Performance of variable selection methods in regression using variations of the Bayesian information criterion. (English) Zbl 1159.62301

Burr, Tom; Fry, Herb; McVey, Brian; Sander, Eric; Cavanaugh, Joseph; Neath, Andrew

Commun. Stat., Simulation Comput. 37, No. 3, 507-520 (2008).

Summary: The Bayesian information criterion (BIC) is widely used for variable selection. We focus on the regression setting for which variations of the BIC have been proposed. A version that includes the Fisher information matrix of the predictor variables performed best in one published study. We extend the evaluation, introduce a performance measure involving how closely posterior probabilities are approximated, and conclude that the version that includes the Fisher information often favors regression models having more predictors, depending on the scale and correlation structure of the predictor matrix. In the image analysis application that we describe, we therefore prefer the standard BIC approximation because of its relative simplicity and competitive performance at approximating the true posterior probabilities.

MSC:

62F15	Bayesian inference
62J05	Linear regression; mixed models
62H35	Image analysis in multivariate analysis

Keywords:

approximate Bayesian posterior probabilities; image analysis; variable selection

References:

[1]	Burr , T. , Fry , H. , McVey , B. , Sander , E. ( 2002 ). Chemical identification using Bayesian model selection . Proceedings of the American Statistical Association. Section on Physical and Engineering Sciences [CD-ROM] , Ann Arbor , Michigan : American Statistical Association , Los Alamos National Laboratory Unclassified Report, LA-UR 02-7281 .
[2]	DOI: 10.3390/s6121721 · doi:10.3390/s6121721
[3]	DOI: 10.1198/004017005000000012 · doi:10.1198/004017005000000012
[4]	DOI: 10.2307/2965554 · Zbl 1050.62520 · doi:10.2307/2965554
[5]	DOI: 10.1117/12.490164 · doi:10.1117/12.490164
[6]	DOI: 10.1093/biomet/87.4.731 · Zbl 1029.62008 · doi:10.1093/biomet/87.4.731
[7]	DOI: 10.2307/2290777 · doi:10.2307/2290777
[8]	DOI: 10.1198/016214503000000828 · Zbl 1047.62003 · doi:10.1198/016214503000000828
[9]	DOI: 10.2307/2291091 · Zbl 0846.62028 · doi:10.2307/2291091
[10]	DOI: 10.1201/9781420035933 · doi:10.1201/9781420035933
[11]	Montgomery D., Introduction to Linear Regression Analysis., 2. ed. (1992) · Zbl 0850.62529
[12]	DOI: 10.1080/03610929708831934 · Zbl 1030.62532 · doi:10.1080/03610929708831934
[13]	DOI: 10.2307/2291462 · Zbl 0888.62026 · doi:10.2307/2291462
[14]	Splus 6 for Linux, Insightful Corp. , Seattle , WA , 2002 .
[15]	DOI: 10.1214/ss/1009212519 · Zbl 1059.62525 · doi:10.1214/ss/1009212519
[16]	DOI: 10.1081/SAC-200033363 · Zbl 1101.62355 · doi:10.1081/SAC-200033363