DOUBLY ROBUST INFERENCE WITH MISSING DATA IN SURVEY SAMPLING

Abstract

Statistical inference with missing data requires assumptions about the population or about the response probability. Doubly robust (DR) estimators use both relationships to estimate the parameters of interest, so that they are consistent even when one of the models is misspecified. In this paper, we propose a method of computing propensity scores that leads to DR estimation. In addition, we discuss DR variance estimation so that the resulting inference is doubly robust. Some asymptotic properties are discussed. Results from two limited simulation studies are also presented