Small Area Estimation of Non-Monetary Poverty with Geospatial Data

Abstract

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.

A Map of the Poor or a Poor Map?

“…In other simulation experiments run, but not shown here, all the covariates' aggregates at the cluster level are used and similar results are obtained. The results shown here are not evident under the model based simulation conducted in Masaki et al [12], page 36, because under the simulation presented here, true welfare is generated from household level covariates as is likely the case in real world scenarios. In Masaki et al [12], the authors chose to model the dependent variable using only 2 subdomain level covariates, which are constant for all households in the subdomain.…”

“…The results from design-based validation show that all small area estimation methods represent an improvement, in terms of MSE, over direct estimates. However, given the inherent high level of bias of methods relying solely on aggregate information (see Nguyen [9]; Lange et al [11], Masaki et al [12]) these should be used with caution and only under certain circumstances.…”

“…Lange et al [11] present an approach similar to that of Nguyen [9], which the authors suggest as an alternative in cases when census and survey data are not from similar periods, though the same issues noted above for the ELL method would likely persist in a model using only area-level covariates. Masaki et al [12] use a similar modeling approach to Nguyen [9], but take measures to address some of the shortcomings of a standard ELL approach and obtain EB estimators of Molina and Rao [5], which appear to be more precise. The authors conduct a design-based validation study using a wealth index constructed with principal component analysis using census data for Sri Lanka and Tanzania.…”

“…Unit-context two-fold nested error Census EB estimators τUC−CEB ac a based on a twofold nested error model with random location effects at the area and cluster level. This estimator follows the approach from Masaki et al [12] and Nguyen [9], where only area means for some of the right hand side variables are used. Specifically, the covariates used in this model are x1ac , x3a , x4ac , x5a and x7ac .…”

“…1. UC − CEB a : Unit-context model originally proposed by Nguyen [9] but with EB, similar to Masaki et al [12]. Model fit is done using Henderson's Method III (with sampling weights) and estimates are obtained using CensusEB as noted in Corral et al [16].…”