110-2009: A SAS® Macro for Adaptive Regression Modeling

A SAS macro called genreg is available from the author for conducting adaptive regression modeling. It is written primarily in the matrix language PROC IML and supports nonparametric linear, logistic, and Poisson regression modeling of expected values and/or of variances/dispersions in terms of fractional polynomials in one or more predictor variables. Fractional polynomial models are compared using kfold likelihood cross-validation (LCV) and adaptively selected through heuristic search. The genreg macro supports modeling of independent outcomes under normal, logistic, and Poisson distributions. For the normal distribution case, it also supports modeling of repeated measurements under either order 1 autoregressive or exchangeable (i.e., constant) correlation structures. For the logistic case, it supports polytomous-valued outcomes under generalized logits. An overview of the macro is presented and its use demonstrated through an adaptive growth curve analysis.