ABSTRACT
Randomization is a method of inference that has been available for 70 years, but has only been practical about 30 years. What should have become apparent from the discussions and examples in this book is that it is mainly of value under three particular circumstances. First, it tends to have better properties than more conventional methods for standard analyses, like regression and analysis of variance, with extremely nonnormal data. Second, randomization can sometimes be applied with situations such as testing for an association between two distance matrices where no alternative seems sensible. Third, there are cases where the observations form the entire population of interest, so that randomization is the only method of inference possible.
