|
Activity Number:
|
240
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 8, 2006 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
IMS
|
| Abstract - #307088 |
|
Title:
|
Sliced Inverse Regression under Data Contamination
|
|
Author(s):
|
Ulrike Genschel*+
|
|
Companies:
|
Iowa State University
|
|
Address:
|
312 Snedecor Hall, Ames, IA, 50011,
|
|
Keywords:
|
dimension reduction ; outliers ; SIR ; subspace estimation ; subspace metric
|
|
Abstract:
|
SIR (Li, 1991) is a dimension reduction procedure for estimating an appropriate regressor subspace. Because fitting a regression curve relies on the correct identification of such a subspace, the sensitivity of SIR to outlying observations becomes crucial to understand and we examine the influence of data contamination on the subspace estimate from SIR. Not only is it possible to overestimate the dimension of the subspace under contamination as stated in the literature, underestimation is possible as well to the extent that none of the true dimension reduction directions of the subspace are recoverable. We also demonstrate that data contamination scenarios producing erroneous subspace estimates in SIR depend on the of the covariance structure of the regressor variables as well as on the knowledge of the dimension of the true subspace. Our findings are supported by a simulation study.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2006 program |
