【文档说明】随机控制实验了解接受远端祈祷心脏病病人的治疗效果课件.ppt，共(47)页，275.625 KB，由小橙橙上传

转载请保留链接：https://www.ichengzhen.cn/view-253314.html

以下为本文档部分文字说明：

QuestionsTherelationshipofinterestTheidealexperimentTheidentificationstrategyThemodeofinference随机控制实验了解接受远端祈祷心脏病病人的治疗效果1Therelation

ofinterestMostquestionsareaboutcasualandeffectExample◼Smallerclassesarebetterforthelearningofthestudents◼Lowercopaymentencourages

thehealthutilizations随机控制实验了解接受远端祈祷心脏病病人的治疗效果2CoefficientEquations11112121ˆnniiniiiixyinxxiniiixyxySSnSSxxn=====

−==−Slopey-intercept01ˆˆˆyx=+PredictionEquation01ˆˆyx=−随机控制实验了解接受远端祈祷心脏病病人的治疗效果3Theestimatesofregressionshowstheextentof“corr

elation”,butweareinterestedtoknow“causation”Keyissueinempiricalanalysisisseparatingcausationfromcorrelation.◼Correlatedmeansthattwoeconomicvariabl

esmovetogether.◼Casualmeansthatoneofthevariablesiscausingthemovementintheother.随机控制实验了解接受远端祈祷心脏病病人的治疗效果4THEIMPORTAN

TDISTINCTIONBETWEENCORRELATIONANDCAUSATIONTherearemanyexampleswherecausationandcorrelationgetconfused.Itiscriticalforgovernmentpolicytou

nderstandthedifference;otherwisepolicymaynothavetheintendedimpact.随机控制实验了解接受远端祈祷心脏病病人的治疗效果5THEIMPORTANTDISTINCTIONBETWEENCORRELATIO

NANDCAUSATIONOneexampleconcernsSATpreparationcourses.◼In1988,HarvardintervieweditsfreshmenandfoundthosewhotookS

AT“coaching”coursesscored63pointslowerthanthosewhodidnot.◼OnedeanconcludedthattheSATcourseswereunhelpfuland“th

ecoachingindustryisplayingonparentalanxiety.”随机控制实验了解接受远端祈祷心脏病病人的治疗效果6TheProblemInbothexamples,thereisaco

mmonproblem:anattempttointerpretacorrelationasacausalrelationship,withoutsufficientthoughttotheunderlyingdatage

neratingprocess.ForanycorrelationbetweentwovariablesAandB,therearethreepossibleexplanationsforacorrela

tion:◼AiscausingB.◼BiscausingA.◼Someotherfactoriscausingboth.随机控制实验了解接受远端祈祷心脏病病人的治疗效果7TheProblemIntheHarvardSATexa

mple,thepossibilitiescouldbe:◼SATprepcoursesworsenpreparationfortheSATs.◼Thosewithpoorertesttakingabilitytakeprepcour

sestotrytocatchup.◼Thosewhoaregenerallynervousbothliketotakeprepcoursesanddotheworstonstandardizedexams.Harvarddeanthoughtthefirstpos

sibilitywascorrect.随机控制实验了解接受远端祈祷心脏病病人的治疗效果8TheProblemAlthoughthepeasantsortheHarvarddeancouldactuallybecorrect,oddsaretheyaremisin

terpretingtheunderlyingprocessatwork.Forpolicypurposes,whatwecareaboutiscausation.Knowingthattwofactorsarecorrelatedgivesyounopredictivepower.随

机控制实验了解接受远端祈祷心脏病病人的治疗效果9TheProblemofBiasInthiscase,theassignmentoftheinterventionwasnotrandom.Thismeansthetreatmentandcontrolgroupsare

notidentical.Non-randomassignment,inturn,couldcausebias.随机控制实验了解接受远端祈祷心脏病病人的治疗效果10TheProblemofBiasBiasrepresentsany

sourceofdifferencebetweentreatmentandcontrolgroupsthatiscorrelatedwiththetreatment,butnotduetothetreatment.◼IntheSATexample,theim

pactofSATcoursesisbiasedbythefactthatthosewhotaketheprepcoursearelikelytodoworseontheSATforotherreasons.随机控制实验了解接受远端

祈祷心脏病病人的治疗效果11TheProblemofBiasBydefinition,suchdifferencesdonotexistinarandomizedtrial,sincethegroupsarenotdiff

erentinanyconsistentfashion.Asaresult,randomizedtrialshavenobias,anditisforthisreasontheyarethe“goldstandard”forempiricall

yestimatingcausaleffects.随机控制实验了解接受远端祈祷心脏病病人的治疗效果12TheidealexperimentWhatsortofexperimentcanideallybeusedtocapturethecas

ualeffect?Goldenstandard:controlledtrials◼Randomizesubjectsintothetreatmentandcontrolgroup,thencomparetheirou

tcomedifferencebetweencontrolledandtreatmentgroup◼Commonlyusedtoanswerquestionsinnaturalscience,butdifficulttoimplementtoanswerq

uestionsinsocialscienceforvariousissues随机控制实验了解接受远端祈祷心脏病病人的治疗效果13MEASURINGCAUSATIONWITHDATAWE’DLIKETOHAVE:RANDOMIZEDTRIALSWithrandomassign

ment,theassignmentoftheinterventionisnotdeterminedbyanythingaboutthesubjects.Asaresult,thetreatmentgroupisident

icaltothecontrolgroupineveryfacetbutone:thetreatmentgroupgetstheintervention.随机控制实验了解接受远端祈祷心脏病病人的治疗效果1

4ExampleI:doesthepayerwork?Example(Harrisetal.):“隨機控制實驗：瞭解接受遠端祈禱心臟病病人的治療效果“随机控制实验了解接受远端祈祷心脏病病人的治疗效果15Harrisetel.實驗設計原則：“隨機,控制,雙盲,事前，同時實驗.“◼

隨機：病人隨機分配到禱告與否◼控制：有些病人沒有禱告◼雙盲：病人或醫師不知道為實驗或對照組◼事前：在治療前隨機分配◼同時：實驗同時進行随机控制实验了解接受远端祈祷心脏病病人的治疗效果16Harrisetel.的設計随机控

制实验了解接受远端祈祷心脏病病人的治疗效果17随机控制实验了解接受远端祈祷心脏病病人的治疗效果18Harrisetel.的結論“結論：遠端禱告有效”随机控制实验了解接受远端祈祷心脏病病人的治疗效果19TANFandlaborsupplyamongsi

nglemothersTANFis“TemporaryAssistanceforNeedyFamilies.”Cashwelfareforpoorfamilies,mainlysinglemothers.◼Fore

xample,inNewMexico,familyofthreereceives$389permonth.Assumethetwo“goods”inutilitymaximizationproblemareleisureandfoo

dconsumption.Whatevertimeisnotdevotedtoleisureisspentworkingandearningmoney.随机控制实验了解接受远端祈祷心脏病病人的治疗效果20RandomizedTrial

sintheTANFContextItisbelievedthatanincreasesinlaborsupplywhenTANFbenefitsarecut,butthemagnitudeoftheeffectisunclear.Onecoulddesignarandomize

dtrialtolearnabouttheelasticityofemploymentwithrespecttoTANFbenefits.随机控制实验了解接受远端祈祷心脏病病人的治疗效果21RandomizedTrialsintheTANFContextImaginealargegroup(sa

y,2000)ofsinglemotherswererandomlyassignedtooneoftwogroupswithacoinflip:◼The“control”groupcontinuestoreceiveaguaranteeof$5,000.◼The“treatment

”groupnowhastheirTANFbenefitcutto$3,000.Followgroupsforaperiodoftime,andmeasuretheworkeffort.随机控制实验了解接受远端祈祷心脏病病

人的治疗效果22RandomizedTrialsintheTANFContextInanexperimentlikethisinCaliforniain1992,theelasticityofemploymentwithrespecttowelfarebenefitswasestimate

dtobe-0.67.Thus,a10%decreaseinbenefitsresultedina6.7%increaseinemployment.随机控制实验了解接受远端祈祷心脏病病人的治疗效果23

WhyWeNeedtoGoBeyondRandomizedTrialsRandomizedtrialspresentsomeproblems:◼Theycanbeexpensive.◼Theycant

akealongtimetocomplete.◼Theymayraiseethicalissues(especiallyinthecontextofmedicaltreatments).◼Theinferencesfromthe

mmaynotgeneralizetothepopulationasawhole.◼Subjectsmaydropoutoftheexperimentfornon-randomreasons,aproblemknownasat

trition.随机控制实验了解接受远端祈祷心脏病病人的治疗效果24WhyWeNeedtoGoBeyondRandomizedTrialsForthesereasons(especiallythefirstone

aboutrandomizedtrialsbeingexpensive),economistsoftentakedifferentapproachestotrytoassesscausalrelationshipsinempiricalresearch.随机控制实验了解接受远端祈

祷心脏病病人的治疗效果25WhatisyouridentificationstrategyHowcanyouobtainthecasualeffectusingyourobservationaldata?Drawingin

ferencefromtheobservationaldataneedstobeverycarefulLikelytosufferfromtheselectionbias随机控制实验了解接受远端祈祷心

脏病病人的治疗效果26ExampleI根據9999泛亞人力銀行調查顯示，有六成八的大學應屆畢業生求職處處碰壁，到現在還找不到工作，而這些失業青年目前生活經費的主要來源，有七成以上是在家靠父母親友養，更令

人憂心的是，有一成的社會新鮮人是以借貸、舉債過日。9999泛亞人力銀行營運長楊肯誠認為，八月份二技、四技二專考試放榜後，部分為被錄取的高職、專科生也將投入就業市場，預估九月份的應屆畢業生失業率更為嚴重，有可能超越去年九月的七

成二。該人力銀行昨天在台北的喜來登酒店舉行記者會，公布八月三日至十四日所進行今年找全職工作的「應屆畢業生就業狀況」調查，針對跟人力銀行資料庫符合條件的九萬二千一百五十二名應屆畢業生發出問卷，有效樣本共一萬九千三百七十八份，回收率百分之廿一，當信心水準為百分之

九十五時，誤差值為正負百分之零點八。受訪樣本集中在大專以上學歷為主，占全體百分之八十七點五。試就上述論點一一評論下列敘述：随机控制实验了解接受远端祈祷心脏病病人的治疗效果27Doyoubelievethenumber?Beforeyoubelieveintheobservationald

ata,youshould◼Examinethenumberswithyourpersonalexperiences◼Examinethenumberswiththegovernmentdata◼Examinethe

numberswithotherstudiesMorethanhalftimesthesenumbersareincorrect,sobettingondisbelievingthesenumber

sisusuallysmarter随机控制实验了解接受远端祈祷心脏病病人的治疗效果28Whatistheproblem?◼Providebasicsurveystatistics:sample,responserate,samplesize,confidenceinterval◼T

hemajorityofthestudieshavetheresponserateislowerthan70%◼Forconfidentialquestions,theresponseratecouldbeevenlowerthan30%随机控制

实验了解接受远端祈祷心脏病病人的治疗效果29Wheredoesthebiascomefrom?◼Thosewhohavefoundthejobisunlikelytorespondtotheemail◼Therefore,themajo

rityofthemshouldbetheoneswhocannotfindthejobs◼Ifwedothereweighingthisfactor,thentheunemploymentratewould

droptolessthan15%随机控制实验了解接受远端祈祷心脏病病人的治疗效果30Whatisthecorrectestimates随机控制实验了解接受远端祈祷心脏病病人的治疗效果31ExampleIINHISasksaboutthehealthstat

us(1bestand5worst)ofindividualsadmittedandnotadmittedforhospitalsHospitalmakesyousicker!随机控制实验了解接受远端祈祷心脏病病人的治疗效果32Howtosolvetheproblem?Quasi-ra

ndomassignmentInstrumentalvariablemethodPropensityscorematchingmethodRegression随机控制实验了解接受远端祈祷心脏病病人的治疗效果33Quasi-ExperimentsEconomiststypically

cannotsetuprandomizedtrialsformanypublicpolicydiscussions.Yet,thetime-seriesandcross-sectionalapproachesareoftenun

satisfactory.Quasi-experimentsarechangesintheeconomicenvironmentthatcreateroughlyidenticaltreatmentandcon

trolgroupsforstudyingtheeffectofthatenvironmentalchange.◼Thisallowsresearcherstotakeadvantageofrandomizationcr

eatedbyexternalforces.随机控制实验了解接受远端祈祷心脏病病人的治疗效果34Quasi-ExperimentsBasicapproachistoletoutsideforcesdotherandomizationforus.Insomecases

,thesituationhappensnaturally.◼Suppose,forexample,thatArkansascutitsTANFbenefitby20%in1997,andthatwehada

largesampleofsinglemothersinArkansasin1996and1998.◼Atthesametime,imaginethatLouisiana’sbenefitsremainedun

changed.随机控制实验了解接受远端祈祷心脏病病人的治疗效果35Quasi-ExperimentsInprinciple,thealterationinthestates’policieshasessentiallyperformedourrandomizationforus.◼Thewo

meninArkansaswhoexperiencedthedecreaseinbenefitsarethetreatmentgroup.◼ThewomeninLouisianawhosebenefitswereunchangedarethecontrol.◼Bycomputi

ngthechangeinlaborsupplyacrossthesegroups,andthenexaminingthedifferencebetweentreatment(Arkansas)andcontrol(Louisiana),wecanobtainanestimateoftheimpa

ctofbenefitsonlaborsupplythatisfreefrombias.随机控制实验了解接受远端祈祷心脏病病人的治疗效果36Quasi-ExperimentsImaginewesimplystudiedsinglemothersinArkansasalone.Arkans

ashasessentiallyperformedan“experiment”wheresinglemothersin1996arethecontrolgroup,andthosein1998arethetreatmentgroup.Inpractice,thiscompariso

nrunsintothecriticismsthatconfrontuswithtimeseriesanalysis.◼Forexample,thenationaleconomywasgrowingexceptionallyfa

stduringthisperiod.随机控制实验了解接受远端祈祷心脏病病人的治疗效果37Quasi-ExperimentsBecauseoftheseconcernsaboutnationaltrends

,thequasi-experimentalapproachincludestheextrastepofcomparingthetreatmentgroupforwhomthepolicychangedtoacontrolgroupforwhomitdidnot.S

inglemothersinLouisianadidnotexperiencetheTANFcut,yetbenefitfromthegrowthintheeconomy.随机控制实验了解接受远端祈祷心脏病病人的治疗效果38Quasi-Exp

erimentsThatis,byexamininghoursofworkinArkansas,weobtain:◼HOURSAR,1998-HOURSAR,1996◼Thiscontainsboththe

treatmenteffectandthebiasfromtheeconomicboom.Incontrast,byexamininghoursofworkinLouisiana,weobtain:◼HOURSLA,1998-HOU

RSLA,1996◼Thiscontainsonlytheeffectoftheeconomicboom.随机控制实验了解接受远端祈祷心脏病病人的治疗效果39Quasi-ExperimentsBysubtractingthechangeinhoursofworkinLouisi

anafromthatinArkansas,wecontrolforthebiascausedbytheeconomicboom.WeobtainacausalestimateoftheeffectofTANFbenefitsonhoursofwork.A

nexampleisgiveninTable1,firstfocusingonArkansasalone.随机控制实验了解接受远端祈祷心脏病病人的治疗效果40Table1UsingQuasi-ExperimentalVariationArkansas19961998DifferenceBen

efitGuarantee$5,000$4,000-$1,000HoursofWorkPerYear1,0001,200200CPSdatashowsactualhoursofworkforsinglemothersinArkansas.At

thesametimebenefitswerebeingcut,hoursofworkincreasedforsinglemothersinArkansas.Benefitsfellby$1,000inArkansasduring1997.随机控制实验了解接受远端祈祷心脏病病人的治疗效果4

1Quasi-ExperimentsWhilebenefitsfellby20%,hoursofworkincreasedby20%;theimpliedelasticityoflaborsupplywithre

specttobenefitslevelsis-1.Thisislargerthanthe-0.67elasticityestimatefoundintherandomizedtrialinCalifornia

.随机控制实验了解接受远端祈祷心脏病病人的治疗效果42Quasi-ExperimentsThereislikelytobebiasinthis“first-difference,”becausetherewasmajoreconomicg

rowthduringthisperiod.◼Thus,singlemothersinArkansasmayhaveincreasedtheirworkeffortevenifTANFbenefitshadnotfallen.Weexaminesinglemothersint

heneighboringstateofLouisiana,inthebottompanelofTable1.随机控制实验了解接受远端祈祷心脏病病人的治疗效果43Table1UsingQuasi-Experimenta

lVariationArkansas19961998DifferenceBenefitGuarantee$5,000$4,000-$1,000HoursofWorkPerYear1,0001,20020

0Louisiana19961998DifferenceBenefitGuarantee$5,000$5,000$0HoursofWorkPerYear1,0501,10050WecangatherthesamekindofdataforLouisiana.Benefitl

evelsdidnotfallinLouisiana.Butlaborsupplystillincreased,perhapsduetothegrowingeconomy.Itappearsthat50hoursofthe200hourincreasewasduetoeco

nomicconditions,notTANF.随机控制实验了解接受远端祈祷心脏病病人的治疗效果44Quasi-ExperimentsThisapproachyieldsthedifference-in-differenceestimator

–thedifferencebetweenthechangesinoutcomesforthetreatmentgroupthatexperiencesaninterventionandacontrolgroupthatdoesnot.Wearetakingthedifferen

ceinlaborsupplychangesinthesestatesinanattempttopurgetheestimateofbias(duetothegrowingeconomy).◼Whilecross-sectionalanalysiswo

uldsuggestthatthereductioninwelfarebenefitsleadstoa100-hourincreaseinwork,thedifference-in-differenceanalysissuggestsa150-hourincrease.随机控制实

验了解接受远端祈祷心脏病病人的治疗效果45Quasi-ExperimentsThedifference-in-differenceestimatoris:Thesecondterm,forLouisiana,netsoutthebiasfromthegro

wingeconomy.Thus,thecausaleffectofTANFbenefitcutswouldbea150-hourincreaseinlaborsupply.()()HOURSHOURSHOURSHOURSAKAKLALA,,,,1998199619981996−−

−随机控制实验了解接受远端祈祷心脏病病人的治疗效果46Quasi-Experiments:Problemswithquasi-experimentalanalysisThisapproachalsohasproblems,however.◼Itispossiblethatt

heeconomicboomaffectedArkansasdifferentlythanitdidLouisiana.◼Moregenerally,singlemothersmaybedifferentacrossstates.Wecanneverbecompletelycertainthat

wehavepurgedthetreatment-controlcomparisonsofbias.随机控制实验了解接受远端祈祷心脏病病人的治疗效果47