Abstract
A criterion is developed from the sum of weighted squared bias and variance for use in selecting regressors. We show that it is a generalization of the Cp criterion and describe a graphical analysis of it. Its relationship with other selection criteria is discussed. Examples illustrating its use are included.