Global poverty estimation using private and public sector big data sources.

Area of Science:

• Geospatial analysis
• Development economics
• Data science

Background:

• Household surveys provide accurate poverty data but are expensive and infrequent.
• Estimating wealth and its changes globally requires innovative, scalable methods.

Purpose of the Study:

• To develop and validate a model using diverse data sources for estimating wealth levels and changes.
• To assess the model's generalizability across a wide range of countries.

Main Methods:

• Trained machine learning models on 63,854 survey cluster locations in 59 countries.
• Integrated data from satellites, Facebook Marketing, and OpenStreetMap.
• Evaluated model performance in explaining wealth variation at cluster and district levels.

Main Results:

• The model explained 55% of wealth level variation at the cluster level and 59% at the district level on average.
• Variation in wealth changes explained was 4% at the cluster level and 6% at the district level.
• Nighttime lights, OpenStreetMap, and land cover data were key predictors for wealth levels.

Conclusions:

• Combining multiple public and private data sources effectively estimates wealth levels and changes.
• The model shows strong performance, particularly in lower-income countries and those with high wealth variance.
• This approach offers a scalable and cost-efficient complement to traditional poverty measurement surveys.