A randomization-test wrapper for SAS PROCs
page Paper 251-27. (2002)3752<m:linebreak></m:linebreak>Resampling.Abstract
SAS/STAT procedures are often used in settings where the underlying model assumptions are not really met. Permutation tests can permit one to assess correct pvalues in many of these cases, but too often the total number of permutations is unmanageable. Instead, a randomization test using a random subset of all permutations can be used. This paper demonstrates a flexible, extensible pair of 'wrapper' macros which allow the user to use almost any SAS PROC and get a randomization p-value for the desired test at the back end. The wrapper is a front macro to create the data set of replicates with a randomized dependent variable, and a back macro to process the results of the SAS PROC and compute the p-value according to a randomization test. These bracket the already-written SAS procedure or process, which requires minimal alteration. We demonstrate the wrapper on several SAS procedures, using standard output data sets as well as tables pulled out using ODS.