Robust model selection in regression via weighted likelihood methodology

Robust model selection procedures are introduced as a robust modification of the Akaike information criterion (AIC) and Mallows Cp. These extensions are based on the weighted likelihood methodology. When the model is correctly specified, these robust criteria are asymptotically equivalent to the classical ones under mild conditions. Robustness properties and the performance of the procedures are illustrated with examples and Monte Carlo simulations. (C) 2002 Elsevier Science B.V. All rights reserved.