Building Multiple Linear Regression Models – Food for Thought

To obtain a parsimonious set of variables that efficiently predicts the response of interest, many people delve straight in and start performing an automated model selection method such as forward selection. This paper covers the fundamentals of what the first step really should be – getting to know your data – and how SAS® can help you do this! This knowledge is then utilized to build multiple linear regression models. Direct comparisons between PROC REG and PROC GLMSELECT are made.

Paper: 

McGahan, Colleen and Mei Chong - Building Multiple Linear Regression Models.pdf