Fence methods for mixed model selection

Jiming Jiang, Jonnagadda S Rao, Zhonghua Gu, Thuan Nguyen

Research output: Contribution to journalArticle

51 Citations (Scopus)

Abstract

Many model search strategies involve trading off model fit with model complexity in a penalized goodness of fit measure. Asymptotic properties for these types of procedures in settings like linear regression and ARMA time series have been studied, but these do not naturally extend to nonstandard situations such as mixed effects models, where simple definition of the sample size is not meaningful. This paper introduces a new class of strategies, known as fence methods, for mixed model selection, which includes linear and generalized linear mixed models. The idea involves a procedure to isolate a subgroup of what are known as correct models (of which the optimal model is a member). This is accomplished by constructing a statistical fence, or barrier, to carefully eliminate incorrect models. Once the fence is constructed, the optimal model is selected from among those within the fence according to a criterion which can be made flexible. In addition, we propose two variations of the fence. The first is a stepwise procedure to handle situations of many predictors; the second is an adaptive approach for choosing a tuning constant. We give sufficient conditions for consistency offence and its variations, a desirable property for a good model selection procedure. The methods are illustrated through simulation studies and real data analysis.

Original languageEnglish
Pages (from-to)1669-1692
Number of pages24
JournalAnnals of Statistics
Volume36
Issue number4
DOIs
StatePublished - Aug 1 2008
Externally publishedYes

Fingerprint

Mixed Model
Model Selection
Model
Mixed Effects Model
Generalized Linear Mixed Model
Model Complexity
Autoregressive Moving Average
Selection Procedures
Search Strategy
Goodness of fit
Linear regression
Asymptotic Properties
Mixed model
Model selection
Predictors
Tuning
Data analysis
Sample Size
Eliminate
Time series

Keywords

  • Adaptive fence
  • Consistency
  • F-B fence
  • Finite sample performance
  • GLMM
  • Linear mixed model
  • Model selection

ASJC Scopus subject areas

  • Mathematics(all)
  • Statistics and Probability

Cite this

Fence methods for mixed model selection. / Jiang, Jiming; Rao, Jonnagadda S; Gu, Zhonghua; Nguyen, Thuan.

In: Annals of Statistics, Vol. 36, No. 4, 01.08.2008, p. 1669-1692.

Research output: Contribution to journalArticle

Jiang, J, Rao, JS, Gu, Z & Nguyen, T 2008, 'Fence methods for mixed model selection', Annals of Statistics, vol. 36, no. 4, pp. 1669-1692. https://doi.org/10.1214/07-AOS517
Jiang, Jiming ; Rao, Jonnagadda S ; Gu, Zhonghua ; Nguyen, Thuan. / Fence methods for mixed model selection. In: Annals of Statistics. 2008 ; Vol. 36, No. 4. pp. 1669-1692.
@article{48582eda356e423388053478ea802357,
title = "Fence methods for mixed model selection",
abstract = "Many model search strategies involve trading off model fit with model complexity in a penalized goodness of fit measure. Asymptotic properties for these types of procedures in settings like linear regression and ARMA time series have been studied, but these do not naturally extend to nonstandard situations such as mixed effects models, where simple definition of the sample size is not meaningful. This paper introduces a new class of strategies, known as fence methods, for mixed model selection, which includes linear and generalized linear mixed models. The idea involves a procedure to isolate a subgroup of what are known as correct models (of which the optimal model is a member). This is accomplished by constructing a statistical fence, or barrier, to carefully eliminate incorrect models. Once the fence is constructed, the optimal model is selected from among those within the fence according to a criterion which can be made flexible. In addition, we propose two variations of the fence. The first is a stepwise procedure to handle situations of many predictors; the second is an adaptive approach for choosing a tuning constant. We give sufficient conditions for consistency offence and its variations, a desirable property for a good model selection procedure. The methods are illustrated through simulation studies and real data analysis.",
keywords = "Adaptive fence, Consistency, F-B fence, Finite sample performance, GLMM, Linear mixed model, Model selection",
author = "Jiming Jiang and Rao, {Jonnagadda S} and Zhonghua Gu and Thuan Nguyen",
year = "2008",
month = "8",
day = "1",
doi = "10.1214/07-AOS517",
language = "English",
volume = "36",
pages = "1669--1692",
journal = "Annals of Statistics",
issn = "0090-5364",
publisher = "Institute of Mathematical Statistics",
number = "4",

}

TY - JOUR

T1 - Fence methods for mixed model selection

AU - Jiang, Jiming

AU - Rao, Jonnagadda S

AU - Gu, Zhonghua

AU - Nguyen, Thuan

PY - 2008/8/1

Y1 - 2008/8/1

N2 - Many model search strategies involve trading off model fit with model complexity in a penalized goodness of fit measure. Asymptotic properties for these types of procedures in settings like linear regression and ARMA time series have been studied, but these do not naturally extend to nonstandard situations such as mixed effects models, where simple definition of the sample size is not meaningful. This paper introduces a new class of strategies, known as fence methods, for mixed model selection, which includes linear and generalized linear mixed models. The idea involves a procedure to isolate a subgroup of what are known as correct models (of which the optimal model is a member). This is accomplished by constructing a statistical fence, or barrier, to carefully eliminate incorrect models. Once the fence is constructed, the optimal model is selected from among those within the fence according to a criterion which can be made flexible. In addition, we propose two variations of the fence. The first is a stepwise procedure to handle situations of many predictors; the second is an adaptive approach for choosing a tuning constant. We give sufficient conditions for consistency offence and its variations, a desirable property for a good model selection procedure. The methods are illustrated through simulation studies and real data analysis.

AB - Many model search strategies involve trading off model fit with model complexity in a penalized goodness of fit measure. Asymptotic properties for these types of procedures in settings like linear regression and ARMA time series have been studied, but these do not naturally extend to nonstandard situations such as mixed effects models, where simple definition of the sample size is not meaningful. This paper introduces a new class of strategies, known as fence methods, for mixed model selection, which includes linear and generalized linear mixed models. The idea involves a procedure to isolate a subgroup of what are known as correct models (of which the optimal model is a member). This is accomplished by constructing a statistical fence, or barrier, to carefully eliminate incorrect models. Once the fence is constructed, the optimal model is selected from among those within the fence according to a criterion which can be made flexible. In addition, we propose two variations of the fence. The first is a stepwise procedure to handle situations of many predictors; the second is an adaptive approach for choosing a tuning constant. We give sufficient conditions for consistency offence and its variations, a desirable property for a good model selection procedure. The methods are illustrated through simulation studies and real data analysis.

KW - Adaptive fence

KW - Consistency

KW - F-B fence

KW - Finite sample performance

KW - GLMM

KW - Linear mixed model

KW - Model selection

UR - http://www.scopus.com/inward/record.url?scp=51049087315&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=51049087315&partnerID=8YFLogxK

U2 - 10.1214/07-AOS517

DO - 10.1214/07-AOS517

M3 - Article

VL - 36

SP - 1669

EP - 1692

JO - Annals of Statistics

JF - Annals of Statistics

SN - 0090-5364

IS - 4

ER -