|
Activity Number:
|
363
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2008 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #302137 |
|
Title:
|
Numerical Identifiability and Convergence for Gibbs Sampling of Nonlinear Models
|
|
Author(s):
|
Seongho Kim*+ and Lang Li+
|
|
Companies:
|
Indiana University School of Medicine and Indiana University
|
|
Address:
|
Division of Biostatistics, Department of Medicine, Indianapolis, IN, 46202, 410 West 10th St., Carmel, IN, 46032,
|
|
Keywords:
|
Convergence ; Gibbs Sampler ; Nonidentifiability
|
|
Abstract:
|
Although formal identifiablility is not an issue in a Bayesian viewpoint, a poorly behaved Gibbs sampler frequently arises. The relationship between the prior selection and the convergence rate of a Gibbs sampler has been mostly investigated on nonidentifiability. The objective of this article, however, is to investigate the relationship between two Gibbs sampling schemes and their convergence rates on the numerical identifiability of parameters of a nonlinear model, given a prior information. We clarify that separate components in a Gibbs sampler result in a more efficient sampling scheme on nonidentifiability, while grouping random components are more efficient on identifiability. We also suggest a scheme for checking numerical identifiability of the parameters at the initial estimates for given experiments on the model.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2008 program |
