MODIFIED AIC AND C-P IN MULTIVARIATE LINEAR-REGRESSION

The Akaike information criterion, Are, and the Mallows' C-p, criterion have been proposed as approximately unbiased estimators for their ris ks or underlying criterion functions. In this paper we propose modifie d AIC. and C-p for selecting multivariate linear regression models. Ou r modified AIC and modified C-p are intended to reduce bias in situati ons where the collection of candidate models includes both underspecif ied and overspecified models. In a simulation study it is verified tha t the modified AIC and modified C-p provide better approximations to t heir risk functions, and better model selection, than AIC and C-p.