資源描述：
《handbook on impact evaluation randomization》由會(huì)員上傳分享，免費(fèi)在線閱讀，更多相關(guān)內(nèi)容在工程資料-天天文庫(kù)。
1、3.RandomizationSummaryAllocatingaprogramorinterventionrandomlyacrossasampleofobservationsisonesolutiontoavoidingselectionbias,providedthatprogramimpactsareexaminedatthelevelofrandomization.Carefulselectionofcontrolareas(orthecounterfactual)isalsoimportantinensuringcomparabilitywithparticipantareas
2、andultimatelycal-culatingthetreatmenteffect(ordifferenceinoutcomes)betweenthetwogroups.Thetreatmenteffectcanbedistinguishedastheaveragetreatmenteffect(ATE)betweenparticipantsandcontrolunits,orthetreatmenteffectonthetreated(TOT),anarrowermeasurethatcomparesparticipantandcontrolunits,conditionalonpa
3、rticipantsbeinginatreatedarea.Randomizationcouldbeconductedpurelyrandomly(wheretreatedandcontrolunitshavethesameexpectedoutcomeinabsenceoftheprogram);thismethodrequiresensuringexternalandinternalvalidityofthetargetingdesign.Inactuality,however,researchershaveworkedinpartialrandomizationsettings,wh
4、eretreatmentandcon-trolsamplesarechosenrandomly,conditionalonsomeobservablecharacteristics(forexample,landholdingorincome).Iftheseprogramsareexogenouslyplaced,condi-tionalontheseobservedcharacteristics,anunbiasedprogramestimatecanbemade.Despitetheclarityofarandomizedapproach,anumberoffactorsstilln
5、eedtobeaddressedinpractice.Theyincluderesolvingethicalissuesinexcludingareasthatsharesimilarcharacteristicswiththetargetedsample,accountingforspilloverstonontar-getedareasaswellasforselectiveattrition,andensuringheterogeneityinparticipationandultimateoutcomes,eveniftheprogramisrandomized.LearningO
6、bjectivesAftercompletingthischapter,thereaderwillbeabletodiscuss■Howtoconstructanappropriatecounterfactual■Howtodesignarandomizedexperiment,includingexternalandinternalvalidity■HowtodistinguishtheATEfromtheTOT■Howtoaddresspracticalissuesinevaluatingrandomizedinterventions,includ-ingaccountingforsp
7、illovers,selectiveattrition,ethicalissues,andselectivehet-erogeneityinprogramimpactsamongthetreatedsample33HandbookonImpactEvaluationSettingtheCounterfactualAsarguedinchapter2,Tndingapropercounterfactualtotreatme
