BensonAdomako/Crop-Yields-and-Conflict

Replication code and data for: Crop Yield Shocks and Conflict in Sub-Saharan Africa # Crop Yield Shocks and Conflict in Sub-Saharan Africa Replication code and data for: > **Crop Yield Shocks and Conflict in Sub-Saharan Africa: Subnational Evidence from Satellite-Based Yield Predictions** > Benson Adomako (2026) ## Overview This paper estimates the effect of satellite-predicted maize yields on violent conflict across seven Sub-Saharan African countries (Burkina Faso, Ethiopia, Mali, Malawi, Niger, Nigeria, Tanzania) using an 18,096 Admin-2 district-year panel from 2010 to 2024. Predicted yields are generated from an XGBoost model trained on GROW-Africa administrative yield records and 224 satellite/environmental features, providing a measure of agricultural productivity that is substantially less contaminated by reverse causality than observed yields. **Key findings:** - Higher predicted maize yields reduce per-capita conflict: a 10% yield increase is associated with an approximately 30.5% reduction in the per-capita conflict rate in the 3-month post-harvest window. - This relationship reverses in the Sahel (Burkina Faso, Mali, Niger), where higher yields are associated with more conflict, consistent with a resource-predation channel in which armed groups appropriate agricultural surplus. - The effect is concentrated in violence against civilians and battles rather than riots, consistent with the opportunity-cost mechanism operating at the armed-actor margin. ## Repository Structure ``` Code/ build_panel_v3.R # Step 2: Build conflict-yields panel from raw data Reg_analysis.r # Step 3: Run all regressions, generate LaTeX tables generate_predictions_v2.py # Step 1: Generate yield predictions (runs on remote server) figures_gen.R # Generate thesis figures extract_feature_importance.py # Extract XGBoost feature importance table build_presentation.py # Build thesis defence slides archive/ # Exploratory and superseded scripts Data/ conflict_yields_panel_v3.csv # Main ana …