Estimation in Nonsmooth Moment Condition Models with Nonignorably Nonresponse
主 题: Estimation in Nonsmooth Moment Condition Models with Nonignorably Nonresponse
报告人: 唐年胜教授 (云南大学)
时 间: 2015-10-22 16:00-17:00
地 点: 理科一号楼1418
This paper focuses on parameter estimation of a general class of possibly over-identified non-smooth moment conditions (MCs) when observations are subject to missingness not at random. We assume a parametric model on the response mechanism, and construct a set of the augmented inverse probability weighted MCs by reformulating MCs through a propensity score-based and kernel-assisted imputation procedure. A class of nonparametric and semiparametric estimates for the parameter of interest is developed via applying the theory of generalized method of moments (GMM) to the modified MCs. Under a set of sufficient conditions, we establish the $\sqrt{n}$-consistency and asymptotic normality of the proposed estimators for the cases that the propensity score is known or unknown or belongs to a correctly specified parametric family. Two methods, including GMM together with a validation sample and semi-empirical likelihood method by incorporating auxiliary information from the calibration constraints under non-ignorability, are proposed to obtain the consistent estimator of the unknown propensity score. With some mild conditions, the provided sufficient conditions are satisfied for the both parametric estimates of the propensity score. A simulation study is conducted to evaluate the performance of the proposed methodologies. A real data set is used to illustrate the proposed methodologies.
