Multi-stage tests based on sequential ranks†

Abstract

In this paper we present a technique for constructing k-stage Wilcoxon rank sum and signed rank tests for testing equality of two distributions. The test statistics suggested for use are functions of the sequential ranks (defined by Parent) of the observations obtained. When the underlying population from which observations are taken is normal, Monte Carlo studies show that a considerable savings in average sample size can be obtained by using multi-stage tests rather than fixed sample size tests while power levels are maintained.