An Alternative Estimation of Sliced Inverse Regression

• Activity Number: 35
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Qin Wang*+ and Xiangrong Yin
• Companies: Virginia Commonwealth University and The University of Georgia
• Keywords: Sufficient dimension reduction ; Sliced inverse regression ; Orthogonal series estimation

Abstract

In this talk, we propose the use of orthogonal series to estimate the inverse mean space. Comparing with the original slicing scheme, it signifcantly improves the estimation accuracy without losing computation efficiency, especially for the heteroscedastic models. Comparing with the local smoothing approach, it is more computationally efficient. The new approach also has the advantage of robustness in selecting the tuning parameter. Permutation test is used to determine the structural dimension. Moreover, a variable selection procedure is incorporated into this new approach, which is particularly useful when the model is sparse. The efficacy of the proposed method is demonstrated through simulations and a real data analysis.