在Stata中实现平均处理效应匹配估计方法研究.pdfVIP

TheStataJournal(2004)

4,Number3,pp.290–311

Implementingmatchingestimatorsforaverage

treatmenteffectsinStata

AlbertoAbadie1DavidDrukkerJaneLeberHerrGuidoW.Imbens2

HarvardUniversityStataCorpUCBerkeleyUCBerkeley

.Thispaperpresentsanimplementationofmatchingestimatorsforaverage

treatmenteffectsinStata.Thennmatchcommandallowsyoutoestimatethe

averageeffectforallunitsoronlyforthetreatedorcontrolunits;tochoosethe

numberofmatches;tospecifythedistancemetric;toselectabiasadjustment;andto

useheteroskedastic-robustvarianceestimators.

Keywords:st0072,nnmatch,averagetreatmenteffects,matching,exogeneity,un-

confoundedness,ignorability

1Introduction

Thispaperprovidesabriefintroductiontomatchingestimatorsforaveragetreatment

effectsanddescribesthenewStatacommandnnmatchwhichimplementstheseesti-

mators.Thecommandimplementsnearest-neighbormatchingestimatorsforaverage

treatmenteffectsforeithertheoverallsampleorasubsampleoftreatedorcontrolunits.

Whilesimplematchingestimatorshavebeenwidelyusedintheprogrameval-uation

literature,nnmatchimplementsthespecificmatchingestimatorsdevelopedinAbadie

andImbens(2002),includingtheirbias-correctedmatchingestimator.Follow-ing

AbadieandImbens(2002),nnmatchallowsindividualobservationstobeusedasa

matchmorethanonce.Comparedwithmatchingwithoutrecement,thismethod

generallylowersthebiasbutincreasesthevariance.

W