A SAS macro for stepwise correlated binary regression.

Several regression methods have been proposed for the analysis of correlated binary data, but none deals with the selection of covariates when there exist a large number of potentially relevant covariates. We present a SAS macro based on a stepwise selection procedure for the analysis of correlated binary data. Using regression methods based on generalized estimating equations originally proposed by Liang and Zeger and extended by Prentice, we describe a score test for forward selection, a Wald's test for backward elimination, and a test for model adequacy based on generalized scores. The methodology and the accompanying computer macro program written in SAS IML are illustrated with data from a prospective study of functional decline in the activities of daily living in a group of elderly patients.