Abstract

The pooled variance t-tests used prominently for comparing means between two groups is usually restricted with the assumptions of normality and homogeneity of variances. However, the violation of the assumptions happens in many real world data. In this study, the conditions where the pooled variance t-test would fail were investigated. The performance of the t-test was evaluated under different conditions. They were sample sizes, type of distributions (normal or non-normal), and unequal group variances. The Type I error rates and power of the pooled variance t-tests for different designs were obtained and compared. The results showed that the test failed dramatically when the group sample sizes were small and unequal with slight departure from homogeneity of variances.

Key words: Pooled variance, power, t-test, type 1 error.