The Beginnings of Randomization Tests
Herbert A. David · The American Statistician · 2008
Randomization (or permutation) tests became known through R.A. Fisher's 1935 demonstration that no assumption of normality was needed for the analysis of Darwin's paired t test data: Assuming normality was simply more convenient than tedious enumeration of cases more removed from the null hypothesis of no difference than were the experimental results. We describe an interesting precursor, often overlooked or merely cited, to Fisher's treatment. Next, we outline the analytical development of randomization theory by Welch and Pitman, including the latter's realization of broad possibilities for tests without distributional assumptions. Finally, we briefly indicate the current status of randomization tests which are no longer tedious to carry out with modern computers.