对于行为方法研究bias的修正方法精.docx

对于行为方法研究bias的修正方法精.docx

《对于行为方法研究bias的修正方法精.docx》由会员分享，可在线阅读，更多相关《对于行为方法研究bias的修正方法精.docx（80页珍藏版）》请在冰豆网上搜索。

对于行为方法研究bias的修正方法精

CommonMethodBiasesinBehavioralResearch:

ACriticalReviewofthe

LiteratureandRecommendedRemedies

PhilipM.Podsakoff,ScottB.MacKenzie,and

Jeong-YeonLee

IndianaUniversity

NathanP.Podsakoff

UniversityofFlorida

Interestintheproblemofmethodbiaseshasalonghistoryinthebehavioralsciences.Despitethis,acomprehensivesummaryofthepotentialsourcesofmethodbiasesandhowtocontrolforthemdoesnotexist.Therefore,thepurposeofthisarticleistoexaminetheextenttowhichmethodbiasesinfluencebehavioralresearchresults,identifypotentialsourcesofmethodbiases,discussthecognitiveprocessesthroughwhichmethodbiasesinfluenceresponsestomeasures,evaluatethemanydifferentproceduralandstatisticaltechniquesthatcanbeusedtocontrolmethodbiases,andproviderecommendationsforhowtoselectappropriateproceduralandstatisticalremediesfordifferenttypesofresearchsettings.

Mostresearchersagreethatcommonmethodvariance（i.e.,variancethatisattributabletothemeasurementmethodratherthantotheconstructsthemeasuresrepresentisapotentialprobleminbehavioralresearch.Infact,discussionsofthepotentialimpactofcommonmethodbiasesdatebackwellover40years（cf.Campbell&Fiske,1959,andinterestinthisissueappearstohavecontinuedrelativelyunabatedtothepresentday（cf.Bagozzi&Yi,1990;Bagozzi,Yi,&Phillips,1991;Campbell&O’Connell,1982;Conway,1998;Cote&Buckley,1987,1988;Kline,Sulsky,&Rever-Moriyama,2000;Lindell&Brandt,2000;Lindell&Whit-ney,2001;Millsap,1990;Parker,1999;Schmitt,Nason,Whitney,&Pulakos,1995;Scullen,1999;Williams&Anderson,1994;Williams&Brown,1994.

Methodbiasesareaproblembecausetheyareoneofthemainsourcesofmeasurementerror.Measurementerrorthreatensthevalidityoftheconclusionsabouttherelationshipsbetweenmea-suresandiswidelyrecognizedtohavebotharandomandasystematiccomponent（cf.Bagozzi&Yi,1991;Nunnally,1978;Spector,1987.Althoughbothtypesofmeasurementerrorareproblematic,systematicmeasurementerrorisaparticularlyseriousproblembecauseitprovidesanalternativeexplanationfortheobservedrelationshipsbetweenmeasuresofdifferentconstructsthatisindependentoftheonehypothesized.BagozziandYi（1991notedthatoneofthemainsourcesofsystematicmeasurementerrorismethodvariancethatmayarisefromavarietyofsources:

Methodvariancereferstovariancethatisattributabletothemeasure-mentmethodratherthantotheconstructofinterest.Thetermmethodreferstotheformofmeasurementatdifferentlevelsofabstraction,

suchasthecontentofspecificitems,scaletype,responseformat,andthegeneralcontext（Fiske,1982,pp.81–84.Atamoreabstractlevel,methodeffectsmightbeinterpretedintermsofresponsebiasessuchashaloeffects,socialdesirability,acquiescence,leniencyeffects,oryea-andnay-saying.（p.426

However,regardlessofitssource,systematicerrorvariancecanhaveaseriousconfoundinginfluenceonempiricalresults,yieldingpotentiallymisleadingconclusions（Campbell&Fiske,1959.Forexample,let’sassumethataresearcherisinterestedinstudyingahypothesizedrelationshipbetweenConstructsAandB.Basedontheoreticalconsiderations,onewouldexpectthatthemeasuresofConstructAwouldbecorrelatedwithmeasuresofConstructB.However,ifthemeasuresofConstructAandthemeasuresofConstructBalsosharecommonmethods,thosemethodsmayexertasystematiceffectontheobservedcorrelationbetweenthemea-sures.Thus,atleastpartially,commonmethodbiasesposearivalexplanationforthecorrelationobservedbetweenthemeasures.Withintheabovecontext,thepurposeofthisresearchisto（aexaminetheextenttowhichmethodbiasesinfluencebehavioralresearchresults,（bidentifypotentialsourcesofmethodbiases,（cdiscussthecognitiveprocessesthroughwhichmethodbiasesin-fluenceresponsestomeasures,（devaluatethemanydifferentproceduralandstatisticaltechniquesthatcanbeusedtocontrolmethodbiases,and（eproviderecommendationsforhowtoselectappropriateproceduralandstatisticalremediesfordifferenttypesofresearchsettings.Thisisimportantbecause,toourknowledge,thereisnocomprehensivediscussionofalloftheseissuesavail-ableintheliterature,andtheevidencesuggeststhatmanyre-searchersarenoteffectivelycontrollingforthissourceofbias.

ExtentoftheBiasCausedbyCommonMethodVariance

Overthepastfewdecades,aconsiderableamountofevidencehasaccumulatedregardingtheextenttowhichmethodvarianceinfluences（ameasuresusedinthefieldand（brelationshipsbetweenthesemeasures.Muchoftheevidenceoftheextenttowhichmethodvarianceispresentinmeasuresusedinbehavioralresearchcomesfrommeta-analysesofmultitrait–multimethod

PhilipM.PodsakoffandJeong-YeonLee,DepartmentofManagement,IndianaUniversity;ScottB.MacKenzie,DepartmentofMarketing,Indi-anaUniversity;NathanP.Podsakoff,DepartmentofManagement,Uni-versityofFlorida.

CorrespondenceconcerningthisarticleshouldbeaddressedtoPhilipM.Podsakoff,DepartmentofManagement,KelleySchoolofBusiness,Indi-anaUniversity,1309EastTenthStreet,Bloomington,Indiana47405-1701.E-mail:

podsakof@indiana.edu

JournalofAppliedPsychologyCopyright2003bytheAmericanPsychologicalAssociation,Inc.

2003,Vol.88,No.5,879–9030021-9010/03/$12.00DOI:

10.1037/0021-9010.88.5.879

879

studies（cf.Bagozzi&Yi,1990;Cote&Buckley,1987,1988;Williams,Cote,&Buckley,1989.Perhapsthemostcomprehen-siveevidencecomesfromCoteandBuckley（1987,whoexam-inedtheamountofcommonmethodvariancepresentinmeasuresacross70MTMMstudiesinthepsychology–sociology,marketing,business,andeducationliteratures.Theyfoundthatapproximatelyonequarter（26.3%ofthevarianceinatypicalresearchmeasuremightbeduetosystematicsourcesofmeasurementerrorlikecommonmethodbiases.However,theyalsofoundthattheamountofvarianceattributabletomethodbiasesvariedconsiderablybydisciplineandbythetypeofconstructbeinginvestigated.Forexample,CoteandBuckley（1987foundthat,onaverage,methodvariancewaslowestinthefieldofmarketing（15.8%andhighestinthefieldofeducation（30.5%.Theyalsofoundthattypicaljobperformancemeasurescontainedanaverageof22.5%methodvariance,whereasattitudemeasurescontainanaverageof40.7%.AsimilarpatternoffindingsemergesfromWilliamsetal.’s（1989studyofjusttheappliedpsychologyliterature.

Inadditiontotheseestimatesoftheextenttowhichmethodvarianceispresentintypicalmeasures,thereisalsoagrowingbodyofresearchexaminingtheextenttowhichmethodvarianceinfluencesrelationshipsbetweenmeasures（cf.Fuller,Patterson,Hester,&Stringer,1996;Gerstner&Day,1997;Lowe,Kroeck,&Sivasubramaniam,1996;Podsakoff,MacKenzie,Paine,&Bach-rach,2000;Wagner&Gooding,1987.Thesestudiescontrastedthestrengthoftherelationshipbetweentwovariableswhencom-monmethodvariancewascontrolledversuswhenitwasnot.Theyfoundthat,onaverage,theamountofvarianceaccountedforwhencommonmethodvariancewaspresentwasapproximately35%versusapproximately11%whenitwasnotpresent.Thus,thereisaconsiderableamountofevidencethatcommonmethodvariancecanhaveasubstantialeffectonobservedrelationshipsbetweenmeasuresofdifferentconstructs.However,itisimportanttorec-ognizethatthefindingssuggestthatthemagnitudeofthebias

producedbythesemethodfactorsvariesacrossresearchcontexts（cf.Cote&Buckley,1987;Crampton&Wagner,1994.

Notonlycanthestrengthofthebiasvarybutsocanthedirectionofitseffect.Methodvariancecaneitherinflateordeflateobservedrelationshipsbetweenconstructs,thusleadingtobothTypeIandTypeIIerrors.ThispointisillustratedinTable1,whichusesCoteandBuckley’s（1987estimatesoftheaverageamountoftraitvariance,theaverageamountofmethodvariance,andtheaveragemethodintercorrelationsandinsertsthemintotheequationbelowtocalculatetheimpactofcommonmethodvari-anceontheobservedcorrelationbetweenmeasuresofdifferenttypesofconstructs（e.g.,attitude,personality,aptitude:

Rx,yϭ（trueRti,tjͱxͱyϩ（trueRmk,mlͱxͱy,

（1

wheretrueRti,tjistheaveragecorrelationbetweentraitiandtraitj,txisthepercentoftraitvarianceinmeasurex,tyisthepercentoftraitvarianceinmeasurey,trueRmk,mlistheaveragecorrelationbetweenmethodkandmethodl,mxisthepercentofmethodvarianceinmeasurex,andmyisthepercentofmethodvarianceinmeasurey.

Forexample,thecorrelation.52inthesecondrowofthefirstcolumnofTable1wascalculatedbymultiplyingthetruecorre-lation（1.00timesthesquarerootofCoteandBuckley’s（1987estimateofthepercentoftraitvariancetypicallyfoundinattitudemeasures（ͱtimesthesquarerootoftheirestimateofthepercentoftraitvariancetypicallyfoundinpersonalitymeasures（ͱ.391plustheaverageoftheirestimatesofthetypicalcorrelationbetweenmethodsforattitude（.556andpersonality（.546con-structsmultipliedbythesquarerootoftheirestimateofthepercentofmethodvariancetypicallyfoundinattitudemeasures（ͱtimesthesquarerootoftheirestimateofthepercentofmethodvariancetypicallyfoundinpersonalitymeasures（ͱ.247.

Table1

RelationshipBetweenTrueandObservedCorrelationforAverageMeasuresbyTypeofConstruct

TrueRti,tjcorrelation（Rti,tj2

TypeofConstructs

1.00（1.00.50（.25.30（.09.10（.01.00（.00Attitude–attitude.52（.27.38（.14.32（.10.26（.07.23（.05Attitude–personality.52（.27.35（.12.28（.08.21（.04.17（.0