When a linear model is chosen by searching for the best subset among a set of candidate predictors, a fixed penalty such as that imposed by the Akaike information criterion may penalize model complexity inadequately, leading to biased model selection. We study resampling-based information criteria that aim to overcome this problem through improved estimation of the effective model dimension. The first proposed approach builds upon previous work on bootstrap-based model selection. We then propose a more novel approach based on cross-validation. Simulations and analyses of a functional neuroimaging data set illustrate the strong performance of our resampling-based methods, which are implemented in a new R package.
|Number of pages
|Annals of the Institute of Statistical Mathematics
|Published - Dec 2012
Bibliographical noteFunding Information:
Acknowledgments The first author’s research is supported in part by National Science Foundation grant DMS-0907017. The authors thank Mike Milham, Eva Petkova, Thad Tarpey, Lee Dicker and Tao Zhang, for illuminating discussions; Zarrar Shehzad, for assistance with the functional connectivity data; and the Associate Editor and referee, whose incisive comments led to major improvements in the paper.
- Adaptive model selection
- Covariance inflation criterion
- Extended information criterion
- Functional connectivity
ASJC Scopus subject areas
- Statistics and Probability