Activity Number:
|
363
- Contributed Poster Presentations: IMS
|
Type:
|
Contributed
|
Date/Time:
|
Wednesday, August 5, 2020 : 10:00 AM to 2:00 PM
|
Sponsor:
|
IMS
|
Abstract #312286
|
|
Title:
|
Block Gibbs Samplers for Logistic Mixed Models: Convergence Properties and Comparison with Full Gibbs Samplers
|
Author(s):
|
Yalin Rao* and Vivekananda Roy
|
Companies:
|
Iowa State Univ and Iowa State University
|
Keywords:
|
Block Gibbs Sampler;
Logistic linear mixed model;
Markov chain Monte Carlo;
Geometric ergodicity;
Data augmentation;
Comparison with full Gibbs
|
Abstract:
|
Logistic linear mixed model (LLMM) is one of the most widely used statistical models. In this presentation, we consider Bayesian LLMMs with both proper and improper priors on the regression coefficients as well as variance parameters of the random effects. Generally Markov chain Monte Carlo algorithms are used to explore the associated posterior densities. We construct efficient block Gibbs samplers for Bayesian LLMMs using Polson et al's (2013) Polya-Gamma data augmentation technique. We also consider conditions guaranteeing geometric ergodicity (GE) of the block Gibbs Markov chains in both proper and improper prior cases. These theoretical results have important practical implications, including honest statistical inference with valid Monte Carlo standard errors. Improved performance of the block Gibbs samplers (compared to full Gibbs samplers) is demonstrated using numerical examples.
|
Authors who are presenting talks have a * after their name.