Robust regression for highly corrupted response by shifting outliers

Yoonsuh Jung, Seung Pil Lee, Jianhua Hu

Research output: Contribution to journalArticle

4 Citations (Scopus)

Abstract

Outlying observations are often disregarded at the sacrifice of degrees of freedom or downsized via robust loss functions (e.g., Huber's loss) to reduce the undesirable impact on data analysis. In this article, we treat the outlying status of each observation as a parameter and propose a penalization method to automatically adjust the outliers. The proposed method shifts the outliers towards the fitted values, while preserve the non-outlying observations. We also develop a generally applicable algorithm in the iterative fashion to estimate model parameters and demonstrate the connection with the maximum likelihood based estimation procedure in the case of least squares estimation. We establish asymptotic property of the resulting parameter estimators under the condition that the proportion of outliers does not vanish as sample size increases. We apply the proposed outlier adjustment method to ordinary least squares and lasso-type penalization procedure and demonstrate its empirical value via numeric studies. Furthermore, we study applicability of the proposed method to two robust estimators, Huber's robust estimator and Huberized lasso, and demonstrate its noticeable improvement of model fit in the presence of extremely large outliers.

Original languageEnglish
Pages (from-to)1-23
Number of pages23
JournalStatistical Modelling
Volume16
Issue number1
DOIs
Publication statusPublished - 2016 Feb 1
Externally publishedYes

Fingerprint

Robust Regression
Outlier
Lasso
Robust Estimators
Penalization Method
Demonstrate
Ordinary Least Squares
Least Squares Estimation
Penalization
Loss Function
Numerics
Asymptotic Properties
Maximum Likelihood
Vanish
Data analysis
Adjustment
Sample Size
Proportion
Degree of freedom
Robust regression

Keywords

  • case-specific parameter
  • extreme outliers
  • Huber's estimator
  • robust lasso
  • robust linear model

ASJC Scopus subject areas

  • Statistics and Probability
  • Statistics, Probability and Uncertainty

Cite this

Robust regression for highly corrupted response by shifting outliers. / Jung, Yoonsuh; Lee, Seung Pil; Hu, Jianhua.

In: Statistical Modelling, Vol. 16, No. 1, 01.02.2016, p. 1-23.

Research output: Contribution to journalArticle

Jung, Yoonsuh ; Lee, Seung Pil ; Hu, Jianhua. / Robust regression for highly corrupted response by shifting outliers. In: Statistical Modelling. 2016 ; Vol. 16, No. 1. pp. 1-23.
@article{dc82d143883b4714a49c39db2544699a,
title = "Robust regression for highly corrupted response by shifting outliers",
abstract = "Outlying observations are often disregarded at the sacrifice of degrees of freedom or downsized via robust loss functions (e.g., Huber's loss) to reduce the undesirable impact on data analysis. In this article, we treat the outlying status of each observation as a parameter and propose a penalization method to automatically adjust the outliers. The proposed method shifts the outliers towards the fitted values, while preserve the non-outlying observations. We also develop a generally applicable algorithm in the iterative fashion to estimate model parameters and demonstrate the connection with the maximum likelihood based estimation procedure in the case of least squares estimation. We establish asymptotic property of the resulting parameter estimators under the condition that the proportion of outliers does not vanish as sample size increases. We apply the proposed outlier adjustment method to ordinary least squares and lasso-type penalization procedure and demonstrate its empirical value via numeric studies. Furthermore, we study applicability of the proposed method to two robust estimators, Huber's robust estimator and Huberized lasso, and demonstrate its noticeable improvement of model fit in the presence of extremely large outliers.",
keywords = "case-specific parameter, extreme outliers, Huber's estimator, robust lasso, robust linear model",
author = "Yoonsuh Jung and Lee, {Seung Pil} and Jianhua Hu",
year = "2016",
month = "2",
day = "1",
doi = "10.1177/1471082X15624040",
language = "English",
volume = "16",
pages = "1--23",
journal = "Statistical Modelling",
issn = "1471-082X",
publisher = "SAGE Publications Ltd",
number = "1",

}

TY - JOUR

T1 - Robust regression for highly corrupted response by shifting outliers

AU - Jung, Yoonsuh

AU - Lee, Seung Pil

AU - Hu, Jianhua

PY - 2016/2/1

Y1 - 2016/2/1

N2 - Outlying observations are often disregarded at the sacrifice of degrees of freedom or downsized via robust loss functions (e.g., Huber's loss) to reduce the undesirable impact on data analysis. In this article, we treat the outlying status of each observation as a parameter and propose a penalization method to automatically adjust the outliers. The proposed method shifts the outliers towards the fitted values, while preserve the non-outlying observations. We also develop a generally applicable algorithm in the iterative fashion to estimate model parameters and demonstrate the connection with the maximum likelihood based estimation procedure in the case of least squares estimation. We establish asymptotic property of the resulting parameter estimators under the condition that the proportion of outliers does not vanish as sample size increases. We apply the proposed outlier adjustment method to ordinary least squares and lasso-type penalization procedure and demonstrate its empirical value via numeric studies. Furthermore, we study applicability of the proposed method to two robust estimators, Huber's robust estimator and Huberized lasso, and demonstrate its noticeable improvement of model fit in the presence of extremely large outliers.

AB - Outlying observations are often disregarded at the sacrifice of degrees of freedom or downsized via robust loss functions (e.g., Huber's loss) to reduce the undesirable impact on data analysis. In this article, we treat the outlying status of each observation as a parameter and propose a penalization method to automatically adjust the outliers. The proposed method shifts the outliers towards the fitted values, while preserve the non-outlying observations. We also develop a generally applicable algorithm in the iterative fashion to estimate model parameters and demonstrate the connection with the maximum likelihood based estimation procedure in the case of least squares estimation. We establish asymptotic property of the resulting parameter estimators under the condition that the proportion of outliers does not vanish as sample size increases. We apply the proposed outlier adjustment method to ordinary least squares and lasso-type penalization procedure and demonstrate its empirical value via numeric studies. Furthermore, we study applicability of the proposed method to two robust estimators, Huber's robust estimator and Huberized lasso, and demonstrate its noticeable improvement of model fit in the presence of extremely large outliers.

KW - case-specific parameter

KW - extreme outliers

KW - Huber's estimator

KW - robust lasso

KW - robust linear model

UR - http://www.scopus.com/inward/record.url?scp=84957879039&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84957879039&partnerID=8YFLogxK

U2 - 10.1177/1471082X15624040

DO - 10.1177/1471082X15624040

M3 - Article

VL - 16

SP - 1

EP - 23

JO - Statistical Modelling

JF - Statistical Modelling

SN - 1471-082X

IS - 1

ER -