Abstract
When testing the equality of the means of J independent normal distributions, two-stage procedures have the advantage of providing exact control over both Type I errors and power even when the variances are unequal. Single-stage procedures have been proposed for handling unequal variances, but recent results (which are briefly reviewed in this paper) have shown that in certain practical situations these approximate solutions give unsatisfactory control over both Type I errors and power. A practical problem with two-stage procedures is the assumption that an equal number of observations is sampled from each treatment group in the first stage. Of course, for various reasons, a researcher might conduct a study resulting in unequal sample sizes. This paper describes a simple yet accurate method for dealing with this problem.
Get full access to this article
View all access options for this article.