Title: Sufficient variable selection using independence measures for continuous responses
Authors: Baoying Yang - Southwest Jiaotong University (China) [presenting]
Abstract: Two sufficient variable selection procedures are proposed: one stage and two stage approaches using independence measures for continuous responses. Although any independence measure can be used, we use distance correlation and Hilbert Schmidt Independence Criterion correlation to illustrate the two procedures. By comparing with some existing marginal screening methods, we show the advantages of the proposed procedures through the simulation studies and a real data analysis. Our procedures are model-free and robust against model mis-specification, especially useful, when the active predictors are marginally independent of the response.