Two level stochastic search variable selection in GLMs with missing predictors
Two level stochastic search variable selection in GLMs with missing predictors
Stochastic search variable selection (SSVS) algorithms provide an appealing and widely used approach for searching for good subsets of predictors, while simultaneously estimating posterior model probabilities and model-averaged predictive distributions. This article proposes a two-level generalization of SSVS to account for missing predictors, while accommodating uncertainty in the relationships between these predictors. Bayesian approaches for allowing predictors that are missing at random require a model on the joint distribution of the predictors. We show that predictive performance can be improved by allowing uncertainty in the specification of this model. The methods are illustrated through simulation studies and analysis of an epidemiologic data set.
missing at random, model averaging, multiple imputation, stochastic search, subset selection, variable selection
Southampton Statistical Sciences Research Institute, University of Southampton
Mitra, Robin
2b944cd7-5be8-4dd1-ab44-f8ada9a33405
Dunson, David D.
de41205b-5af1-49a9-b8de-ee36a8b02081
17 February 2009
Mitra, Robin
2b944cd7-5be8-4dd1-ab44-f8ada9a33405
Dunson, David D.
de41205b-5af1-49a9-b8de-ee36a8b02081
Mitra, Robin and Dunson, David D.
(2009)
Two level stochastic search variable selection in GLMs with missing predictors
(S3RI Methodology Working Papers, M09/04)
Southampton, UK.
Southampton Statistical Sciences Research Institute, University of Southampton
28pp.
Record type:
Monograph
(Working Paper)
Abstract
Stochastic search variable selection (SSVS) algorithms provide an appealing and widely used approach for searching for good subsets of predictors, while simultaneously estimating posterior model probabilities and model-averaged predictive distributions. This article proposes a two-level generalization of SSVS to account for missing predictors, while accommodating uncertainty in the relationships between these predictors. Bayesian approaches for allowing predictors that are missing at random require a model on the joint distribution of the predictors. We show that predictive performance can be improved by allowing uncertainty in the specification of this model. The methods are illustrated through simulation studies and analysis of an epidemiologic data set.
Text
65460-01.pdf
- Author's Original
More information
Published date: 17 February 2009
Keywords:
missing at random, model averaging, multiple imputation, stochastic search, subset selection, variable selection
Identifiers
Local EPrints ID: 65460
URI: http://eprints.soton.ac.uk/id/eprint/65460
PURE UUID: b911a37a-e89b-4fd0-9f6c-2cd3bae4cd98
Catalogue record
Date deposited: 17 Feb 2009
Last modified: 20 Feb 2024 03:17
Export record
Contributors
Author:
David D. Dunson
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics