Exact Randomization Technique

Abstract

Abstract The exact randomization technique is a distribution‐free, computationally‐intensive statistical method of dealing with small samples and uncertain distributions. The technique was first proposed by Fisher in 1935 as a method of dealing with small biological samples. Because of the computational demands, only the recent development of powerful desktop computers permits researchers to regularly use this technique. Suppose two sets of observations are recorded and labeled according to whether they come from a treatment or a control group. The concept of randomization is based on the observation that a conventional test statistic (e.g. a t statistic) can be interpreted as a typical draw from the distribution of test statistics that would be generated from randomly shuffling the treatment and control labels among the observed data. In fact, the exact randomization test is essentially a controlled example of bootstrapping.