Nonnormal Data and Unequal Variance

Nonnormality is likely to widen confidence intervals and decrease power. The robustness of Anova applies mainly to the false alarm parameter, α. Power, in contrast, can be seriously affected by nonnormality.