Bayesian variable selection with shrinking and diffusing priors (Q118687)

The authors are using the linear regression setup with high dimensional covariates where the number of covariates can be large relative to the sample size. Only a small number of covariates, called active covariates, have nonzero coefficients. The aim of this paper is to develop a new Bayesian methodology for selecting the active covariates that is asymptotically consistent and computationally convenient. A common notion of consistency for Bayesian variable selection is defined in terms of pairwise Bayes factors. Another notion of consistency for both frequentist and Bayesian methods is that the selected model equals the true model with probability converging to one. It is shown that the strong selection consistency of the proposed method in the sense that the posterior probability of the true model converges to one even when the number of covariates grows nearly exponentially with the sample size. We find important contributions to variable selection in this article. The conditions on the prior parameters and motivation for these conditions are given. The last part of the paper provides a discussion on the conditions assumed for proving the previous results. Some computational aspects of the proposed method, as well as simulation studies regarding the comparing of the proposed method with some existing methods are presented. At the end of this important paper we find also 35 references.