Sliced inverse regression(SIR) is statistical tool for dimension reduction. It identifies appropriate factors through simple statistical tests for determining the number of significant factors. We propose a refined implementation of SIR method by allowing slice overlapping. It allows to take more advantage of the available data while the number of parameters keeps the same . The refined SIR method can estimate the e.d.r. space (the subspace with intrinsic lower dimensionality) and determine the number of effective components more accurately. We verified the effectiveness of the method by simulations and applications.

海报