Do administrative boundary reforms in China reduce pollution or just reallocate it?

Two reforms move authority in opposite directions

China's county-level territories hold one of three administrative statuses, and the status determines the fiscal contract with the prefecture, land-quota allocation, who approves industrial projects, and where the local environmental protection bureau sits in the hierarchy. The paper studies reforms that change this status while leaving territory, population, and firms physically in place, which is what makes the setting useful: the governance boundary moves, the geographic boundary does not.

County-to-district reform (chedi she qu) absorbs a county into its prefecture-level city and converts it into an urban district, shifting planning, land, and fiscal authority upward. The authors treat this as administrative centralization. County-to-city reform (chedi she shi) upgrades a county to a county-level city, keeping county-level authority but adding urban status, more autonomy over land conversion and industrial development, and explicit urbanization mandates. The authors treat this as local growth empowerment. A third reform, city-to-district, also shifts authority upward and is used as a robustness arm rather than a third primary treatment.

The two mechanisms predict opposite pollution responses. Centralization can internalize environmental externalities and strengthen enforcement, so incumbent firms should cut emission intensity through cleaner production without necessarily changing scale. Empowerment strengthens growth incentives, so pollution should rise through dirty entry and intensity-increasing expansion. Because both operate on the governance boundary, local changes need not correspond to regional ones.

A border-pair design compares firms across one border

The core evidence comes from a firm-year pollution panel covering 1998-2013 with 694,835 firm-year observations, built from the Chinese Annual Survey of Industrial Firms matched to the Environmental Statistics Database. It records six pollutants: SO2, COD, NOx, industrial dust, wastewater, and waste gas. SO2, COD, NOx, and dust are measured in kilograms, wastewater in tons, and waste gas as a volume measure that is excluded from monetary damage calculations.

The border-pair design compares firms on opposite sides of the same county border before and after reform, absorbing border-pair-by-year shocks and isolating the local treatment effect from anything common to both sides of that border. This matters for reading the numbers: the paper's headline local-abatement claim is anchored on the intensity margin, NOx per worker, which passes the joint pre-trend test (F=1.25, p=0.289), while NOx levels are only marginal (p=0.073).

Identification is uneven across the two reforms, and the paper says so. Of 168 county-to-city reforms in the roster, 71 of the 77 in the analytic sample occurred in 1993-1997, before the firm panel begins, so county-to-city effects are identified off a small set of late reforms. The stacked county-to-city level specification fails parallel trends for SO2, COD, dust, and wastewater (joint F=30-307), so those level effects are reported only from the border-pair design.

Centralization cuts NOx intensity without shrinking scale

County-to-district reform reduces NOx levels by 29.4 log points (-0.294, p<0.01) with no significant effect on SO2, COD, dust, or wastewater levels. The effect is concentrated in the clean-production margin: NOx per worker falls by 14.8 log points (-0.148, p<0.01), SO2 per worker by 14.9 log points (-0.149, p<0.01), and wastewater per worker by 12.9 log points (-0.129, p<0.10). Sales-denominated intensities point the same way.

The intensity framing disciplines the interpretation. This is not a story about shutting production down: employment rises (+0.052, p<0.01), sales fall only modestly (-0.071, p<0.10), firm count falls (-0.196, p<0.01), and exits fall (-0.208, p<0.01). The authors read this as restructuring toward a cleaner composition rather than broad contraction.

Industry heterogeneity reinforces that reading. The county-to-district intensity reductions are concentrated in high-polluting industries (DID x high-polluting industry = -0.153, p<0.01 for COD per worker), and the county-to-city intensity increases are concentrated in the same industries (+0.129, p<0.05). Abatement and dirty growth both land where pollution is already concentrated.

Empowerment raises COD, dust, and wastewater levels

County-to-city reform moves in the opposite direction. It raises COD, dust, and wastewater levels by 86.9, 94.5, and 92.6 log points respectively, and raises intensity as well: COD per worker (+0.275, p<0.05), dust per worker (+0.644, p<0.10), COD per sales (+0.045, p<0.05), and dust per sales (+0.151, p<0.05). The authors describe this as dirty growth.

The scale and selection margins differ from the centralizing reform. County-to-city reform generates firm churn rather than net contraction: entry rises (+0.231, p<0.01) and exit rises (+0.492, p<0.01), with no significant net change in firm count.

The mirror-image pattern across the two reform types is the paper's central fact. The same umbrella of administrative boundary adjustment produces opposite-signed effects depending on whether authority shifts upward or local growth autonomy is enhanced. The institutional path divergence, the difference between the two reform coefficients, is negative for NOx and positive for COD, dust, and wastewater.

Neighbor NOx rises along administrative, not geographic, lines

Local abatement is not net abatement. Following county-to-district reform, a reformed county's same-prefecture neighbors see a significant NOx increase (+0.773, p=0.004), while cross-prefecture neighbors do not (+0.034, not significant). The distance-band-by-prefecture interaction sharpens the pattern: the positive spillover is concentrated in same-prefecture counties 20-50 km away (+0.890, p<0.05), while cross-prefecture counties immediately across the border at 0-20 km experience a significant decline (-2.215, p<0.001).

That contrast is the paper's key spatial claim: reallocation follows administrative boundaries rather than geographic distance. Two independent data chains point the same way. In satellite ambient data, the 0-20 km ring shows ambient NO2 rising by 1.184 (p<0.01) and PM2.5 by 1.514 (p<0.01), while locally NO2 falls after reform (event-time coefficient at t=1 of -0.589, p<0.05).

Physical transport is ruled out as the explanation. Wind-cone falsifications assign neighbors to downwind, upwind, and crosswind cones of plus or minus 45 degrees from MERRA2 data; downwind effects are zero and crosswind placebo effects are zero, and the apparent upwind effect vanishes once fixed compass sectors are controlled for, which the authors attribute to bimodal monsoon winds under vector averaging. River data show no downstream deterioration for COD or ammonia, and the county-level far-downstream positive result disappears at the reach level. The reallocation is specific to air, and specifically to NOx.

Incumbent neighbor firms get dirtier, not new entrants

The reallocation does not run through firms moving or new plants appearing. Within same-prefecture neighbor counties, true entry and true polluting entry are insignificant when measured from business-registration founding years, a correction that matters because 92.4% of firms recorded as 1998 entrants were actually pre-existing, given the panel's left-censoring at its 1998 start.

Instead, incumbent neighbor firms raise NOx (+0.820, p<0.01) while holding output constant (+0.005, not significant) and reducing employment (-0.253, p<0.10). The authors characterize this as a deterioration of the clean-production margin among incumbents: same output, fewer workers, more NOx.

The enforcement channel cannot be tested in this window, and the paper reports that as a null rather than a finding. Environmental penalty records begin in 2011, and the central environmental inspection of 2016 and the environmental vertical-management reforms of 2016-2018 postdate the panel. The 11th Five-Year Plan total-control targets (2006-2010) serve as a within-window regulatory proxy, but the interaction effects are null or opposite-signed, with stricter targets associated with smaller reform-induced abatement, consistent with targets being assigned to initially dirtier counties. The mechanism analysis therefore rests on scale, selection, and clean production.

The welfare gain is indistinguishable from zero

The paper monetizes emission changes in a back-of-the-envelope calculation, converting kilograms to tons and applying China-specific health-damage parameters: the GEMM concentration-response function applied to age-specific NCD plus LRI mortality from GBD, population-weighted PM2.5 exposure, and a value of a statistical life of roughly 1 million yuan, with COD damages from aggregate water-pollution damage estimates. Marginal damages are in yuan per ton at 2003 prices.

For county-to-district reform, the only significant abatement is NOx, which monetizes to 47 million yuan per year, but the 95% confidence interval spans -3.3 billion to +6.0 billion yuan and crosses zero, so the welfare gain is statistically indistinguishable from zero. For county-to-city reform, the significant increases in COD and dust monetize to a damage of -2.3 billion yuan per year.

Neither reform produces a significant output loss, so a unit abatement cost cannot be computed. The authors treat that as a finding in itself: the observed abatement is not achieved at a measurable cost to local output. They also flag that these magnitudes are sensitivity calculations, because the air-pollutant marginal damages rely on generic intake fractions rather than China-specific source-receptor estimates.

Where the local abatement actually shows up

Exposure-weighted damage outcomes, constructed as emissions multiplied by county population, downwind population, and downstream population with seasonal splits, are uniformly insignificant for county-to-district reform. Reforms do not produce detectable population-weighted damage reductions, which is consistent with local abatement being offset by reallocation.

The one clean exposure finding is geographic within the county: local NOx abatement is concentrated among near-urban firms, within 10 km of built-up area, where NOx falls by 0.279 (p<0.10), versus -0.031 (not significant) for firms far from urban areas.

Land supply moves in the same direction as the reform. County-to-district reform significantly increases industrial land area (+0.628, p<0.01), residential area (+0.558, p<0.01), total area (+0.546, p<0.01), parcels (+0.522, p<0.01), and industrial land value (+1.309, p<0.01), with a stacked-DID specification passing pre-trend tests. The authors read this as consistent with a reform-to-development-to-pollution-reallocation chain, though the aggregate regional NOx increase they initially estimate (+0.50, p<0.05) disappears after matching on pre-treatment NOx levels and trends, because treated regions show a positive selection pre-trend. That is why the reallocation claim is anchored on the within-neighbor, administrative-boundary comparison, which is identified off never-treated counties.

The direction of the authority shift decides

The paper's practical priority is not whether boundary reform helps or hurts the environment in general, but which direction authority moves. Centralizing county-to-district reform lowers local NOx intensity by 14.8 log points through cleaner production without shrinking output or employment, yet the same-prefecture neighbors absorb a NOx increase of 0.773, the reallocation tracks administrative borders rather than distance, and the monetized welfare gain from that abatement has a confidence interval crossing zero.

Empowering county-to-city reform is the clearer case for a reader weighing costs: it raises COD, dust, and wastewater levels by 87-95% through dirty growth and monetizes to roughly 2.3 billion yuan per year in damages. The asymmetry is the point. If the question is whether these reforms reduce pollution, the evidence says they reshape its spatial configuration across administrative boundaries rather than reducing it, and the reform that produces measurable damage is the one that grants local growth autonomy.

Two boundaries on that reading are worth carrying forward. County-to-city effects rest on a small set of late reforms, since 71 of the 77 analytic-sample reforms occurred in 1993-1997 before the panel begins, and the enforcement channel is untestable in a 1998-2013 window whose penalty records start in 2011. Within those limits, the paper's most decision-relevant distinction is between the two institutional paths, not between reformed and unreformed places.

Sources used in this answer