Assessing the convergence of markov chain monte carlo methods: an example from evaluation of diagnostic tests in absence of a gold standard

N. Toft, G.T. Innocent, G. Gettinby, S.W.J. Reid

Research output: Contribution to journalArticle

68 Citations (Scopus)

Abstract

The accessibility of Markov Chain Monte Carlo (MCMC) methods for statistical inference have improved with the advent of general purpose software. This enables researchers with limited statistical skills to perform Bayesian analysis. Using MCMC sampling to do statistical inference requires convergence of the MCMC chain to its stationary distribution. There is no certain way to prove convergence; it is only possible to ascertain when convergence definitely has not been achieved. These methods are rather subjective and not implemented as automatic safeguards in general MCMC software. This paper considers a pragmatic approach towards assessing the convergence of MCMC methods illustrated by a Bayesian analysis of the Hui-Walter model for evaluating diagnostic tests in the absence of a gold standard. The Hui-Walter model has two optimal solutions, a property which causes problems with convergence when the solutions are sufficiently close in the parameter space. Using simulated data we demonstrate tools to assess the convergence and mixing of MCMC chains using examples with and without convergence. Suggestions to remedy the situation when the MCMC sampler fails to converge are given. The epidemiological implications of the two solutions of the Hui-Walter model are discussed.
LanguageEnglish
Pages244-256
Number of pages12
JournalPreventive Veterinary Medicine
Volume79
Issue number2-4
DOIs
Publication statusPublished - 16 May 2007

Fingerprint

Monte Carlo Method
Markov Chains
Monte Carlo method
Routine Diagnostic Tests
diagnostic techniques
gold
Bayes Theorem
Bayesian theory
Software
samplers
Markov chain
researchers
Research Personnel

Keywords

  • Bayesian analysis
  • MCMC
  • Markov chain Monte Carlo
  • diagnostic tests
  • latent class analysis

Cite this

@article{9ecfdd7a4a9d44baafc6c1448f85b086,
title = "Assessing the convergence of markov chain monte carlo methods: an example from evaluation of diagnostic tests in absence of a gold standard",
abstract = "The accessibility of Markov Chain Monte Carlo (MCMC) methods for statistical inference have improved with the advent of general purpose software. This enables researchers with limited statistical skills to perform Bayesian analysis. Using MCMC sampling to do statistical inference requires convergence of the MCMC chain to its stationary distribution. There is no certain way to prove convergence; it is only possible to ascertain when convergence definitely has not been achieved. These methods are rather subjective and not implemented as automatic safeguards in general MCMC software. This paper considers a pragmatic approach towards assessing the convergence of MCMC methods illustrated by a Bayesian analysis of the Hui-Walter model for evaluating diagnostic tests in the absence of a gold standard. The Hui-Walter model has two optimal solutions, a property which causes problems with convergence when the solutions are sufficiently close in the parameter space. Using simulated data we demonstrate tools to assess the convergence and mixing of MCMC chains using examples with and without convergence. Suggestions to remedy the situation when the MCMC sampler fails to converge are given. The epidemiological implications of the two solutions of the Hui-Walter model are discussed.",
keywords = "Bayesian analysis, MCMC, Markov chain Monte Carlo, diagnostic tests, latent class analysis",
author = "N. Toft and G.T. Innocent and G. Gettinby and S.W.J. Reid",
year = "2007",
month = "5",
day = "16",
doi = "10.1016/j.prevetmed.2007.01.003",
language = "English",
volume = "79",
pages = "244--256",
journal = "Preventive Veterinary Medicine",
issn = "0167-5877",
number = "2-4",

}

Assessing the convergence of markov chain monte carlo methods: an example from evaluation of diagnostic tests in absence of a gold standard. / Toft, N.; Innocent, G.T.; Gettinby, G.; Reid, S.W.J.

In: Preventive Veterinary Medicine, Vol. 79, No. 2-4, 16.05.2007, p. 244-256.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Assessing the convergence of markov chain monte carlo methods: an example from evaluation of diagnostic tests in absence of a gold standard

AU - Toft, N.

AU - Innocent, G.T.

AU - Gettinby, G.

AU - Reid, S.W.J.

PY - 2007/5/16

Y1 - 2007/5/16

N2 - The accessibility of Markov Chain Monte Carlo (MCMC) methods for statistical inference have improved with the advent of general purpose software. This enables researchers with limited statistical skills to perform Bayesian analysis. Using MCMC sampling to do statistical inference requires convergence of the MCMC chain to its stationary distribution. There is no certain way to prove convergence; it is only possible to ascertain when convergence definitely has not been achieved. These methods are rather subjective and not implemented as automatic safeguards in general MCMC software. This paper considers a pragmatic approach towards assessing the convergence of MCMC methods illustrated by a Bayesian analysis of the Hui-Walter model for evaluating diagnostic tests in the absence of a gold standard. The Hui-Walter model has two optimal solutions, a property which causes problems with convergence when the solutions are sufficiently close in the parameter space. Using simulated data we demonstrate tools to assess the convergence and mixing of MCMC chains using examples with and without convergence. Suggestions to remedy the situation when the MCMC sampler fails to converge are given. The epidemiological implications of the two solutions of the Hui-Walter model are discussed.

AB - The accessibility of Markov Chain Monte Carlo (MCMC) methods for statistical inference have improved with the advent of general purpose software. This enables researchers with limited statistical skills to perform Bayesian analysis. Using MCMC sampling to do statistical inference requires convergence of the MCMC chain to its stationary distribution. There is no certain way to prove convergence; it is only possible to ascertain when convergence definitely has not been achieved. These methods are rather subjective and not implemented as automatic safeguards in general MCMC software. This paper considers a pragmatic approach towards assessing the convergence of MCMC methods illustrated by a Bayesian analysis of the Hui-Walter model for evaluating diagnostic tests in the absence of a gold standard. The Hui-Walter model has two optimal solutions, a property which causes problems with convergence when the solutions are sufficiently close in the parameter space. Using simulated data we demonstrate tools to assess the convergence and mixing of MCMC chains using examples with and without convergence. Suggestions to remedy the situation when the MCMC sampler fails to converge are given. The epidemiological implications of the two solutions of the Hui-Walter model are discussed.

KW - Bayesian analysis

KW - MCMC

KW - Markov chain Monte Carlo

KW - diagnostic tests

KW - latent class analysis

UR - http://dx.doi.org/10.1016/j.prevetmed.2007.01.003

U2 - 10.1016/j.prevetmed.2007.01.003

DO - 10.1016/j.prevetmed.2007.01.003

M3 - Article

VL - 79

SP - 244

EP - 256

JO - Preventive Veterinary Medicine

T2 - Preventive Veterinary Medicine

JF - Preventive Veterinary Medicine

SN - 0167-5877

IS - 2-4

ER -