Title: Semiparametric Adaptive Estimation With Nonignorable Nonresponse Data

Abstract: When the response mechanism is believed to be nonignorable or not missing at random (NMAR), a valid analysis requires stronger assumptions about the data than do standard statistical methods. Semiparametric estimators have been developed under the correct model specification assumption for the response mechanism. In this paper, we consider a scheme for obtaining the optimal estimation for the parameters such as the mean and propose two semiparametric adaptive estimators that do not require any model assumptions except for the response mechanism. Asymptotic properties of proposed estimators are discussed, and we present an application to Korean Labor and Income Panel Survey data.