methods of Miao and Geng.(2)Based on regression weighting method of propensity scoring approach,we obtain an effective estimator of the unknown parameters in partial linear model.According to the empirical likelihood methods provided by Qin,Leung and shao,we estimate the missing probability function with the nonignorable missing data.(3)We report the results of a small simulation study for the proposed methods and prove their feasibility.