Remedial measures for violations of assumptions in ANOVA
( \newcommand{\kernel}{\mathrm{null}\,}\)
A basic question when one suspects that the variances of the treatment groups may be unequal or the distribution of measurements may be non-normal, one question is: How to make up for unequal variance and/or nonnormality and still apply the ANOVA framework?
- Transformation of the response variable to make its distribution closer to normal and to stabilize the variance.
- When the departures are too extreme such that transformations do not work: use nonparametric tests instead of the usual F-test for testing the equality of means.