Microlevel structural poverty estimates for southern and eastern Africa.

Area of Science:

• Environmental Science
• Computer Science
• Development Economics

Background:

• Poverty estimation in the Global South is often infrequent and spatially coarse, hindering effective policy interventions.
• Existing Earth observation and machine learning methods excel at asset-based wealth but struggle with consumption-based poverty measures.
• Accurate, granular poverty data is crucial for understanding socioeconomic dynamics and targeting aid.

Purpose of the Study:

• To pilot a novel two-step approach for granular poverty estimation using Earth observation and machine learning.
• To bridge the gap between asset-based wealth indices and consumption-based poverty measures.
• To enhance the policy-relevance and interpretability of machine learning-based poverty estimation.

Main Methods:

• Combined Earth observation data with accessible machine learning techniques.
• Employed an asset-based structural poverty measurement framework.
• Utilized a two-step approach for poverty estimation in four southern and eastern African countries.

Main Results:

• The approach explained 72–78% of cluster-level variation in a pooled model.
• Out-of-country predictions achieved 40–54% explanatory power.
• The method preserves interpretability and policy-relevance of consumption-based poverty measures.

Conclusions:

• The piloted approach effectively enhances granular poverty estimation in data-scarce regions.
• This method offers a viable solution for improving the targeting of humanitarian and development interventions.
• Integrating Earth observation and machine learning provides a powerful tool for socioeconomic analysis.