Some Properties of Probability Lattice Sampling

Abstract

The accuracy of three estimators of sampling variance using non-replacement probability lattice samples was determined empirically on a microuniverse. A simple split-sample scheme, although biased, was superior to the unbiased Horvitz-Thompson and Yates-Grundy estimators. A weighted ANOVA type of model for representing two-way universes with unequal cell sizes is presented and tested empirically to determine how well it can account for sources of sampling variability using 0, 1 and 2-way controls with EP (equal probability) and PNR (probability proportional to size with non-replacement) samples.