Partially Collapsed Gibbs Sampling for Linear Mixed-effects Models

Taeyoung Park, Seunghyun Min

Research output: Contribution to journalArticle

12 Citations (Scopus)

Abstract

This article presents a novel Bayesian analysis for linear mixed-effects models. The analysis is based on the method of partial collapsing that allows some components to be partially collapsed out of a model. The resulting partially collapsed Gibbs (PCG) sampler constructed to fit linear mixed-effects models is expected to exhibit much better convergence properties than the corresponding Gibbs sampler. In order to construct the PCG sampler without complicating component updates, we consider the reparameterization of model components by expressing a between-group variance in terms of a within-group variance in a linear mixed-effects model. The proposed method of partial collapsing with reparameterization is applied to the Mertons jump diffusion model as well as general linear mixed-effects models with proper prior distributions and illustrated using simulated data and longitudinal data on sleep deprivation.

Original languageEnglish
Pages (from-to)165-180
Number of pages16
JournalCommunications in Statistics: Simulation and Computation
Volume45
Issue number1
DOIs
Publication statusPublished - 2016 Jan 1

Fingerprint

Linear Mixed Effects Model
Gibbs Sampling
Gibbs Sampler
Sampling
Reparameterization
Collapsing
Jump-diffusion Model
Partial
Sleep
Longitudinal Data
Bayesian Analysis
Component Model
Prior distribution
Convergence Properties
Update

All Science Journal Classification (ASJC) codes

  • Statistics and Probability
  • Modelling and Simulation

Cite this

@article{9d8a98ab273940518a47a7a288661f14,
title = "Partially Collapsed Gibbs Sampling for Linear Mixed-effects Models",
abstract = "This article presents a novel Bayesian analysis for linear mixed-effects models. The analysis is based on the method of partial collapsing that allows some components to be partially collapsed out of a model. The resulting partially collapsed Gibbs (PCG) sampler constructed to fit linear mixed-effects models is expected to exhibit much better convergence properties than the corresponding Gibbs sampler. In order to construct the PCG sampler without complicating component updates, we consider the reparameterization of model components by expressing a between-group variance in terms of a within-group variance in a linear mixed-effects model. The proposed method of partial collapsing with reparameterization is applied to the Mertons jump diffusion model as well as general linear mixed-effects models with proper prior distributions and illustrated using simulated data and longitudinal data on sleep deprivation.",
author = "Taeyoung Park and Seunghyun Min",
year = "2016",
month = "1",
day = "1",
doi = "10.1080/03610918.2013.857687",
language = "English",
volume = "45",
pages = "165--180",
journal = "Communications in Statistics Part B: Simulation and Computation",
issn = "0361-0918",
publisher = "Taylor and Francis Ltd.",
number = "1",

}

Partially Collapsed Gibbs Sampling for Linear Mixed-effects Models. / Park, Taeyoung; Min, Seunghyun.

In: Communications in Statistics: Simulation and Computation, Vol. 45, No. 1, 01.01.2016, p. 165-180.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Partially Collapsed Gibbs Sampling for Linear Mixed-effects Models

AU - Park, Taeyoung

AU - Min, Seunghyun

PY - 2016/1/1

Y1 - 2016/1/1

N2 - This article presents a novel Bayesian analysis for linear mixed-effects models. The analysis is based on the method of partial collapsing that allows some components to be partially collapsed out of a model. The resulting partially collapsed Gibbs (PCG) sampler constructed to fit linear mixed-effects models is expected to exhibit much better convergence properties than the corresponding Gibbs sampler. In order to construct the PCG sampler without complicating component updates, we consider the reparameterization of model components by expressing a between-group variance in terms of a within-group variance in a linear mixed-effects model. The proposed method of partial collapsing with reparameterization is applied to the Mertons jump diffusion model as well as general linear mixed-effects models with proper prior distributions and illustrated using simulated data and longitudinal data on sleep deprivation.

AB - This article presents a novel Bayesian analysis for linear mixed-effects models. The analysis is based on the method of partial collapsing that allows some components to be partially collapsed out of a model. The resulting partially collapsed Gibbs (PCG) sampler constructed to fit linear mixed-effects models is expected to exhibit much better convergence properties than the corresponding Gibbs sampler. In order to construct the PCG sampler without complicating component updates, we consider the reparameterization of model components by expressing a between-group variance in terms of a within-group variance in a linear mixed-effects model. The proposed method of partial collapsing with reparameterization is applied to the Mertons jump diffusion model as well as general linear mixed-effects models with proper prior distributions and illustrated using simulated data and longitudinal data on sleep deprivation.

UR - http://www.scopus.com/inward/record.url?scp=84947737019&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84947737019&partnerID=8YFLogxK

U2 - 10.1080/03610918.2013.857687

DO - 10.1080/03610918.2013.857687

M3 - Article

VL - 45

SP - 165

EP - 180

JO - Communications in Statistics Part B: Simulation and Computation

JF - Communications in Statistics Part B: Simulation and Computation

SN - 0361-0918

IS - 1

ER -