The tail functions approach to confidence estimation in survey sampling

Abstract

This paper shows how the method of tail functions (Puza and O'Neill, 2005) can be applied in survey sampling so as to produce improved confidence intervals for a finite population quantity of interest. The method works best in the presence of prior information and involves choosing a tail function that leads to the interval which is optimal in some sense, such as minimising prior expected length. The method is illustrated by application to inference on the finite population mean, and tested via a Monte Carlo study.