2015 QPRC

Penalized regression for variable screening in designed experiments

Clay Barker, SAS Institute

Penalized regression techniques like such as the Lasso have become very popular for modeling observational data. Using penalized regression to analyze designed experiments presents a different challenge. For these data, variable screening may be a more important goal than prediction accuracy. We will look at the performance of the Lasso for screening variables in designed experiments with a specific focus on supersaturated designs.