Retooling Poverty Targeting Using Out-of-Sample Validation and Machine Learning

McBride, Linden ; Nichols, Austin

World Bank, Washington, DC

关键词: targeting;    proxy means testing;    poverty;    poverty assessment;   

DOI  :  10.1596/1813-9450-7849
RP-ID  :  WPS7849

学科分类：社会科学、人文和艺术（综合）

来源: World Bank Open Knowledge Repository

【 摘 要 】

Proxy means test (PMT) poverty targetingtools have become common tools for beneficiary targeting andpoverty assessment where full means tests are costly.Currently popular estimation procedures for generating thesetools prioritize minimization of in-sample predictionerrors; however, the objective in generating such tools isout-of-sample prediction. This paper presents evidence thatprioritizing minimal out-of-sample error, identified throughcross-validation and stochastic ensemble methods, in PMTtool development can substantially improve the out-of-sampleperformance of these targeting tools. The USAID povertyassessment tool and base data are used for demonstration ofthese methods; however, the methods applied in this papershould be considered for PMT and other poverty-targetingtool development more broadly.