Bayesian model comparison through
posterior deviances (2pm)
Murray Aitkin
Department of Mathematics and Statistics, University of Melbourne
This talk gives an overview of the use of the posterior
distributions of likelihoods and likelihood ratios for model comparisons, as an
alternative to frequentist likelihood ratio tests and Bayesian Bayes factors.
Examples are given of a Bayesian version of the t-test, the goodness of fit of
a Poisson model to count data, and an evaluation of the number of components
needed for a mixture of normal distributions for the Roeder galaxy recession
velocity data.
Prof. Murray Aitkin is a Professorial Fellow in
the Department of Mathematics and Statistics at the University of
Melbourne. He trained in Statistics
at Sydney University and did postdoctoral work at the Thurstone
Psychometric
Laboratory at Chapel Hill. Since then he has held senior positions in
Macquarie
University, the Educational Testing Center, Lancaster University, Tel
Aviv
University, the University of Newcastle (UK), Australian National
University
and the University of Western Australia, Education Statistics Services
Institute (US) and now at Melbourne University. He is a fellow of the
American
Statistical Association and an elected member of the International
Statistical
Institute. Prof. Aitkin has a range of research interests in
theoretical and
applied statistics, including Bayesian, frequentist and likelihood
theories of
inference, modelling in survey sampling and item-response modelling in
large-scale testing.