A COMPUTATIONAL APPROACH TO ONE-WAY ANOVA AND ITS MODEL ROBUSTNESS

Ching-Hui Chang (Taiwan, R. O. C.) and Nabendu Pal (U. S. A.)

Received December 20, 2007

Abstract

A recently developed simple ‘computational approach test’ is applied to the traditional one-way ANOVA, i.e., testing the equality of several population means assuming homoscedasticity. Some theoretical properties of our proposed approach such as the size and the power are studied under both the normal model as well as the t-model. It appears that our computational approach can challenge the usual F-test method in terms of power and model robustness. In our study we also investigate some features of Analysis of Means (ANOM) method, and make some important observations. The whole idea behind this computational approach is to provide applied researchers a statistical tool to carry out a comparison of several population means, in a parametric set-up, without worrying about the sampling distribution of the test statistic.

Keywords and phrases: homoscedasticity, parametric bootstrap, t-model.