The simulated randomization test.

Abstract

Non-parametric statistical methods have been the subject of renewed interest. They are particularly useful in the behavioral sciences as they can be applied to a large class of distributions, and because the investigator is not forced into making any erroneous assumptions about a Gaussian distribution for the parent population or about equal variances in the contrasted groups. The randomization test is the most powerful non-parametric test [1] but it is rarely used because it calls for a prohibitive amount of computation. However, two tools now available make it more feasible: simulation and high-speed computer calculations. This test's importance lies in the extremely simple logic by which it is set up. The program described here is written in BASIC language for a microcomputer and carries out an approximate randomization test using a simulated distribution instead of the entire distribution. This version is slightly less powerful, a small price to pay for reducing the enormous calculation capacity otherwise required.