Non-normal Empirical Bayes prediction of local welfare

Estimates of the area- and household idiosyncratic error distributions from household income and consumption regression models across 142 household surveys from 16 different countries, the type of models that underpin poverty maps, points to significant deviations from normality. The poverty mapping approach put forward by Elbers et al. (2003) demonstrates the value of accommodating this non-normality. Unfortunately, their non-parametric estimator for the error distribution functions does not lend itself for Empirical Bayes (EB) prediction. This paper puts forward an approach that accommodates both EB prediction and non-normally distributed errors. Accounting for non-normality in Empirical Best estimation of local welfare is found to increase precision relative to normal-EB estimation. While the gains in precision range between substantial and marginal, it is always positive. Being able to obtain precise local estimates of welfare help governments and development agencies identify and monitor lagging areas and allocate their resources in a more efficient manner.