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

Introduction There is limited research evaluating 20 mph speed limit interventions, and long-term assessments are seldom conducted either globally or within the UK. This study evaluated the impact of the phased 20 mph speed limit implementation on road traffic collisions and casualties in the City of Edinburgh, UK over approximately 3 years post implementation. Methods We used four sets of complementary analyses for collision and casualty rates. First, we compared rates for road segments changing to 20 mph against those at 30 mph. Second, we compared rates for the seven implementation zones in the city against paired control zones. Third, we investigated citywide casualty rate trends using generalised additive model. Finally, we used simulation modelling to predict casualty rate changes based on changes in observed speeds. Results We found a 10% (95% CI −19% to 0%) greater reduction in casualties (8% for collisions) for streets that changed to 20 mph compared with those staying at 30 mph. However, the reduction was similar, 8% (95% CI −22% to 5%) for casualties (10% collisions), in streets that were already at 20 mph. In the implementation zones, we found a 20% (95% CI −22% to −8%) citywide reduction in casualties (22% for collisions) compared with control zones; this compared with a predicted 10% (95% CI −18% to −2%) reduction in injuries based on the changes in speed and traffic volume. Citywide casualties dropped 17% (95% CI 13% to 22%) 3 years post implementation, accounting for trend. Conclusion Our results indicate that the introduction of 20 mph limits resulted in a reduction in collisions and casualties 3 years post implementation. However, the effect exceeded expectations from changes in speed alone, possibly due to a wider network effect.

The full list of factors considered in the matching were the seven separate domains of SIMD, the 6 category urban-rural classification and the area population density.For the domains of SIMD, the ranks were used.Subsequently, apart from urban rural classification, where frequencies were used, we looked as mean and median as well as histograms to identify the most similar match, it was 1:1.With all these complications and the challenge to find a matching geography for the implementation areas the methods were simple and thus the final selections came down to a manual process.

Implementation zones and the number of possible matches
Implementation zone 1A City centre 1B 2 3 4 5 6 Number of areas to find a match from 14 16 3 2 7 10 10 Implementation-control zone pairings for the 20mph speed limit program in the City of Edinburgh.
The last column with maps of the control zones shows the exact datazones selected from every city based on SIMD, urban-rural classification and the area population density.

Supplementary file 4: GIS manipulation: Explanation of correcting the nearest line method in junctions
We looked in junctions where multiple road segments may be joined (and selected the two closest road segments from the collision's location).For these injuries we looked up their road class information (from the STATS19 1 st road class variable).This variable describes the class of the road that the accident happened with values A, B, C or unclassified and it is the most direct connection STATS19 offers for identifying the exact location.The closest match tended not to "find" the major roads at junctions.We investigated the bias of the closest match in junctions by comparing the proportion of major roads (1 st road class = A class) with minor roads (road segment = "20mph existing streets" & "20mph local streets").We corrected these injuries by reassigning them to the major closest road segment.

Supplementary file 8:
Basic analysis comparing the mean number of casualties and collisions before and after the 20mph limit change in Edinburgh.The data before the limit changes relate to the period August 2013 through July 2016, while the after data relate to the period March 2018 to December 2019.Figures in bold are statistically significant differences (p<0.05) according to a two-sample t-test with unequal variances.

*
As it was not possible to identify a control zone for Zone 2, these rows combine the other intervention zones and their matched control zones ** Only one post intervention observation so unable to undertake T-test Comparison of collision ratesimplementation vs control zones with 95% CI (Poisson distribution and delta method).Years pre for all zones: 3 years; years post: zone 1a and 1b [3.42 years], 2 and 3 [2.83years], 4 and 5 [2.63 years], and 6 [1.83 years] BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) J Epidemiol Community Health doi: 10.1136/jech-2023-221612 -443.

Difference in mean number of collisions after limit change compared to before
As it was not possible to identify a control zone for Zone 2, these rows combine the other intervention zones and their matched control zones ** Only one post intervention observation so unable to undertake T-test *