Spring Seminar Series - June 11, 2003
University of
Minnesota
School of Statistics
College of Liberal
Arts
On the Relationship between Bayesian and
Frequency Theory Prediction
Paul H. Kvam
Georgia Tech
University
Wednesday, June 11, 2003
3:30 PM, 127
Ford Hall
Minneapolis, East Bank Campus
Social at 3:00 PM, 300
Ford Hall
Abstract
Given
a sufficient statistic, basic predictive inference based on frequency theory
actually implies the existence of a prediction distribution function, conditional
on the sufficient statistic. Unlike Bayesian posterior predictive functions,
the derived distribution is not necessarily a valid one. If it is, the prediction
distribution function is necessarily a Bayesian predictive function. In such
cases, the frequency theory prediction method implies a particular Bayesian
prior on the nuisance parameter, thus these prediction methods represent
a special case of Bayesian predictive inference.