| Abstract: | D. W. Levine and W. P. Dunlap (see record 1982-27158-001) concluded that tests on skewed data lack power. The present author contends that this conclusion is incorrect. What they have rediscovered is that curvilinear transformations change population means and population variances, as well as the form (skewness and kurtosis) of the population. Thus, such transformations change the noncentrality parameter, φ. When the null hypothesis is false, the power entries in the rows of their Table 1 are influenced by all of these characteristics. To attribute this change to skewness alone is incorrect. Many distinctions are pertinent to the choice of whether to use such transformations: informal data analysis vs formal inference on experiments based on strong theory, the size of the error term relative to the size of the treatment effects, considerations of interactions, simplicity of results, and statistical assumptions. Skewness is the least important of these factors. (21 ref) (PsycINFO Database Record (c) 2010 APA, all rights reserved) |
