To pool or not to pool: A comparison between two commonly used test statistics

Abstract

When it comes to testing hypotheses regarding two population means, the most commonly used test is the two-sample t-test. There are two versions of this test, one is used when the variances of the two populations are equal (the pooled test) and the other one is used when the variances of the two populations are unequal (the unpooled test). The pooled test seems to have fallen into some disfavor because of its 'claimed' sensitivity to departures from the assumptions of equal population variances. Through a simulation study, this paper demonstrates that although both the pooled and the unpooled test underperform at times in their allocated settings, the overall performance of the pooled t-test is significantly superior to that of the unpooled t-test. © 2013 Academic Publications, Ltd.