OPTIMAL VOLUME-CORRECTED LAPLACE- METROPOLIS METHOD
Resource
ANNALS OF THE INSTITUTE OF STATISTICAL MATHEMATICS v.55 n.3 pp.655-670
Journal
ANNALS OF THE INSTITUTE OF STATISTICAL MATHEMATICS
Journal Volume
v.55
Journal Issue
n.3
Pages
655-670
Date Issued
2003
Date
2003
Author(s)
HSIAO, CHUHSING, K.
Abstract
The article provides a refinement for the volume-corrected Laplace-Metropolis estimator of the marginal likelihood of DiCiccio et al. The correction volume of probability a in DiCiccio et al. is fixed and suggested to take the value alpha = 0.05. In this article a is selected based on an asymptotic analysis to minimize the mean square relative error (MSRE). This optimal choice of a is shown to be invariant under linear transformations. The invariance property leads to easy implementation for multivariate problems. An implementation procedure is provided for practical use. A simulation study and a real data example are presented.
Subjects
Bayes factor
Laplace approximation
marginal probability
Markov chain Monte carlo
Type
journal article