Mapping poverty at multiple geographical scales

TL;DR

This paper presents a multi-scale Bayesian approach combining survey and remote sensing data for detailed poverty mapping across different geographical resolutions, improving accuracy and coherence.

Contribution

It introduces a novel hierarchical Bayesian model with a benchmarking algorithm for multi-resolution poverty estimation, integrating diverse data sources.

Findings

Effective in simulation studies

Provides coherent estimates across scales

Applied successfully to Bangladesh data

Abstract

Poverty mapping is a powerful tool to study the geography of poverty. The choice of the spatial resolution is central as poverty measures defined at a coarser level may mask their heterogeneity at finer levels. We introduce a small area multi-scale approach integrating survey and remote sensing data that leverages information at different spatial resolutions and accounts for hierarchical dependencies, preserving estimates coherence. We map poverty rates by proposing a Bayesian Beta-based model equipped with a new benchmarking algorithm that accounts for the double-bounded support. A simulation study shows the effectiveness of our proposal and an application on Bangladesh is discussed.