SCHOOL OF STATISTICS
and
THE COLLEGE OF LIBERAL ARTS
UNIVERSITY OF MINNESOTA
BUEHLER-MARTIN DISTINGUISHED
LECTURER SERIES
April 10, 11 and 12, 2007
Established by Mr. and Mrs. Thomas Martin
in Memory of
Robert J. Buehler, Professor of Statistics (1963-1988)
Mike
Titterington
Department of Statistics
University of Glasgow
Bayesian
Measures of Complexity, and Model Selection, Based on Incomplete Data
Thursday, April 12, 2007
3:30 PM
Ford Hall 115
(Refreshments: 3:00 PM, Ford Hall 300)
A fairly recent paper by Spiegelhalter
et al. (2002, JRSSB) introduced the so-called Deviance Information
Criterion (DIC) as a possible medium for model choice in a Bayesian
context. However, extension of the formulation beyond simple data
structures is not straightforward. The talk will consider a range
of plausible ways of defining DICs in the presence of incomplete data,
the various versions will be assessing in the context of mixture
distributions, and some general points will be made about the future
prospects for DIC.