In testing the significance of treatment effects in randomized clinical trials (RCTs), randomization-based inference is distinguished from population-based parametric and nonparametric inference, such as the t-test or permutation tests, taking into account three properties: preservation of type I error rate, relation of power to the randomization procedure, and flexibility in choosing the test statistic. In this paper, we revisit rationale of the properties and provide justification through simulations. We propose that the choice of randomization procedures and the analysis of RCTs can be facilitated by the application of randomization-based inference.
Randomization tests Population tests Randomization procedure Type I error rate Statistical power
This is a preview of subscription content, log in to check access.