中介效应分析：方法和模型发展

2014,Vol.22,No.5,731–745AdvancesinPsychologicalScienceDOI:10.3724/SP.J.1042.2014.00731731·主编特邀(Editor-In-ChiefInvited)·编者按：,,,“”,,,2004,Google10,,,,,,,,()中介效应分析：方法和模型发展*1,23(1/,510631)(2,)(3,330022)在心理学和其他社科研究领域,大量实证文章建立中介效应模型,以分析自变量对因变量的影响过程和作用机制。检验中介效应最流行的方法是Baron和Kenny的逐步法,但近年来不断受到批评和质疑,有人甚至呼吁停止使用其中的依次检验,改用目前普遍认为比较好的Bootstrap法直接检验系数乘积。本文对相关的议题做了辨析,并讨论了中介分析中建立因果关系的方法。综合新近的研究成果,总结出一个中介效应分析流程,并分别给出显变量和潜变量Mplus程序。最后介绍了中介效应模型的发展。中介效应;间接效应;逐步法;Bootstrap法;因果B841,,(mediationeffect),Rucker,Preacher,TormalaPetty(2011),20052009Journalof:2014-02-06*(31271116)(11JJD190005)(2013M540535)(13YJC190029)(2013KY08):,E-mail:[e-mail]ychology(JPSP)PersonalityandSocialPsychologyBulletin(PSPB),59%65%,,,,,Baron73222Kenny(1986)(causalstepsapproach),(,Edwards&Lambert,2007;Hayes,2009;Spencer,Zanna,&Fong,2005;Zhao,Lynch,&Chen,2010)(piecemeal)(),Bootstrap(Zhaoetal.,2010),,;,;1中介效应模型及检验方法介绍XY,XMY,M,“”“”,“”,,(,0)(0,1),(1)1YcXe=+(1)2MaXe=+(2)3YcXbMe′=++(3)(1)cXY;(2)aXM;(3)bX,MY;c′M,XY;e1~e3,(indirecteffect),ab,(MacKinnon,Warsi,&Dwyer,1995)ccab′=+(4)(Baron&Kenny,1986;Judd&Kenny,1981;,,,,2004),(i)(1)c(H0:c=0);(ii)(2)a(H0:a=0)(3)b(H0:b=0),(testofjointsignificance,Hayes,2009)(i)c,(ii)ab,(iii)(3)c′BaronKenny(1986),XY;(H0:ab=0),ab;,(H0:ab=0),,,,(H0:ab=0),15:733,,0a≠0b≠,0ab≠,,H0:ab=0,,(0.05)(MacKinnon,Lockwood,Hoffman,West,&Sheets,2002;,2004),ab,,ab(power),(Fritz&MacKinnon,2007;MacKinnonetal.,2002)H0:ab=0Sobel(1982)ˆˆ/abzabs=,ˆaˆbab,2222ˆˆabbasasbs=+ˆˆab,sasbˆaˆb,Sobel(MacKinnonetal.,2002;,2004)ˆˆab,,,abs,,Sobel(,,2012;Hayes,2009;MacKinnon,2008;MacKinnon,Lockwood,&Williams,2004)SobelH0:ab=0(,,2012),Bootstrap(MCMC)ˆˆab,(Fritz&MacKinnon,2007;MacKinnonetal.,2004;Tofighi&MacKinnon,2011)Bootstrap,()Bootstrap,,Bootstrap,BootstrapBootstrap(Wen,Marsh,&Hau,2010),500Bootstrap,,Bootstrap(500)Bootstrap(1000),1000Bootstrap,1000,ˆˆ{}ab,2.597.5ab95%,0,(,,2012;Preacher&Hayes,2008;Preacher,Rucker,&Hayes,2007;,,,2012),Bootstrap,Sobel(Fritz&MacKinnon,2007;MacKinnonetal.,2004),Bootstrap(Edwards&Lambert,2007;,,2012;Fritz&MacKinnon,2007;MacKinnon,2008;Preacher&Hayes,2008;Taylor,MacKinnon,&Tein,2008;,2012)Bootstrap“”,Bootstrap(),,,(MCMC)(Ntzoufras,2009;Yuan&MacKinnon,2009),MCMC,,(,,2012)MCMC,(priordistribution),,,BootstrapMCMCSobel,(,,2012;Hayes&Scharkow,2013;MacKinnon,2008;Preacher&Hayes,2004;Yuan&MacKinnon,2009)Mplus(Muthén&Muthén,2012),BootstrapMCMC,BootstrapMCMCBootstrapMCMC,,73422Bootstrap(Biesanz,Falk,&Savalei,2010;Cheung&Lau,2008;Fritz,Taylor,&MacKinnon,2012;Hayes&Scharkow,2013;MacKinnonetal.,2004;Pituch&Stapleton,2008;Pituch,Stapleton,&Kang,2006;Tayloretal.,2008),MCMC,,BootstrapSobel,Bootstrap(0.05)(,,2012;Fritz&MacKinnon,2007;MacKinnonetal.,2004),Bootstrap(Fritzetal.,2012)abcc′=−,0:0Hcc′−=(Clogg,Petkova,&Shihadeh,1992;Freedman&Schatzkin,1992),,(0.05),(MacKinnonetal.,2002;,2004),,,H0:ab=0,0:Hcc′−0=H0:ab=0H0:ab=0,SobelBootstrapMCMCH0:ab=0H0:ab=0Bootstrap,Sobel,Bootstrap(Fritzetal.,2012;Hayes&Scharkow,2013)Bootstrap,,2对Baron和Kenny逐步法的质疑和辨析BaronKenny(1986),,Zhao(2010),2.1依次检验还有用吗？,(2)a(3)b,Bootstrap,,,(Fritz&MacKinnon,2007;Hay,2009;MacKinnonetal.,2002),,,Bootstrap,,(H0:ab=0),,Bootstrap,(1),,BootstrapSobel(Fritz&MacKinnon,2007;MacKinnonetal.,2004),(MacKinnonetal.,2002);()(2),XM,MY,;,,XM,MY(3),,,0.05,0.05(MacKinnonetal.,2002;,2004),Bootstrap0.05(Fritz&MacKinnon,2007;MacKinnonetal.,2004),,,Bootstrap,“”(),BootstrapBootstrap,35:7352.2要先检验总效应吗？(1)c,(,MacKinnon,Krull,&Lockwood,2000;Zhaoetal.,2010),(ab)(c′),(c),;,,,,,cc,XYXY,M,XY,XY,XY,“”c,,,,,c,,“”,,BaronKenny(1986),c,XY,“XY”,“XY”(MacKinnon&Fairchild,2009;,,,2005)“XY()”c,XY,“XY”“XY”,,“XY”,,“”(suppressingeffects)(Kenny,2003;MacKinnon,2008;MacKinnonetal.,2000,2002;Shrout&Bolger,2002),,(2012),c“”,,,“”,,c,,,“,”,“”,,(),,,c,,,,c,,c,,,,,;(Mathieu&Taylor,2006;,2004)2.3区分完全中介和部分中介是否合适？,(3)c′c′,(James&Brett,1984)BaronKenny(1986),(Preacher&Kelley,2011),,(),,,,,,(Preacher&Hayes,2008),(Baron&Kenny,1986;Iacobucci,2008),MXY,73622(Pituch,Whittaker,&Stapleton,2005)PreacherHayes(2008),Zhao(2010),2.4逐步法能验证因果关系吗？“causalstepsapproach”“”,,,(Stone-Romero&Rosopa,2008),,,XM,MY,,,,,,,“X→Y”,“X→M”,“M→Y”,,,,,,,“X→M”“M→Y”,“X→Y”,XY,,“XY”,“XY”,3(Cook&Campbell,1979)(1);(2);(3),,,,XY(),“X→Y”(1)XY,XY(),,,,,,,,,,,,,,“”(2)XY,“Y→X”“X→Y”,,,;“”,→→;→→,,,,,(3)(,,,,2004),,,(),,,,,,,(,)(),,,,,,(,,,2007),“X→M”“M→Y”“X→Y”(,2.2),5:737XM,MY,(Spenceretal.,2005),,,(moderation-of-processdesign),“X→Y”;,M,X,Y(Spenceretal.,2005),,,,,,,,,,,,(Cole&Maxwell,2003)(),,(,,2005),,Mackinnon,FairchildFritz(2007)3,(autoregressivemodeling),(latentgrowthmodeling),(latentdifferencescoremodeling),,,,,(,2009),,,(),,,,,,,,,,3新的中介效应检验流程,(2004),,Sobel,,,Sobel,SobelBootstrap,,(2),,(1)c,,,,,(2)a(3)b,,,;,,BootstrapH0:ab=0,,;,,(3)c′,,,,,,abc′,,,ab/c,,|ab/c′|,(1),,,abc′,,,,,(2),ab/c|ab/c′|,PreacherKelley(2011)10,73822MacKinnon(2008,Fairchild,MacKinnon,Taborga,&Taylor,2009)R2,(ab,),PreacherKelley(2011)2κ,,(Wen,underreview),(3),,,abBootstrapSobel,,Bootstrap,Bootstrap,Bootstrapab(,Bootstrapab)(3)Bootstrap,,,,,1Mplus,,Bootstrap(4),,Zhao(2010)4基于结构方程的中介分析(0.9),(25:7390.7