Finite Population Sampling with Multivariate Auxiliary Information

Abstract

This article examines strategies that are approximately design-unbiased and nearly optimal, assuming a large-sample survey and a regression superpopulation model. A new class of predictors is proposed to link certain features of optimal design-unbiased and model-unbiased predictors. Generalized regression predictors are shown to pervade the subclass of asymptotically design-unbiased (ADU) predictors. Generalized regression predictors are combined with model-based stratification to construct highly efficient ADU strategies.