Studying Effect Of Subsamples On Estimated Power Of Moses'S Test Used To Test Variances

Abstract

This study compared two tests to test for difference in estimated powers for two underlying distributions (normal and exponential): Moses and F-test. Moses test can vary depending on the subsamples chosen, while F-test is not suitable for exponential distributions. Powers are estimated for sample sizes of 18 with subsamples of sizes 3 and 6. Powers are estimated for sample sizes of 36 with subsamples of sizes 3, 6, and 9 and for sample sizes of 64 with subsamples of sizes 4, 8, and 16. Powers were estimated where just the variances were different and then where both means and variances were different. Unequal sample sizes were also considered. The study recommends that subsample sizes for normal and exponential distributions vary with sample size when using the Moses test. Power estimates were higher when subsample sizes were 1/3 to 1/4 of the original sample size, but results were inconsistent with exponential distributions.