Bureaucrat incentives reduce crop burning and child mortality in South Asia

Air pollution kills over 7 million people globally each year and impairs child development, education, and productivity. South Asia faces one of the largest public health emergencies from air pollution: a quarter of the world’s population is exposed to hazardous air. Crop residue burning contributes 40–60% of air pollution during the winter harvest months in northern India and is widely practiced despite being illegal. Yet, how state institutions and administrators control crop burning remains poorly studied. Prior work has paid limited attention to the conditions under which governments successfully implement environmental policy, and the bureaucracy literature is only beginning to examine whether bureaucrats can innovate to address environmental degradation. Theoretically, dispersed farmers and limited marginal gains from penalizing any single violator may weaken bureaucrats’ incentives. In practice, many consider crop burning control intractable. This paper studies whether and under what conditions bureaucrats exert effort to reduce pollution. Using 10 years of satellite and administrative data at a 5 km² grid (about 18 million observations) across India and Pakistan, we examine how bureaucratic incentives shaped by wind-driven pollution spillovers affect enforcement, whether punitive actions deter future burning, and how exposure to crop-burning-induced PM2.5 in utero affects infant and child mortality. With estimates of fires’ impacts on mortality, we provide a back-of-the-envelope calculation of potential mortality reductions if bureaucrats acted uniformly where incentives to protect their own jurisdiction are strongest.

The study builds on and contributes to several strands of literature. First, it relates to research on environmental externalities and cross-jurisdiction spillovers, which shows that regulators may under-enforce when pollution harms others (e.g., industrial and water pollution at administrative borders). Second, it contributes to the political economy of the state and bureaucracy, particularly on career incentives, overload, and local embeddedness shaping bureaucratic performance. Third, it connects to a large literature on air pollution’s impacts on infant and child mortality and broader human capital outcomes. The paper addresses gaps by focusing on the conditions under which subnational bureaucrats act to manage an environmental problem created by dispersed agents, and by linking bureaucratic incentives directly to environmental outcomes (fires) and to health outcomes (mortality) using high-frequency, spatially resolved data and an instrumental variables strategy.

Setting and data: The analysis covers northern India and Pakistan from 2012–2022, when high-resolution fire data are available. Data are aggregated to a 5 km² grid observed monthly, yielding approximately 17,979,600 grid-month observations. The study combines satellite fire detections, wind direction and speed, administrative district boundaries, road networks, crop type/area (e.g., rice), and district-month records of severe enforcement actions (criminal cases/FIRs) against stubble burning. For health, the authors use geolocated Demographic and Health Surveys (DHS) birth histories, Copernicus Atmosphere Monitoring Service (CAMS) PM2.5, and Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) atmospheric dispersion modeling to trace fire-originating particulate trajectories and to construct in utero exposure measures.

Bureaucratic incentives design: Treatment is defined by exogenous temporal variation in expected wind direction interacting with the geometry of districts. For a given grid cell and month, the authors determine whether potential smoke from a fire would predominantly affect the home district (upwind relative to district interior) or a neighbouring district (downwind relative to district interior). The same grid cell can switch status across months as wind patterns change. This yields within-cell treatment variation independent of local secular trends.

Estimation for fires: A difference-in-differences model compares grid cells as they switch from primarily polluting a neighbour (control) to primarily polluting the home district (treatment), with grid and time fixed effects and controls as described in the Methods (parallel trends supported by event-study pre-trends). An event-study specification quantifies dynamics before and after treatment switches (sample size ~12,883,548 within a 6-month window). A border-distance analysis on a grid-cell-by-border-segment monthly dataset (~115,149,943 observations) interacts treatment with distance quintiles to district borders to test whether effects intensify where incentives are strongest (closest to borders). Heterogeneity tests examine capacity constraints (share under rice), legibility (proximity to main roads), harvest months, and rice-growing areas. Robustness includes alternative functional forms, transformations, and relaxed identifying assumptions.

Deterrence analysis: District-month measures of severe bureaucratic punishment (criminal cases for burning) are linked to subsequent fires. A triple-difference/event-study design compares fires in areas that primarily pollute the home district (where attention/incentives are higher) versus areas that pollute neighbours, before and after a penalty event within the same district. Seasonality in penalties (visibility-driven) is documented.

Mortality analysis: For each birth, the authors construct in utero exposure to local PM2.5 (CAMS) across the pregnancy months and separately compute exposure to crop-burning particulates by summing HYSPLIT-traced particles from all upwind fires intersecting the DHS cluster’s trajectory footprint during gestation. Two approaches are used: (1) reduced form linking HYSPLIT particle exposure to infant/child mortality; and (2) instrumental variables where the instrument is the log number of HYSPLIT-traced particles from upwind fires affecting the pregnancy, and the endogenous regressor is the log of in utero PM2.5. Regressions include district, birth month-year, and birth order fixed effects and control for fires within 10 km of the DHS cluster; standard errors are clustered at district and birth-date levels. The main sample includes n=542,150 births.

Bureaucratic incentives and fires:

• Event-study shows flat pre-trends and immediate post-switch reductions in fires when wind shifts to pollute the home district; a 9% decrease right after the switch (effect = −0.0086, P=0.024) and a 22.24% decrease two months after (−0.0214, P<0.001).
• Difference-in-differences pooled effects: fires decrease by 10–13% (P<0.001) when a location switches from polluting a neighbour to polluting the home district, translating to roughly 54–72 fewer fires per district annually. Effects are about four times larger in harvest months and in rice-producing areas (treatment effect = −0.046, P=0.003).
• Border-distance design: Approaching the upwind border (maximum home-district exposure) reduces fires by 14.47% (−0.0113, P<0.001), while approaching the downwind border (maximum neighbour exposure) increases fires by 15.11% relative to the upwind effect (difference = 0.0231, P<0.001). These effects intensify monotonically nearer to borders.
• International borders amplify responses: at the India–Pakistan border, fires fall by 56.22% near the upwind border (P<0.001) and rise by 146.37% near the downwind border (P<0.001), consistent with limited cross-border coordination.
• Capacity and legibility: Effects are larger where capacity constraints bind (higher rice share), with an additional 12% change in the quintile closest to the upwind border (P=0.003), and are stronger where bureaucrats have greater legibility (closer to main roads).

Deterrence and enforcement:

• Bureaucratic penalties are 61.63% higher (P=0.000) during winter harvest months when pollution is most visible, with no corresponding increase in summer harvest months.
• Punishment deters future burning: following a penalty, areas that pollute the home district experience an additional 9–13% reduction in fires (concentrated in the first 3 months) compared to other areas within the same district, above and beyond baseline incentive effects (pooled post-period coefficient = −0.0284, P=0.008; pre-period not different from zero).

Health impacts of crop burning:

• Reduced-form: a 1-log increase in particles from crop burning increases infant deaths in the first year by 0.17 (P=0.008) and under-five deaths by 0.21 (P=0.001).
• Instrumental variables: a 1-log increase in in utero PM2.5 from crop burning raises infant mortality by 24–26 per 1,000 births (P=0.016 to 0.010) and child mortality by 30–36 per 1,000 births (P=0.005 to 0.004). Relative risk increases are 64.3–69.8% (infant) and 74.2–86.2% (child). A 10 µg/m³ increase in average daily PM2.5 raises infant mortality by 7.8% and child mortality by 8.5%.
• Back-of-the-envelope: 1.8–2.7 deaths per 1,000 (4.4–6.6% of average child mortality) could be prevented if bureaucrats enforced against burning throughout their jurisdictions as they do where smoke would affect their own district.

The findings directly address whether bureaucratic incentives can mitigate crop residue burning and improve public health. When winds imply that burning will pollute the bureaucrat’s own district, fires fall, while they rise when pollution is externalized to neighbours, indicating that bureaucrats respond to incentives. Enforcement actions, though targeting only a subset of farmers, create meaningful deterrence for others, easing the challenge of addressing a problem generated by many dispersed actors. The mortality estimates demonstrate that pollution from crop burning during pregnancy has large impacts on infant and child survival, underscoring the health benefits of improved enforcement. Together, these results challenge the view that crop burning control is intractable and show that existing state capacity—when properly incentivized—can meaningfully reduce fires and associated mortality. The study highlights the importance of aligning bureaucratic incentives and attention with public health objectives and suggests that strategic, well-timed enforcement may yield outsized benefits.

This paper provides large-scale evidence that bureaucratic incentives reduce crop residue burning and, through lower in utero exposure to PM2.5, can reduce infant and child mortality in South Asia. Fires decrease when winds direct pollution to a bureaucrat’s own jurisdiction and increase when pollution is externalized; enforcement actions produce measurable deterrence; and crop-burning-driven PM2.5 substantially raises infant and child mortality, implying notable potential health gains from improved governance. Future research should examine: (1) how bureaucrats dynamically calibrate actions at high frequency as incentives change; (2) the internal organization and cross-department coordination within district administrations that drive success; and (3) the incentive structures for farmers, including the optimal design and timing of penalties and alternatives (carrots and sticks) that are scalable and cost-effective, and how these interact with major agricultural policies.

The analysis is observational and relies on plausibly exogenous variation from wind-direction changes and district geometry; while event-study pre-trends support parallel trends, causality cannot be established with experimental certainty. Some measures (e.g., PM2.5) are subject to measurement error; the instrumental variables strategy mitigates this but identifies a local average treatment effect for populations whose PM2.5 exposure is shifted by upwind fires. Generalization beyond the study region and period should be made cautiously, and coordination challenges at international borders may limit policy transferability.