1 The Causal Question
Environmental regulation imposes direct compliance costs on firms. But does it also reduce productivity—does the "pollution haven" argument hold at the plant level? This question is not merely academic: if regulation significantly depresses industrial output, the benefit–cost calculus of clean air standards changes substantially.
The challenge is that regulated industries and unregulated ones differ in many ways beyond their pollution status. Simple comparisons of productivity across regulated and unregulated plants confound the effect of regulation with the industries' pre-existing characteristics. Greenstone [2002] addresses this identification problem by exploiting the geographic discontinuity embedded in the 1970 Clean Air Act (CAA) to estimate the causal effect of environmental regulation on manufacturing productivity and employment.
2 The Identification Strategy
2.1 The Clean Air Act's Attainment Designations
The 1970 CAA and its 1977 amendments required the Environmental Protection Agency (EPA) to establish National Ambient Air Quality Standards (NAAQS) for six pollutants—sulphur dioxide, particulate matter, carbon monoxide, ozone, nitrogen dioxide, and lead. Counties that violated these standards were designated nonattainment counties; counties that met the standards were designated attainment counties.
The key institutional feature is that nonattainment counties faced substantially more stringent regulations on new and modified industrial plants. Plants in nonattainment counties had to install the best available control technology (BACT), subjected them to emissions offset requirements, and placed a higher burden on plant managers seeking permits. Plants in attainment counties faced looser requirements under the prevention of significant deterioration programme.
The designation of a county as attainment or nonattainment was determined by whether measured pollution levels crossed a regulatory threshold. This threshold creates an identification opportunity: counties just above the threshold (nonattainment) are, by assumption, similar in most respects to counties just below the threshold (attainment) except for the regulatory burden they face. The discontinuity in regulation at the threshold is thus a natural experiment in the spirit of an RD design.
2.2 From County Designations to Plant-Level Analysis
Greenstone uses plant-level data from the Census of Manufactures (1972–1987) linked to county-level attainment designations. The key variation is: did a plant operate in a nonattainment county, and if so, was the county nonattainment for a pollutant emitted by the plant's industry?
The main regression compares changes in total factor productivity (TFP) and other outcomes (employment, capital stock, output) at plants in nonattainment counties to those in attainment counties, controlling flexibly for plant-level and county-level characteristics:
where Yijt is a plant-level outcome, j indexes counties, t indexes time, and δit are industry-year fixed effects. The identifying assumption is that, conditional on controls, county nonattainment status is uncorrelated with plant-level productivity shocks.
Importantly, Greenstone refines this by focusing on the relevant nonattainment designation: a steel plant cares about SO2 nonattainment, not ozone nonattainment. Using industry-pollutant specific designations sharpens the treatment definition and provides within-county variation that helps purge confounding from unobservable county characteristics.
3 Data and Setting
The data are drawn from the Census of Manufactures for 1972, 1977, 1982, and 1987, covering the near-universe of U.S. manufacturing plants with at least five employees. This produces approximately 1.75 million plant-year observations across 450 four-digit SIC industries. TFP is computed as the Solow residual from a production function estimated within each four-digit industry.
The county attainment designations vary by year and by pollutant, creating rich variation in treatment intensity across both space and time. This panel structure allows for a differences-in-differences interpretation: the identifying comparison is between plants in counties that switched from attainment to nonattainment (or vice versa) across census years, relative to plants in counties whose status did not change.
4 Key Findings
Greenstone reports three main results.
- Productivity losses. Plants in nonattainment counties experience statistically significant declines in TFP relative to plants in attainment counties. The estimated effect is approximately −2.5 percentage points per year of nonattainment exposure for the most-affected industries. Over the sample period, this implies economically meaningful productivity losses for regulated manufacturers.
- Capital and labour effects. Nonattainment status reduces employment growth and capital investment at affected plants, consistent with both direct compliance costs (installing pollution control equipment diverts resources from productive investment) and a relative competitiveness disadvantage against plants in attainment counties.
- Industry and pollutant heterogeneity. The productivity effects are concentrated among industries that are the heaviest emitters of the regulated pollutants. Plants in industries with low emissions of regulated pollutants show no significant effects even when located in nonattainment counties, providing a useful placebo test for the identification strategy.
5 Limitations and What We Learn
5.1 Identification Assumptions
The main identifying assumption—that conditional on controls, nonattainment designation is as good as randomly assigned—is more plausible near the threshold than far from it. Greenstone acknowledges this and provides evidence of balance on pre-treatment plant characteristics between attainment and nonattainment counties. However, the design is not as clean as a sharp RD: there is no formal continuity test, and the threshold separating attainment from nonattainment counties is not always sharply observed in the data.
A second concern is general equilibrium: the CAA affected all manufacturing plants, so there is no clean "control" group in the usual sense. Plants in attainment counties may have benefited from the competitive disadvantage imposed on nonattainment plants, biasing the productivity comparison upward.
5.2 Complementary Evidence
The findings of Greenstone have been supported and extended by subsequent research. Chay and Greenstone [2003] use a similar county-level discontinuity to estimate the effect of the 1970–1972 air quality improvements on infant mortality, finding that a one-unit reduction in total suspended particulates reduces infant mortality by approximately 0.5 per cent—evidence that the health benefits of the CAA were large. Walker [2013] studies the employment effects of the CAA amendments on individual workers, using a worker-level panel linked to county attainment designations and finding earnings losses for workers displaced from regulated plants.
Taken together, this body of work suggests a consistent picture: the CAA reduced air pollution and improved health, but it did so at a real cost to industrial productivity and employment in regulated areas. This is not evidence against regulation per se—the health benefits may outweigh the productivity costs—but it is essential information for evaluating the net welfare effects of environmental policy.
6 Conclusion
Greenstone [2002] is a landmark paper in the economics of environmental regulation. Its contribution is methodological as well as empirical: it demonstrates how regulatory discontinuities can be leveraged to identify causal effects on firm behaviour that would be unidentifiable from simple cross-sectional comparisons. The CAA's attainment designation system creates a natural treatment-control comparison that, while imperfect, is far more credible than the counterfactual of unregulated states.
The broader lesson is that well-designed regulatory systems often contain identification opportunities that patient empirical researchers can exploit. The CAA was not designed as an experiment, but its arbitrary regulatory thresholds produced the variation needed to estimate its causal effects. Similar strategies have been applied to evaluate air quality regulations in China [Cai et al., 2016] and to study the effects of environmental permits on firm productivity in Europe, making the Greenstone design a template for the environmental economics literature.
References
- Cai, H., Chen, Y., and Gong, Q. Polluting thy neighbor: Unintended consequences of China's pollution reduction mandates. Journal of Environmental Economics and Management, 76:86–104, 2016.
- Chay, K. Y. and Greenstone, M. 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, 2003.
- Greenstone, M. 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, 2002.
- Walker, W. R. The transitional costs of sectoral reallocation: Evidence from the Clean Air Act and the workforce. Quarterly Journal of Economics, 128(4):1787–1835, 2013.