Decomposing Causal Effects in a Dynamic Interdependence System: Pollution, Income, and Conservation in U.S. Watersheds

This paper estimates the dynamic relationship among conservation policy, water quality, and local income using county-year data on USDA Environmental Quality Incentives Program (EQIP) spending, nitrogen concentration, and income. We develop a five-stage identification strategy that combines two-way fixed effects, pre-program double/debiased machine learning residualization, instrumental variables, a triangular 3SLS system, local projections, and a longrun decomposition of direct and income-feedback channels. Preliminary results show that EQIP water-practice spending has a negative but statistically insignificant contemporaneous effect on nitrogen concentration. Local projection estimates provide limited evidence of persistent pollution reductions. However, alternative specifications that restrict the sample to later stages of the program, particularly the post-1996 and post-2001 periods, suggest that EQIP is associated with a small but persistent increase in county income. The income-feedback channel may contribute to an initial increase in county-level pollution, and these results hold after controlling for upstream nitrogen pollution and population changes.