摘要：We have developed an enhanced spike and slab model for vari-able selection in linear regression models via restricted nal prediction error(FPE) criteria; classic examples of which are AIC and BIC. Based on ourproposed Bayesian hierarchical model, a Gibbs sampler is developed to sam-ple models. The special structure of the prior enforces a unique mappingbetween sampling a model and calculating constrained ordinary least squaresestimates for that model, which helps to formulate the restricted FPE crite-ria. Empirical comparisons are done to the lasso, adaptive lasso and relaxedlasso; followed by a real life data example.
关键词：FPE analysis; model exploration; rescaled spike and slab model;variable selection.
