A Purposeful Selection of Variables Macro for Logistic Regression

The main problem in any model-building situation is to choose from a large set of covariates those that should be included in the “best” model. A decision to keep a variable in the model might be based on the clinical or statistical significance. There are several variable selection algorithms embedded in SAS PROC LOGISTIC. Those methods are mechanical and as such carry some limitations. Hosmer and Lemeshow describe a purposeful selection of covariates algorithm within which an analyst makes a variable selection decision at each step of the modeling process. In this paper we introduce a macro, %PurposefulSelection, which automates that process. The macro is based on the following algorithm: (1) fit a univariate model with each covariate, (2) select as candidates for a multivariate model those significant at some chosen alpha level, (3) identify those variables that are not significant in the multivariate model at some arbitrary alpha level, (4) fit a reduced model and evaluate confounding by change in parameter estimates, (5) repeat steps 3 and 4 until the model contains significant covariates and/or confounders and (6) add back in the model, one at a time, any variable not originally selected, keep any that are significant, and reduce the model following steps 3 and 4. At the end of step 6, the analyst will have a “main effects model.” Performance of the macro is illustrated with the application to the Hosmer and Lemeshow Worchester Heart Attack Study (WHAS) data.