1 The Causal Question
What is the causal effect of long-run exposure to air pollution on human health and longevity? This question is one of the most consequential in environmental economics, yet it is extraordinarily difficult to answer. The fundamental problem is confounding: areas with high air pollution tend to differ from clean-air areas in income, industrial structure, infrastructure, and many other dimensions that independently affect health. Simple cross-sectional comparisons of health outcomes in polluted versus clean regions confound the effect of pollution with these other factors.
An elegant natural experiment arises from an unusual Chinese government policy: the Huai River Heating Policy [Chen et al., 2013, Ebenstein et al., 2017]. During China's central planning era, the government subsidised coal for winter heating but only in the regions north of the Huai River. This policy created a sharp geographic discontinuity in particulate air pollution along the river's latitude, providing a regression discontinuity design for estimating the health effects of sustained exposure to coal combustion products.
2 The Identification Strategy
2.1 The Huai River Policy
China's central heating policy, established in the 1950s, provided free coal for winter heating to all areas north of the Huai River and Qin Mountains line (approximately 33°N latitude). Regions south of this line received no coal subsidy and instead relied on inefficient and relatively cleaner alternatives (wood, charcoal, small electric heaters).
The policy was not designed as a health experiment, and the Huai River is not a natural boundary for disease, culture, or economic development. The river runs roughly east-west through central China, and there is no reason to expect sharp discontinuities in health-relevant variables other than winter heating pollution on either side of the river.
2.2 The Geographic RDD
Let ri denote the latitude of city i relative to the Huai River (positive = north, negative = south, ri = 0 at the river). The identifying assumption is:
where Yi(0) is the potential health outcome absent the coal heating policy. This continuity assumption is plausible for a geographic boundary defined by a river: absent the subsidy policy, health outcomes should vary smoothly across the Huai River line.
The RDD estimand is the local average effect at the boundary:
In practice, the discontinuity is estimated by fitting local polynomial regressions on each side of the boundary, using optimal bandwidth selection procedures [Calonico et al., 2014].
2.3 The Running Variable and First Stage
The running variable is latitude relative to the river. The "first stage" is the discontinuity in total suspended particulates (TSP) concentration at the Huai River boundary. Chen et al. [2013] document that TSP concentrations are approximately 184 μg/m3 higher in northern than southern cities near the boundary a massive discontinuity, representing roughly 55% higher pollution in the north.
This large first stage is critical for interpretation: the reduced-form effect of crossing the boundary on health outcomes is driven primarily through the coal-heating pollution channel.
3 Data and Estimation
The primary data source in Chen et al. [2013] is Chinese city-level administrative data, including mortality records from the Disease Surveillance Point (DSP) system for the period
1981-2000, combined with air quality monitoring data. The unit of observation is a city (prefecture-level), with latitude serving as the running variable.
Ebenstein et al. [2017] extend the original analysis in two important ways. First, they use a richer air quality dataset from 2004 to 2012, comprising approximately 80 monitoring stations, which allows more precise first-stage estimates. Second, they match air quality data to individual health records from the Chinese Longitudinal Healthy Longevity Survey, allowing person-level rather than city-level estimation.
4 Key Findings
4.1 Air Pollution Discontinuity
Chen et al. [2013] find that TSP concentrations jump sharply at the Huai River line: northern cities have approximately 184 μg/m3 higher TSP on average, a discontinuity that is visually dramatic and statistically decisive. Ebenstein et al. [2017] corroborate this, finding a PM10 discontinuity of roughly 46 μg/m3 using monitoring station data.
4.2 Life Expectancy
The reduced-form estimate the discontinuity in life expectancy at the Huai River boundary is approximately 5.5 years in the north relative to the south, as estimated by Chen et al. [2013]. That is, individuals living just north of the Huai River have life expectancy about 5.5 years shorter than otherwise comparable individuals just south of the river.
Ebenstein et al. [2017] refine this estimate using individual-level data and find that each additional 10 μg/m3 of PM10 exposure reduces life expectancy by approximately 0.60 years, with most of the mortality burden operating through cardiorespiratory disease disease, stroke, and chronic obstructive pulmonary disease (COPD).
4.3 Mortality Channels
The effect is concentrated in cardiorespiratory mortality, consistent with the epidemiological literature on the mechanisms of particulate matter toxicity. Mortality from accidents and other non-respiratory causes shows no discontinuity at the Huai River line, providing a falsification check: if unobserved confounders were driving the result, they should affect all causes of death, not selectively cardiorespiratory causes.
Table 1: Huai River Policy: Summary of Key Findings
5 Placebo Tests and Robustness
Chen et al. [2013] and Ebenstein et al. [2017] conduct several robustness checks:
Latitude placebo tests. Using alternative "cutoffs" at other latitudes (where no policy boundary exists) produces no significant discontinuities in TSP or life expectancy, confirming that the effects at the true Huai River boundary are not artifacts of local polynomial extrapolation.
Non-respiratory mortality placebo. As noted above, the boundary effect is specific to cardiorespiratory causes, consistent with air pollution as the mechanism.
Income and development controls. When city-level income, industrial structure, and other development indicators are added as controls, the estimated discontinuity in health outcomes is largely unchanged, suggesting they are not confounding the relationship.
Alternative bandwidths. The life expectancy discontinuity is robust to a range of bandwidth choices and to both local linear and local quadratic polynomial specifications.
6 Limitations
Spillovers and air quality diffusion. Air pollution does not respect administrative boundaries. Cities just south of the Huai River may receive some pollution from northern coal combustion, particularly in winter. This "spillover" would attenuate the measured discontinuity in both pollution and health outcomes, suggesting that the true effect could be even larger.
Compound treatment. The Huai River policy affected not just air quality, but also indoor heating, indoor air quality, and potentially access to warm public buildings. Attributing the entire health discontinuity to outdoor TSP exposure may overstate the role of ambient particulates relative to indoor pollution.
Historical policy changes. The free-coal policy was not perfectly stable over the study period. Reforms in the 1990s and 2000s changed the programme structure, and the coal
heating advantage of the north has declined over time. This temporal variation complicates the cross-sectional RD interpretation.
Selection and migration. If healthier individuals selectively migrate south of the Huai River (or unhealthy individuals are more likely to migrate north to access heating subsidies), the RDD estimate conflates the effect of pollution with the effect of selective in- and outmigration.
7 Policy Implications
The Huai River studies provide some of the most credible evidence on the health costs of particulate air pollution. Chen et al. [2013] calculate that the policy cost approximately 2.5 billion life-years in total for the population north of the Huai River a staggering welfare cost for a heating subsidy that operated for several decades.
These findings have direct relevance for Chinese clean-air policy. China's ongoing coal-to-gas and coal-to-electricity transition programmes in northern cities, if implemented at scale, could generate health benefits commensurate with these estimates. The results also inform the global debate on the social cost of carbon and particulate matter regulation, suggesting that the marginal value of air quality improvements may be substantially larger than traditional cost-benefit analyses have assumed.
8 Conclusion
The Huai River design is a textbook illustration of geographic regression discontinuity applied to a consequential policy question. An administrative boundary drawn for energy policy reasons not health reasons happens to create a sharp discontinuity in sustained air pollution exposure. By comparing health outcomes just north and south of this boundary, Chen et al. [2013] and Ebenstein et al. [2017] estimate that long-run particulate exposure reduces life expectancy by approximately 0.6 years per 10 μg/m3 of PM10, with the entire effect concentrated in cardiorespiratory disease. The study stands as one of the most influential pieces of evidence linking ambient air pollution to human mortality.
References
- Calonico, S., Cattaneo, M. D., and Titiunik, R. (2014). Robust nonparametric confidence intervals for regression-discontinuity designs. Econometrica, 82(6):2295-2326.
- Chay, K. Y. and Greenstone, M. (2003). The impact of air pollution on infant mortality: Evidence from geographic variation in pollution shocks induced by a recession. Quarterly Journal of Economics, 118(3):1121-1167.
- Chen, Y., Ebenstein, A., Greenstone, M., and Li, H. (2013). Evidence on the impact of sustained exposure to air pollution on life expectancy from China's Huai River policy. Proceedings of the National Academy of Sciences, 110(32):12936-12941.
- Ebenstein, A., Fan, M., Greenstone, M., He, G., and Zhou, M. (2017). New evidence on the impact of sustained exposure to air pollution on life expectancy from China's Huai River Policy. Proceedings of the National Academy of Sciences, 114(39):10384-10389.
- Greenstone, M. (2002). The impacts of environmental regulations on industrial activity: Evidence from the 1970 and 1977 Clean Air Act Amendments and the Census of Manufactures. Journal of Political Economy, 110(6):1175-1219.
- Imbens, G. W. and Lemieux, T. (2008). Regression discontinuity designs: A guide to practice. Journal of Econometrics, 142(2):615-635.