SELECTION CRITERIA BASED ON MONTE CARLO SIMULATION AND CROSS VALIDATION IN MIXED MODELS

Junfeng Shang (U. S. A.)

Received October 17, 2007

Abstract

In the mixed modeling framework, Monte Carlo simulation and cross validation are employed to develop an “improved” Akaike information criterion, AICi, and the predictive divergence criterion, PDC, respectively, for model selection. The selection and the estimation performance of the criteria is investigated in a simulation study. Our simulation results demonstrate that PDC outperforms AIC and AICi in choosing an appropriate mixed model as a selection criterion, and AICi is less biased than AIC and PDC in estimating the Kullback-Leibler discrepancy between the true model and a fitted candidate model.

Keywords and phrases: Akaike information criterion (AIC), improved AIC (AICi), predictive divergence criterion (PDC), Kullback-Leibler discrepancy.