STAT 425 Week 09 - Variable Selection

In selecting predictor variables to include in a regression model, an important consideration is the bias-variance tradeoff: adding variables to the model generally reduces the bias of the model, but at the cost of increased variance or uncertainty in estimating the coefficients. This suggests the parsimony principle, that we try to find the simplest adequate model for the data. We discuss both testing based approaches and criterion based approaches for selecting which variables to include in the model.