Comparing Measures of Location: Some Small-Sample Results When Distributions Differ in Skewness and Kurtosis Under Heterogeneity of Variances

Abstract

This article considers the problem of comparing two independent groups in terms of some measure of location. Nine methods, including a new method and a new robust estimator, were compared in terms of actual significance level and power. Simulations were performed using data generated from both theoretical distributions and data stemming from actual studies. For the theoretical distributions, the bootstrap-t B 2 t method (with 20% trimming) and Yuen's test (with 10% trimming) performed best. Otherwise, the bootstrap-t B 2 t method (with 25% trimming) and a percentile bootstrap method with trimmed means (20% and 25% trimming) or one-step M-estimator performed best.