Independent Verification of the European Climate Law's Effect on Nitrogen Dioxide Pollution: A Satellite-Derived, Difference-in-Differences Analysis
Year: 2026
Type: preprint
Abstract: Testing the European Climate Law (Regulation (EU) 2021/1119) against a deliberately constructed external control group of nine non-EU countries (the United Kingdom, Norway, Switzerland, Iceland, Albania, Bosnia and Herzegovina, Montenegro, North Macedonia and Serbia), using a two-group Difference-in-Differences (DiD) model with country-clustered standard errors on a 36-country panel, produces this study's headline result: no statistically significant pooled EU-specific effect at the conventional 5% level (coefficient = −2.22 × 10⁻⁶, p = 0.101, cluster-robust). It's not all one-sided, however: splitting the sample by baseline pollution level shows a statistically significant decline for the fourteen higher-baseline, more industrialized EU countries (coefficient = −5.46 × 10⁻⁶, p = 0.003), while the lower-baseline countries show none — a finding corroborated by a 23-quarter event-study analysis, which finds four individual post-treatment quarters where the coefficient was negative and significant, more than the roughly one expected by chance. Reaching this corrected, null pooled model took first discovering that an earlier, single-cohort EU-27-only model had found a statistically significant decrease (p = 0.026) — until a placebo test using a counterfactual treatment date showed this same "effect" turning up just as strongly on a date with no real policy change, revealing the model was simply picking up a general, ongoing European pollution-decline trend rather than anything specific to the Climate Law, which is exactly why the external control group was built in the first place. This study tests all of this using Sentinel-5P satellite data (2019–2024) as the independent verification layer, rather than the self-reported data governments typically use to claim policy success — an approach that remains uncommon in EU climate-policy evaluation. None of a further series of robustness checks — a bad-control test, a log-transformed outcome, treatment-date sensitivity, and the baseline-pollution split itself — overturn the pooled null as this study's headline result, but together they point to a real, heterogeneous effect rather than random noise, given the statistical power this design has. An augmented synthetic control, weighting a seven-country donor pool rather than averaging it equally, reaches the same near-zero, same-sign pooled estimate through a methodologically distinct method, and a Moran's I spatial-autocorrelation diagnostic confirms that while raw country-level NO₂ readings are strongly spatially clustered (I = 0.570, p = 0.001), the DiD model's residuals are not (I = 0.069, p = 0.135). Applying this same two-group, cluster-robust design to a secondary outcome — NDVI vegetation health, previously assessed only with the original, since-invalidated single-cohort design — revealed a statistically significant relative decline in EU-27 vegetation health versus the control group (coefficient = −0.0145, p = 0.007), reported as an exploratory finding not attributable to the Climate Law itself. These findings suggest that any EU-specific NO₂ effect from the Climate Law, if real, is smaller and more concentrated in specific countries and periods than a pooled year-round EU-27 average can cleanly resolve, while the NDVI result shows the risk of applying validation rigor unevenly across a study's outcomes. The study's validation methodology is presented as a methodological contribution as significant as the substantive findings themselves.
DOI: 10.31223/x5gb87
Author: SAKSHI MASKE
Language: English
Subfield: Economics and Econometrics
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7208828554