倾向得分匹配方法的敏感性分析-刘凤芹

©1994-2010ChinaAcademicJournalElectronicPublishingHouse.Allrightsreserved.:2009-03-24:(2008LZ032):(1975-),,,,,:1,2(1.,100875;2.,710061):(propensityscorematching):(1),,(2)61%(3),,(4),,50%:;;;:F222:A:1007-3116(2009)10-0007-07,:,,,,,[1]:()():,RosenbaumRubin1983[2],,,:Johar,M.[3];Perkins[4],,,,,:?,:Heckman,,[5]SmithTod,,DID(difference-in-differencesmatch2ing)[6],Dehejia,RWahba,S.[7]Rubin,[8]7©1994-2010ChinaAcademicJournalElectronicPublishingHouse.Allrightsreserved.[9]n,m(),;n-m,:Y1,Y0T,1,0X(AverageTreatmentEffectontheTreated):ATT=E(Y1|T=1)-E(Y0|T=1)ATT:,,,(counterfactual),Y0,E(Y0|T=0)E(Y0|T=1)[10]():A^TTM=1n1iI1Sp[y1i-E^(y0i|Ti=1,Pi)](1)E^(y0i|Ti=1,Pi)=jI0W(i,j)y0j,I1,I0,Sp(),n1I1Sp,y1iy0iY1Y0ipii,:ijI0W(i,j)y0j,W(i,j)iPijPj()RosenbaumRubin1983A^TTMATT:1X,(Y0,Y1)T2X,T1011,T(Y0,Y1),12f(p|T=1)f(p|T=0)0,,,ATT,,LogisticProbit,A^TTM()[11],(2)(5):Y1=a10+a11X1+a12X2+1=Xa1+1(2)Y0=a00+a01X1+a02X2+0=Xa0+0(3)T3=h(X,b)+(4)T=1T300T30(5)(2)(3),(4)(5)Y0Y1TX=(X1X2),a1=(a10a11a12),a0=(a00a01a02)T3,h(X,b)(10)8©1994-2010ChinaAcademicJournalElectronicPublishingHouse.Allrightsreserved.()h(X,b),,ProbitLogistic[12]:11;210,1logistic;31,N(-2,1)N(2,1);41,N(0,1+015X21+015X22);51=(1+012X21)3Z,Zlogistic1,(E(Y1|T=1)-E(Y0|T=0))Logistic(Logistic)Probit(Probit)ATT1:,,:Logistic,,ATT01988,1,LogisictProbit,,,2,ATT1,1,,[13]1LogisticProbit3.5210.1131.0710.4971.0760.491Logistic2.8320.1241.0350.2531.0430.2522.8190.1211.0420.2511.0380.2512.8280.1201.0320.2491.0290.249logistic2.0330.1320.9880.1970.9880.199:()h(X,b),h(X,b),h(X,b)[14]h(X,b):11h(X,b)=b0+b1X1+b2X2+b3X1X2h^(X,b)=b0+b1X1+b2X221h(X,b)=b0+b1X1+b2X2+b3X21+b4X22h^(X,b)=b0+b1X1+b2X231h(X,b)=b0+b1X1+b2ln(X22)h^(X,b)=b0+b1X1+b2X241h(X,b)=b0X1exp(b1+b2X2)h^(X,b)=b0+b1X1+b2X2+b3X21+b4X2251h(X,b)=b0X1exp(b1+b2X2)h^(X,b)=b0+b1X1+b2X2h(X,b),h^(X,b)h(X,b);;;();()22:,,,,h(X,b)h^(X,b)9,:©1994-2010ChinaAcademicJournalElectronicPublishingHouse.Allrightsreserved.(X,b)LogisticProbit3.5210.1081.2370.4951.2300.4973.5130.1091.2470.4891.2440.4993.0500.0941.6070.6521.5770.684()2.7060.1100.9120.4810.9200.493()2.7060.1101.4320.4401.4000.466,,,,(),:(RadiusMatching),,1,,1:,;;;,[0.8,1),(0,0.8)134561,3456:,,:,11064,11243;,11907,11481,,,01©1994-2010ChinaAcademicJournalElectronicPublishingHouse.Allrightsreserved.(Logistic)01026010490115301698;01134011560124301395,,:,;,,,,:0114401146012710129501295;:0119701207012090122501321,,,probitLogistic,LogisticProbit,()(2)(5),10,,(10),3LogisticProbit1.0010.1681.0010.1671.0060.1321.0050.1311.0410.1231.0320.1241.0700.1211.0690.1211.0130.1271.0090.132:k5;0.014LogisticProbit1.0150.1981.0050.1991.0730.1581.0730.1591.0600.1571.0620.1551.0810.1521.0780.1511.0160.1141.0130.112:k10;0.025LogisticProbit1.0640.2941.0640.2951.2430.2111.2400.2201.2000.2181.1850.2191.1960.2211.1790.2231.0640.2691.0580.276:k10;0.0511,:©1994-2010ChinaAcademicJournalElectronicPublishingHouse.Allrightsreserved.=10.050.050.0510.50.050.5121=10.10.10.110.50.10.5131=10.20.20.210.50.20.5141=10.50.50.510.50.50.5177:10,,0.05,50%;0.1,70%,,(Y1Y0)T3,,,,:DID7LogisticProbit13.8370.1011.5410.6561.7440.65423.9390.1001.7830.6451.9250.64234.0380.0972.0820.6272.3430.62944.3710.0892.9790.5582.7420.558,:,,,,ATT,,,,61%,,,,,?,,;,,,,,,,,,,:,:,21©1994-2010ChinaAcademicJournalElectronicPublishingHouse.Allrightsreserved.(balanc2ingtest):[1].[J].,2004(9):13-17.[2]RosenbaumP,RubinD.Thecentralroleofthepropensityscoreinobservationalstudiesforcausaleffects[J].Biometrika,1983(70):41-55.[3]JoharM.TheimpactoftheIndonesianhealthcardprogram:amatchingestimatorapproach[J].[2009-03-10].JournlalofHealthEconomics,2008doi:10.1016/j.jhealeco.2008.10.001.[4]PerkinsSM,TuW,UnderhillMG,etal.Theuseofpropensityscoresinpharmacoepidemiologicresearch[J].PharmacoepidemiologyandDrugSafety,2000(9):93-101.[5]HeckmanJamesJ.Econometriccausality[J].InternationalStatisticalReview,2008(1):1-27.[6]SmithJeffreyA,ToddPetraE.DoesmatchingovercomeLaLondescritiqueofnonexperimentalestimators[J].JournalofEconometrics,2005(125):305-353.[7]DehejiaR,WahbaS.Propensityscorematchingmethodsfornonexperimentalcausalstudies[J].ReviewofEconomicsandStatistics,2002(1):151-161.[8]RubinDB.Forobjectivecausalinference:designtrumpsanalysis[J].TheAnnualofAppliedAtatistics,2008(3):808-840.[9]HeckmanJamesJ.TheScienticmodelofcausality[J].SociologicalMethodology,2005(1):1-97.[10]RubinD.Causalinferenceusingpotentialoutcomes:design,modeling,decisions[J].JournaloftheAmericanStatisticalAssociation,2005(100):322-331.[11]ZhaoZ.Usingmatchingtoestimatetreatmenteffects:datarequirements,matchingmetricsandMonteCarloevidence[J].ReviewofEconomicsandStatistics,2004(1):91-107.[12]HorowitzJoelL.Semiparametricandnonparametricestimationofquantalresponsemodels[C]/