On a bayesian criterion for choosing predictive sub-models in linear regression |
| |
Authors: | A S Young |
| |
Institution: | (1) Department of Mathematics, University of Benin, Benin City, Nigeria, Africa |
| |
Abstract: | Summary We treat the model selection problem in regression as a decision problem in which the decisions are the alternative predictive
distributions based on the different sub-models and the parameter space is the set of possible future values of the regressand.
The loss function balances out the conflicting needs for a predictive distribution with mean close to the true value ofy but without too great a variation. The treatment is Bayesian and the criterion derived is a Bayesian generalization of Mallows
(1973)C
p
, the Bivar criterion (Young 1982) and AIC (Akaike 1974). An application using a graphical sensitivity analysis is presented. |
| |
Keywords: | Selection Criterion Model Choice Regression Bayesian Analysis Predictive distribution |
本文献已被 SpringerLink 等数据库收录! |