安慰劑檢驗(yàn)介紹、操作及舉例

1、安慰劑檢驗(yàn)介紹（Placebotest）安慰劑是一種附加實(shí)證檢驗(yàn)的思路，并不存在一個(gè)具體的特定的操作方法。一般存在兩種尋找安慰劑變量的方法。比如，在已有的實(shí)證檢驗(yàn)中，發(fā)現(xiàn)自變量Xi會(huì)影響自變量Zi與因變量Yi之間存在相關(guān)關(guān)系。在其后的實(shí)證檢驗(yàn)中，采用其他主體（國(guó)家，省份，公司）的Xj變量作為安慰劑變量，檢驗(yàn)Xj是否影響Zi與Yi之間的相關(guān)關(guān)系。如果不存在類(lèi)似于Xi的影響，即可排除Xi的安慰劑效應(yīng)，使得結(jié)果更為穩(wěn)健。另一種尋找安慰劑變量的方法。已知，Xi是虛擬變量，Xi=1,iftT；Xi=0iftT+n；Xi=0iftT+n，其中n根據(jù)實(shí)際情況取值，可正可負(fù)。檢驗(yàn)Xi是否影響Zi與Yi之間的相

2、關(guān)關(guān)系。如果不存在類(lèi)似于Xi的影響，即可排除Xi的安慰劑效應(yīng),使得結(jié)果更為穩(wěn)健。舉例：以美國(guó)市場(chǎng)某種政策沖擊識(shí)別策略的因果關(guān)系考察，在最后部分選取英國(guó)同期的因變量，檢驗(yàn)是否有類(lèi)似的特征，就是安慰劑檢驗(yàn)。以中國(guó)2007年所得稅改革作為減稅的政策沖擊以驗(yàn)證減稅對(duì)企業(yè)創(chuàng)新的影響。亦可以通過(guò)把虛擬的政策實(shí)施時(shí)間往前往后推幾年，作為虛擬的政策時(shí)點(diǎn)，如果檢驗(yàn)發(fā)現(xiàn)沒(méi)有類(lèi)似的因果，文章的主要結(jié)論就更加可信了。以下是詳細(xì)的例題，安慰劑檢驗(yàn)在最后。SurvivingGraduateEconometricswithR:Difference-in-DifferencesEstimation2of8Thefollowi

3、ngreplicationexercisecloselyfollowsthehomeworkassignment#2inECNS562.Thedataforthisexercisecanbefoundhere.ThedataisabouttheexpansionoftheEarnedIncomeTaxCredit.Thisisalegislationaimedatprovidingataxbreakforlowincomeindividuals.Forsomebackgroundonthesubject,seeEissa,Nada,andJeffreyB.Liebman.1996.LaborS

4、upplyResponsestotheEarnedIncomeTaxCredit.QuarterlyJournalofEconomics.111(2):605-637.Thehomeworkquestions(abbreviated):Describeandsummarizedata.Calculatethesamplemeansofallvariablesfor(a)singlewomenwithnochildren,(b)singlewomenwith1child,and(c)singlewomenwith2+children.Createanewvariablewithearningsc

5、onditionalonworking(missingfornon-employed)andcalculatethemeansofthisbygroupaswell.Constructavariableforthe“treatment”calledANYKIDSandavariableforaftertheexpansion(calledPOST93shouldbe1for1994andlater).Createagraphwhichplotsmeanannualemploymentratesbyyear(1991-1996)forsinglewomenwithchildren(treatme

6、nt)andwithoutchildren(control).Calculatetheunconditionaldifference-in-differenceestimatesoftheeffectofthe1993EITCexpansiononemploymentofsinglewomen.Nowrunaregressiontoestimatetheconditionaldifference-in-differenceestimateoftheeffectoftheEITC.Useallwomenwithchildrenasthetreatmentgroup.Reestimatethism

7、odelincludingdemographiccharacteristics.Addthestateunemploymentrateandallowitseffecttovarybythepresenceofchildren.Allowthetreatmenteffecttovarybythosewith1or2+children.Estimatea“placebo”treatmentmodel.Takedatafromonlythepre-reformperiod.Usethesametreatmentandcontrolgroups.Introduceaplacebopolicythat

8、beginsin1992(so1992and1993bothhavethisfakepolicy).Areview:LoadingyourdataRecallthecodeforimportingyourdata:STATA:/*Lastmodified1/11/2011*/*Thefollowingblockofcommandsgoatthestartofnearlyalldofiles*/*Bracketcommentswith/*/orjustuseanasteriskatlinebeginningclear/*Clearsmemory*/setmem50m/*Adjustthisfor

9、yourparticulardataset*/cdC:DATAEcon562homework/*Changethisforyourfilestructure*/logusing,replace/*Logfilerecordsallcommands&results*/display$S_DATE$S_TIMEsetmoreoffinsheetusing,clearR:#KevinGoulding#ECNS562-Assignment2#Loadtheforeignpackagerequire(foreign)#Importdatafromwebsite#update:firstdownloadt

10、hefilefromthislink:10#Thenimportfromyourharddrive:11eitc=(C:/link/to/my/download/folder/)12NotethatanycommentscanbeembeddedintoRcode,simplybyputtinga#totheleftofyourcom13fromyourharddrive:1415eitc=(C:DATACoursesEcon562homework)DescribeandsummarizeyourdataRecallfrompart1ofthisseries,thefollowingcodet

11、odescribeandsummarizeyourdata:STATA:dessumR:InR,eachcolumnofyourdataisassignedaclasswhichwilldeterminehowyourdataistreatedinvariousfunctions.ToseewhatclassRhasinterpretedforallyourvariables,runthefollowingcode:sapply(eitc,class)summary(eitc)source()sumstats(eitc)TooutputthesummarystatisticstabletoLa

12、TeX,usethefollowingcode:require(xtable)#xtablepackagehelpscreateLaTeXcodefromR.xtable(sumstats(eitc)Note:Youwillneedtore-runthecodeforsumstats()whichyoucanfindinanearlierpost.CalculateConditionalSampleMeansSTATA:summarizeifchildren=0summarizeifchildren=1summarizeifchildren=1summarizeifchildren=1&yea

13、r=1994meanworkifpost93=0&anykids=1R:#Thefollowingcodeutilizesthesumstatsfunction(youwillneedtore-runthiscode)TOC o 1-5 h zsumstats(eitceitc$children=0,)sumstats(eitceitc$children=1,)sumstats(eitceitc$children=1,)sumstats(eitceitc$children=1&eitc$year=1994,)#Alternately,youcanusethebuilt-insummaryfun

14、ctionsummary(eitceitc$children=0,)summary(eitceitc$children=1,)summary(eitceitc$children=1,)summary(eitceitc$children=1&eitc$year=1994,)#Anotherexample:Summarizevariableworkforwomenwithonechildfrom1993onwards.summary(subset(eitc,year=1993&children=1,select=work)Thecodeaboveincludesallsummarystatisti

15、cs一butsayyouareonlyinterestedinthemean.Youcouldthenbemorespecificinyourcoding,likethis:mean(eitceitc$children=0,work)mean(eitceitc$children=1,work)mean(eitceitc$children=1,work)Tryoutanyoftheotherheadingswithinthesummaryoutput,theyshouldalsowork:min()forminimumvalue,max()formaximumvalue,stdev()forst

16、andarddeviation,andothers.CreateaNewVariableTocreateanewvariablecalled“”equaltoearningsconditionalonworking(if“work”=1),“NA”otherwise(“work”=0)一usethefollowingcode:STATA:gencearn=earnifwork=1R:121234567z=names(eitc)X=lapply(X,function(x)replace(x,x=0,NA)eitc=cbind(eitc,X)eitc$=NULLnames(eitc)=zConst

17、ructaTreatmentVariableConstructavariableforthetreatmentcalled“anykids”=1fortreatedindividual(hasatleastonechild);andavariableforaftertheexpansioncalled“post93”=1for1994andlater.STATA:genanykids=(children=1)genpost93=(year=1994)R:1eitc$post93=(eitc$year=1994)2eitc$anykids=(eitc$children0)CreateaplotC

18、reateagraphwhichplotsmeanannualemploymentratesbyyear(1991-1996)forsinglewomenwithchildren(treatment)andwithoutchildren(control).STATA:preservecollapsework,by(yearanykids)genwork0=workifanykids=0labelvarwork0Singlewomen,nochildrengenwork1=workifanykids=1labelvarwork1Singlewomen,childrentwoway(linewor

19、k0year,sort)(linework1year,sort),ytitle(LaborForceParticipationRates)graphsaveGraphhomework,replaceR:#Takeaveragevalueofworkbyyear,conditionalonanykidsminfo=aggregate(eitc$work,list(eitc$year,eitc$anykids=1),mean)#renamecolumnheadings(variables)names(minfo)=c(YR,Treatment,LFPR)#Attachanewcolumnwithl

20、abelsminfo$Group1:6=Singlewomen,nochildrenminfo$Group7:12=Singlewomen,childrenminforequire(ggplot2)#packageforcreatingniceplotsqplot(YR,LFPR,data=minfo,geom=c(point,line),colour=Group,xlab=Year,ylab=LaborForceParticipationRate)+-_/-tffI備l=2)genpostXone=post93*onekidgenpostXtwo=post93*twokidR:#Thesta

21、teunemploymentrateinteractedwithnumberofchildreneitc$=eitc$urate*eitc$anykids#Creatinganewtreatmentterm:#First,wellcreateanewdummyvariabletodistinguishbetweenonechildand2+.eitc$manykids=(eitc$children=2)#Next,wellcreateanewvariablebyinteractingthenewdummy#variablewiththeoriginalinteractionterm.eitc$tr2=eitc$*eitc$manykidsEstimateaPlaceboMod