Extended Bayesian information criteria for model selection with large model spaces

摘要: SUMMARY The ordinary Bayesian information criterion is too liberal for model selection when the space large. In this paper, we re-examine paradigm and propose an extended family of criteria, which take into account both number unknown parameters complexity space. Their consistency established, in particular allowing covariates to increase infinity with sample size. performance various situations evaluated by simulation studies. It demonstrated that criteria incur a small loss positive rate but tightly control false discovery rate, desirable property many applications. are extremely useful variable problems moderate size huge covariates, especially genome-wide association studies, now active area genetics research.