Construction of Non-extreme Weighted Regression Weights for Sample Surveys

Abstract

As a method for using auxiliary information to estimate the population quantities, we consider a simple weighted regression procedure that has fewer extreme weights than a general regression procedure or the raking ratio procedure. The proposed weighted regression procedure reduces the effect of outlier in deriving regression weights, and thus, the range of weighted regression weights is narrower than that of general regression weights. We briefly discuss the asymptotic properties of the weighted regression estimator and suggest a possible variance estimator. Further, we compare the weighted regression estimator with several calibration procedures through a simulation study. Unlike other non-extreme or range-restricted procedures, the proposed method does not require an iterative procedure but still shows comparable performance.