Sufficient dimension reduction (SDR) turns out to be a useful dimension reduction tool in high-dimensional regression analysis. Weisberg (2002) developed the dr-package to implement the four most popular SDR methods. However, the package does not provide any clear guidelines as to which method should be used given a data. Since the four methods may provide dramatically different dimension reduction results, the selection in the dr-package is problematic for statistical practitioners. In this paper, a basis-adaptive selection algorithm is developed in order to relieve this issue. The basic idea is to select an SDR method that provides the highest correlation between the basis estimates obtained by the four classical SDR methods. A real data example and numerical studies confirm the practical usefulness of the developed algorithm.
Bibliographical noteFunding Information:
For the corresponding author Jae Keun Yoo, this work was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Korean Ministry of Education (NRF-2017R1A2B1004909)
© 2018 The R Journal.