Publication:
Locally and globally robust estimators in regression

Thumbnail Image
Identifiers
URI: https://hdl.handle.net/10016/6348
Files
ws996021.PDF (1.22 MB)
Publication date
1999-07
Defense date
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Google Scholar
Export
Research Projects
Organizational Units
Journal Issue
Abstract
A new class of estimates for the linear model is introduced. These estimators, that we eaU C-estimators, are defined as a linear convex combination of the Rousseeuw's least median squares (LMS-) estimator and any other estimate, T2• We prove that C-estimators retain the high breakdown point of the LMS-estimator, but inherit the asymptotic properties and the behaviour in terms of local robutness of T2• In particular, a Cestimators will have --In-asymptotics and bounded contamination sensitivity if T2 does. In addition, efficiency, local robustness properties and the maximum bias curve of C-estimators are investigated for different choices ofT2•
Description
Keywords
Linear regression, robust estimates, maximum bias function, high breakdown point, contamination sensivity, high efficiency
Bibliographic citation
Collections
DES - Working Papers. Statistics and Econometrics. WS