Accelerating computation in Markov random field models for spatial data via structured MCMC

Murali Haran, James S. Hodges, Bradley P. Carlin

Research output: Contribution to journalReview article

10 Citations (Scopus)

Abstract

Spatial Poisson models for areal count data use nonstationary "intrinsic autoregressions," also often referred to as "conditionally autoregressive" (CAR) models. Bayesian inference for these models has generally involved using single parameter updating Markov chain Monte Carlo algorithms, which often exhibit slow mixing (i.e., poor convergence) properties. These spatial models are richly parameterized and lend themselves to the structured Markov chain Monte Carlo (SMCMC) algorithms. SMCMC provides a simple, general, and flexible framework for accelerating convergence in an MCMC sampler by providing a systematic way to block groups of similar parameters while taking full advantage of the posterior correlation structure induced by the model and data. Among the SMCMC strategies considered here are blocking using different size blocks (grouping by geographical region), reparameterization, updating jointly with and without model hyperparameters, "oversampling" some of the model parameters, and "pilot adaptation" versus continuous tuning techniques for the proposal density. We apply the techniques presented here to datasets on cancer mortality and late detection in the state of Minnesota. We find that, compared to univariate sampling procedures, our techniques will typically lead to more accurate posterior estimates, and they are sometimes also far more efficient in terms of the number of effective samples generated per second.

Original languageEnglish (US)
Pages (from-to)249-264
Number of pages16
JournalJournal of Computational and Graphical Statistics
Volume12
Issue number2
DOIs
StatePublished - Jun 1 2003

Fingerprint

Spatial Data
Markov Chain Monte Carlo
Random Field
Markov Chain Monte Carlo Algorithms
Spatial Model
Updating
Oversampling
Reparameterization
Autoregression
Hyperparameters
Count Data
Poisson Model
Correlation Structure
Bayesian inference
Autoregressive Model
Mortality
Model
Grouping
Convergence Properties
Univariate

All Science Journal Classification (ASJC) codes

  • Statistics and Probability
  • Discrete Mathematics and Combinatorics
  • Statistics, Probability and Uncertainty

Cite this

@article{d305a1f27e7346f9b6f351957fd0b0f4,
title = "Accelerating computation in Markov random field models for spatial data via structured MCMC",
abstract = "Spatial Poisson models for areal count data use nonstationary {"}intrinsic autoregressions,{"} also often referred to as {"}conditionally autoregressive{"} (CAR) models. Bayesian inference for these models has generally involved using single parameter updating Markov chain Monte Carlo algorithms, which often exhibit slow mixing (i.e., poor convergence) properties. These spatial models are richly parameterized and lend themselves to the structured Markov chain Monte Carlo (SMCMC) algorithms. SMCMC provides a simple, general, and flexible framework for accelerating convergence in an MCMC sampler by providing a systematic way to block groups of similar parameters while taking full advantage of the posterior correlation structure induced by the model and data. Among the SMCMC strategies considered here are blocking using different size blocks (grouping by geographical region), reparameterization, updating jointly with and without model hyperparameters, {"}oversampling{"} some of the model parameters, and {"}pilot adaptation{"} versus continuous tuning techniques for the proposal density. We apply the techniques presented here to datasets on cancer mortality and late detection in the state of Minnesota. We find that, compared to univariate sampling procedures, our techniques will typically lead to more accurate posterior estimates, and they are sometimes also far more efficient in terms of the number of effective samples generated per second.",
author = "Murali Haran and Hodges, {James S.} and Carlin, {Bradley P.}",
year = "2003",
month = "6",
day = "1",
doi = "10.1198/1061860031833",
language = "English (US)",
volume = "12",
pages = "249--264",
journal = "Journal of Computational and Graphical Statistics",
issn = "1061-8600",
publisher = "American Statistical Association",
number = "2",

}

Accelerating computation in Markov random field models for spatial data via structured MCMC. / Haran, Murali; Hodges, James S.; Carlin, Bradley P.

In: Journal of Computational and Graphical Statistics, Vol. 12, No. 2, 01.06.2003, p. 249-264.

Research output: Contribution to journalReview article

TY - JOUR

T1 - Accelerating computation in Markov random field models for spatial data via structured MCMC

AU - Haran, Murali

AU - Hodges, James S.

AU - Carlin, Bradley P.

PY - 2003/6/1

Y1 - 2003/6/1

N2 - Spatial Poisson models for areal count data use nonstationary "intrinsic autoregressions," also often referred to as "conditionally autoregressive" (CAR) models. Bayesian inference for these models has generally involved using single parameter updating Markov chain Monte Carlo algorithms, which often exhibit slow mixing (i.e., poor convergence) properties. These spatial models are richly parameterized and lend themselves to the structured Markov chain Monte Carlo (SMCMC) algorithms. SMCMC provides a simple, general, and flexible framework for accelerating convergence in an MCMC sampler by providing a systematic way to block groups of similar parameters while taking full advantage of the posterior correlation structure induced by the model and data. Among the SMCMC strategies considered here are blocking using different size blocks (grouping by geographical region), reparameterization, updating jointly with and without model hyperparameters, "oversampling" some of the model parameters, and "pilot adaptation" versus continuous tuning techniques for the proposal density. We apply the techniques presented here to datasets on cancer mortality and late detection in the state of Minnesota. We find that, compared to univariate sampling procedures, our techniques will typically lead to more accurate posterior estimates, and they are sometimes also far more efficient in terms of the number of effective samples generated per second.

AB - Spatial Poisson models for areal count data use nonstationary "intrinsic autoregressions," also often referred to as "conditionally autoregressive" (CAR) models. Bayesian inference for these models has generally involved using single parameter updating Markov chain Monte Carlo algorithms, which often exhibit slow mixing (i.e., poor convergence) properties. These spatial models are richly parameterized and lend themselves to the structured Markov chain Monte Carlo (SMCMC) algorithms. SMCMC provides a simple, general, and flexible framework for accelerating convergence in an MCMC sampler by providing a systematic way to block groups of similar parameters while taking full advantage of the posterior correlation structure induced by the model and data. Among the SMCMC strategies considered here are blocking using different size blocks (grouping by geographical region), reparameterization, updating jointly with and without model hyperparameters, "oversampling" some of the model parameters, and "pilot adaptation" versus continuous tuning techniques for the proposal density. We apply the techniques presented here to datasets on cancer mortality and late detection in the state of Minnesota. We find that, compared to univariate sampling procedures, our techniques will typically lead to more accurate posterior estimates, and they are sometimes also far more efficient in terms of the number of effective samples generated per second.

UR - http://www.scopus.com/inward/record.url?scp=0038344437&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0038344437&partnerID=8YFLogxK

U2 - 10.1198/1061860031833

DO - 10.1198/1061860031833

M3 - Review article

AN - SCOPUS:0038344437

VL - 12

SP - 249

EP - 264

JO - Journal of Computational and Graphical Statistics

JF - Journal of Computational and Graphical Statistics

SN - 1061-8600

IS - 2

ER -