A robust test based on bootstrapping for the two-sample scale problem

Abstract

For testing the homogeneity of variances, modifications of well-known tests are known which combine rigorous theory with resampling (bootstrap). We propose versions of these tests, which are computationally simpler (although asymptotically equivalent). The earlier procedures used the smooth bootstrap with two thousand bootstrap replications per sample whereas our proposals use only the classical bootstrap (or percentile method) with just one thousand bootstrap replications per sample, and also required much less computing time. Our proposals cover the Ansari-Bradley-, Mood- and Klotz-tests. We explain their superiority over the existing methodologies available in textbooks and packages