A Monte Carlo comparison of several high breakdown and efficient estimators

✍ Scribed by Jiazhong You

• Publisher: Elsevier Science
• Year: 1999
• Tongue: English
• Weight: 137 KB
• Volume: 30
• Category: Article
• ISSN: 0167-9473
• DOI: 10.1016/s0167-9473(98)00087-5

✦ Synopsis

High breakdown point, bounded in uence and high e ciency at the Gaussian model are desired properties of robust regression estimators. Several estimators have been proposed to achieve at least some of these properties and their asymptotic behavior has been derived in the literature. In this article, the comparison of the ÿnite sample performance of these estimators is carried out by a Monte Carlo study for a number of outlier-generating models. Some striking ÿnite sample behavior is found in most of the high breakdown point estimators. Only S1S-estimator provides consistent performance in both zero-slope and non-zero-slope cases and is the best for the latter case.