Paper 40

Modernizing Predictive Modeling in SAS: When to Move Beyond PROC GLM

September 23, 2025: 4:00 PM - 4:30 PM

Careers, Training, and Education (CTE), Breakout Room #7

Authors	Abstract
Duha Khan, Sriya Venkat, Sydney McDermott	SAS offers many procedures that can be used to conduct predictive analysis. PROC GLM is one of the older tools used to fit general linear models to datasets. While it includes many options that allow for the implementation of classes, subgroups, weighting, and more, there are times when another, newer procedure may be preferred. For example, since PROC GLM is not a high performance procedure, it is sometimes better to use a high-performance alternative like PROC HPREG, which is better suited for larger datasets. Additionally, PROC GLM requires manual variable selection from the programmer. While this can be useful at times, there are other cases where the automated selection with PROC GLMSELECT can prove to be far more efficient. This paper explores the pros and cons of these procedures with examples to help users choose the most appropriate predictive modeling tool for real-world applications.

Duha Khan, Sriya Venkat, Sydney McDermott

SAS offers many procedures that can be used to conduct predictive analysis. PROC GLM is one of the older tools used to fit general linear models to datasets. While it includes many options that allow for the implementation of classes, subgroups, weighting, and more, there are times when another, newer procedure may be preferred. For example, since PROC GLM is not a high performance procedure, it is sometimes better to use a high-performance alternative like PROC HPREG, which is better suited for larger datasets. Additionally, PROC GLM requires manual variable selection from the programmer. While this can be useful at times, there are other cases where the automated selection with PROC GLMSELECT can prove to be far more efficient. This paper explores the pros and cons of these procedures with examples to help users choose the most appropriate predictive modeling tool for real-world applications.