Political economy of the environment · Field lab
Drop a fire anywhere on the plain, then turn the wind and see who inherits the smoke. A district officer who breathes their own farmers' stubble has a reason to stop it; one whose smoke crosses a border does not. This is the incentive Dipoppa & Gulzar measured — here it is as a dial and a draggable match.
Built on Dipoppa, G. & Gulzar, S. “Bureaucrat incentives reduce crop burning and child mortality in South Asia.” Nature 634, 1125–1131 (2024).
Distance from centre = share of this district's smoke that settles on its own people. The needle marks the current wind.
One row per district. Retained is the share of a district's own burning smoke that lands on its own population — the fraction of the externality its officer internalises. Click any row to inspect it on the map; click a column head to sort.
| District | Population | Fires / mo | Retained | Leaves region | Imported PM | Δ fires | PM2.5 | Infant deaths |
|---|
Every number this app moves is anchored to a published estimate. These are the paper's, not the model's.
The authors hold the farmer, the field and the officer fixed and let the wind move. The same 5 km² cell is “treated” in a month when the downwind half of its district is larger than the upwind half — smoke will mostly fall on the officer's own jurisdiction — and “control” when the wind turns and that smoke would blow next door.
Because a monsoon reversal is not chosen by any district officer, the parallel-trends assumption is far more credible than in a usual difference-in-differences. The event study in Fig. 2d shows a flat pre-trend and a drop bottoming out at −22.2% two months after the switch.
That is the whole argument: the same people, the same fields, a different set of lungs downwind — and 10–13% fewer fires. Institutions are not fixed; incentives are.
Work through these with the controls above. Answers should cite specific districts and numbers from the ledger.
Treatment. For every grid cell the app draws a 180° line through the cell perpendicular to the wind and clips the district polygon against it — exactly the authors' definition. If the downwind slice is the larger one, the cell is treated.
Response, in two parts. The level comes from the paper's pooled estimate: a treated cell burns 11.5% less, the midpoint of its 10–13% range. The pattern comes from the border gradient — each cell finds its nearest district border, asks whether that border is upwind or downwind, and takes the paper's coefficient ramped linearly across the five distance-to-border quintiles: −14.47% at the closest upwind quintile, +15.11% at the closest downwind one, fading to zero at the farthest. Cells whose nearest border is the international one use the far larger cross-national coefficients, which is why Lahore and Sheikhupura can burn more even while keeping much of their own smoke.
Exposure. Each burning cell emits a Gaussian plume downwind whose reach scales with wind speed, scored against every population node — plus a ring of receptors outside the frame, so smoke that leaves the region is counted as having gone somewhere rather than quietly disappearing. The result is a who-pollutes-whom matrix. Retained is its diagonal.
Mortality. Δln PM2.5 over a 45 µg/m³ background, multiplied by the paper's instrumental-variables estimates (25.0 infant and 30.8 child deaths per 1,000 births per log point). Read these as this month's exposure held across a whole pregnancy — an upper bound, not an annual figure.