Outlier Robust Finite Population Estimation under Spatial non-Stationarity

Abstract

When survey data shows spatial non-stationarity then geographically weighted regression (GWR) approach explains the data more effectively than standard global regression model. In this article, two outlier robust geographically weighted regression (RGWR) estimators have been proposed to estimate the finite population total under spatial nonstationarity. The first RGWR estimator is based on winsorization whereas second one is based on filtering of outliers. In order to compare the statistical performance of proposed estimators with standard non-robust GWR estimator and a robust estimator proposed by Chamber (1986), a simulation study was carried out. It has been observed that proposed estimator based on winsorization of sampled data performs fairly well in a scenario where spatial non-stationarity appears in population and the survey data contains outliers