Local-mass preserving prior distributions for nonparametric bayesian models

Juhee Lee, Steven N. MacEachern, Yiling Lu, Gordon Mills

Research output: Contribution to journalArticle

4 Citations (Scopus)

Abstract

We address the problem of prior specification for models involving the two-parameter Poisson-Dirichlet process. These models are sometimes partially subjectively specified and are always partially (or fully) specified by a rule. We develop prior distributions based on local mass preservation. The robustness of posterior inference to an arbitrary choice of overdispersion under the proposed and current priors is investigated. Two examples are provided to demonstrate the properties of the proposed priors. We focus on the three major types of inference: clustering of the parameters of interest, estimation and prediction. The new priors are found to provide more stable inference about clustering than traditional priors while showing few drawbacks. Furthermore, it is shown that more stable clustering results in more stable inference for estimation and prediction. We recommend the local-mass preserving priors as a replacement for the traditional priors.

Original languageEnglish (US)
Pages (from-to)307-330
Number of pages24
JournalBayesian Analysis
Volume9
Issue number2
DOIs
StatePublished - Jan 1 2014
Externally publishedYes

Fingerprint

Nonparametric Model
Bayesian Model
Prior distribution
Clustering
Specifications
Overdispersion
Dirichlet Process
Prediction
Poisson process
Preservation
Replacement
Two Parameters
Specification
Robustness
Arbitrary
Model
Demonstrate

Keywords

  • Clustering
  • Dirichlet process
  • Local mass
  • Nonparametric bayes
  • Prior misspecification
  • Two-parameter poisson-dirichlet process

ASJC Scopus subject areas

  • Statistics and Probability
  • Applied Mathematics

Cite this

Local-mass preserving prior distributions for nonparametric bayesian models. / Lee, Juhee; MacEachern, Steven N.; Lu, Yiling; Mills, Gordon.

In: Bayesian Analysis, Vol. 9, No. 2, 01.01.2014, p. 307-330.

Research output: Contribution to journalArticle

Lee, Juhee ; MacEachern, Steven N. ; Lu, Yiling ; Mills, Gordon. / Local-mass preserving prior distributions for nonparametric bayesian models. In: Bayesian Analysis. 2014 ; Vol. 9, No. 2. pp. 307-330.
@article{4c4490b07bab495c80f68432176e9677,
title = "Local-mass preserving prior distributions for nonparametric bayesian models",
abstract = "We address the problem of prior specification for models involving the two-parameter Poisson-Dirichlet process. These models are sometimes partially subjectively specified and are always partially (or fully) specified by a rule. We develop prior distributions based on local mass preservation. The robustness of posterior inference to an arbitrary choice of overdispersion under the proposed and current priors is investigated. Two examples are provided to demonstrate the properties of the proposed priors. We focus on the three major types of inference: clustering of the parameters of interest, estimation and prediction. The new priors are found to provide more stable inference about clustering than traditional priors while showing few drawbacks. Furthermore, it is shown that more stable clustering results in more stable inference for estimation and prediction. We recommend the local-mass preserving priors as a replacement for the traditional priors.",
keywords = "Clustering, Dirichlet process, Local mass, Nonparametric bayes, Prior misspecification, Two-parameter poisson-dirichlet process",
author = "Juhee Lee and MacEachern, {Steven N.} and Yiling Lu and Gordon Mills",
year = "2014",
month = "1",
day = "1",
doi = "10.1214/13-BA857",
language = "English (US)",
volume = "9",
pages = "307--330",
journal = "Bayesian Analysis",
issn = "1936-0975",
publisher = "Carnegie Mellon University",
number = "2",

}

TY - JOUR

T1 - Local-mass preserving prior distributions for nonparametric bayesian models

AU - Lee, Juhee

AU - MacEachern, Steven N.

AU - Lu, Yiling

AU - Mills, Gordon

PY - 2014/1/1

Y1 - 2014/1/1

N2 - We address the problem of prior specification for models involving the two-parameter Poisson-Dirichlet process. These models are sometimes partially subjectively specified and are always partially (or fully) specified by a rule. We develop prior distributions based on local mass preservation. The robustness of posterior inference to an arbitrary choice of overdispersion under the proposed and current priors is investigated. Two examples are provided to demonstrate the properties of the proposed priors. We focus on the three major types of inference: clustering of the parameters of interest, estimation and prediction. The new priors are found to provide more stable inference about clustering than traditional priors while showing few drawbacks. Furthermore, it is shown that more stable clustering results in more stable inference for estimation and prediction. We recommend the local-mass preserving priors as a replacement for the traditional priors.

AB - We address the problem of prior specification for models involving the two-parameter Poisson-Dirichlet process. These models are sometimes partially subjectively specified and are always partially (or fully) specified by a rule. We develop prior distributions based on local mass preservation. The robustness of posterior inference to an arbitrary choice of overdispersion under the proposed and current priors is investigated. Two examples are provided to demonstrate the properties of the proposed priors. We focus on the three major types of inference: clustering of the parameters of interest, estimation and prediction. The new priors are found to provide more stable inference about clustering than traditional priors while showing few drawbacks. Furthermore, it is shown that more stable clustering results in more stable inference for estimation and prediction. We recommend the local-mass preserving priors as a replacement for the traditional priors.

KW - Clustering

KW - Dirichlet process

KW - Local mass

KW - Nonparametric bayes

KW - Prior misspecification

KW - Two-parameter poisson-dirichlet process

UR - http://www.scopus.com/inward/record.url?scp=84901480495&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84901480495&partnerID=8YFLogxK

U2 - 10.1214/13-BA857

DO - 10.1214/13-BA857

M3 - Article

AN - SCOPUS:84901480495

VL - 9

SP - 307

EP - 330

JO - Bayesian Analysis

JF - Bayesian Analysis

SN - 1936-0975

IS - 2

ER -