A TEST FOR COMPARING k INDEPENDENT GROUPS WITH A MEASURE OF LOCATION

Abstract

It is well known that the conventional method for comparing k independent groups, one-way ANOVA F test, depends on normal theory assumptions. In this work, a new test is proposed which is based on trimmed means and a bootstrap t method. The proposed test, ANOV A F test and its rank- based nonparametric counterpart Kruskal-Wallis test were compared in terms of saving nom nal Type I error via Monte Carlo simulations by using moderate sample sizes. It was fo und that the proposed test copes quite well under normality and homogeneity and it was much more preferable than those two traditional tests under non-normality and heterogeneity of variances