A Reaction to “Consequences of Failure to Meet Assumptions Underlying the Fixed Effects Analysis of Variance and Covariance”

Abstract

Glass, Peckham, and Sanders discourage the use of nonparametric counterparts of the t-test even when it is known that data were sampled from non-normal (e.g., highly skewed) distributions. This paper contends that Glass et al. erred in taking this position and that their error was due, at least in part, to their failure to consider the relative power of the t-test and its nonparametric counterparts under various population shapes. It is further contended that Wilcoxon’s rank-sum test has power properties that make it preferable to the t-test in many, perhaps even most, non-normal population situations.