An analysis of statistical power for parametric ancova and rank transform ancova

The power of the parametric ANCOVA and of the rank transform ANCOVA is estimated empirically for a variety of situations in which the data are conditionally non-normal and heteroscedastic. The results indicate that the power of the rank transform ANCOVA is affected by conditional non-normality and heteroscedasticity, but the power of the parametric ANCOVA is not. The power difference between the rank transform and parametric ANCOVA can be positive or negative and can be trivial or important depending on the size of the treatment effect, the degree of posttest-covariate correlation, the type and extent of non-normality, and the degree of heteroscedasticity.