Two Robust Alternatives to Least-Squares Regression
作者:
RichardW. Hill,
PaulW. Holland,
期刊:
Journal of the American Statistical Association
(Taylor Available online 1977)
卷期:
Volume 72,
issue 360
页码: 828-833
ISSN:0162-1459
年代: 1977
DOI:10.1080/01621459.1977.10479965
出版商: Taylor & Francis Group
关键词: Robust regression;Mestimators;One-step sine estimator;L1 estimator;Monte Carlo
数据来源: Taylor
摘要:
We give Monte Carlo results on the performance of two robust alternatives to least-squares regression estimation—least-absolute residuals and the one-step sine estimator. We show how to scale the residuals for the sine estimator to achieve nearly constant efficiency for the normal distribution across various choices of the design matrix. We compare the two estimators to least squares for nine scale-contaminated normal distributions.
点击下载:
PDF (603KB)
返 回