The REGSELECT Procedure

The functionality of the REGSELECT procedure closely resembles that of the HPREG procedure, which is a high-performance procedure. The REGSELECT procedure is the next generation of the HPREG procedure, and it was developed specifically for SAS Viya. Both procedures are designed to run on a cluster of machines that distribute the data and the computations.

Both the REGSELECT and HPREG procedures fit and perform model selection for ordinary least squares regression models, which you can specify as general linear models that include classification variables. You request model selection by using the SELECTION statement.

The REGSELECT and HPREG procedures offer the same methods of effect selection, but the REGSELECT procedure also produces selection plots by using ODS Graphics. The REGSELECT and HPREG procedures offer the same methods of customizing the model selection, and the same set of regression diagnostics. The main differences in functionality are that the REGSELECT procedure provides the ability to specify constructed effects with the EFFECT statement, and it computes Type 3 tests of effects.

The functionality of the REGSELECT procedure also closely resembles that of the GLMSELECT procedure. Both procedures fit and perform model selection for ordinary least squares regression models, which you can specify as general linear models that include classification variables. Both procedures offer the same methods of effect selection (including the LAR and LASSO methods), the ability to use external validation data and cross validation as selection criteria, and extensive options to customize the selection process. Both procedures provide the ability to specify constructed effects with the EFFECT statement.

Unlike the GLMSELECT procedure, the REGSELECT procedure does not perform model selection by default. If you request model selection by using the SELECTION statement, then the default selection method is stepwise selection based on the Schwarz Bayesian information criterion (SBC). This default matches the default method in PROC GLMSELECT.

With the REGSELECT procedure—but not with the GLMSELECT procedure—you can request observationwise residual and influence diagnostics in the OUTPUT statement and variance inflation and tolerance statistics for the parameter estimates. If the fitted model has been obtained by performing model selection, then these statistics are conditional on the selected model and do not take into account the variability introduced by the selection process.

A major functional difference between the REGSELECT procedure and the REG procedure is that the REGSELECT procedure enables you to specify general linear models that include classification variables. In this respect it is similar to the GLM and GLMSELECT procedures in SAS/STAT.

Unlike the REG procedure, the REGSELECT procedure supports the LAR and LASSO methods, the ability to use external validation data and cross validation as selection criteria, and extensive options to customize the selection process. PROC REGSELECT does not support the all-subset-based methods that you find in PROC REG, nor does it support the MINR method.

Like the REG procedure, PROC REGSELECT does not perform model selection by default. If you request model selection by using the SELECTION statement in PROC REGSELECT, then the default selection method is stepwise selection based on the Schwarz Bayesian information criterion (SBC).

With the REGSELECT procedure, as with the REG procedure, you can request observationwise residual and influence diagnostics in the OUTPUT statement, and variance inflation and tolerance statistics for the parameter estimates. If the fitted model has been obtained by performing model selection, then these statistics are conditional on the selected model and do not take into account the variability introduced by the selection process.