This computer simulation study evaluates the robustness of the nonparametric Levene test of equal variances (Nordstokke & Zumbo, 2010) when sampling from populations with unequal (and unknown) means. Testing for population mean differences when population variances are unknown and possibly unequal is often referred to as the Behrens-Fisher problem when the populations are normally distributed, and the generalized Behrens-Fisher problem when the populations are non-normal. The nonparametric Levene test was developed to overcome reductions in power of the original Levene test of equal variances in the case of the generalized Behrens-Fisher problem. We use a Monte Carlo computer simulation to demonstrate that sampling from populations with unequal and unknown means can lead to incorrect (either inflated or decreased) Type I error rates of the nonparametric Levene test. Centering samples using either sample means or medians does not correct the Type I error rates. This note is intended to make applied researchers aware of this problem when testing for the equality of population variances with the NPL test and in general. Accessed 1,444 times on https://pareonline.net from September 17, 2018 to December 31, 2019. For downloads from January 1, 2020 forward, please click on the PlumX Metrics link to the right.
Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.
Shear, Benjamin R.; Nordstokke, David W.; and Zumbo, Bruno D.
"A Note on Using the Nonparametric Levene Test When Population Means Are Unequal,"
Practical Assessment, Research, and Evaluation: Vol. 23
, Article 13.
Available at: https://scholarworks.umass.edu/pare/vol23/iss1/13