On permutation tests for predictor contribution in sufficient dimension reduction

Abstract

To test predictor contribution in a model-free fashion, marginal coordinate tests based on sliced inverse regression (SIR) and sliced average variance estimation (SAVE) have been studied in Cook (2004), and Shao et al. (2007) respectively. Estimating the null distributions of the test statistics is a critical step for such tests. We propose a novel permutation test approach to facilitate the marginal coordinate tests, which applies to existing tests based on SIR and SAVE, and can be readily extended to a new marginal coordinate test based on directional regression (Li and Wang, 2007).