Testing variance equality with randmization tests

Abstract

A simulation experiment was conducted to explore the operating characteristics of randomization tests for testing the equality of two population variances. The F test, and adjusted testing procedure suggested by Box and Andersen (1955) were also included in this study. The F statistic was conservative for samples from Uniform populations and anti-conservative for samples from Double Exponential populations. The Box–Anderson procedure provided better control of the significance level though this procedure tended to be anti-conservative. A partial randomization test was seen to perform with approximately the same characteristics as the based on all of the permutations. Both of the randomization tests maintained the nominally specified significance levels, and had comparable power to the other procedures.