← PLSC 441 · The Politics of Climate Change in Developing States River Lab

Political economy of the environment · Field lab

The Exit Border

A river runs through several jurisdictions. Each official weighs the harm to their own residents downstream, and nobody else's. That is why the dirtiest water sits just upstream of a border. Redraw the county lines, as Brazil did, or change what gets officials promoted, as China did in 2006, and watch where the enforcement and the pollution go.

Lipscomb, M. & Mobarak, A. M. “Decentralization and Pollution Spillovers: Evidence from the Re-drawing of County Borders in Brazil.” Review of Economic Studies 84 (2017): 464–502. Kahn, M. E., Li, P. & Zhao, D. “Water Pollution Progress at Borders: The Role of Changes in China's Political Promotion Incentives.” AEJ: Economic Policy 7(4) (2015): 223–242.
Case
—
River pollution
0
Where officials enforce, and why

What the river carries

County map—

Click the map to draw a new county border. Drag a border's handle to move it, or double-click it to erase. Handles also take arrow keys and Delete.

Start from
Pollutant
Where people live
Station pairS1 → S2

01What the papers found

Every effect the simulation shows is tied to one of these published estimates. The numbers below come from the papers, not from the model.

Brazil: Lipscomb & Mobarak (2017)
+2.1%
BOD for every kilometre a station sits closer to its county's exit border (Table 2; 3.1–3.4% with station-specific trends). This is Prediction 1.
1.4→2.0%
Increase per km at 10 km versus 1 km from the border. Pollution rises at an increasing rate as the river nears the exit (Prediction 2).
+2.1 / −3.3
The slope of the pollution function in % per km on either side of the border. It flips sign once the river enters the next county (Prediction 3, p ≈ 0.07–0.10).
+3.2%
Extra BOD for each additional county border crossed between two stations (Table 3). It is 4.8–5.5% when county splits are controlled for directly (Table 6). This is Prediction 4.
0.3%
The per-km border gradient where a water-basin committee operates, against 2.1% where none does (Table 11).
none
No border pattern for turbidity, dissolved solids or conductivity (Table 10). These come from diffuse erosion, which a county cannot steer.
lights
Night-time lights grow fastest in the downstream 10% of a county after a split (Table 8). This fits informal settlement being allowed near the exit.

Data: 372 station pairs, 5,989 observations, 1990–2007. Brazil's county count grew from 4,492 (1991) to 5,807 (2005).

China: Kahn, Li & Zhao (2015)
12.6 vs 7.4
Mean COD (mg/L) at provincial-boundary stations versus interior stations in 2004. This is the free-riding baseline.
−54% / −39%
Fall in COD from 2004 to 2010 at boundary and non-boundary stations (Table 1).
−1.89
Boundary × Post-2005 (mg/L COD), Table 2. It is about −2.0 with station fixed effects (Table 3). The effect grows each year to −3.9 in 2010 (Table 4).
+0.063
Boundary × trend × governor age (Table 5): −4.199 + 0.063·age per year. The border effect fades out for governors near 67. Party secretaries' age does not matter.
COD only
COD and BOD improve at borders. Petroleum, mercury and phenol, which were not in the promotion criteria, show no differential progress (Table 4).
32.6%
Pulp and paper's share of industrial COD, from 2.5% of output. Mills opened in 2006–08 sit farther from boundary stations than those opened in 2003–05 (Fig. 4).

Data: 499 central-government monitoring stations, 127 of them on a provincial boundary, 2004–2010, 3,377 observations.

02Your assignment

Work through these with the controls above. Answers should cite the numbers the simulation gives you and the estimate from the paper they correspond to.

Questions
  1. The exit spike. In Brazil, choose One county, uniform population. Where on the enforcement chart does the blue “own residents” band thin out? Where does BOD rise fastest? State Predictions 1 and 2 in your own words using what you see.
  2. Split it. Choose Split at 50 km. Drag Station 1 from 20 km toward the border at 50 km and read the local slope at each step. Then place Station 2 just past the border. Where is the structural break (Prediction 3)? Why does the paper find a negative slope just downstream of a border?
  3. Count the crossings. Keep S1 at 20 km and S2 at 90 km. Add borders between them one at a time and record Δ ln BOD. The model's effect per crossing is much larger than the paper's +3.2%. Lower “discharge the county can steer” until the two agree. What does that number say about how much of BOD a Brazilian county really controls?
  4. Endogenous splits. Switch population to City at the split, then Two towns. Which of the four predictions still hold? Appendix D argues that a spurious “splits happen where density is rising” story could not produce all four. Use the charts to explain why.
  5. Coase on the river. Raise the basin-committee weight to 50%, then 100%. What happens to the exit spike and to river-mean BOD? Table 11 says the border gradient drops from 2.1% to 0.3% where committees operate. What is the committee buying?
  6. The 2006 switch. In China, set the year to 2004 and note the boundary–interior gap. Press Play. Which band of the enforcement chart appears in 2006, and where along each province is it concentrated? Compare your simulated differential change with the paper's −1.9 mg/L.
  7. Career horizons. Set Chongqing's governor to 48 and Hubei's to 64. Which boundary station improves most by 2010? Use Table 5's −4.199 + 0.063·age to compute each governor's predicted yearly effect. Why would age matter for the governor but not for the party secretary?
  8. Teaching to the test. Switch the pollutant to mercury, then phenol. Relate what you see to Holmström and Milgrom's multitask problem. Then compare it with Brazil's turbidity result. Both are “no effect” findings. Why do they mean different things?
  9. Where the mills went. Watch the pulp-mill glyphs from 2005 to 2010. Moving a mill upstream lowers the boundary reading. Does it lower pollution, or just move it onto the province's own residents? Kahn et al. also consider hidden pipelines that bypass the station (fn. 21). How would each story show up in the station data?
  10. Two fixes. Brazil shows what decentralisation costs. China shows a central government using career incentives to buy cleaner borders. Using both cases, argue for or against this claim: “Border spillovers are a monitoring problem, not a jurisdiction problem.”
How this model works

The official's problem. This is Lipscomb & Mobarak's model, their eq. 1–2. A river runs along a line and people live along it with density f(x). At each point the official picks a discharge allowance q, trading residents' utility (log) against the harm to people downstream within the same jurisdiction. Harm decays at rate α as the water flows. The first-order condition sets a shadow price 1/q = c + ∫xb e−α(t−x) f(t) dt / f(x). That shadow price is the stacked area in the enforcement chart. Near the exit border b the integral shrinks to nothing, enforcement collapses to the private cost c, and discharge spikes.

What each control adds to the price. The basin committee adds a weight w on residents past the border (green). In China, the post-2006 rules add two terms. One is a province-wide COD target (ochre). The other is a penalty on the reading at the downstream boundary station (red). Its marginal effect is e−α(b−x), so it bites hardest exactly where free-riding was worst. The boundary penalty scales with (66.6 − age)/11.6, from Table 5, and phases in from 2006 to 2010 as the paper's year coefficients do. Untargeted pollutants get no red or ochre term.

Point versus diffuse sources. A share κ of each pollutant follows the official's allowance. The rest is diffuse and simply scales with population (farm runoff, erosion). Turbidity and ammonia get a low κ, as they do in the papers.

Units. Brazilian BOD is scaled so a single, undivided county averages the sample mean of 3.5 mg/L. The 100 km river makes mid-county slopes land near 2%/km. Chinese readings are scaled so 2004 interior stations match each pollutant's Table 1 mean. Pulp mills sit at the quantiles of the allowance each province offers, since a mill goes where enforcement is lax.