Presenter: Run Wang, PhD candidate in Statistics

Abstract: In ultra-high dimensional linear regressions, variable screening is usually conducted before performing a refined variable selection method to save on computation. In this presentation, two novel variable screening methods in Bayesian and frequentist paradigms will be introduced. The Bayesian iterative screening method can incorporate prior information on effect sizes and the number of true variables while not requiring more computational cost than many frequentist screening methods. The ridge partial correlation screening removes the effects of conditioned variables in screening. Both methods are computationally competitive to existing screening methods and under mild conditions, both are screening consistent, that is, they include the important variables with overwhelming probabilities. Simulation studies and real data examples show the fine screening performance of the proposed methods.