欢迎来到天天文库
浏览记录
ID:7279624
大小:370.77 KB
页数:20页
时间:2018-02-10
《heteroscedastic regression models for the systematic analysisof residual variances》由会员上传分享,免费在线阅读,更多相关内容在工程资料-天天文库。
1、Chapter8HeteroscedasticRegressionModelsfortheSystematicAnalysisofResidualVariancesHuiZheng,YangYang,andKennethC.LandAbstractConventionallinearregressionmodelsassumehomoscedasticerrorterms.Thisassumptionoftenisviolatedinempiricalapplications.Variousmethodsforevalua
2、tingtheextentofsuchviolationsandforadjustingtheestimatedmodelparametersifnecessaryaregenerallyavailableinbooksonregressionmethodology.Recentdevelopmentsinstatisticshavetakenadifferentapproachbyexaminingthedatatoascertainwhethertheestimatedheteroscedasticresiduals(
3、fromaÞrst-stageregressionmodeloftheconditionalmeanofanoutcomevariableasafunctionofasetofexplanatoryvariablesorcovariates)arethemselvessystematicallyrelatedtoasetofexplanatoryvariablesinasecond-stageregression.Theseextensionsoftheconventionalmodelshavebeengivenvari
4、ousnamesbut,mostgenerally,areheteroscedasticregressionmodels(HRMs).Insteadoftreatingheteroscedasticityasanuisancetobeadjustedoutofexistencetoreduceoreliminateitsimpactonregressionmodelparameterestimates,thebasicideaofHRMsistomodeltheheteroscedasticityitself.Thisch
5、aptersystematicallyreviewsthespeciÞcationofHRMsinbothlinearandgeneralizedlinearmodelforms,describesmethodsofestimationofsuchmodels,andreportsempiricalapplicationsofthemodelstodataonchangesoverrecentdecadesintheUSincomedistributionandinself-reportedhealth/healthdis
6、parities.Aconcludingsectionpointstosimilaritiesandcomplementaritiesofthegoalsofthecounterfactualapproachtocausalinferenceandheteroscedasticregressionmodels.IntroductionConventionalnormalerrorlinearregressionmodelsandtheirextensionstogeneralizedlinearmodels,linearm
7、ixed(Þxedandrandomeffects)models,andgeneralizedlinearmixedmodelsareessentialworktoolsforsocialresearchthatcanbemodiÞedandadaptedinvariouswaystoembodyvariousaspectsofthecounterfactualparadigmforcausalinference(MorganandWinship2007;GuoandFraser2010).However,conventi
8、onallinearregressionmodelsassumehomoscedasticerrorterms.H.Zheng()DepartmentofSociology,OhioStateUniversity,Columbus,OH,USAe-mail:zheng.64@sociology.osu
此文档下载收益归作者所有