Evaluation Of Two Refined Mann–Whitney Procedures
Loading...
Date
2009-01
Authors
Lai, Choo Heng
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
Testing for the equality of means across two independent groups is a common inferential problem especially in education and psychology. One of the most frequently used tests is either the classical t test or the Mann-Whitney test. But these classical tests are not without flaws as their performance depends on underlying assumptions. A plethora of test statistics and procedures have since appeared, designed to be less sensitive to violation of the underlying assumptions. Hence selecting the appropriate robust statistical test will be tedious. This study intends to facilitate this by identifying broader robust tests and/or providing boundary conditions of popular statistical tests. This study adopts the Monte Carlo computer simulation which generates data under experimental conditions to evaluate the small-sample behaviours of the refinement procedures proposed by Babu and Padmanabhan (2002) and its
alternatives in terms of Type I error rates and statistical power. The experimental conditions that were systematically manipulated are multiple combinations of various distribution shapes, variance heterogeneity and group sample sizes. The alternatives are Welch’s (1974) test, the Mann-Whitney test, Johnson’s (1978) transformation of the Welch’s test and Hall’s (1992) transformation of the Welch’s test. The findings of this study have demonstrated that there is no statistical test that is superior to the others in all test conditions. All the identified statistical tests and procedures are specified tests. However, the proposed Refinement Procedure 1 is found to be generally more robust as it is capable of producing broader probability coverage of maintaining the Type 1 error. Furthermore the Refinement Procedure 1 is comparatively a powerful test in conditions where it is appropriate. Recommendation for the other statistical tests and procedures are made based on their respective boundary conditions discovered in this study. Statisticians will be able to utilize these boundary conditions and incorporate them into an adaptive approach of selecting a more flexible and robust statistical test.
Description
Keywords
Evaluation Of Two Refined , Mann–Whitney Procedures