Algorithms for Bayesian variable selection in regression

Conference · Wed Jan 01 04:00:00 EST 1986

OSTI ID:5934514

Mitchell, T J; Beauchamp, J J

We have developed a Bayesian approach to the problem of deciding which subset of a proposed set of k predictor variables to include in a linear regression model that is to be used for prediction. The direct implementation of this method requires the computation of the usual regression statistics for each of the 2/sup k/ possible submodels. We present here a branch-and-bound method which yields the same results much more quickly by eliminating from consideration those submodels which are destined to have negligible posterior probability. Implementation of the algorithm on the Cray X-MP supercomputer is discussed. 2 refs.

• Research Organization: Oak Ridge National Lab., TN (USA)
• DOE Contract Number: AC05-84OR21400
• OSTI ID: 5934514
• Report Number(s): CONF-8603103-1; ON: DE86008884
• Country of Publication: United States
• Language: English

Related Subjects

658000* -- Mathematical Physics-- (-1987)
71 CLASSICAL AND QUANTUM MECHANICS
GENERAL PHYSICS
ALGORITHMS
CALCULATION METHODS
COMPUTERS
CRAY COMPUTERS
DEC COMPUTERS
FORTRAN
MATHEMATICAL LOGIC
MATHEMATICS
PROGRAMMING LANGUAGES
REGRESSION ANALYSIS
STATISTICS