85%CI:0.13,0.35).Fortheassociationbetweenavo-idantversusnonavoidantatt - PDF document

85%CI:0.13,0.35).Fortheassociationbetweenavo-idantversusnonavoidantattachmentsandexternal-izing(=33),thecombinedeffectssizewas=0.12(.01;85%CI:0.06,0.21),andfortheaggressionoutcomes(=20)thecombinedeffectsizeamountedto=0.28(.05;85%CI:0.08,0.49).Fortheassociationbetweenresistantversusnonresistantattachmentsandexternalizing(=34),thecombinedeffectssizewas=0.10(&#x-334;&#x.900;.05;85%0.01,0.21),andfortheaggressionoutcomes=22)thecombinedeffectsizewas=0.05&#x-334;&#x.900;.05;85%CI:0.10,0.21).Finally,fortheassoci-ationbetweendisorganizedattachmentsandexter-nalizing(=33),thecombinedeffectsizewas=0.37(.01;85%CI:0.25,0.49),whereasfortheaggressionoutcomes(=16)thiscombinedeffectsizewas=0.12(&#x-334;&#x.900;.05;85%CI:0.30).The85%CIsdidoverlapforeachofthesecomparisons,indicatingthatstudyoutcomeswithexternalizingbehaviororwithaggressiondidnotresultinsigniÞcantlydifferenteffectsizes.Becausethevariouscomparisons(securevs.insecure,avoidantvs.nonavoidant,anddisorga-nizedvs.nondisorganized)werebasedonvaryingnumbersofstudiesandparticipants,wealsoselectedacoresetof19studiesthatprovideddataonallfourcomparisons.Fortheassociationbetweenattachmentsecurityandexternalizing,the Table5DisorganizedAttachmentandExternalizingBehavior:TotalSetandSetofSSPStudieskNdinterval95%HomogeneityTotalset343,7780.34**0.18,0.5099.66**0.75.39Nonclinical273,1840.30**0.12,0.4983.00**Clinical75940.43**0.26,0.6111.916.58.04Boys88390.35*0.03,0.6620.84**Girls67020.015.62Mixed202,2370.44**0.26,0.6142.82**0.42.52Low91,2660.44**0.28,0.6013.42high252,5120.31**0.12,0.5080.22**SSP243,1610.27**0.10,0.4569.00**95730.50**0.16,0.8419.41*Other1441.10**0.41,1.79ObserverextMother242,7580.20*0.02,0.3952.94**Teacher44150.48*0.06,0.9111.73**Observed00Other43450.62**0.39,0.845.32Combined22600.61*0.08,1.137.32**SSPstudiesAllSSPstudies243,1610.27**0.10,0.4569.00**0.08.77Nonclinical202,8170.29**0.09,0.4867.25**Clinical43440.26*0.03,0.381.55.01Boys35680.120.12,0.365.17Girls46690.051.84Mixed171,9240.39**0.22,0.5631.38**2.15.14Low91,2660.44**0.28,0.6013.42high151,8950.170.03,0.3843.81**.SES=socioeconomicstatus;SSP=StrangeSituationprocedure;CassMarvin=CassidyandMarvinMacArthurPreschoolAttachmentCodingSystem.Subgroupwith4excludedfromcontrast.Insecure-otherincludedindisorganized..05.**.01.AttachmentandExternalizingBehaviorProblems447 duringthepreschoolandearlyschool-ageperiod.DevelopmentandPsychopathology,511Ð525.Moss,E.,Smolla,N.,Cyr,C.,DuboisComtois,K.,Mazza-rello,T.,&Berthiaume,C.(2006).Attachmentandbehaviorproblemsinmiddlechildhoodasreportedbyadultandchildinformants.DevelopmentandPsychopa-,425Ð444.Mullen,B.(1989).AdvancedBASICmeta-analysis.Hills-dale,NJ:Erlbaum.Mullen,B.,&Rosenthal,R.(1985).BASICmeta-analysisHillsdale,NJ:Erlbaum.*Munson,J.A.,McMahon,R.J.,&Spieker,S.J.(2001).Structureandvariabilityinthedevelopmentaltrajec-toryofchildrenÕsexternalizingproblems:Impactofinfantattachment,maternaldepressivesymptomatol-ogy,andchildsex.DevelopmentandPsychopathologyNationalInstituteofChildHealthandHumanDevel-opment(NICHD)EarlyChildCareResearchNet-work.(2006).Infant-motherattachmentclassiÞcation:Riskandprotectioninrelationtochangingmaternalcaregivingquality.DevelopmentalPsychology,38ÐOrobiodeCastro,B.,Veerman,J.W.,Koops,W.,Bosch,J.D.,&Monshouwer,H.J.(2002).Hostileattributionofintentandaggressivebehavior:Ameta-analysis.,916Ð934.*PerezCorres,A.(2006).MotherÕsattachmentrepresenta-tionsasacontextualdeterminantofchildrenÕsmentalhealth.Asixteen-yearlongitudinalstudy.Unpublisheddoctoraldissertation,UniversityofLondon,United*Pierrehumbert,B.,Miljkovitch,R.,Plancherel,B.,Halfon,O.,&Ansermet,F.(2000).Attachmentandtempera-mentinearlychildhood;ImplicationsforlaterbehaviorInfantandChildDevelopment,17Ð32.*RadkeYarrow,M.,McCann,K.,DeMulder,E.,Belmont,B.,etal.(1995).Attachmentinthecontextofhigh-riskDevelopmentandPsychopathology,247ÐRaine,A.,Venables,P.H.,&Mednick,S.A.(1997).Lowrestingheartrateatage3yearspredisposestoaggres-sionatage11years:EvidencefromtheMauritiusChildHealthProject.JournaloftheAmericanAcademyofChildandAdolescentPsychiatry,1457Ð1464.Renken,B.,Egeland,B.,Marvinney,D.,Mangelsdorf,S.,etal.(1989).Earlychildhoodantecedentsofaggressionandpassive-withdrawalinearlyelementaryschool.JournalofPersonality,257Ð281.Rosenthal,R.(1991).Meta-analyticproceduresforsocial.NewburyPark,CA:Sage.*Rothbaum,F.,Rosen,K.S.,Pott,M.,&Beatty,M.(1995).Earlyparent-childrelationshipsandlaterproblembehavior:Alongitudinalstudy.Merrill-PalmerQuar-,133Ð151.Rutter,M.(1995).Clinicalimplicationsofattachmentcon-cepts:Retrospectandprospect.JournalofChildPsychol-ogyandPsychiatry,549Ð572.Sameroff,A.J.(2000).Developmentalsystemsandpsy-DevelopmentandPsychopathology*Schmidt,M.E.,Demulder,E.K.,&Denham,S.(2002).Kindergartensocial-emotionalcompetence:Develop-mentalpredictorsandpsychosocialimplications.ChildDevelopmentandCare,451Ð462.Schneider-Rosen,K.,&Rothbaum,F.(1993).Qualityofparentalcaregivingandsecurityofattachment.opmentalPsychology,358Ð367.*Seifer,R.,LaGasse,L.L.,Lester,B.,Bauer,C.R.,Shanka-ran,S.,Bada,H.S.,etal.(2004).Attachmentstatusinchildrenprenatallyexposedtococaineandothersub-ChildDevelopment,850Ð868.*Shaw,D.S.,Owens,E.B.,Vondra,J.I.,Keenan,K.,&Winslow,E.B.(1996).Earlyriskfactorsandpathwaysinthedevelopmentofearlydisruptivebehaviorprob-DevelopmentandPsychopathology,679Ð699.Shaw,D.S.,&Vondra,J.I.(1995).Infantattachmentsecurityandmaternalpredictorsofearlybehaviorproblems:Alongitudinalstudyoflow-incomefamilies.JournalofAbnormalChildPsychology,335Ð357.*Smeekens,S.,RiksenWalraven,J.M.,&vanBakel,H.(2007).Multipledeterminantsofexternalizingbehaviorin5-year-olds:Alongitudinalmodel.JournalofAbnor-malChildPsychology,347Ð361.Solomon,J.,&George,C.(1999).Theplaceofdisorgani-zationinattachmenttheory:Linkingclassicobserva-tionswithcontemporaryÞndings.InJ.Solomon&C.George(Eds.),Attachmentdisorganization(Vol.23,pp.3Ð32).NewYork:Guilford.*Solomon,J.,George,C.,&DeJong,A.(1995).ChildrenclassiÞedascontrollingatagesix:Evidenceofdisorga-nizedrepresentationalstrategiesandaggressionathomeandatschool.DevelopmentandPsychopathology*Speltz,M.L.,DeKlyen,M.,&Greenberg,M.T.(1999).Attachmentinboyswithearlyonsetconductproblems.DevelopmentandPsychopathology,269Ð285.*Speltz,M.L.,Greenberg,M.T.,&DeKlyen,M.(1990).Attachmentinpreschoolerswithdisruptivebehavior:Acomparisonofclinic-referredandnon-problemchil-DevelopmentandPsychopathology,31Ð46.Sroufe,L.A.(1983).Infant-caregiverattachmentandpatternsofadaptationinpreschool:Therootsofmaladaptionand(Vol.16).Hillsdale,NJ:Erlbaum.Sroufe,L.A.,Egeland,B.,Carlson,E.,&Collins,W.A.Thedevelopmentoftheperson:TheMinnesotastudyofriskandadaptationfrombirthtoadulthood.NewYork:Guilford.Sroufe,L.A.,Egeland,B.,Carlson,E.,&Collins,W.A.(2005b).Placingearlyattachmentexperiencesindevel-opmentalcontext:TheMinnesotalongitudinalstudy.InK.E.Grossmann,K.Grossmann,&E.Waters(Eds.),Attachmentfrominfancytoadulthood:Themajorlongitudi-nalstudies(pp.48Ð70).NewYork:Guilford.*Stams,G.J.,Juffer,F.,&vanIJzendoorn,M.H.(2002).Maternalsensitivity,infantattachment,andAttachmentandExternalizingBehaviorProblems455 Severalstudiespresenteddataon(partly)over-lappingsamples,suchasShawandcolleagues(Shaw,Owens,Vondra,Keenan,&Winslow,1996;Shaw&Vondra,1995)andthestudiesreportedbyMossandcolleagues(Moss,Bureau,Cyr,Mongeau,&StLaurent,2004;Moss,Cyr,etal.,2004;Moss,Parent,Gosselin,&Rousseau,1996).Becausepartic-ipantscannotbeincludedinameta-analysismorethanonce,thepapersthatreportedonthelargestgroupsofparticipantswereincludedinourmeta-analysis(e.g.,Moss,Bureau,etal.,2004).Intotal,afterexcludingreportsinvolvingoverlappingsamples,wefound53studiesthatyielded69inde-pendentsamplesthatcouldbeincludedinourmeta-analyses,withsamplesizesrangingfrom26to1,075(seeTable1).Inmanycases,outcomesta-tisticswereonlypresentedfortheavoidantandresistantclassiÞcationscombined,orindeedfortheresistant,avoidant,anddisorganizedcasescom-bined.Consequently,wefocusedourprimaryanal-ysesontheoverallcontrastbetweensecurityandinsecurity,withinsecurityrepresentedbytheavoi-dant,resistantand(inthecaseswheredisorganiza-tionhadbeencoded)disorganizedclassiÞcations.Intheseanalyses,wealsotestedwhetheritmadeadifferencetotheoveralleffectsizeforsecurityifdisorganizationhadbeencoded.Inaddition,anumberofstudiesusedtheAQStomeasureattach-mentsecurity,whichdoesnotyielddataonthedifferentsubtypesofinsecurity.Asaresult,thesestudiesonlyappearinthemeta-analysesinvolvingtheoverallcontrastbetweensecurityandinsecu-rity.Subsequently,wealsoextractedmorefocusedcontraststargetingspeciÞcinsecurecategoriesfromthesmallersetofstudieswherethesecouldbeidentiÞed.Thenumbersofstudiesinvolvedinthesesubanalysesareindicatedinthetext.CodingSystemWeusedastructuredcodingsystemforassess-ingthecharacteristicsofthesamplesandtheirstudydesigns.Themeasurementofattachmentwascodedstraightforwardly,asallstudiesincludedoneofseveralwell-knownattachmentassessments(SSP,AQS,PreschoolAttachmentAssessment,Cas-sidy&Marvin;Main&Cassidy).Ineachcase,thecoderextractedeffectsizesattheleveloftheindi-vidualattachmentclassiÞcationwherepossible(i.e.,A,B,C,andD).Inadditiontoanumberofback-groundvariableslikeyearofpublicationanddatasource(journal,bookchapter,unpublisheddata),wecodedseveralimportantpotentialmoderatorsrelatedtothesample:gender(%male),SESmiddlevs.low),andclinicalstatus(clinical-child,clinical-parent,nonclinical).Wherethegen-dercompositionofthesamplewasnotpreciselyreportedweassumeda50%split.Furthermore,whenSESwasnotnoted,adefaultofhighmiddleclasswasrecorded(thisoccurredinÞvecases).ClinicalstatuswasrecordedifeithertheparentorthetargetchildwasidentiÞedashavingaclinicaldiagnosisoriftheyhadbeenselectedusingaclini-calcutoffscoreonavalidatedinstrument.Inaddi-tiontothemeasureusedtoassessattachment,fourotherdesigncharacteristicswerecoded:(a)ageofattachmentassessment,(b)ageofexternalizingassessment,(c)typeofoutcomemeasure,and(d)observerorreporterofexternalizingbehavior.Whenmeasurementsofexternalizingproblemsatdifferentpointsintimewerereported,weselectedtheÞrstmeasurement.Whenmeasurementsatdif-ferentpointsintimewereinsomewaymerged(averagedoratrajectoryextracted),werecordedthemidpointoftherangeastheageoftheexternal-izingassessment(e.g.,Keller,Spieker,&Gilchrist,2005).Toassessintercoderreliability,20randomlyselectedstudieswerecodedbytwocoders.Theagreementbetweenthecodersacrossthemoderatorvariableswas97%(correlationsbetweencontinuousmoderatorswere�.95).Meta-AnalyticProceduresAnumberofstudiesreportedresultsseparatelyforboysandgirlsandfourstudiesreportedonsamplesinvolvingonlyboysoronlygirls.Inthesecases,wecalculatedseparateeffectsizesforeachgender,andthesubsamplesweretreatedasinde-pendentoutcomesintheanalyses.Whenmultiplemeasuresofaggressionorexternalizingbehaviorswereusedwithinonestudy(e.g.,Solomon,George,&DeJong,1995),weselectedtheoutcomeforexternalizingbehaviorfortheprimarysetofstud-ies.Inaseparatesetofmeta-analyses,wetestedwhetheroutcomesweredifferentwhenaggressionoutcomeswereselected(seethefollowing).StatisticalAnalysesFoursetsofmeta-analyseswereconducted,onefortherelationbetweenattachmentinsecurityandexternalizingoraggressivebehavior,onefortherelationbetweenavoidanceandexternalizingoraggressivebehavior,onefortherelationbetweenresistanceandexternalizingoraggressivebehavior,andonefortherelationbetweenattachmentdisorganizationandexternalizingoraggressiveAttachmentandExternalizingBehaviorProblems439 betweendisorganizedattachmentandexternalizingbehavior.Inthesetof24SSPstudieswith=3,161partic-ipants,wefoundasigniÞcantcombinedeffectsize=0.27,whichwassimilartotheeffectsizecomputedforthetotalset.NoothersigniÞcantmoderatorswerefoundinthisset(seeTable5forthosecontraststhatcouldbetested).AdditionalAnalysesThecoresetofstudiesonexternalizingincludedassessmentsofexternalizingaswellasofaggres-sivebehaviorinthecaseofstudiesthatdidnotpresentdataonexternalizingproblems.Inordertoexaminewhetherthemorefocusedaggressionstudieswouldresultinhighereffectsizes,wedecidedtoconducttwosetsofmeta-analyses,oneforstudieswithdataonexternalizingandoneforstudiespresentingaggressiondata.Thesetwosetsofstudiespartiallyoverlapped(assomestudiesreportedboth),anditwasthereforeimpossibletocomparedirectlyeffectsizesacrossthesesets(seetheMethodsection).Wecomputed85%CIsforthepointestimatesofthecombinedeffectsizesandcomparedtheseintervalsacrossthetwosetsofstudies:Nonoverlapping85%CIsindicatedasig-niÞcantdifferenceincombinedeffectsizes(Baker-mans-Kranenburgetal.,2003).Fortheassociationbetweensecureversusinsecureattachmentandexternalizingbehavior(=65),acombinedeffectsizeof=0.32(.01;85%CI:0.26,0.38)wasfound.Thecomparablecombinedeffectsizefortheaggressionoutcomes(=32)was=0.24(.01; Table4ResistantAttachmentandExternalizingBehavior:TotalSetandSetofSSPStudieskNdinterval95%ContrastTotalset353,5680.110.04,0.2666.32**1.36.24Nonclinical293,1650.060.10,0.2348.76**Clinical64030.300.05,0.6414.27*0.11.95Boys98830.160.02,0.347.47Girls108090.160.20,0.5222.88**Mixed161,8760.100.12,0.3235.45**Low38570.040.17,0.240.66high322,7110.110.06,0.2865.17**SSP242,9100.050.13,0.2435.54*CassMarvin95590.25*0.05,0.4512.90Other2990.390.23,1.0113.24**ObserverextMother272,8310.080.12,0.2854.62**Teacher33340.210.29,0.717.14*Observed1480.66,0.52Other32500.240.02,0.491.03Combined11050.100.29,0.49SSPstudiesAllSSPstudies242,9100.050.13,0.2435.54*0.56.75Boys56370.110.14,0.364.17Girls77660.170.19,0.5316.59*Mixed121,5070.010.13,0.1414.03Low38570.040.17,0.240.66high212,0530.050.14,0.2334.87*.SES=socioeconomicstatus;SSP=StrangeSituationprocedure;CassMarvin=CassidyandMarvinMacArthurPreschoolAttachmentCodingSystem.Subgroupswith4excludedfromcontrast..05.**.01.446Fearon,Bakermans-Kranenburg,vanIJzendoorn,Lapsley,andRoisman James,W.(1890).Principlesofpsychology.NewYork:HenryHolt.Kagan,J.(2007).Atrioofconcerns.PerspectivesonPsycho-logicalScience,361Ð376.*Keller,T.E.,Spieker,S.J.,&Gilchrist,L.(2005).Patternsofriskandtrajectoriesofpreschoolproblembehaviors:Aperson-orientedanalysisofattachmentincontext.DevelopmentandPsychopathology,349Ð384.*KleinVelderman,M.,Bakermans-Kranenburg,M.J.,Juffer,F.,&vanIJzendoorn,M.H.(2006).Effectsofattachment-basedinterventionsonmaternalsensitivityandinfantattachment:Differentialsusceptibilityofhighlyreactiveinfants.JournalofFamilyPsychologyKobak,R.,Cassidy,J.,LyonsRuth,K.,&Ziv,Y.(2005).Attachment,stressandpsychopathology:Adevelop-mentalpathwaysmodel.InD.Cicchetti&J.D.CohenDevelopmentalpsychopathology,Vol.I:Theoryand(2nded.,pp.333Ð369).Hoboken,NJ:Wiley.Kochanska,G.(1997).Mutuallyresponsiveorientationbetweenmothersandtheiryoungchildren:Implica-tionsforearlysocialization.ChildDevelopment,94ÐLamb,M.E.,Thompson,R.A.,Gardner,W.P.,Charnov,E.L.,&Estes,D.(1984).SecurityofinfantileattachmentasassessedintheÔÔStrangeSituationÕÕ:Itsstudyandbiologicalinterpretation.BehavoralandBrainSciences*Lewis,M.,Feiring,C.,McGuffog,C.,&Jaskir,J.(1984).Predictingpsychopathologyinsix-year-oldsfromearlysocialrelations.ChildDevelopment,123Ð136.*Lieberman,A.F.,Weston,D.R.,&Pawl,J.H.(1991).Preventiveinterventionandoutcomewithanxiouslyattacheddyads.ChildDevelopment,199Ð209.Liotti,G.(1992).Disorganizeddisorientedattachmentintheetiologyofthedissociativedisorders.,196Ð204.Lipsey,M.W.,&Wilson,D.B.(2001).Practicalmeta-anal-ysis.Appliedsocialresearchmethodsseries(Vol.49).Thou-sandOaks,CA:Sage.Loeber,R.,&Hay,D.(1997).Keyissuesinthedevelopmentofaggressionandviolencefromchildhoodtoearlyadult-AnnualReviewofPsychology,371Ð410.Loeber,R.,&Schmaling,K.B.(1985).Empiricalevidenceforovertandcovertpatternsofantisocialconductproblems:Ameta-analysis.JournalofAbnormalChild,337Ð353.Lorber,M.F.(2004).Psychophysiologyofaggression,psychopathy,andconductproblems:Ameta-analysis.PsychologicalBulletin,531Ð552.*LyonsRuth,K.,Alpern,L.,&Repacholi,B.(1993).Dis-organizedinfantattachmentclassiÞcationandmaternalpsychosocialproblemsaspredictorsofhostile-aggres-sivebehaviorinthepreschoolclassroom.ChildDevelop-,572Ð585.LyonsRuth,K.,Zeanah,C.H.,&Benoit,D.(2003).Disorderandriskfordisorderduringinfancyandtoddlerhood.InE.J.Mash&R.A.Barkley(Eds.),(2nded.,pp.589Ð631).NewYork:*Madigan,S.,Moran,G.,Schuengel,C.,Pederson,D.R.,&Otten,R.(2007).Unresolvedmaternalattachmentrepresentations,disruptedmaternalbehavior,anddis-organizedattachmentininfancy:Linkstotoddlerbehaviorproblems.JournalofChildPsychologyandPsy-,1042Ð1050.Main,M.,&Cassidy,J.(1988).Categoriesofresponsetoreunionwiththeparentatage6:PredictablefrominfantattachmentclassiÞcationsandstableovera1-monthperiod.DevelopmentalPsychology,415Ð426.Main,M.,Kaplan,N.,&Cassidy,J.(1985).Securityininfancy,childhoodandadulthood:Amovetothelevelofrepresentation.InI.Bretherton&E.Waters(Eds.),GrowingpointsofattachmenttheoryandMonographsoftheSocietyforResearchinChild(1Ð2,SerialNo.209),66Ð104.Main,M.,&Morgan,H.(1996).Disorganizationanddis-orientationininfantstrangesituationbehavior:Pheno-typicresemblancetodissociativestates.InL.K.Michelson&W.J.Ray(Eds.),Handbookofdissociation:Theoretical,empirical,andclinicalperspectives(pp.107Ð138).NewYork:Plenum.Main,M.,&Solomon,J.(1986).Discoveryofaninsecure-disorientedattachmentpattern.InT.B.Brazelton&M.W.Yogman(Eds.),Affectivedevelopmentininfancy(pp.95Ð124).Norwood:Ablex.Main,M.,&Solomon,J.(1990).ProceduresforidentifyinginfantsasdisorganizeddisorientedduringtheAinsworthStrangeSituation.InM.T.Greenberg&D.Cicchetti(Eds.),Attachmentinthepreschoolyears:Theory,research,andintervention.TheJohnD.andCatherineT.MacArthurFoundationseriesonmentalhealthanddevelopment(Vol.19,pp.121Ð160).Chicago:UniversityofChicagoPress.*Marchand,J.F.,&Hock,E.(1998).Therelationofprob-lembehaviorsinpreschoolchildrentodepressivesymptomsinmothersandfathers.JournalofGenetic,353Ð366.*Matas,L.,Arend,R.A.,&Sroufe,L.A.(1978).Continu-ityofadaptationinthesecondyear:TherelationshipbetweenqualityofattachmentandlatercompetentChildDevelopment,547Ð556.McCartney,K.,&Rosenthal,R.(2000).Effectsize,practi-calimportance,andsocialpolicyforchildren.,173Ð180.Moss,E.,Bureau,J.F.,Cyr,C.,Mongeau,C.,&StLau-rent,D.(2004).Correlatesofattachmentatage3:Con-structvalidityofthepreschoolattachmentclassiÞcationDevelopmentalPsychology,323Ð334.*Moss,E.,Cyr,C.,&Dubois-Comtois,K.(2004).Attach-mentatearlyschoolageanddevelopmentalrisk:Examiningfamilycontextsandbehaviorproblemsofcontrolling-caregiving,controlling-punitive,andbehav-iorallydisorganizedchildren.DevelopmentalPsychology,519Ð532.Moss,E.,Parent,S.,Gosselin,C.,&Rousseau,D.(1996).Attachmentandteacher-reportedbehaviorproblems454Fearon,Bakermans-Kranenburg,vanIJzendoorn,Lapsley,andRoisman externalizingbehaviorproblems;(b)strongereffectswouldbefoundinlow-socioeconomic-status(SES)samplesthanhigh-SESsamples;(c)strongerassociationswouldbefoundinboysthaningirls;and(d)attachmentdisorganizationwouldpredictexternalizingproblemsmorestronglythanavoid-anceorinsecuritygenerally.Weaddedtotheseafocusonwhethereffectsofattachment-relatedvaria-tionweremoderatedbyageofassessmentofexter-nalizingproblems.TheclaimthatearlyattachmenthasenduringÑratherthanmerelytransientÑeffectsondevelopmentrequiresthatthemagnitudeofsuchassociationsarenotreducedtonilovertime.Finally,wealsoexaminedarangeofrelevantmethodologi-calfactorsthatmightaccountforsystematicbetween-studyvariabilityineffectsizes,includingthemethodofassessmentofattachmentandthetypeandcontextofoutcomemeasurement.LiteratureSearchWesystematicallysearchedtheelectronicdata-basesPsychInfo,WebofScience,MEDLINE,Sci-enceCitationIndexExpanded,SocialSciencesCitationIndex,andArt&HumanitiesCitationIndexwiththekeywordsexternalizing,aggressi*,conduct,psychopathology,opposition*,compe-tence,socialfunctioning,prosocial,antisocial,anti-social,behaviorproblem*,behaviourproblem*inthetitleorabstract(theasteriskindicatesthatthesearchcontainedthewordorwordfragment).Thislargesetwasnarroweddownbyaddingthecon-straintthatthepapersmustalsocontainthewordattachmentandchild*orinfan*inthetitleorabstract.Thissearchreturnedover1,200articlesineachofthedatabases.Twofurtherseparatesearch-narrowingstrategieswereadopted,whichyieldedtwopartiallyoverlappingstudysets.Inonesearch,wetargetedempiricalstudiesbyrequiringtheabstracttocontainthewordssampleor.Inanother,werequiredthatatleastoneofthewordssecur*,avoidan*,resistan*,ordisorgani*appearedinthetitleorabstract.Ineachcase,thisreducedtheformersearchbyaround60%.Whenthesetwosetsofsearchresultsweremerged,thisresultedin856candidatearticles.TheseweresubjectedtoabstractreviewintheÞrstinstance,fromwhichalargenumberofclearlyirrelevantarticlesweredis-carded(e.g.,nonempiricalpapers,studiesnotinvolvingchildren).Afurther115articlesremained.Thesewereexaminedindividuallybytheauthorsaccordingtocriteriadescribednext.Second,thereferencelistsofthecollectedempiricalpapersandinßuentialreviewsweresearchedforrelevantstudies(e.g.,Kobaketal.,2005).Third,datasetsavailabletotheauthorssincetheywereinthepublicÞeld(NICHDStudyofEarlyChildCareandYouthDevelopment)orpartoftheirongoingresearch(SCRIPT;VanZeijletal.,2006)wereana-lyzedwithregardtotheassociationsbetweenattachmentandexternalizingbehavior.Studieswereincludediftheyreportedontherelationbetweenattachmentandexternalizingoraggressivebehaviorinchildren12yearsofageoryounger.ExternalizingbehaviorwasdeÞnedasaggression,oppositionalproblems,conductprob-lemsorhostility(eitheraloneorincombination),asindicatedinthedescriptionsprovidedinthemethodsectionsoftherespectivearticles.Studiesthatdidnotdifferentiatebetweenexternalizingandinternalizingproblems(e.g.,justtotalproblemsscoreoftheCBCL)wereexcluded.Externalizingoraggressivebehaviorwasassessedusingobservation(e.g.,Matas,Arend,&Sroufe,1978;Turner,1991),questionnaires(CBCL,PBQ)orclinicalinterviews(e.g.,Speltz,DeKlyen,&Greenberg,1999),com-pletedbyparents(e.g.,Aviezer,Sagi,Resnick,&Gini,2002),teachers(e.g.,Egeland&Heister,1995),orcliniciansortrainedobservers(e.g.,Turner,1991).Werestrictedthereviewtostudiesusingobservationalmeasuresofattachment,suchastheSSP(Ainsworthetal.,1978),theCassidy&MarvinPreschoolAttachmentsystem(Cassidy,Marvin,andTheMacArthurWorkingGrouponAttach-ment,1989),theAttachmentQ-Sort(AQS;Waters&Deane,1985),andtheMainandCassidysystem(Main&Cassidy,1988).Incaseswheremorethanoneattachmentassessmentwasemployed(e.g.,theSSPfollowedbyCassidy&Marvinatalaterage),theearliestattachmentassessmentwasselected.Wedidnotincludestudiesthatreportedonrepre-sentationalmeasuresofattachment(e.g.,Attach-mentStoryCompletionTask;Verschueren&Marcoen,1999).WheninterventionstudieswereidentiÞed,weonlyincludeddatafromthenon-treatedcontrolsample(e.g.,Lieberman,Weston,&Pawl,1991).Onlyonestudythatmetourentrycri-teriaalsoreportedonoutcomedataforfatherÐchildattachmentsecurity(Aviezeretal.,2002).Asthiswouldnotallowameaningfulcomparisonofeffectsizesbetweenmotherandfatherattachment,thisstudywasexcluded.Themeta-analysesreportedhereinthereforeonlypertaintomotherÐchildattachment.Alsonoteworthyisthefactthatonlyfourstudieswerebasedonsamplesfrompredomi-nantlyminority-ethniccommunities.438Fearon,Bakermans-Kranenburg,vanIJzendoorn,Lapsley,andRoisman inthreeandfouryearolds:Codingguidelines.Unpublishedscoringmanual,UniversityofVirginia,Charlottesville.*Cicchetti,D.,Rogosch,F.A.,&Toth,S.L.(1998).Mater-naldepressivedisorderandcontextualrisk:Contribu-tionstothedevelopmentofattachmentinsecurityandbehaviorproblemsintoddlerhood.Developmentand,283Ð300.*Cohn,D.A.(1990).Child-motherattachmentofsix-year-oldsandsocialcompetenceatschool.ChildDevelop-,152Ð162.Crick,N.R.,&Grotpeter,J.K.(1995).Relationalaggres-sion,gender,andsocial-psychologicaladjustment.,710Ð722.Cummings,E.M.,Iannotti,R.J.,&ZahnWaxler,C.(1989).Aggressionbetweenpeersinearlychildhood:Individualcontinuityanddevelopmentalchange.,887Ð895.DeKlyen,M.,&Greenberg,M.T.(2008).Attachmentandpsychopathologyinchildhood.InJ.Cassidy&P.R.Shaver(Eds.),Handbookofattachment:Theory,researchandclinicalapplications(2nded.,pp.637Ð665).NewYork:Guilford.*DeMulder,E.K.,Denham,S.,Schmidt,M.,&Mitchell,J.(2000).Q-sortassessmentofattachmentsecurityduringthepreschoolyears:Linksfromhometoschool.opmentalPsychology,274Ð282.*DeVito,C.,&Hopkins,J.(2001).Attachment,parenting,andmaritaldissatisfactionaspredictorsofdisruptivebehaviorinpreschoolers.DevelopmentandPsychopathol-,215Ð231.Dodge,K.A.,&Coie,J.D.(1987).Social-information-pro-cessingfactorsinreactiveandproactiveaggressioninchildrenÕspeergroups.JournalofPersonalityandSocial,1146Ð1158.Duval,S.,&Tweedie,R.(2000a).AnonparametricÔÔtrimandÞllÕÕmethodforaccountingforpublicationbiasinJournaloftheAmericanStatisticalAssocia-,89Ð98.Duval,S.,&Tweedie,R.(2000b).TrimandÞll:Asimplefunnel-plot-basedmethodoftestingandadjustingforpublicationbiasinmeta-analysis.,455Ð*Edwards,E.P.,Eiden,R.D.,&Leonard,K.E.(2006).Behaviorproblemsin18-to36-month-oldchildrenofalcoholicfathers:Securemother-infantattachmentasaprotectivefactor.DevelopmentandPsychopathology*Egeland,B.,&Heister,M.(1995).Thelong-termconse-quencesofinfantday-careandmother-infantattach-ChildDevelopment,474Ð485.Eisenberg,N.,Cumberland,A.,Spinrad,T.L.,Fabes,R.A.,Shepard,S.A.,Reiser,M.,etal.(2001).TherelationsofregulationandemotionalitytochildrenÕsexternaliz-ingandinternalizingproblembehavior.ChildDevelop-,1112Ð1134.Erickson,M.F.,Sroufe,L.A.,&Egeland,B.(1985).Therelationshipbetweenqualityofattachmentandbehav-iorproblemsinpreschoolinahigh-risksample.InI.Bretherton&E.Waters(Eds.),Growingpointsofattachmenttheoryandresearch.MonographsoftheSocietyforResearchinChildDevelopment(1Ð2,SerialNo.209),147Ð166.*Fagot,B.I.,&Leve,L.D.(1998).Teacherratingsofexternalizingbehavioratschoolentryforboysandgirls:Similarearlypredictorsanddifferentcorrelates.JournalofChildPsychologyandPsychiatryandAlliedDis-,555Ð566.Fonagy,P.(2004).Early-lifetraumaandthepsychogene-sisandpreventionofviolence.AnnalsoftheNewYorkAcademyofSciences,181Ð200.Freud,S.(1908).UeberinfantileSexualtheorien.Wissens-chaft:StudienausgabeBandV.Fischer.Gilbody,S.M.,Song,F.,Eastwood,A.J.,&Sutton,A.J.(2000).Thecauses,consequencesanddetectionofpub-licationbiasinpsychiatry.ACTAPsychiatricaScandinav-,241Ð249.*Gilliom,M.,Shaw,D.S.,Beck,J.E.,Schonberg,M.A.,&Lukon,J.L.(2002).Angerregulationindisadvantagedpreschoolboys:Strategies,antecedents,andthedevel-opmentofself-control.DevelopmentalPsychologyGoldberg,S.(1997).Attachmentandchildhoodbehaviorproblemsinnormal,at-riskandclinicalsamples.InL.Atkinson&K.J.Zucker(Eds.),Attachmentandpsycho-(Vol.3,pp.171Ð195).NewYork:Guilford.*Goldberg,S.,Corter,C.,Lojkasek,M.,&Minde,K.(1990).Predictionofbehaviorproblemsin4-year-oldsbornprematurely.DevelopmentandPsychopathology*Goldberg,S.,Gotowiec,A.,&Simmons,R.J.(1995).Infant-motherattachmentandbehaviorproblemsinhealthyandchronicallyillpreschoolers.andPsychopathology,267Ð282.Goldstein,H.,&Healy,M.J.R.(1995).Thegraphicalpre-sentationofacollectionofmeans.JournaloftheRoyalStatisticalSociety,SeriesA(StatisticsinSociety),175Ð*Greenberg,M.T.,Speltz,M.L.,Deklyen,M.,&Endriga,M.C.(1991).Attachmentsecurityinpreschoolerswithandwithoutexternalizingbehaviorproblems:Arepli-DevelopmentandPsychopathology,413Ð430.Guttmann-Steinmetz,S.,&Crowell,J.A.(2006).Attach-mentandexternalizingdisorders:Adevelopmentalpsy-chopathologyperspective.JournaloftheAmericanAcademyofChildandAdolescentPsychiatry,440Ð451.Hedges,L.V.,&Olkin,I.(1985).Statisticalmethodsfor.SanDiego,CA:AcademicPress.*Howes,C.,Matheson,C.C.,&Hamilton,C.E.(1994).Maternal,teacher,andchildcarehistorycorrelatesofchildrenÕsrelationshipswithpeers.ChildDevelopment,264Ð273.*HubbsTait,L.,Osofsky,J.D.,Hann,D.M.,&McDonaldCulp,A.(1994).Predictingbehaviorproblemsandsocialcompetenceinchildrenofadolescentmothers.FamilyRelations:InterdisciplinaryJournalofAppliedFam-ilyStudies,439Ð446.AttachmentandExternalizingBehaviorProblems453 Inamannerthatwastobecomesomewhatcharac-teristicofthetopic,theÞrstreportsoftheseÞndingswerefollowedimmediatelybyanonreplication.Inthesamemonograph,Bates,Maslin,andFrankel(1985)reportedonalongitudinalfollow-upof120infantswhohadpreviouslybeenobservedintheSSPat12monthsandfoundnoassociationbetweenattachmentsecurityandparentreportsofexternal-izingbehaviorproblemsatage3.Anynumberofmethodologicalfactorscouldbeconsideredwheninterpretingtheseearly,apparentlycontradictoryÞndings.Notably,Ericksonetal.Õs(1985)studyexcludedcasesthatwerenotstableintermsofattachmentclassiÞcationsbetween12and18months,whileBatesetal.(1985)onlycollectedattachmentdataat12months.Thereareobviousrea-sonswhythestabilityofattachmentmightbeafactorinitspredictivepower.Furthermore,Sroufeetal.Õsstudywasdrawnfromasubstantiallymoreimpover-ishedpopulationthantheBatesetal.study,whichwaspredominantlymiddleclass.AnumberofauthorshavearguedthatattachmentsecurityshouldbethoughtofasaninteractiveriskfactorthatismoresigniÞcantwhenotherpsychosocialstressorsarepresentinthefamilyecology(Belsky&Fearon,2002;Kobaketal.,2005).Thetwostudiesalsoemployeddifferentoutcomemeasures(teachervs.parentreport),whichinturnmayindexcontex-tualdifferencesintheexpressionofexternalizingbehaviororinthevalidityoftheassessments.Thus,evenintheearliestphaseofresearchintothelongitudinaloutcomesofattachmentsecurityandinsecurity,positiveÞndings,negativeÞndings,andinteractionsemergedinalmostequalmeasure.Asimilarmixofresultsemergedfromlaterstudiesconductedinthe1980sandearly1990s.Withsuchacomplexpatternofstudyoutcomes,narrativereviewstookdivergingpositionsregardingthesta-tusoftheevidenceforanassociationbetweenattachmentandchildrenÕsbehaviorproblems(seeBelsky&Nezworski,1988).TheidentiÞcationofdisorganizedattachment(Main&Solomon,1986,1990)ledtorenewedinter-estinthepotentialforattachmenttopredictrobustlyexternalizingbehaviorproblems(Carlson,1998;LyonsRuth,Alpern,&Repacholi,1993;Moss,Cyr,&DuboisComtois,2004).Theseseeminglyinexplicable,contradictory,andfragmentarybehav-iorsobservedduringtheSSPareconsideredbymanytorepresentrelationalprocessesatspecialriskforpsychopathology,particularlyinthedomainofchildhoodaggression(Liotti,1992;LyonsRuth,Zea-nah,&Benoit,2003;Main&Morgan,1996;Moss,Cyr,etal.,2004).Severalauthorshaveoutlinedhypothesesregardingthemechanismsbywhichdis-organizedattachmentmayleadtoaggression,withconsiderableattentionfocusingonstatesofemo-tionaldysregulationanddissociativeprocessesthatmayblockthepersonÕsawarenessofhisorhervio-lentactions(e.g.,Fonagy,2004;Liotti,1992;Solomon&George,1999).Subsequently,asizablebodyofevidenceemergedthatwasconsistentwiththeviewthatdisorganizedattachmentmaybeassociatedwithincreasedriskforexternalizingbehaviorprob-lemsandaggression.Ameta-analysisof12studiescarriedoutin1999(vanIJzendoorn,Schuengel,&Bakermans-Kranenburg,1999)showedthatthisassociationwasrobust,withameaneffectsizeof=.29(=734).Nevertheless,since1999asigniÞ-cantnumberofnewstudieshavebeenconducted,includingthelargesteverlongitudinalstudyofattachment(withmorethan1,000participants),theNationalInstituteofChildHealthandHumanDevelopment(NICHD)StudyofEarlyChildCareandYouthDevelopment,whichfailedtoÞndstrongevidenceofgreaterexternalizingbehaviorproblemsindisorganizedchildren(Belsky&Fearon,2002;NICHDEarlyChildCareResearchNetwork,2006).Thequestionofwhetherattachmentinsecurityplaysacausalroleinthedevelopmentofexternal-izingpsychopathologyisavitalonefortheÞeld,butthereisclearlylittlecaseforcausalityifthereisnoassociation.Withthesheervolume,range,anddiversityofstudiesthathaveexaminedtheassocia-tionbetweenattachmentsecurityandchildrenÕsexternalizingbehaviorproblems,ithasbecomevir-tuallyimpossibletoprovideaclearnarrativeaccountofthestatusoftheevidenceconcerningthiscriticalissueindevelopmentalscience.Giventhatsamplevariabilityaroundaneffectofzerocanleadtofalsepositives,andsamplingvariabilityaroundapositiveeffectcanleadtofalsenegatives,thequestionofwhethertheexistingevidenceisconsistentwithpositiveassociationiscriticalforafullappreciationofthepredictivesigniÞcanceofattachmentforlaterexternalizingbehaviorproblems.Meta-analysisprovidesastructured,principledmethodologyforresolvingÑwithinlimitsÑtheseessentialscientiÞcquestions.Inthecurrentstudy,weanalyzedover60independentstudiesthathaveconductedassessmentsofattach-mentsecurityandinsecurityusingstandardizedobservationaltoolsandrelatedthemtomeasuresofchildrenÕsexternalizingbehaviorproblems.Inlinewithexpectationsderivedfromtheliterature,wesetouttotestseveralhypotheses,namely,that(a)attachmentinsecurity,inparticularavoidantattachment,wouldbesigniÞcantlyassociatedwithAttachmentandExternalizingBehaviorProblems437 thesameregressionwassigniÞcantwhenAQSstudieswereexcluded(slope=.005,=.03).Itisalsonoteworthythattheoveralleffectsizeforsecu-ritydidnotvaryasafunctionofwhetherornotdisorganizationhadbeencoded(=0.10,=.78).NotethatthislatteranalysisexcludesAQSstudiesbydeÞnition,asthisproceduredoesnotyieldascoreorclassiÞcationfordisorganizedattachment.BecausetheSSPisconsideredtobethegoldstan-dardformeasuringattachmentsecurity,andthesetofSSPstudieswassufÞcientlylarge(=43),wedecidedtoconductcomplementaryanalysesforsecureversusinsecureattachmentonthissubset(seeTable2).Inthishomogeneoussetofstudies,wedidnotÞndadifferenceincombinedeffectsizebetweenclinicalandnonclinicalsamples.Thedifferencebetweensampleswithonlygirlsversusonlyboysormixedsamples,however,wassigniÞcantagain,withall-femalesamplesshowingnoassociationbetweenattachmentsecurityandexternalizingproblems0.06).Also,weconÞrmedthelargercombinedeffectsizeforthoseSSPstudiesthatincludedobser-vationalmeasuresofexternalizing(=0.61)com-paredtotheothertypesofassessment.Interestingly,amongtheSSPstudiestheageatwhichtheassessmentofexternalizingbehaviorwastakenyieldedasigniÞcantregressionweight,withaslopeof.002(=.02),indicatingthattheassociationbetweenattachmentsecurityandexternalizingbecamestrongerwithage.BecausetheSSPsareusuallyconductedwithinasmallagewindowof12Ð18months,thissigniÞcantslopesuggestedthatthepredictionofexternalizingfromattachmentsecuritywasbetterwithlaterÑnotearlierÑassessmentsofexternalizing,whichissurprisinggiventhelongerintervalbetweenassessmentsinthesestudies.IsInsecure-AvoidantAttachmentAssociatedWithMoreExternalizingProblems?In34studiesinvolving=3,675participants,theinsecure-avoidantattachmentclassiÞcationsweredifferentiatedfromtheotherclassiÞcations,andinthissubsetofstudiesthecombinedeffectsizewas=0.12,whichwassigniÞcantbutsmallinmagnitude(seeTable3).Insecure-avoidantlyattachedchildrendisplayedsomewhatmoreexter-nalizingbehaviorsthancomparisons.WithonestudytrimmedandÞlledtheresultingsigniÞcanteffectsizewas=0.11,butthefail-safenumberofstudiesneededtobringtheeffectdownbelowsigniÞcancewasonly24.BecausethisnumberisbelowtheRosenthal(1991)criterionof5+10,thisoutcomeshouldbeconsideredwithcaution.NosigniÞcantmoderatoreffectswerefound.Theeffectdidnotvaryaccordingtowhetherornotdisorgani-zationhadbeencoded.WithintheSSPsubsetof25studies,thecombinedeffectsizeforinsecure-avoidantattachmentwas=0.13,andthemodera-toranalysesonthissubsetconvergedwiththeanalysesonthetotalsetofstudiesforavoidance(seeTable3forthosecontraststhatcouldbetested).IsInsecure-ResistantAttachmentAssociatedWithMoreExternalizingProblems?In35studiesinvolving=3,568participants,theinsecure-resistantattachmentclassiÞcationsweredifferentiatedfromtheotherclassiÞcations.Inthissubsetofstudies,thecombinedeffectsizewasnotsigniÞcant,=0.11(seeTable4),andnosignif-icantmoderatoreffectswerefound.WithintheSSPsubsetof24studies,thecombinedeffectsizeforinsecure-resistantattachmentwasonly=0.05,andthecontrastsonthissubsetshowednosigniÞ-cantmoderators(seeTable4forthosecontraststhatcouldbetested).IsDisorganizedAttachmentAssociatedWithMoreExternalizingProblems?In34studiesincluding=3,778participants,asigniÞcantcombinedeffectsizeof=0.34(seeTable5)wasfoundfortheassociationbetweendis-organizedattachmentandexternalizingbehavior.Asexpected,disorganizedattachmentwasassoci-atedwithahigherriskforexternalizingbehaviorlaterinchildhood.However,eightstudieshadtobetrimmedandÞlled,witharecomputedsigniÞcantcombinedeffectsizeof=0.18(95%CI:0.01,0.34).Thefail-safenumberamountedto=407,whichwasabovetheRosenthalcriterion,suggestingthattheÞle-drawerproblemwasnotresponsiblefortheassociationfoundinthecurrentsetofstudies.Thesetofstudieswasheterogeneous,andmod-eratoranalysesshowedthatgenderwasasigniÞ-cantmoderator(seeTable5).Remarkably,inthesampleswithfemalesonlytheassociationbetweendisorganizedattachmentandexternalizingbehav-iorswassigniÞcantlydifferentfromthesampleswithonlyboysorwithmixedgender,andinfact,therelationwasnegative;thatis,disorganizedattachmentwasassociatedwithlessexternalizingbehavior.However,thecombinedeffectsizeofthesixfemalesamples(=702)wasmodest,0.20.Clinicalstatus,SES,ortypeofassess-ment(forbothSSPandexternalizingbehavior)didnotappeartomoderatetheassociationAttachmentandExternalizingBehaviorProblems445 ReferencesReferenceswithasterisksareincludedinthemeta-analy-ticdataset.Ainsworth,M.S.,Blehar,M.C.,Waters,E.,&Wall,S.Patternsofattachment:Apsychologicalstudyofthestrangesituation(Vol.18).Hillsdale,NJ:Erlbaum.Alink,L.R.A.,VanIJzendoorn,M.H.,Bakermans-Kran-enburg,M.J.,Mesman,J.,Juffer,F.,&Koot,H.M.(2008).Cortisolandexternalizingbehaviorinchildrenandadolescents:Mixedmeta-analyticevidencefortheinverserelationofbasalcortisolandcortisolreactivitywithexternalizingbehavior.DevelopmentalPsychobiol-,427Ð450.*Anan,R.M.,&Barnett,D.(1999).Perceivedsocialsup-portmediatesbetweenpriorattachmentandsubse-quentadjustment:AstudyofurbanAfricanAmericanDevelopmentalPsychology,1210Ð1222.*Aviezer,O.,Sagi,A.,Resnick,G.,&Gini,M.(2002).Schoolcompetenceinyoungadolescence:Linkstoearlyattachmentrelationshipsbeyondconcurrentself-perceivedcompetenceandrepresentationsofrelation-ships.InternationalJournalofBehavioralDevelopment*Bakermans-Kranenburg,M.J.,&vanIJzendoorn,M.H.(2006).GeneÐenvironmentinteractionofthedopamineD4receptor(DRD4)andobservedmaternalinsensitiv-itypredictingexternalizingbehaviorinpreschoolers.DevelopmentalPsychobiology,406Ð409.Bakermans-Kranenburg,M.J.,vanIJzendoorn,M.H.,&Juffer,F.(2003).Lessismore:Meta-analysesofsensitiv-ityandattachmentinterventionsinearlychildhood.PsychologicalBulletin,195Ð215.Bakermans-Kranenburg,M.J.,vanIJzendoorn,M.H.,Mesman,J.,Alink,L.R.A.,&Juffer,F.(2008).Effectsofanattachment-basedinterventionondailycortisolmoderatedbyDRD4:Arandomizedcontroltrialon1-3-year-oldsscreenedforexternalizingbehavior.opment&Psychopathology,805Ð820.Baldwin,J.M.(1895).Mentaldevelopmentofthechildandtherace:Methodsandprocesses.NewYork:Macmillan.*Bates,J.E.,&Bayles,K.(1988).Attachmentandthedevel-opmentofbehaviorproblems.InJ.Belsky&T.Nezwor-ski(Eds.),Clinicalimplicationsofattachment.Child(Vol.7,pp.253Ð299).Hillsdale,NJ:Erlbaum.Bates,J.,Maslin,C.,&Frankel,K.(1985).Attachmentsecurity,mother-childinteractions,andtemperamentaspredictorsofbehaviorproblemratingsatagethreeyears.InI.Bretherton&E.Waters(Eds.),Growingpointsinattachmenttheoryandresearch.oftheSocietyforResearchinChildDevelopmentSerialNo.209),167Ð193.Belsky,J.,&Fearon,R.M.P.(2002).Infant-motherattach-mentsecurity,contextualrisk,andearlydevelopment:Amoderationalanalysis.DevelopmentandPsychopathol-,293Ð310.Belsky,J.,Fearon,R.M.P.,&Bell,B.(2007).Parenting,attentionandexternalizingproblems:Testingmedia-tionlongitudinally,repeatedlyandreciprocally.ofChildPsychologyandPsychiatry,1233Ð1242.Belsky,J.,&Isabella,R.(1988).Maternal,infant,andsocial-contextualdeterminantsofattachmentsecurity.InJ.Belsky&T.Nezworski(Eds.),Clinicalimplicationsofattachment(pp.41Ð94).Hillsdale,NJ:Erlbaum.Belsky,J.,&Nezworski,T.(Eds.).(1988).Clinicalimplica-tionsofattachment.Hillsdale,NJ:Erlbaum.*Booth,C.L.,RoseKrasnor,L.,McKinnon,J.A.,&Rubin,K.H.(1994).Predictingsocialadjustmentinmiddlechildhood:Theroleofpreschoolattachmentsecurityandmaternalstyle.SocialDevelopment,189Ð204.*Booth,C.L.,RoseKrasnor,L.,&Rubin,K.H.(1991).RelatingpreschoolersÕsocialcompetenceandtheirmothersÕparentingbehaviorstoearlyattachmentsecu-rityandhigh-riskstatus.JournalofSocialandPersonal,363Ð382.Borenstein,M.,Rothstein,D.,&Cohen,J.(2005).Compre-hensivemeta-analysis:Acomputerprogramforresearchsynthesis[Computersoftware].EnglewoodCliffs,NJ:Biostat.Bowlby,J.(1944).Fourty-fourjuveniletheives:Theirchar-actersandhomelife.InternationalJournalofPsycho-,19Ð52.Bowlby,J.(1969).Attachmentandloss:Vol.1.AttachmentLondon:HogarthPressandtheInstituteofPsycho-Bradley,R.H.,&Corwyn,R.F.(2002).Socioeconomicstatusandchilddevelopment.AnnualReviewofPsychol-,371Ð399.Bretherton,I.(1995).Internalworkingmodels:Cognitiveandaffectiveaspectsofattachmentrepresentations.InD.Cicchetti&S.Toth(Eds.),4thRochesterSymposiumonDevelopmentalPsychopathologyonÔÔEmotion,Cognition,andRepresentationÕÕ(pp.231Ð259).Hillsdale,NJ:Erlbaum.Bretherton,I.(1997).BowlbyÕslegacytodevelopmentalChildPsychiatryandHumanDevelopment,33Ð43.*Burgess,K.B.,Marshall,P.J.,Rubin,K.H.,&Fox,N.A.(2003).Infantattachmentandtemperamentaspredic-torsofsubsequentexternalizingproblemsandcardiacJournalofChildPsychologyandPsychiatry,819Ð831.Cairns,R.B.,&Cairns,B.D.(1984).Predictingaggressivepatternsingirlsandboys:Adevelopmentalstudy.AggressiveBehavior,227Ð242.*Carlson,E.A.(1998).Aprospectivelongitudinalstudyofattachmentdisorganization,1107Ð1128.Cassidy,J.(1988).Child-motherattachmentandtheselfinsix-year-olds.ChildDevelopment,121Ð134.Cassidy,J.(1994).Emotionregulation:Inßuencesofattachmentrelationships.InN.A.Fox(Ed.),Thedevel-opmentofattachmentregulation(pp.228Ð249).oftheSocietyforResearchinChildDevelopmentNo.240).Cassidy,J.,Marvin,R.S.,&TheMacArthurWorkingGrouponAttachment.(1989).Attachmentorganization452Fearon,Bakermans-Kranenburg,vanIJzendoorn,Lapsley,andRoisman signiÞcant,=0.35and=0.36,respectively,andsigniÞcantlydifferentfromthenulleffectinthegirls-onlysamples.Withinthemixedsamples,thepercentageofmalesinthestudytendedtobepositivelyrelatedtothemagnitudeoftheattach-ment-externalizingproblemseffectsize(slope==.11).Clinicalsamplesalsoshowedasignif-icantlylargercombinedeffectsize(=0.49)thannonclinicalsamples(=0.26).Thismoderatoreffectwasnotdependentongenderdifferencesasonlytwooftheclinicalsamplesconsistedofboysonly(Greenberg,Speltz,Deklyen,&Endriga,1991;Speltzetal.,1999).Contrarytoexpectations,SESwasnotasigniÞcantmoderator(seeTable2).Thetypeofassessmentofexternalizingbehav-iorsandofattachmentappearedtomakeadiffer-encetotheeffectsizes.Whenexternalizingbehaviorwasobserveddirectly(sevenstudies)orwasindexedbyaclinicaldiagnosis(sixstudies),thecombinedeffectsizeswerelargerthanincaseswhereaparentorteacherratedthelevelofexter-nalizingbehaviors.Thegroupofstudiesbasedonobservationsofexternalizingbehaviorwashomo-geneous.Itscombinedeffectsizeof=0.58wasthereforeanadequateestimateoftheaverageeffect,whichamountedtoaquitestrongassociationbetweenattachmentsecurityandexternalizingbehavior.ThewayinwhichattachmentsecuritywasassessedalsomadeasigniÞcantdifference.Inparticular,studiesconductedwiththeAQSshowedthelargesteffectsizes(=0.70),whereasstudiesusingtheSSPyieldedthelowestcombinedeffectsize(=0.18),thoughthislattereffectwasnever-thelesssigniÞcant.Becausethedifferentattachmentmeasuresaretypicallyconductedatdifferentages,wealsoconductedameta-regressionanalysiswithageattheassessmentofattachmentasapredictor.Asexpected,theregressionwassigniÞcant(slope==.01).Notablythough,theeffectofagewasnotsigniÞcantwithintheSSP,AQS,orCassidyandMarvinstudies,suggestingthatageandattachmentmeasurewereconfounded.Furthermore,theregressionwithagedidnotappeartoresultsolelyfromthelargereffectsassociatedwiththeAQS,as Table3AvoidantAttachmentandExternalizingBehavior:TotalSetandSetofSSPStudieskNdinterval95%HomogeneityTotalset343,6750.12**0.03,0.2140.540.93.33Nonclinical283,2730.10*0.01,0.2035.62Clinical64020.22*0.01,0.433.981.24.54Boys98900.19*0.03,0.365.32Girls77630.130.09,0.348.45Mixed182,0220.080.04,0.2025.532.42.12Low51,0010.20,0.176.81high292,6740.15**0.06,0.2531.31SSP253,0540.13**0.03,0.2326.01CassMarvin75230.130.11,0.3712.78*Other2980.69,0.280.01SSPstudiesAllSSPstudies253,0540.13**0.03,0.2326.010.03.98Boys56470.120.08,0.331.09Girls67530.160.06,0.381.77Mixed141,6540.15*0.01,0.3323.03*2.95.10Low51,0010.20,0.176.81high202,0530.18**0.07,0.2916.25.SES=socioeconomicstatus;SSP=StrangeSituationprocedure;CassMarvin=CassidyandMarvinMacArthurPreschoolAttachmentCodingSystem.Subgroupswith4excludedfromcontrast..05.**.01.444Fearon,Bakermans-Kranenburg,vanIJzendoorn,Lapsley,andRoisman childrenrespondtosourcesofthreatandchallenge,andtheextenttowhichchildrenareabletodrawonparentalsupportandcomfortasameansofcop-ing(Kobak,Cassidy,LyonsRuth,&Ziv,2005).Securechildren,itismaintained,havehadrepeatedexperiencesofacaregiverwhoisresponsivewhensupportandproximityareneededandexpectthecaregiver(s)tobeavailableandcomfortingwhencalledupon.Incontrast,childrenwithinsecureattachmentrelationshipsmayhavehadexperiencesinwhichbidsforproximityhavebeendiscouraged,rejected,orinconsistentlyrespondedtoandrelymoreheavilyonsecondarycopingprocessestodealwithstressandchallenge.Developmentalcontinuitiesbetweentheorgani-zationoftheattachmentrelationshipandfunc-tioningbeyondit(intimeorspace)havegenerallybeenconceptualizedwithreferencetotheinternalworkingmodelsconstruct.ThisimportantconceptualheuristicisthoughtofasasetoforganizedcognitiveÐaffectivepsychologicalstructuresthatorganizethinking,feeling,andbehaviorvis-a-vistheattachmentÞgureasapotentialhavenofsafetyandcomfortintimesofstress(Bretherton,1995;Main,Kaplan,&Cassidy,1985).Thesemodelsarethoughttobecomegener-alizedovertimeandinßuencefunctioninginwiderinterpersonalrelationshipsacrossthelifespanandformthebasisofageneralizedsenseoftheselfasworthyofloveandcareandothersasavailableandresponsive(Cassidy,1988;Sroufe,Egeland,Carlson,&Collins,2005a).Inadditiontothisprimaryexplanatoryconstruct,severalotherfactors(possiblyrelatedtointernalworkingmod-els)havebeendiscussedaspotentialmediatorsbetweenasecureattachmentrelationshipandloweredriskformentalhealthproblemsgenerallyandexternalizingproblemsmorespeciÞcally,suchas(a)adevelopingsenseofself-conÞdencethroughrepeatedexperiencesofsupportandcom-fortandthrougheffectiveexplorationoftheenvi-ronment(Goldberg,1997),(b)generalizedpositivesocialexpectations(asopposedtomistrustandperceivedhostility;seeDodge&Coie,1987),(c)thesocializationofmoralemotionsandvalueswithinasecureattachmentrelationship(Ko-chanska,1997;vanIJzendoorn,1997),(d)model-ingofprosocialbehaviorbyasensitivecaregiver(Guttmann-Steinmetz&Crowell,2006),(e)conti-nuityinthequalityandsupportivenessofongo-ingparentalcare(Lambetal.,1984),and(f)thecapacityforeffectiveemotionregulation(e.g.,Cassidy,1994).Otherpossibilitiesexistthathavebeengivenlessattention,suchasthesocialmod-ulationofbiologicalsystemsmediatingstressandarousalregulation(e.g.,Suomi,2003;Weaveretal.,2004).AlthoughBowlbyÕsworkwashighlyinßuential,thedevelopmentofastandardizedprocedureforthesystematicstudyofattachmentbehavior,asobservedinanaturalisticsetting,wasamajorfur-therstepforwardintheestablishmentofanempiri-calknowledgebaseconcerningthedevelopmentalsigniÞcanceofattachment(Ainsworth,Blehar,Waters,&Wall,1978;Sroufe,1983).Ainsworthetal.Õs(1978)StrangeSituationprocedure(SSP)hasbecomeoneofthemostwidelyusedÑifnotthemostwidelyusedÑstandardizedlabassessmentofearlychildhoodbehaviorbasedondirectobserva-tionandrepresentsaparadigmexampleofhowtosystematicallystudynaturallyoccurringbehaviorinquasi-naturalisticcontexts.TheidentiÞcationofindividualdifferencesinpatternsofreunionbehav-iorfollowingseparationintheSSPtriggeredapro-gramofresearchstudiesaimedatuncoveringtheirdevelopmentalantecedentsandsequelae(Belsky&Isabella,1988;Schneider-Rosen&Rothbaum,1993).Despitetheaccumulationofanimpressivevolumeofdataovertheyears,thepicturethatunfoldedregardingthedevelopmentalconsequencesofattachmenthasprovedcomplexandoftencontra-dictory,particularlyinthedomainofmentalhealthandpsychopathology(Goldberg,1997).Oneoftheearliestandmostinßuentiallongitu-dinalstudiesofthepsychosocialoutcomesofchil-drenobservedintheSSPininfancywaslaunchedinMinnesotabyByronEgeland,AlanSroufe,andcolleagues(seeSroufe,Egeland,Carlson,&Collins,2005a,2005b).Erickson,Sroufe,andEgeland(1985)followedtheirrelativelylargesampleofhigh-riskinfantsfrom12monthstopreschoolandcollectedextensiveassessmentsofchildrenÕsbehaviorusingobserverandteacherratingsofsocialcompetence,egocontrol,peerconÞdence,andexternalizingbehaviorproblemsintheschoolsetting.Mostimportantlyforthecurrentpurposes,securechil-drenscoredlowerthaninsecurechildrenonassess-mentsofbehaviorproblemsandavoidantchildrenstoodoutasbeingparticularlyatrisk,aÞndingechoedinseverallaterstudies(Burgess,Marshall,Rubin,&Fox,2003;Goldberg,Gotowiec,&Sim-mons,1995;Munson,McMahon,&Spieker,2001).Interestingly,alaterreportfromtheMinnesotastudyatGrades1Ð3(Renken,Egeland,Marvinney,Mangelsdorf,etal.,1989)foundassociationsbetweenattachmentinsecurityandexternalizingproblemsinboysbutnotgirls(seealsoLewis,Feiring,McGuffog,&Jaskir,1984).436Fearon,Bakermans-Kranenburg,vanIJzendoorn,Lapsley,andRoisman thanattachmentinsecuritygenerally(vanIJzendo-ornetal.,1999).However,theÞndingsofthecur-rentmeta-analysiswereonlypartiallysupportiveofthespecialstatussometimesaccordedtodisorga-nizedattachmentasaprecursorofchildrenÕsexter-nalizingproblems.Theglobaleffectsizeforattachmentsecurity(=0.31)wasverysimilartothatfordisorganization(=0.34).Itshouldbenotedthattheseeffectsizesdonotreßecttheout-comesofthesamesetofstudiessodirectcompari-sonsaredifÞcult.However,inananalysisof19studieswheredatafromallfourattachmentgroupswereavailabletheresultscorrespondedtotheaforementionedglobaleffects.Theeffectsforsecu-rityanddisorganizationweresigniÞcantandofsimilarmagnitude(=0.27and=0.29,respec-tively),whiletheeffectsforavoidanceandresis-tanceweresmallandinthecaseofavoidancenonsigniÞcant(=0.06and=0.11,respectively).The85%CIsforsecurityanddisorganizationdidnotoverlapwiththatforavoidance,suggestingsigniÞcantlystrongereffectsforinsecurityanddis-organizationthanforavoidance.Theeffectforresistancedidnotdifferfromanyoftheothercon-trasts,inpartbecausethiseffectshowedmarkedheterogeneity.Furthermore,inthelargersetofstudies,resistantattachmentwasnotsigniÞcantlyassociatedwithexternalizingproblemsandshowedaneffectsizenumericallysimilartothatforavoid-ance.Whenonlystudiesthatprovideddirectpair-wisecomparisonsbetweengroupswereconsidered,securechildrenscoredloweronmea-suresofexternalizingproblemsthandisorganized=0.27),avoidant(=0.12),andresistant=0.19)children,andwhiletheCIsfortheseeffectsoverlapped,disorganizedchildrendemon-stratednumericallylargereffectsthantheotherinsecurecategories.Althoughtheeffectsizesfordisorganizationandsecuritycomparefavorablywithotherprominentriskfactorsforexternalizingbehavior,theyalsoleaveroomforthepossibilitythattheassociationmaybemoderatedbyotherfactors.Itishighlynoteworthythatfewifanyofthestudiesreviewedinthisstudydirectlyaddressedquestionsofmech-anism.Arguably,untilmediatingmechanismsarebetterunderstood,wemaystruggletoÞndtherele-vantmoderators.Giventhatanassociationclearlyexists,thereisanobviousneedforstrongtheory-drivenstudiesthataddressmediatingprocesses,particularlythosedrawingonmethodsderivedfromotherÞelds.Somewell-studiedriskprocessesworthyofconsideration(inadditiontothetradi-tionalinternalworkingmodelsconstruct)includeimpulsivity,negativityemotionality,affectregula-tion,hostileattributionalbiases,andphysiologicalhypo-arousal(Belsky,Fearon,&Bell,2007;Dodge&Coie,1987;Eisenbergetal.,2001;Raine,Ven-ables,&Mednick,1997).Riskfactorssuchasthese,situatedatthebiological,cognitive,oraffectivelevelmaybeconsideredproximaldeterminantsofexternalizingbehavior,withthequalityoftheattachmentrelationshipwithaprimarycaregiverconceptualizedasamoredistaldeterminant.Acru-cialquestioninthatcontexthaslongbeen,andremains,theextenttowhichlongitudinalcontinu-itiesintheeffectsofattachmentrepresenttheongo-ingsupportivefunctionprovidedbytheattachmentrelationshipasopposedtotheearlyeffectsofattachmentexperiencesontheemergenceofstablepsychologicalstructures,suchasinternalworkingmodels.ItisnotablethatBelskyandFea-ronÕs(2002)analysisoftheNICHDdataatage1Ð3yearssuggestedthattheeffectsofattachmenttendedtopersistprimarilywhentherewasconti-nuityinthequalityofmaternalcare.Giventhecen-tralityofthisissueandthecogentargumentsmadebyseveralauthorsonthissubject(e.g.,Sameroff,2000;Sroufeetal.,2005a,2005b),itisunfortunatethatfewstudieshaveattemptedtoaddressthis.Relatedtothis,fewstudieshavedirectlyconsid-eredthepossibilitythatcertainparentingcharacter-isticsmaybecommondeterminantsofbothattachmentinsecurityandexternalizingbehaviorproblems.Thus,theextenttowhichattachmentprocessespersecanbeshowntomakeaspeciÞcandcausalcontributiontochildrenÕsexternalizingproblems,eitherindependentlyofparentingorasacausalmediatorofitseffects,remainstobeseen.Thereisalsoaclearneedforabetterunder-standingofcausation.Inthatregard,theabsenceofstudiesthathaverepeatedlyassessedbothattachmentandoutcomeisasigniÞcantbarrier.Longitudinalstudiesthatemploycross-laggedpaneldesignscouldprovideeleganttestsofcau-sationbyfocusingontemporalordering.How-ever,inordertodothisthereisanurgentneedtoestablishrobustmeasurementprotocolsthatallowformeaningfulrepeatedassessmentsofattachmentandhencethedocumentationofchange.Ofcourse,causalhypothesescanalsobepowerfullyaddressedbyinterventionstudies(Bakermans-Kranenburg,vanIJzendoorn,Mes-man,Alink,&Juffer,2008)andinthefuturethesemaybecriticalfordeterminingtheroleofattachmentinchildrenÕsbehaviorproblems,aswellasthatofthecausalmediatorsandmodera-torsofitseffects.AttachmentandExternalizingBehaviorProblems451 oftheÞle-drawerproblemwasevidentthroughthetrim-and-Þllapproach,showingthatonlyÞvestudieshadtobetrimmedandÞlled,witharesult-ingsigniÞcantcombinedeffectsizeof=0.27(95%CI:0.18,0.36).Asthetotalsetofstudiesforsecureversusinse-cureattachmentwasheterogeneous(seeTable3),welookedforsigniÞcantmoderatorsthatmightaccountforbetween-studyvariabilityinoutcome.GenderappearedtobeasigniÞcantmoderator(seeTable2).Insampleswithonlygirls,thecombinedeffectsizewas0.03(),whereasinthesam-pleswithonlyboysandinthemixedsamples(i.e.,withboysandgirls)thecombinedeffectsizeswere Table2InsecureAttachmentandExternalizingBehavior:TotalSetandSetofSSPStudieskNdinterval95%HomogeneityTotalset695,9470.31**0.23,0.40135.18**5.48.02Nonclinical564,8120.26**0.17,0.3589.06**Clinical131,1350.49**0.32,0.6629.98**8.49.01Boys141,2100.35**0.17,0.5432.41**Girls129070.16,0.1113.93Mixed433,8300.36**0.26,0.4669.79**0.05.82Low141,8010.25**0.15,0.3515.49high554,1460.31**0.21,0.41119.69**14.49.01SSP434,4880.18**0.12,0.2457.31AQS74640.70**0.51,0.905.57CassMarvin127080.37**0.16,0.5732.18**Other72870.39**0.14,0.6410.93ObserverextMother444,1290.22**0.12,0.3272.37**Teacher109220.30**0.17,0.4415.72Observed72110.58**0.30,0.864.12Other64250.62**0.35,0.8915.83**Combined22600.45**0.20,0.712.21SSPstudiesAllSSPstudies434,4880.18**0.12,0.2457.310.03.86Nonclinical363,8990.22**0.13,0.3155.33*Clinical75890.21*0.04,0.381.8313.08.01Boys68360.18*0.04,0.324.27Girls67530.21,0.098.52Mixed312,8990.24**0.17,0.3231.440.90.34Low121,6350.22**0.12,0.3211.47high312,8530.19**0.09,0.2944.87*ObserverextMother313,4990.13**0.06,0.2039.35Teacher44630.22*0.03,0.400.79Observed51710.61**0.29,0.920.80Other1950.320.10,0.73Combined22600.45**0.20,0.712.21.SES=socioeconomicstatus;SSP=StrangeSituationprocedure;AQS=AttachmentQ-Sort;CassMarvin=CassidyandMacArthurPreschoolAttachmentCodingSystem.Subgroupwith4excludedfromcontrast..05.**.01.AttachmentandExternalizingBehaviorProblems443 associationsbygender.Afurtherimportantpossi-bilitytoconsider,whichmaynotbeindependentofthepossibilitiesdescribedearlier,isthatinsecurityingirlsismoreassociatedwithinternalizingprob-lemsratherthanexternalizingproblems(seeDeKlyen&Greenberg,2008).InlightofclaimsmadebymanyintheÞeldthatattachmenthaspersistenteffectsonfuturedevelop-ment,wealsoexpectedtheassociationbetweenattachmentandbehaviorproblemstobeevidentinstudieswithrelativelylong-termfollow-upperiods.Consistentwiththisexpectation,wefoundthattheageofthechildatthepointwhenexternalizingproblemswereassessedwasnotsigniÞcantlyasso-ciatedwiththemagnitudeofstudyeffectsizes.Evenmoreintriguingly,amongstudiesemployingtheSSP(=43studies),appliedwithinasmalltimewindow,effectsizesappearedtoincreasesigniÞcantlyovertime.Inadditiontotheseapriorihypotheses,methodofmeasurementprovedasigniÞcantfactorthatdis-tinguishedstudieswithrelativelysmallandlargeeffectsintwodistinctways.First,differentassess-mentsofattachmentappearedtobeassociatedwithreliabledifferencesinthemagnitudeoftheattach-ment-behaviorproblemsassociation.Broadlyspeak-ing,theSSPproducedsmallereffectsizesthantheotherattachmentassessmentsandtheAQSpro-ducedcomparativelylargereffectsizes.Critically,however,itwasnotpossibletodisentangleentirelytheroleplayedbymeasurementtype(e.g.,SSPvs.otherassessments)fromtheageatwhichtheassess-mentwasconducted(infancyorlater),asthesewereessentiallyconfounded.Notably,agewasnotasig-niÞcantregressorwithintheSSPstudiesorwithinthenon-SSPstudies,soasubstantialportionoftheoveralleffectcameexclusivelyfromthedifferencebetweentheSSPstudiesontheonehandandtheotherassessmenttypesontheother.AlthoughtheAQSevidencedthestrongesteffectsizesbysomemargin(=0.70),themetaregressionofageoneffectsizeremainedsigniÞcantevenwhentheAQSwasremovedfromtheanalysis.Thus,whileitwasclearthatsecurityassessedbytheAQSwasrathermorestronglyassociatedwithbehaviorproblemsthanthatderivedfromtheSSP,laterassessmentsofattachmentbeyondtheSSPalsoappearedtoyieldstrongereffectsizes.Onthatbasis,itwasnotpossi-bletodeterminewhetherthecriticalfactorwassomemethodologicalorcodingvariationthatdistin-guishestheSSPfromtheotherattachmentassess-ments,orsomefactormorecloselyrelatedtodevelopmentitself.Itiscertainlyplausiblethatimportantdevelopmentalchangestakeplacearoundthebeginningofthe3rdyearthatamplifythelinkbetweenattachmentandexternalizingproblems.Thesecondcriticalmeasurementfactorthatemergedfromthismeta-analysisconcernedthemeasurementofoutcome.Studiesthatassessedexternalizingbehaviorproblemsviadirectobserva-tionidentiÞedreliablylargereffectsizes(=0.58)thanthosethatreliedonquestionnairesfrompar-entsorteachers(=0.22and=0.30,respec-tively).ThisisapotentiallyimportantÞndingfromavarietyofperspectives.First,thevastmajorityofthestudiesthatinvestigatedexternalizingbehaviorinrelationtoattachmentreviewedheredidnotuseanyobjectivesourceofinformationconcerningtheoutcome.Notonlydoesthisstrategylimitthecon-clusionsthatcanbedrawn(seeKagan,2007),itcouldalsopartiallyexplainthemixedeffectsacrossattachment-outcomestudiesasawhole.Itisnote-worthythattheeffectsizescitedearlierareappar-entlypositivelycorrelatedwiththedegreeofobjectivityoftheobserver.Nevertheless,althoughtheobjectivityofmea-surementisonepotentialaccountofthestrongereffectsizesassociatedwithdirectobservation,itcouldalsobethatdirectobservationrevealsquali-tativelydistinctfeaturesofbehaviororqualitativelydistinctcontextsinwhichbehaviortakesplace.Notably,studiesthatusedobservationalmethodstypicallyfocusedonthepeersetting(e.g.,Booth,RoseKrasnor,&Rubin,1991;Matasetal.,1978;Suess,Grossmann,&Sroufe,1992)andhencemayhavebeentappingbehaviormorecloselyconnectedwithsocialcompetencethanthosethatreliedonparentorteacherquestionnaires.However,theseeffectsareunderstood,thelimitedagreementbetweenobjectiveobservers,teachers,andparentsconcerningchildrenÕsbehaviorproblemscreatesastrongimperativetousemultiplesourcesofout-comedatainfuturestudiesofattachmentsecurityanditssequelae.Atthesametime,moredatasourcespersewillnotsolvethisproblem.Whatisarguablyneededisabetterunderstandingoftheprecisecircumstancesandmechanismsunderwhichaggressionandotherantisocialactsmaybetriggeredandthecontributionthatattachmentpro-cessesmaketothis.ItishardtoimaginethatthisdegreeofmechanisticspeciÞcitycouldbeachievedwithoutgreaterrelianceonÞeldstudiesthatinvolvedirectandextensiveobservationsinarangeofrelevantsocialsettings.Argumentsputforwardintheliterature,andresultsofanearliermeta-analysis,ledustoexpectthatattachmentdisorganizationwouldbeastron-gerpredictorofexternalizingbehaviorproblems450Fearon,Bakermans-Kranenburg,vanIJzendoorn,Lapsley,andRoisman TheSigniÞcanceofInsecureAttachmentandDisorganizationinthe DevelopmentofChildrenÕsExternalizingBehavior:AMeta-AnalyticStudy R.PascoFearon UniversityofReading MarianJ.Bakermans-Kranenburg andMarinusH.vanIJzendoorn UniversityofLeiden Anne-MarieLapsley Barnet,EnÞeld&HaringeyMentalHealth NHSTrust GlennI.Roisman UniversityofIllinoisatUrbana-Champaign Thisstudyaddressestheextenttowhichinsecureanddisorganizedattachmentsincreaseriskforexternaliz- ingproblemsusingmeta-analysis.From69samples( N =5,947),theassociationbetweeninsecurityandexter- nalizingproblemswassigniÞcant, d =0.31(95%CI:0.23,0.40).Largereffectswerefoundforboys( d =0.35), clinicalsamples( d =0.49),andfromobservation-basedoutcomeassessments( d =0.58).Largereffectswere foundforattachmentassessmentsotherthantheStrangeSituation.Overall,disorganizedchildrenappeared atelevatedrisk( d =0.34,95%CI:0.18,0.50),withweakereffectsforavoidance( d =0.12,95%CI:0.03,0.21) andresistance( d =0.11,95%CI: ) 0.04,0.26).TheresultsarediscussedintermsofthepotentialsigniÞcance ofattachmentformentalhealth. AlthoughthesigniÞcanceoftheparentÐchildrela- tionshipwasrecognizedbyscientistsandclinicians sincetheearliestdaysofformalpsychological inquiry(e.g.,Baldwin,1895;Freud,1908;James, 1890),twomajoradvancesoccurredinthe1960s and1970sthatcreatedlastinglegaciesforthestudy ofhumandevelopment.JohnBowlbyÕs(1969)the- oryofparentÐchildattachmentwasrevolutionary inthewayitintegratedevolutionary,biological, developmental,andcognitiveconceptsintoauni- Þedaccountofhumanattachmentbehavior.This remarkableachievementpavedthewayforthe scientiÞcstudyofattachment,largelybecauseit createdaconceptualframeworkfordeveloping testablehypothesesaboutcausalinßuences,devel- opmentalprocesses,andexpectedlong-termconse- quencesofattachmentformentalhealth (Bretherton,1997).Criticalamongthesenovelcon- tributionsweretheclearcharacterizationofthe proximalbehavioralfunctionsassociatedwith attachmentandtheirinterplaywithotherbiologi- callysigniÞcantbehavioralsystems,theuseof comparativeevidenceascrucialsourcesoftheory- development,andtheconceptofaninternal workingmodelasaframeworkforunderstanding continuitiesinattachmentbehavioracrosscontext andovertime.Thenotionthatthequalityor organizationofattachmentbehaviorinearly infancyorchildhoodmighthaveimplicationsfor latersocioemotionaldevelopmentandmental healthisarguablyoneofattachmenttheoryÕsmost well-knownandcontestedpredictions(Lamb, Thompson,Gardner,Charnov,&Estes,1984; Rutter,1995).NowhereisthisissuemoresigniÞcant thaninthedomainofaggressionandexternalizing behaviorproblems,wherethesocialcostsare substantial(Loeber&Hay,1997). ResearchersworkingintheattachmentÞeld, followingBowlbyandothers,haveconsidereda numberofmechanismsthatmightexplainwhy attachmentexperiencesinearlylifemightbeasso- ciatedwithlateradaptationandmentalhealth. Severaltheoristshavesuggestedthattherole ofattachmentmaycenteronthewayinwhich SupportfromtheNetherlandsOrganizationforScientiÞc ResearchtoMarinusvanIJzendoorn(NWOSPINOZAaward) andtoMarianBakermans-Kranenburg(NWOVIDIgrant)is gratefullyacknowledged.Thestudywasalsosupportedinpart byagrantfromtheNationalScienceFoundationtoGlennI. Roisman(BCS-0720538). Correspondenceconcerningthisarticleshouldbeaddressedto R.PascoFearon,SchoolofPsychologyandClinicalLanguage Sciences,UniversityofReading,3EarleyGate,Whiteknights, ReadingRD66AL,UnitedKingdom.Electronicmailmaybesent top.fearon@reading.ac.uk. ChildDevelopment,March/April2010,Volume81,Number2,Pages435Ð456  2010,CopyrighttheAuthor(s) JournalCompilation  2010,SocietyforResearchinChildDevelopment,Inc. Allrightsreserved.0009-3920/2010/8102-0002 studiesthatareassociatedwithvariationsinprocedures,measures,andsettingsthatgobeyondsubject-levelsamplingerrorandthuspointtodifferentstudypopulations(Lipsey&Wilson,2001).TotestthehomogeneityoftheoverallandspeciÞcsetsofeffectsizes,wecom-statistics(Borensteinetal.,2005).Inaddition,wecomputed95%conÞdenceintervals(CIs)aroundthepointestimateofeachsetofeffectsizes.Whenthesetwashomogeneous,CIswerebasedonÞxedestimates.Incaseswheretherewasheterogeneityacrossstudies,webasedCIsonrandomestimates.statisticsandueswerealsocomputedtoassessdifferencesbetweencombinedeffectsizesforspeciÞcsubsetsofstudiesgroupedbymoderators.Again,Þxedeffectsmodeltestswereusedinthecaseofhomogeneoussetsofoutcomes,andmoreconser-vativerandomeffectsmodeltestswereusedinthecaseofheterogeneousoutcomes.Inthepres-entstudy,randommodelsweretestedunlessotherwisespeciÞed.Contrastswereonlytestedwhenatleasttwoofthesubsetsconsistedofatleastfourstudies(Bakermans-Kranenburg,vanIJzendoorn,&Juffer,2003).Whenthechildrenintwosetsofstudies(par-tially)overlapped(e.g.,somestudiesreportedbothaggressionandexternalizing,andwewantedtocomparethecombinedeffectsforaggressionandexternalizing),itwasimpossibletocomparedirectlyeffectsizesacrossthesesets.Wecomputed85%CIsforthepointestimatesofthecombinedeffectsizesinthetwosets:Non-overlapping85%CIsindicateasigniÞcantdiffer-encebetweencombinedeffectsizes(Bakermans-Kranenburgetal.,2003).Thisapproachofcom-paring85%CIsservedasaconservativesigniÞ-cancetest(Goldstein&Healy,1995).WeusedtheÔÔtrim-and-ÞllÕÕmethod(Duval&Tweedie,2000a,2000b)tocalculatetheeffectofpotentialdatacensoring(orpublicationbias)ontheout-comeofthemeta-analyses.Usingthismethod,afunnelplotisconstructedofeachstudyÕseffectsizeagainstthesamplesizeorthestandarderror(usuallyplottedas1,orprecision).Itisexpectedthatthisplothastheshapeofafunnelbecausestudieswithsmallersamplesizes(largerstandarderrors)haveincreasinglylargervariationinestimatesoftheireffectsizeasrandomvaria-tionbecomesincreasinglyinßuential,whereasstudieswithlargersamplesizeshavesmallervariationineffectsizes(Duval&Tweedie,2000b;Sutton,Duval,Tweedie,Abrams,&Jones,2000).Theplotsshouldonlybeshapedlikeafunnelifnodatacensoringispresent.However,assmallerornonsigniÞcantstudiesarelesslikelytobepublished(theÔÔÞle-drawerÕÕproblem;Mullen,1989),studiesinthebottomleft-handcorneroftheplotareoftenomitted(Suttonetal.,2000).Inourmeta-analyses,theright-moststudiescon-sideredtobesymmetricallyunmatchedweretrimmed.ThetrimmedstudiescanthenbereplacedandtheirmissingcounterpartsimputedorÔÔÞlledÕÕasmirrorimagesofthetrimmedout-comes.ThisthenallowsforthecomputationofanadjustedoveralleffectsizeandCI(Gilbody,Song,Eastwood,&Sutton,2000;Suttonetal.,Foreachmeta-analysis,wealsocalculatedthenumberofstudieswithaveragesamplesizeandnonsigniÞcantoutcomethatwouldberequiredtobringthecombinedeffectsizeofthemeta-analysistoanonsigniÞcantlevel(fail-safenumber;Mullen,1989).Rosenthal(1991,p.106)suggestedthatafail-safenumberofmorethan5+10(=numberofstudiesincluded)maybeconsideredageneralcriterionforrobustness.Foreachstudy,FisherÕsscoreswerecomputedaswell-distributedequivalentsfortheeffectsizeandthescoreswerestandardizedtotestforout-liers.Nooutliers(standardizedvaluessmaller3.29orlargerthan3.29;Tabachnik&Fidell,2001)werefoundforstudyeffectsizes.ResultsIsInsecureAttachmentAssociatedWithMoreExternalizingProblems?TheÞrstsetofmeta-analysesconcernedthedif-ferenceinexternalizingbehaviorsbetweenchildrenratedassecureversuschildrenclassiÞedasinse-cure.In69studiesincluding=5,947participants,theassociationbetweenattachmentsecurityandexternalizingbehaviorswasreported.Anystudyassessingattachmentandexternalizingwasincludedinthistotalset,regardlessoftypeofmea-suresused.Ifdisorganizedattachmentwasassessed,thiscategorywasincludedintheinsecuregroup.Inthisoverallset,wefoundasigniÞcantcombinedeffectsizeof=0.31(seeTable2).Chil-drenratedasinsecureshowedhigherlevelsofexternalizingbehaviorsthanchildrenratedassecure.Theeffectsizewasmodestbutrobust,asmorethan1,700studieswithnullresults(fail-safenumber)wouldbeneededtoreducethiseffecttononsigniÞcance.Furthersupportfortheabsence442Fearon,Bakermans-Kranenburg,vanIJzendoorn,Lapsley,andRoisman largenumberofmoderatingrelationshipsthatcouldamplifyorattenuatetheassociationbetweenattachmentandexternalizingproblems(e.g.,Belsky&Fearon,2002;NICHDEarlyChildCareResearchNetwork,2006).Relatedtothispoint,theassocia-tionforchildrenfromnonclinicalpopulationswashighlyheterogeneous,indicatinglargebetween-studydifferencesineffectsize,whichcouldhaveresultedfromanumberofmethodologicalfactors,suchasthecompositionofthepopulationsstudied,thelengthofthefollow-upperiodorthemeasure-mentstrategiesthatwereadopted.Indeed,thelargereffectfoundinclinicalsamples(=0.49)illustratesthispoint,andatthesametimedemon-stratesthesigniÞcanceofattachmentforclinicalgroups.Withinclearconstraints,theanalysesreportedinthisstudywereabletoconsidertheroleofanumberofotherpotentiallyimportantmoderatorsthathavebeenhighlightedintheliterature.Cen-tralamongthesewereSES(withlow-risksamplesanticipatedtoyieldlargereffects),gender(inse-cureboysexpectedtoshowmorebehaviorprob-lems),andageofoutcome(witheffectsexpectedtopersistovertime).Incontrasttoourexpecta-tions,themagnitudeoftheassociationbetweenattachmentandbehaviorproblemswasrelativelyconsistentacrosshigh-andlow-SESsamples=0.25vs.=0.31,respectively).SESrepresentsaratherbluntapproximationofanumberofimportantproximalpsychosocialriskprocessesthatmaybemoreclearlyimplicatedintheassoci-ationbetweenattachmentandchildrenÕsexternal-izingproblems.Nevertheless,SESdoesaccountforaconsiderableportionofthevarianceinexter-nalizingproblemsinchildhood(Bradley&Cor-wyn,2002),andtheabsenceofmoderationinthismeta-analysisisapotentiallyimportantresult.TheÞndingcertainlysuggeststhatattachmentinsecurityisassociatedwithhigherlevelsofbehaviorproblemseveninapparentlylow-riskpsychosocialcircumstances.However,possibleeffectsofSESshouldnotberuledout,asabroaddivisionofsamplesintolowversushighormid-dleclassisclearlylimitedinprecisionandinsev-eralcasesthereviewedstudiesinvolvedmixturesofdifferinglevelsofsocioeconomicdisadvantagethatwerenotcapturedinourcodingscheme.Turningtogender,theresultswereconsistentwithourexpectationsandtheÞndingsofsomeearlylongitudinalstudies(Lewisetal.,1984;Ren-kenetal.,1989),inthatattachmentwasmorestronglyassociatedwithexternalizingbehaviorproblemsinsamplesofboysthaninsamplesofgirls(=0.35vs.0.03,respectively).Thesetofstudiesongirlswashomogeneous;therefore,theabsenceofasigniÞcantassociationwasnotduetooneorafewoutlyingoutcomesinthelowerrange.Furthermore,withinthemixedsample(boysandgirls),ametaregressionshowedatrendforstrongereffectsinsampleswithcomparativelymoreboysthangirls.Giventhatthevariationingendercom-positionwithinthissubsetreßectedaverynarrowrange,thisÞndingÑwhencombinedwiththosefromthesingle-genderstudiesÑprovidesquitecompellingevidenceoftheelevatedsigniÞcanceofattachmentforbehaviorproblemsinboys.Never-theless,themixedsamples,constitutingthelargemajorityofthestudies,showedequallystrongeffectsizesasthosewithonlyboys,suggestingthattheassociationbetweenattachmentandexternaliz-ingcannotbeascribedtoboysonly.Thereareanumberofplausiblewaysofinter-pretingthisgendereffect.First,onthefaceofittheeffectmaynotbesurprising,giventhesubstantiallyhigherriskofexternalizingproblemsinboys(Loe-ber&Hay,1997).However,thisbroadexplanationsubsumesseveraldistinctpossibilitiesthatarewor-thyofattention.First,thelowerratesofexternaliz-ingbehaviorproblemsingirlsmayimposearangerestrictiononthedependentvariable,whichinturnwouldattenuatetheeffectsize(DeKlyen&Green-berg,2008).Ontheotherhand,thelowerratesofexternalizingproblemsmayrepresentamanifesta-tionofadifferingsetofetiologicalmechanismsingirlsthanboys,withattachmentprocessesÞguringmoresigniÞcantlyinthedevelopmentaltrajectoriesofboys.Certainly,somebehaviorgeneticstudieshavesuggestedpartiallyindependentgeneticandenvironmentalcontributionstoexternalizingsymp-toms(e.g.,Vierikko,Pulkkinen,Kaprio,Viken,&Rose,2003).Furthermore,anumberofnongeneticstudieshavedocumenteddistinctriskfactorsforgirlsandboys(Cairns&Cairns,1984;Cummings,Iannotti,&ZahnWaxler,1989).Afurtherimportantpossibilitytoconsideristhatattachment(andindeedotherriskfactors)maycontributetoacom-monlatentprocessthathasdistinctivebehavioralmanifestations,whichinturnaremorecommonlyassociatedwithboysthangirls.Pertinentcandidateexamplesmightincludethebehavioralandcontex-tualdifferencesbetweenphysicalandrelationalaggression(Crick&Grotpeter,1995)orovertver-suscovertantisocialbehavior(Loeber&Schmaling,1985).Totheextentthatthesebehavioralprocessesaremeasuredbydifferentinstruments,areevidentindifferentcontexts,orareweighteddifferentlybydifferentobservers,wemayexpectdifferentialAttachmentandExternalizingBehaviorProblems449 behavior.Themeta-analyseswereperformedusingtheComprehensiveMeta-Analysis(CMA)program(Version2;Borenstein,Rothstein,&Cohen,2005).Foreachstudy,aneffectsize()wascalculatedasthestandardizeddifferencebetweenthetwoperti-nentgroups(e.g.,securevs.insecure).Inthosecaseswherecontinuousattachmentscoreswerecorrelatedwithexternalizingscores(e.g.,whenthestudyreportedontheAQS),werecomputedthestatisticintoCohenÕs(seeMullen,1989;Mullen&Rosenthal,1985,chap.6,fortheformulaefortrans-formationofvariousstatisticsintoCohenÕsEffectsizesindicatingapositiverelationbetweenexternalizingbehaviorandinsecurity,avoidance,anddisorganization,respectively(higherlevelsofexternalizingbehaviorintheinsecure,avoidant,ordisorganizedgroupcomparedtothereferencegroup),weregivenapositivesign.Thus,apositivecombinedeffectforthesetofstudiescomparingdisorganizedchildrenwithsecurechildrenonexternalizingbehaviorswouldmeanthatacrossthesestudiesthelevelofexternalizingbehaviorsindisorganizedchildrenwashigherthaninsecurechildren.Inthemainanalyses,wecomparedexter-nalizingbehaviorsofthechildrenineachattach-mentclassiÞcationwithallotherclassiÞcationscombined.Inanadditionalanalysisonasmallersetofstudieswithpertinentdata,wealsocom-paredeachclassiÞcationwiththesecureclassiÞca-tionasthemostÔÔpureÕÕreferencecategory.UsingCMA,combinedeffectsizeswerecom-puted.SigniÞcancetestsandmoderatoranalyseswereperformedthroughÞxedorrandomeffectsmodels,dependingonthehomogeneityofthestudyoutcomes.Fixedeffectsmodelsarebasedontheassumptionthateffectsizesobservedinastudyestimatethecorrespondingpopulationeffectwithrandomerrorthatstemsonlyfromthechancefactorsassociatedwithsubject-levelsamplingerrorinthatstudy(Lipsey&Wilson,2001;Rosenthal,1991).Thisassumptionisnotmadeinrandomeffectsmodels(Hedges&Olkin,1985).Randomeffectsmodelsallowforthepossi-bilitythattherearerandomdifferencesbetween Table1SourceSubsampledescriptionmeasureOutcomeRothbaumetal.(1995)SSPCBCL32Schmidtetal.(2002)AQSCBCL49CSchuengel,M.J.Bakermans-Kranenburg,&M.H.vanIJzendoorn,unpublisheddataSSPCBCL38Seiferetal.(2004)SSPCBCL732Shawetal.(1996)SSPCBCL77Smeekensetal.(2007)SSPCBCL105Solomonetal.(1995)MCOther40Speltzetal.(1990)CassMarvDiag50Speltzetal.(1999)CassMarvDiag160Stamsetal.(2002)SSPCBCL155Suessetal.(1992)SSPObs35Turner(1991)BoyssubsampleCassMarvObs18GirlssubsampleCassMarvObs22vanIJzendoornetal.(1992)SSPOther68VanZeijletal.(2006)1-yearSSPboysCassMarvCBCL251-yearSSPgirlsCassMarvCBCL182-yearSSPboysCassMarvCBCL282-yearSSPgirlsCassMarvCBCL103-yearSSPboysCassMarvCBCL183-yearSSPgirlsCassMarvCBCL16Vondraetal.(2001)SSPCBCL165Weiss&Seed(2002)AQSCBCL110Woodetal.(2004)AQSCBCL37.AQS=WatersandDeane(1985)AttachmentQ-Set;CassMarv=CassidyandMarvinMacArthurPreschoolAttachmentCodingSystem;SES=socioeconomicstatus;SSP=StrangeSituationprocedure;PAA=CrittendenPreschoolAttachmentAssessment;MC=MainandCassidyAge6ScoringSystem;Obs=externalizingdirectlyobserved;CBCL=ChildBehaviorChecklist;Other=otherexternalizingquestionnaire;Diag=clinicaldiagnosis.AttachmentandExternalizingBehaviorProblems441 temperamentinearlychildhoodpredictadjustmentinmiddlechildhood:Thecaseofadoptedchildrenandtheirbiologicallyunrelatedparents.,806Ð821.*Suess,G.J.,Grossmann,K.,&Sroufe,L.A.(1992).Effectsofinfantattachmenttomotherandfatheronqualityofadaptationinpreschool:Fromdyadictoindi-vidualorganisationofself.InternationalJournalofBehav-ioralDevelopment,43Ð65.Suomi,S.J.(2003).Gene-environmentinteractionsandtheneurobiologyofsocialconßict.InJ.A.King,C.F.Ferris,&I.I.Lederhendler(Eds.),Rootsofmentalillnessinchildren(pp.132Ð139).NewYork:NewYorkAcad-emyofSciences.Sutton,A.J.,Duval,S.J.,Tweedie,R.L.,Abrams,K.R.,&Jones,D.R.(2000).Empiricalassessmentofeffectofpublicationbiasonmeta-analyses.BritishMedicalJour-,1574Ð1577.Tabachnik,B.G.,&Fidell,L.S.(2001).Usingmultivariate(4thed.).Boston:Allyn&Bacon.*Turner,P.(1991).Relationsbetweenattachment,gender,andbehaviorwithpeersinthepreschool.ChildDevel-,1475Ð1488.vanIJzendoorn,M.H.(1997).Attachment,emergentmorality,andaggression:Towardadevelopmentalsocioemotionalmodelofantisocialbehaviour.tionalJournalofBehavioralDevelopment,703Ð727.*vanIJzendoorn,M.H.,Sagi,A.,&Lambermon,M.W.E.(1992).Themultiplecaretakerparadox:DatafromHol-landandIsrael.InR.C.Pianta(Ed.),Beyondtheparent:TheroleofotheradultsinchildrenÕslives(pp.5Ð24).SanFrancisco:Jossey-Bass.vanIJzendoorn,M.H.,Schuengel,C.,&Bakermans-Kranenburg,M.J.(1999).Disorganizedattachmentinearlychildhood:Meta-analysisofprecursors,concomit-ants,andsequelae.DevelopmentandPsychopathology*VanZeijl,J.,Mesman,J.,VanIjzendoorn,M.H.,Baker-mansKranenburg,M.J.,Juffer,F.,Stolk,M.N.,etal.(2006).Attachment-basedinterventionforenhancingsensitivedisciplineinmothersof1-to3-year-oldchil-drenatriskforexternalizingbehaviorproblems:Aran-domizedcontrolledtrial.JournalofConsultingandClinicalPsychology,994Ð1005.Verschueren,K.,&Marcoen,A.(1999).Representationofselfandsocioemotionalcompetenceinkindergart-ners:Differentialandcombinedeffectsofattachmenttomotherandfather.ChildDevelopment,183Ð201.Vierikko,E.,Pulkkinen,L.,Kaprio,J.,Viken,R.,&Rose,R.J.(2003).Sexdifferencesingeneticandenvironmen-taleffectsonaggresion.AggressiveBehavior,55Ð68.*Vondra,J.I.,Shaw,D.S.,Swearingen,L.,Cohen,M.,&Owens,E.B.(2001).AttachmentstabilityandemotionalandbehavioralregulationfrominfancytopreschoolDevelopmentandPsychopathology,13Ð33.Waters,E.,&Deane,K.E.(1985).DeÞningandassessingindividualdifferencesinattachmentrelationships:Q-methodologyandtheorganizationofbehaviorininfancyandearlychildhood.InI.Bretherton&E.Waters(Eds.),Growingpointsofattachmenttheoryandresearch.MonographsoftheSocietyforResearchinChildDevelopment(1Ð2,SerialNo.209),41Ð65.Weaver,I.C.G.,Cervoni,N.,Champagne,F.A.,DÕAles-sio,A.C.,Sharma,S.,Seckl,J.R.,etal.(2004).Epige-neticprogrammingbymaternalbehavior.,847Ð854.*Weiss,S.J.,&Seed,M.S.(2002).Precursorsofmentalhealthproblemsforlowbirthweightchildren:Thesal-ienceoffamilyenvironmentduringtheÞrstyearoflife.Childpsychiatryandhumandevelopment,3Ð27.*Wood,J.J.,Emmerson,N.A.,&Cowan,P.A.(2004).Isearlyattachmentsecuritycarriedforwardintorelation-shipswithpreschoolpeers?BritishJournalofDevelop-mentalPsychology,245Ð253.456Fearon,Bakermans-Kranenburg,vanIJzendoorn,Lapsley,andRoisman combinedeffectsizewas=0.27(.01;85%CI:0.15,0.39).Foravoidance,thiscombinedeffectsizeamountedto=0.06(&#x-334;&#x.900;.05;85%CI:0.01,0.13),forresistanceitwas=0.11(.05;85%CI:0.03,0.19),andfordisorganizationthecombinedeffectsizewas=0.29(.01;85%CI:0.13,0.45).The85%CIsofthecombinedeffectsizesforattachmentsecurityanddisorganizationdidnotoverlapwiththe85%CIofthecombinedeffectsizeforavoidance,butfortheothercomparisonsthe85%CIsover-lapped.Resistantattachmentdidnotshowadiffer-entcombinedeffectsizecomparedtoallotherattachmentclassiÞcations.Insecureanddisorga-nizedattachmentsthusimplicateahigherriskforexternalizingbehaviorthanavoidantattachment.WealsocomparedeachofthenonsecureclassiÞca-tionswiththesecureclassiÞcation.Inacoresetof18studiesthatprovideddataonallthreecomparisons,thecombinedeffectsizefortheassociationbetweenavoidanceversussecurityandexternalizingwas=0.12(.05;85%CI:0.04,0.20).Forresistantversussecureattachment,itwas=0.19(.01;85%CI:0.11,0.28),andfordisorganizedversussecureattachment,thecombinedeffectsizewas=0.27(.01;85%CI:0.13,0.41).Forallthreecomparisons,the85%CIsoverlapped,implyingthattheeffectsofthevarioustypesofinsecurityshowedsimilarassociationswithexternalizingbehavior.DiscussionSinceBowlbyÕsearliestworkonattachmentandseparation,therehavebeenpersistentsuggestionsintheliteraturethatattachmentinsecuritymayplayanimportantroleinthedevelopmentofaggressionandantisocialbehavior(Bowlby,1944;Lewisetal.,1984;LyonsRuthetal.,1993;Renkenetal.,1989;vanIJzendoorn,1997).Thebodyofresearchthatsubsequentlytestedthisassociationisimpressiveinitssheersize.However,despiteanextensiveaccumulationofdata,aclearviewontheempiricalstandingofthisimportanthypothesishasbeenelusive.ApparentlycontradictoryÞndings,nonre-plications,andadiversityofstudydesignsandsamplesizeshavecreatedabodyofworkthatisdifÞculttointegratecoherentlyinnarrativereviews.Thecentralquestionwethusposedinthismeta-analysiswaswhetherattachmentinsecuritywasassociatedwithexternalizingbehaviorsacrossallthestudiesconductedtodate.TheresultsshowedquiteclearlythattheanswertothisquestionisaÞrmyes.Drawingfromdataonnearly6,000chil-drentestedinstandardizedobservationalassess-mentsofmotherÐchildattachmentsecurity,theaverageeffectsizeforthecontrastbetweensecureandinsecurechildrenwas=0.31(95%CI:0.23,0.40).Over1,700studiesofaveragesamplesizewithnullresultswouldneedtobeaddedtothedatabasetoreducethiseffecttononsigniÞcance.Forclinicalsamples,theaverageeffectsizeamountedto=0.49(95%CI:0.32,0.66).Onthefaceofit,theserobustÞndingslenddirectsupporttothenotionthatattachmentplaysasigniÞcantroleintheevolutionofchildrenÕsbehaviorprob-lems,fortypicallydevelopingchildrenaswellasforclinicalgroups.Itshouldbenotedthatformeta-analyticresultstheevaluationofcombinedeffectsizesintermsofabsolutemagnitudeisproblematic(McCartney&Rosenthal,2000).Itisarguablymoremeaningfultoconsidertheglobalassociationbetweeninsecurityandexternalizingproblemsinthecontextofotherstudiesexaminingsimilarphenomenaandemploy-ingsimilarmethodologies(McCartney&Rosenthal,2000).Inthatrespect,thecombinedeffectsizeof=0.31reportedhereisofsimilarmagnitudetometa-analyticresultsconcerningtheassociationbetweenaggressionandhostileattributionalbiases=0.35;seeOrobiodeCastro,Veerman,Koops,Bosch,&Monshouwer,2002)orrestingheartrate0.38,Lorber,2004)andsubstantiallyhigherthanthatbetweenaggressionandbasalcortisol0.10;seeAlinketal.,2008).Thecombinedeffectfoundinthisanalysisgainsevengreatersig-niÞcanceinlightofthefactthatamajorityofthestudiesincludedintheanalysiswerelongitudinalinvestigations,whereindependentanddependentvariablesweremeasuredoftenseveralyearsapart(withaninterveningperiodof25monthsonaver-Threeperhapsobviousadditionalpointsabouttheeffectsshouldbemadeattheoutset.First,thereweresofewoutcomestudiesthatexaminedfatherÐchildattachmentsecuritythatwewereunabletoincludetheminthismeta-analysis.Thereisclearlyanurgentneedforfurtherresearchintothecontri-butionoffatherÐchildattachmentsecurityandinse-curitytochildrenÕsdevelopment.Second,theeffectsreportedinthismeta-analysisreßectstatisti-calassociation,notcausationandÑalthoughtheyprovideevidencerelevanttoeffortstodeterminecausalityanditsmechanismsÑalonetheyaremuteonthisissue.Third,theeffectsareuncorrectedfortheinßuenceofrelevantthirdvariables.Controlsforthesecouldreduceorincreasethemagnitudeoftheglobalmeta-analyticeffects.Furthermore,theeffectsrepresentanaverageacrossapotentially448Fearon,Bakermans-Kranenburg,vanIJzendoorn,Lapsley,andRoisman Table1SampleCharacteristicsforPrimarySetofStudiesSourceSubsampledescriptionAttachmentmeasureOutcomeAnan&Barnett(1999)CassMarvCBCL56Aviezeretal.(2002)SSPCBCL63Bakermans-Kranenburg&vanIJzendoorn(2006)BoyssubsampleSSPCBCL23GirlssubsampleSSPCBCL24Bates&Bayles(1988)BoyssubsampleSSPCBCL28GirlssubsampleSSPCBCL27Boothetal.(1991)High-risksubsampleSSPObs20Low-risksubsampleSSPObs16Boothetal.(1994)SSPCBCL69Burgessetal.(2003)SSPCBCL140Carlson(1998)Minnesotastudy:EffectsizeforDversusnon-DonlySSPCBCL78Cicchettietal.(1998)AQSCBCL128Cohn(1990)BoyssubsampleSSPOther34GirlssubsampleSSPOther46DeMulderetal.(2000)BoyssubsampleAQSOther51GirlssubsampleAQSOther43DeVito&Hopkins(2001)PAACBCL58Edwardsetal.(2006)ClinicalfatherssubsampleSSPCBCL82NonclinicalfatherssubsampleSSPCBCL94Egeland&Heister(1995)Minnesotastudy:Bversusnon-BeffectsizeSSPCBCL64Fagot&Leve(1998)SSPCBCL136R.P.Fearon,unpublisheddataTwinsampleSSPCBCL27Gilliometal.(2002)SSPCBCL189Goldbergetal.(1995)CysticÞbrosissampleSSPCBCL40CoronaryheartdiseasesampleSSPCBCL54NonclinicalsampleSSPCBCL51Goldbergetal.(1990)SSPOther69Greenbergetal.(1991)ODD+ControlsCassMarvDiag50Howesetal.(1994)SSPOther74HubbsTaitetal.(1994)SSPCBCL44Kelleretal.(2005)SSPCBCL169KleinVeldermanetal.(2006)SSPCBCL26Lewisetal.(1984)BoyssubsamplemodiÞedSSPSSPCBCL51GirlssubsamplemodiÞedSSPSSPCBCL57Liebermanetal.(1991)InterventionÑcontrolsampleonlySSPObs52LyonsRuthetal.(1993)SSPOther62Madiganetal.(2007);.SSPCBCL64Marchand&Hock(1998)AQSCBCL46Matasetal.(1978)SSPObs48Moss,Cyr,etal.(2004)CassMarvOther220Munsonetal.(2001)SSPCBCL101NICHDBoyshigh-SESsubsampleSSPCBCL490Boyslow-SESsubsampleSSPCBCL55Girlshigh-SESsubsampleSSPCBCL460Girlslow-SESsubsampleSSPCBCL70F.D.Pannebakker,unpublisheddataSSPCBCL115PerezCorres(2006)SSPCBCL51Pierrehumbertetal.(2000)SSPCBCL40RadkeYarrowetal.(1995)SSPDiag95440Fearon,Bakermans-Kranenburg,vanIJzendoorn,Lapsley,andRoisman