A working likelihood approach for robust regression.
Robust approach is often desirable in presence of outliers for more efficient parameter estimation. However, the choice of the regularization parameter value impacts the efficiency of the parameter estimators. To maximize the estimation efficiency, we construct a likelihood function for simultaneous...
| Published in: | Statistical Methods in Medical Research Vol. 29; no. 12; pp. 3641 - 3653 |
|---|---|
| Main Authors: | , , |
| Format: | research Journal Article |
| Published: |
Sage Publications Inc.
Dec2020
|
| Online Access: | View this record in EBSCOhost |
| fields | @attributes: recordID: 1 pdfLink: plink: https://search.ebscohost.com/login.aspx?direct=true&db=ccm&AN=146317549&site=ehost-live header: @attributes: shortDbName: ccm uiTerm: 146317549 longDbName: CINAHL Complete uiTag: AN controlInfo: bkinfo: dissinfo: jinfo: jid: 09622802 31F jtl: Statistical Methods in Medical Research issn: 09622802 maglogo: Y pubinfo: dt: Dec2020 vid: 29 iid: 12 pid: 344 pub: Sage Publications Inc. place: Thousand Oaks, California artinfo: ui: 146317549 146283176 146317549 NLM32662336 146317549 10.1177/0962280220936310 NLM32662336 146317549 ppf: 3641 ppct: 12 formats: tig: atl: A working likelihood approach for robust regression. aug: au: Fu, Liya Wang, You-Gan Cai, Fengjing affil: School of Mathematics and Statistics, Xi'an Jiaotong University, Xi'an, China sug: subj: Probability Computer Simulation Comparative Studies Multicenter Studies Evaluation Research Validation Studies ab: Robust approach is often desirable in presence of outliers for more efficient parameter estimation. However, the choice of the regularization parameter value impacts the efficiency of the parameter estimators. To maximize the estimation efficiency, we construct a likelihood function for simultaneously estimating the regression parameters and the tuning parameter. The "working" likelihood function is deemed as a vehicle for efficient regression parameter estimation, because we do not assume the data are generated from this likelihood function. The proposed method can effectively find a value of the regularization parameter based on the extent of contamination in the data. We carry out extensive simulation studies in a variety of cases to investigate the performance of the proposed method. The simulation results show that the efficiency can be enhanced as much as 40% when the data follow a heavy-tailed distribution, and reaches as high as 468% for the heteroscedastic variance cases compared to the traditional Huber's method with a fixed regularization parameter. For illustration, we also analyzed two datasets: one from a diabetics study and the other from a mortality study. pubtype: Academic Journal doctype: research Journal Article ougenre: Article language: English refInfo: holdings: @attributes: islocal: N |
|---|