RHEA — Scottish LEZ Evaluation

What did we find?

We evaluated all four Scottish Low Emission Zones using difference-in-differences — comparing NO2 changes at monitoring sites inside each LEZ against sites outside. We used block bootstrap for honest confidence intervals and ran the analysis separately for each hour of the day to reveal when effects occur.

Edinburgh LEZ

−6.6 µg/m³

Evening rush (16–18), p = 0.006
Daily DiD: p = 0.004 (spatially verified)
Traffic pattern confirmed

Glasgow LEZ

−1.5 µg/m³

Not significant (p = 0.65)
No effect detected

Aberdeen LEZ

−2.8 µg/m³

Overnight only (p = 0.042)
Non-traffic pattern

Dundee LEZ

−1.7 µg/m³

Not significant (p = 0.40)
No effect detected

The headline

Edinburgh is the only Scottish LEZ showing a statistically significant, traffic-related air quality improvement. The effect is concentrated during afternoon and evening hours (12:00–22:00), peaking at 18:00 with a 7.4 µg/m³ reduction. Overnight hours show no significant change — confirming the effect is from reduced traffic, not weather or regional trends.

Why the others didn't show effects

Glasgow: First Bus deployed 150 electric buses before LEZ enforcement, removing ~70% of Hope Street NOx. The DiD correctly shows the marginal LEZ enforcement effect is near zero — because the buses were already clean.
Aberdeen: The diurnal fingerprint reveals an overnight signature (22:00–03:00), not a traffic pattern. This is consistent with Union Street construction works (started Q2 2024), which involve overnight road closures. Without the fingerprint, this would have been reported as a borderline effect (p=0.089).
Dundee: The smallest LEZ zone (~0.5 km²) with the least traffic impact. Genuine null — the zone is likely too small to produce a measurable change.

One analysis, four cities, same legislation — each demonstrating a different capability: detecting a genuine effect (Edinburgh), confirming the mechanism via temporal signature (the fingerprint), diagnosing confounders that would otherwise produce false positives (Aberdeen), and explaining policy-relevant nulls with contextual evidence (Glasgow). Site classifications were spatially verified against boundary polygons; Edinburgh's result strengthened under scrutiny (p = 0.038 → 0.004).

Monitoring Sites

Treatment sites (red) are inside the LEZ boundary. Control sites (blue) are outside. Click any marker for site details. The LEZ boundary is shown as a shaded polygon.

Site counts

City	Treatment sites	Control sites	LEZ area	Enforcement
Edinburgh	4 (ED7, EDNS, ED6, ED)	4 (ED3, ED2, ED11, ED5)	~3.1 km²	Jun 2024
Glasgow	5 (GLA4, GHSR, GLAS, GLA, GLKP)	5 (GLA5, GLA6, GLA7, GLA8, PAI2)	~2.6 km²	Jun 2023
Aberdeen	4 (ABD2, ABD0, ABD3, AD1)	3 (ABD9, ABD1, ABD8)	~0.9 km²	Jun 2024
Dundee	4 (DUN3, DUN5, DUNM, DUN6)	3 (DUN1, DUN4, DUN7)	~0.5 km²	May 2024

Site counts shown are from the original hand-coded classification. Spatial verification against boundary polygons identified 7 sites within 100m of boundaries — see sensitivity analysis. Edinburgh's result strengthens under corrected classification (p = 0.038 → 0.004). Glasgow and Dundee remain null. Aberdeen has no sites unambiguously inside the narrow zone.

Data source: AURN + SAQN hourly parquets (2019–2026 data lake). Sites classified against LEZ boundary polygons from Transport Scotland GeoJSON.

Cross-Intervention Results

Daily DiD (block bootstrap, ±365 days)

City	ρ₁	Block	DiD Effect	95% CI	p-value	PT p
Edinburgh	0.61	13d	−5.13 µg/m³ (−21%)	[−8.93, −1.43]	0.004	0.33
Glasgow	0.68	15d	−1.48 µg/m³ (−5%)	[−7.74, +4.95]	0.645	0.08†
Aberdeen	0.43	8d	−2.94 µg/m³ (−12%)	[−6.35, +0.62]	0.089	—
Dundee	0.61	13d	−1.68 µg/m³ (−6%)	[−5.64, +2.25]	0.402	—

Sub-daily DiD by time-of-day window

City	Full day	Morning rush	Evening rush	Overnight	Pattern
Edinburgh	−3.80*	−3.58	−6.58**	−1.92	Traffic
Glasgow	−1.48	−1.81	−2.90	−0.52	None
Aberdeen	−2.94	−3.61	−2.72	−2.77*	Non-traffic
Dundee	−1.68	−1.68	−2.63	−0.54	None

* p < 0.05 ** p < 0.01. All effects in µg/m³. Block bootstrap, 2000 resamples. Sub-daily window results use the original site classification; daily DiD row above uses spatially verified classification (see sensitivity analysis). † Glasgow PT p = 0.08 with original classification; improves to 0.93 with spatially corrected sites.

Edinburgh: pre vs post diurnal profile

Figure 2: Mean hourly NO2 at Edinburgh treatment sites before (grey) and after (blue) LEZ enforcement. The afternoon/evening peak flattens post-LEZ. ±1 SD shading. Rush-hour periods highlighted in amber.

Edinburgh: effect by time window

Figure 3: Forest plot of DiD effect estimates by time-of-day window. Evening rush shows the largest, most significant effect. Overnight is non-significant.

Diurnal Fingerprints

Each bar shows the DiD effect estimate for one hour of the day. Dark blue = p < 0.01, light blue = p < 0.05, grey = not significant. Rush-hour periods shaded.

Edinburgh LEZ: clear traffic signature. 12 of 24 hours significant, peaking at 18:00 (−7.4 µg/m³, p=0.003). Overnight hours not significant. This pattern is consistent with reduced through-traffic on the Nicolson Street corridor during afternoon/evening commute.

Glasgow LEZ: flat fingerprint. No significant effect at any hour. Consistent with bus fleet electrification (150 electric buses deployed 2020–2023) having already cleaned up the dominant emission source before LEZ Phase 2 enforcement.

Aberdeen LEZ: overnight signature. Significant reductions at 22:00–03:00 but not during rush hours. This is NOT a traffic pattern — consistent with Union Street Central construction works (started Q2 2024) involving overnight road closures and traffic diversions.

Dundee LEZ: no significant effect at any hour. The smallest zone (~0.5 km²) with limited traffic impact. One borderline hour (13:00, p=0.07) but no coherent pattern.

How It Works

Load hourly NO2 from UK regulatory monitoring networks (AURN, SAQN) via the Aeolus data library. Filter to ±365 days around each LEZ enforcement date.

Classify sites as treatment (inside LEZ boundary) or control (outside). No boundary buffer exclusion due to sparse Scottish networks.

Aggregate to daily zone means. No deweathering applied — we found that daily-resolution deweathering suppresses intervention signals regardless of training strategy (documented in detail). Raw DiD is the recommended estimator for daily data.

Difference-in-differences with block bootstrap (2000 resamples). Block length auto-estimated from autocorrelation using Lahiri (2003). This produces correctly-sized CIs — validated at 90,000 Monte Carlo trials showing 93–100% coverage at ρ ≤ 0.6.

Diurnal fingerprint: repeat step 4 for each hour of the day (0–23), producing 24 effect estimates. Classify the pattern as "traffic" (rush-hour effects), "non-traffic" (overnight), or "none". This provides mechanistic evidence from monitoring data alone.

Interpret against policy context: bus fleet changes, construction, tram extensions. The fingerprint diagnosis + context explains each city's result.

Why block bootstrap?

Daily NO2 data has temporal autocorrelation (ρ ≈ 0.4–0.7): today's value predicts tomorrow's. Standard (i.i.d.) bootstrap ignores this, producing confidence intervals that are too narrow and a false positive rate of 24% (at ρ = 0.6) instead of the nominal 5%.

Figure 4: Monte Carlo validation (90,000 trials). Block bootstrap (blue) achieves correct coverage; i.i.d. bootstrap (red) fails severely at ρ ≥ 0.6. Green band = target 93–97%.

Why no deweathering?

We tested three deweathering approaches for daily DiD and found all suppress the intervention signal. The root cause: daily aggregation destroys the within-day variance that deweathering preserves. The DiD already controls for common weather trends (same city, same weather affects treatment and control). See docs/deweathering-intervention-interaction.md for the full analysis.

Policy Context

Edinburgh — the tram and the LEZ

Edinburgh's Tram to Newhaven extension opened 7 June 2023 — one year before LEZ enforcement (June 2024). This is a pre-period change that improved the baseline before the LEZ took effect, making our estimate conservative. The LEZ effect we detect is on top of any tram-driven improvements. The George Street redesign and City Centre Transformation programme (£314M) may also contribute, but Nicolson Street (our primary treatment site) is not part of that redesign.

Glasgow — the bus fleet story

First Bus deployed 150 BYD ADL electric double-decker buses to their Caledonia depot, progressively from 2020 through March 2023 — before LEZ Phase 2 enforcement. Buses were responsible for ~70% of NOx emissions on Hope Street (our primary treatment site). The UBDC study (Shin et al. 2024) found Hope Street NO2 fell 9–13% from fleet change alone, with no change in traffic volume. The DiD correctly shows the marginal enforcement effect is near zero.

Policy insight: The LEZ announcement and grants programme may have been more effective than enforcement itself, by motivating fleet investment.

Aberdeen — construction on LEZ streets

Union Street Central construction started Q2 2024 — an 18-month project directly on the LEZ's main street, coinciding exactly with enforcement. New market building construction from February 2024 (£20M). South Harbour expansion opened September 2023. The overnight diurnal signature (22:00–03:00) is consistent with construction traffic management — overnight road closures diverting traffic away from treatment monitors.

Dundee — smallest zone

At ~0.5 km², Dundee's LEZ is the smallest of the four. Xplore Dundee deployed 12 new electric double-deckers (smaller scale than Glasgow's 150). The null result is most likely explained by the zone being too small to affect enough traffic to produce a measurable change at monitoring sites.

Shared confounders

ScotZEB: The Scottish Zero Emission Bus Challenge Fund funded electric bus deployment across all four cities. Glasgow received the largest allocation (150 buses), others smaller numbers. This is a shared confounder that biases all DiD estimates toward zero.

National fleet turnover: UK-wide diesel vehicle scrapping and EV adoption affects all sites equally. Captured by the DiD design (common trend assumption).