On the development of intrinsic Bayes factors
Abstract
The Intrinsic Bayes Factor (IBF), a Bayesian model selection criterion recently proposed by Berger and Pericchi, possesses many attractive properties such as broad generality of application and automatic implementation. It was shown to perform well for a variety of situations involving sequences of IID data. The first part of the thesis suggests a modification of IBFs for model selection with dependent data structures, autoregressive models in particular. We also present numerical problems arising in direct computations of IBFs and discuss their solutions. The second part of the work suggests an effective computational scheme for the evaluation of the integrals involved in the IBF and discusses several reasonable approaches towards reducing the amount of the computations. The last part of the thesis concentrates on the development of IBFs for nonnormal linear models.
Degree
Ph.D.
Advisors
Berger, Purdue University.
Subject Area
Statistics
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.