Effect of 20 mph speed limits on traffic injuries in Edinburgh, UK: a natural experiment and modelling study

Road traffic collisions are a major public health problem, causing approximately 1.3 million deaths and tens of millions of injuries annually worldwide. Despite declining fatalities in the UK and Scotland, the Vision Zero approach seeks to end fatal and serious injuries. Traffic speed is a key risk factor for both the occurrence and severity of collisions; higher speeds reduce drivers’ field of vision and increase stopping distance, thereby increasing collision risk and injury severity. The City of Edinburgh progressively expanded 20 mph limits from residential areas and around schools to a citywide network covering about 80% of roads between 2016 and 2018, while retaining 30–40 mph roads in suburbs. This study aims to evaluate the impact of Edinburgh’s phased citywide 20 mph speed limit intervention on collisions and casualties and their severity over approximately 3 years post implementation, using multiple complementary analyses at street, zone, and city levels.

Prior evidence on 20 mph interventions is limited and often lacks long-term evaluations. Reviews have generally found that traffic calming measures, including speed reductions, are associated with decreased collisions and casualties. An umbrella review reported associations of 20 mph policies with collision and injury reductions, while broader syntheses included effects on air quality, active travel, noise, inequalities, and liveability. Empirical studies show mixed findings: Portsmouth reported about a 22% reduction in casualties post 20 mph limits without trend adjustment or controls; Bristol reported reductions in injury severities but methods were not clearly reported and trends were not accounted for; Belfast showed smaller impacts, potentially due to more limited geographic coverage and lack of enforcement and awareness campaigns. Overall, the literature suggests benefits of lower speed limits but highlights gaps in robust, controlled, and long-term casualty evaluations.

Design: Natural experiment with phased implementation (stepped-wedge) of 20 mph limits across seven zones in Edinburgh between 31 July 2016 and 5 March 2018. The intervention comprised policy, signage, education, and enforcement components without physical traffic calming (e.g., humps). Outcomes: Police-recorded collisions (STATS19) and casualties (persons injured), including severity (slight, serious, fatal). Data/GIS: Implementation zones were matched with Scottish Data Zones using the Scottish Index of Multiple Deprivation (2016), urban–rural classification, and population density. Each implementation zone was paired 1:1 with a matched control zone elsewhere in Scotland via manual selection informed by ranks, distributions, and histograms of these characteristics. Road speed limit shapefiles and zone boundaries were linked to STATS19 collision locations (eastings/northings), assigning each collision to the nearest road segment within 12 m and to an implementation or control zone. Statistical software: R (v4.2.2); code available at https://github.com/20mph-study. Analyses: 1) Road segment analysis: Annual collision and casualty rates calculated for pre- (3 years) and post- (1.83 years) periods for three street categories within implementation zones: existing 20 mph, stayed at 30 mph, and changed from 30 to 20 mph (local and main combined). Percentage changes computed and difference-in-differences (DiD) calculated relative to streets that stayed at 30 mph. 2) Implementation vs control zones: Annual rates and percentage changes for collisions and casualties computed for each of seven implementation zones and their matched controls over zone-specific pre/post periods; DiD = % change in implementation − % change in control. 3) Citywide time series: Generalised additive model (GAM) of monthly casualties (273 observations; 30 post-intervention) with smooth functions for secular trend, seasonality, and intervention effect; offset for days per month. A sub-model using data up to first intervention date predicted post-intervention casualties; observed vs predicted compared to estimate intervention impact. 4) Simulation modelling: Elvik’s exponential speed models (re-parameterised power model) used to predict changes in casualties by severity (slight/injury, serious, fatal) based on observed speed changes (in the 20–30 mph range), with collision severity defined by the most severe casualty in each collision. Combined speed+volume linear adjustment applied: predicted injuries from speed multiplied by relative change in traffic volume. Notes: Speed decreased on average by 1.3 mph (23.63 to 22.29) and traffic volume by 2% (3641 to 3555) in implementation zones. Reporting change: In mid-2019, Scottish police revised severity reporting, affecting slight/serious counts for 6 months of the post period.

• Road segments: Collisions fell 42% on streets that changed to 20 mph and 44% on existing 20 mph streets; 35% on streets that stayed at 30 mph. DiD for collisions was 8% (95% CI −2% to 18%). Casualties fell 43% on streets that changed to 20 mph, and 42% on existing 20 mph; 33% on streets that stayed at 30 mph. DiD for casualties was 10% (95% CI 0% to 19%). • Implementation vs control zones: Across seven zones, casualty rates decreased by 33% vs 13% in controls, yielding an overall DiD of −20% (95% CI −22% to −8%). Collisions decreased by 34% vs 12% in controls, DiD −22% (95% CI −40% to −2%). Most zones showed larger reductions than controls; exceptions noted (e.g., zone 1b for collisions, zone 4 where controls saw an increase). • Citywide trend (GAM): Predicted post-intervention annual casualties from pre-intervention trend were 1306 (95% CI 1261 to 1352), while observed were 1079 (95% CI 1037 to 1120), a difference of 228 (95% CI 166 to 289), corresponding to a 17% (95% CI 13% to 22%) additional reduction beyond trend. Overall observed pre vs post difference was −332 casualties (−23%; 95% CI −27% to −20%) when comparing 3 years pre to 2.42 years post. • Simulation modelling: With average speed reduction of 1.3 mph and volume reduction of ~2%, Elvik’s model predicted ~10% reduction in all casualties, smaller than observed reductions. For injury (slight) category, predicted annual rate 817 vs observed post 593; serious injuries predicted 101 vs observed 107; fatalities predicted 4.9 vs observed 4.7. Observed reductions exceeded modelled expectations, especially for slight injuries.

The study demonstrates that Edinburgh’s citywide 20 mph speed limit intervention was associated with significant reductions in collisions and casualties beyond the underlying declining trend and relative to matched controls. Consistent reductions at city, zone, and street levels suggest a robust effect. The observed impact exceeded expectations based on speed and traffic volume changes alone, indicating possible broader network or behavioural effects such as increased driver caution, spillover of lower speeds to non-20 mph streets, or cumulative benefits accruing over time. The triangulated approach—street segment comparisons, zone-level DiD, citywide GAM, and simulation modelling—strengthens causal inference in a natural experiment context. These findings address the research question by showing that 20 mph limits contribute meaningfully to reductions in collision and casualty rates within approximately three years of implementation and underscore the public health relevance of speed limit policies as part of comprehensive road safety strategies.

Citywide 20 mph speed limits in Edinburgh were associated with substantial reductions in road traffic collisions and casualties over roughly three years post implementation, with effects larger than predicted from speed reductions alone. The results support incorporating 20 mph limits into broader urban safety and liveability policies. Future research should assess longer-term impacts, explore mechanisms behind network-level effects, evaluate interactions with enforcement and complementary traffic calming measures, and examine potential changes in active travel and modal shifts.

• Use of STATS19 police data likely underestimates collisions and, to a lesser extent, injuries. • No data on active travel; changes in walking and cycling could confound casualty trends. • Concurrent policy changes (e.g., Scotland’s drink-driving law change) may have contributed to citywide reductions, though unlikely to explain larger reductions in implementation zones versus controls. • Lack of street-segment speed measurements limits exploration of local speed-behaviour mechanisms and wider area spillover effects. • Mid-2019 change in police severity reporting likely affected slight/serious counts for part of the post period. • Limited pool of matched control geographies in Scotland; manual matching may introduce selection bias.