Andrew Womack
Department of Statistics
University of Florida

TITLE:  Some Uses of Information Theory in Model Selection

ABSTRACT:  
In Bayesian Model selection the quantities of interest are
marginal distributions and Bayes' Factors.  Though these have been
shown to behave well when prior distributions are proper (and have
nice coherency properties), they also give rise to the
Jeffreys-Lindley paradox when improper priors are used.  We will look
at several methods that have been developed for use with improper
priors, discuss those that can provide consistency, and derive some
properties of a new class of selectors that provides nice consistency
results (and are asymptotically equivalent to Bayes' Factors)