资源描述:
《missing data techniques for structural equation modeling结构方程模型的缺失数据技术》由会员上传分享,免费在线阅读,更多相关内容在教育资源-天天文库。
1、JournalofAbnormalPsychologyCopyright2003bytheAmericanPsychologicalAssociation,Inc.2003,Vol.112,No.4,545–5570021-843X/03/$12.00DOI:10.1037/0021-843X.112.4.545MissingDataTechniquesforStructuralEquationModelingPaulD.AllisonUniversityofPennsylvaniaAswitho
2、therstatisticalmethods,missingdataoftencreatemajorproblemsfortheestimationofstructuralequationmodels(SEMs).Conventionalmethodssuchaslistwiseorpairwisedeletiongenerallydoapoorjobofusingalltheavailableinformation.However,structuralequationmodelersarefor
3、tunatethatmanyprogramsforestimatingSEMsnowhavemaximumlikelihoodmethodsforhandlingmissingdatainanoptimalfashion.Inadditiontomaximumlikelihood,thisarticlealsodiscussesmultipleimputation.Thismethodhasstatisticalpropertiesthatarealmostasgoodasthoseformaxi
4、mumlikelihoodandcanbeappliedtoamuchwiderarrayofmodelsandestimationmethods.VirtuallyallmethodsofstatisticalanalysisareplaguedbydelinquencyandXisyearsofschooling,theMARassumptionproblemswithmissingdata,andstructuralequationmodelingisnowouldbesatisfiedif
5、theprobabilitythatdelinquencyismissingexception.Itiswellknownthattheuseofinappropriatemethodsdependsonyearsofschooling,butwithineachlevelofschoolingforhandlingmissingdatacanleadtobiasinparameterestimatestheprobabilityofmissingdelinquencydoesnotdependo
6、ndelin-(Jones,1996),biasinstandarderrorsandteststatistics(Glasser,quency.Inessence,MARallowsmissingnesstodependonthings1964),andinefficientuseofthedata(Afifi&Elashoff,1966).Thisthatareobserved,butnotonthingsthatarenotobserved.Clearly,articlesurveysvar
7、iousmethodsthatareavailableforhandlingifthedataaremissingcompletelyatrandom,theyarealsomissingmissingdataintheestimationofstructuralequationmodelsatrandom.(SEMs).AfterreviewingsuchconventionalmethodsaslistwiseItisstraightforwardtotestwhetherthedataare
8、missingcom-deletion,pairwisedeletion,andregressionimputation,Ifocusonpletelyatrandom.Forexample,onecouldcomparemenandtheimplementationoftwonewermethods,maximumlikelihoodwomentotestwhethertheydifferintheproportionofcaseswithandmultipleimputatio