世界银行-利用调查对调查的推断以低成本填补贫困数据缺口：来自随机调查实验的证据（英）-2024.3-78页

PublicDisclosureAuthorizedPolicyResearchWorkingPaper10738PublicDisclosureAuthorizedUsingSurvey-to-SurveyImputationtoFillPovertyDataGapsataLowCostPublicDisclosureAuthorizedEvidencefromaRandomizedSurveyExperimentHai-AnhDangTalipKilicVladimirHlasnyKseniyaAbanokovaCalogeroCarlettoPublicDisclosureAuthorizedDevelopmentEconomicsDevelopmentDataGroupMarch2024PolicyResearchWorkingPaper10738Abstractandalternative(quarterlyormonthly)ConsumerPriceIndexdeflators.TheproposedapproachtoimputationalsoSurveydataonhouseholdconsumptionareoftenperformsbetterthanmultipleimputationandarangeofunavailableorincomparableovertimeinmanylow-andmachinelearningtechniques.Inthecaseofatargetsurveymiddle-incomecountries.Basedonauniquerandomizedwithmodified(shortenedoraggregated)foodornon-foodsurveyexperimentimplementedinTanzania,thisstudyconsumptionmodules,imputationmodelsincludingfoodoffersnewandrigorousevidencedemonstratingthatornon-foodconsumptionaspredictorsdowellonlyifthesurvey-to-surveyimputationcanfillconsumptiondatadistributionsofthepredictorsarestandardizedvis-à-visgapsandprovidelow-costandreliablepovertyestimates.thebasesurvey.Forthebest-performingmodelstoreachBasicimputationmodelsfeaturingutilityexpenditures,acceptablelevelsofaccuracy,theminimumrequiredsampletogetherwithamodestsetofpredictorsondemographics,sizeshouldbe1,000forbothth...