An Akaike Criterion based on Kullback Symmetric Divergence in the Presence of Incomplete-Data

Abstract

This paper investigates and evaluates an extension of the Akaike information criterion, KIC, which is an approximately unbiased estimator for a risk function based on the Kullback symmetric divergence. KIC is based on the observed-data empirical
log-likelihood which may be problematic to compute in the presence of incompletedata. We derive and investigate a variant of KIC criterion for model selection in settings where the observed-data is incomplete. We examine the performance of our
criterion relative to other well known criteria in a large simulation study based on bivariate normal model and bivariate regression modeling.