In the Pursuit of Balanced Groups: A SAS® Macro for an Adaptive Randomization Test

In adaptive randomized experiments, researchers verify that random allocation yielded equivalent experimental groups before proceeding with data analysis. Given a precise definition of balanced groups (e.g., Cohen’s d < 0.5 for a continuous covariate), researchers may plan to keep reshuffling participants until balance is achieved. However, reshuffling can invalidate standard parametric tests, such as t-tests, F-tests, and -tests. This paper describes a non-parametric analog of the two-sample t-test, called an adaptive randomization test, which can preserve Type I error control in adaptive randomized experiments. Although such tests are well established in mainstream statistics (Morgan & Rubin, 2015), they are not readily available in SAS making them difficult to implement. This paper provides an overview of adaptive randomization tests, presents a macro for doing them in base SAS, and concludes with an executed example of the macro.