A Study on Robustness of the Paired Sample Tests

• Published in: Industrial Engineering & Management Systems Vol. 19; no. 2; pp. 386 - 397
• Main Authors: Park, Chanseok, Wang, Heng, Hwang, Wook-Yeon
• Format: Journal Article
• Language: English
• Published: 대한산업공학회 01.06.2020
• Subjects: 산업공학; Robustness; Statistical Power; Paired Test; t-Test
• ISSN: 1598-7248; 2234-6473
• DOI: 10.7232/iems.2020.19.2.386

The paired sample t-test is one of the widely-used statistical procedures for comparing the equality of the means of the two paired populations. The basic underlying assumption of the test is that observations are normally distributed and uncontaminated, whereas this assumption is easily violated in practice, which could result in an improper effect on the result. There exist other paired sample tests without the assumption so that extensive comparisons among them in terms of robustness can be a guideline for practitioners. In this regard, we investigate the robustness properties of several widely-used paired sample tests and evaluate these tests by comparing their performances under the possibility of (i) data contamination and (ii) normal model departure. These tests include the paired sample t-test, Wilcoxon test, Yuen’s t-test, and two robustified t-tests. It is shown that the robustified t-tests perform well even when the basic underlying assumption is valid, and clearly outperform the other tests in the case of data contamination and normal model departure as well. KCI Citation Count: 0