Relative Power of the Wilcoxon Test, the Friedman Test, and Repeated-Measures ANOVA on Ranks
Donald W. Zimmerman,Bruno D. Zumbo
DOI: https://doi.org/10.1080/00220973.1993.9943832
1993-07-01
The Journal of Experimental Education
Abstract:Many introductory statistics textbooks in education, psychology, and the social sciences consider the Friedman test to be a nonparametric counterpart of repeated-measures ANOVA, just as the Kruskal-Wallis test is a counterpart of oneway ANOVA. However, it is known in theoretical statistics that the Friedman test is a generalization of the sign test and possesses the modest statistical power of the sign test for normal as well as many nonnormal distributions. Although not familiar to researchers, another significance test that can be regarded as a nonparametric counterpart of repeated-measures ANOVA is a rank-transformation procedure, in which the usual parametric statistical analysis is performed on ranks replacing the original scores. In the present computer simulation study we compared the ordinary paired-samples Student t test, the Wilcoxon signed-ranks test, and the sign test for correlated samples from normal, uniform, mixed-normal, exponential, Laplace, and Cauchy distributions, for which the relative efficiency of the methods is known. We also compared repeated-measures ANOVA, repeated-measures ANOVA on ranks, and the Friedman test for k mutually correlated samples from the same distributions, where k = 2,3, and 4. Power functions revealed that the Friedman test performed like the sign test for all distributions, whereas ANOVA on ranks performed like the Wilcoxon test These comparisons emphasize that classification of these statistical tests in introductory textbooks should be revised and that more attention should be paid to the rank transformation concept.