Pull Your Small Area Estimates Up by the Bootstraps

Corral, Paul ; Molina, Isabel ; Nguyen, Minh

World Bank, Washington, DC

关键词: POVERTY MAPPING;    SMALL AREA ESTIMATE;    ELL;    EMPIRICAL BEST;    PARAMETRIC BOOTSTRAP;   

DOI  :  10.1596/1813-9450-9256
RP-ID  :  WPS9256

学科分类：社会科学、人文和艺术（综合）

来源: World Bank Open Knowledge Repository

【 摘 要 】

After almost two decades of poverty mapsproduced by the World Bank and multiple advances in theliterature, this paper presents a methodological update tothe World Bank's toolkit for small area estimation. Thepaper reviews the computational procedures of the currentmethods used by the World Bank: the traditional approach byElbers, Lanjouw and Lanjouw (2003) and the EmpiricalBest/Bayes (EB) addition introduced by Van der Weide (2014).The addition extends the EB procedure of Molina and Rao(2010) by considering heteroscedasticity and includes surveyweights, but uses a different bootstrap approach, herereferred to as clustered bootstrap. Simulation experimentscomparing these methods to the original EB approach ofMolina and Rao (2010) provide empirical evidence of theshortcomings of the clustered bootstrap approach, whichyields biased point estimates. The main contributions ofthis paper are then two: 1) to adapt the original MonteCarlo simulation procedure of Molina and Rao (2010) for theapproximation of the extended EB estimators that includeheteroscedasticity and survey weights as in Van der Weide(2014); and 2) to adapt the parametric bootstrap approachfor mean squared error (MSE) estimation considered by Molinaand Rao (2010), and proposed originally by González-Manteigaet al. (2008), to these extended EB estimators. Simulationexperiments illustrate that the revised Monte Carlosimulation method yields estimators that are considerablyless biased and more efficient in terms of MSE than thoseobtained from the clustered bootstrap approach, and that theparametric bootstrap MSE estimators are in line with thetrue MSEs under realistic scenarios.