A Combined Resampling Test for Scale in Two-Sample Designs

Situations where scale parameters are not nuisance factors to control but outcomes to explain arise in many contexts. A test which usually stands out in terms of power and robustness against non normality is the modified Levene test, however no test is found to be uniformly the best. Our aim is to identify an effective method for comparing scales in two sample designs. We propose a combined resampling test that, even if was no uniformly the best, has good overall performance under every type of distribution. Note that non combined tests show good performance only in particular situations. Size and power of the test are studied via simulation and compared with many robust tests. A practical application to quality control is discussed.