| Summary: | The classical independent t-test is often in jeopardy when the assumptions of normality or homogeneity of variances being violated. The test performs worsen when these violations occur simultaneously. Alexander-Govern test offers the alternative solution to the classical t-test when dealing with heterogeneous variances conditions. However, it produces good control of Type I error rates only if the data are normally distributed, which is a known fact that normality is hardly achieved in real life situation. As a remedy, in this study, we modify the Alexander-Govern test using trimmed mean and Winsorized mean as the location measures. Generally, the modified test using trimmed mean performs better compared to the original test in terms of Type I error rates. However, the test using Winsorized mean failed to control the Type I error rate well under most condition considered
|
|---|