Variable Selection in Regression Analysis Using LASSO, LARS, Elastic Net, and Best Subsets - Brenda Gillespie, University of Michigan

With so many variable selection procedures available now, it's hard to know which one to use.  This talk will describe the newer methods (LASSO, LAR(S), Elastic Net), with contrasts to older methods (ridge regression, stepwise, and best subset selection).  Methods will be compared in terms of performance with collinearity, modeling with a medium versus huge number of covariates,  importance of parsimony, use of cross-validation, and more.  SAS procedures illustrated will include GLMSelect and HPreg (for newer methods), and PHREG and LOGISTIC (for older methods).

About the Presenter

Brenda Gillespie works at the University of Michigan, and is the Associate Director of the campus unit, Consulting for Statistics, Computing, and Analytics Research (CSCAR). She works on several medical research projects, mainly in kidney disease. She has used various types of regression analyses for over 40 years, and has lived through many advances in variable selection methodology. In this exciting new era, she is happy to share what she has learned.