Using machine learning to assess the livelihood impact of electricity access

Area of Science:

• Utilizes advancements in remote sensing and artificial intelligence for economic policy evaluation.

Background:

• Sparse economic data hinders effective public policy development and evaluation globally.
• Traditional methods struggle with data limitations in developing regions.

Purpose of the Study:

• To demonstrate how satellite imagery and machine learning (ML) can overcome data scarcity for policy analysis.
• To measure the causal impact of electricity access on local livelihoods in Uganda.
• To improve the reliability of causal inference in data-sparse environments.

Main Methods:

• Employs satellite imagery and computer vision to create local-level livelihood measurements.
• Applies machine learning (ML) inference techniques for causal impact estimation.
• Analyzes data from an electrical grid expansion in rural Uganda.

Main Results:

• Estimates that grid access increases village-level asset wealth in rural Uganda by 0.15 standard deviations.
• Electrification more than doubled the growth rate of asset wealth in treated areas compared to untreated areas.
• Demonstrates ML-based inference yields more reliable causal estimates than traditional methods.

Conclusions:

• Provides country-scale evidence on the economic impact of grid infrastructure investment.
• Offers a low-cost, generalizable methodology for policy evaluation in data-limited settings.
• Highlights the potential of integrated geospatial and ML approaches for socioeconomic research.