Collisions in Wales before and after the 20 mph default

Main points

MODELLED, ratios of MEASURED counts, Wales against matched England, Oct 2021 to Dec 2025

Each figure is a ratio of ratios: the change in Wales's monthly rate from October 2021 to August 2023 against October 2023 to December 2025, divided by the same change in 80 English districts matched to the Welsh authorities. Below 1 means Wales fell further than the comparison. The interval is a 95 % month-block bootstrap. September 2023, the month of the change, is left out.

The chart

The placebo Passed

Roads at 40 mph and above kept their limits, so if the comparison is fair the ratio there sits close to 1. The band was fixed before the code ran: 0.97 to 1.08, with an interval that includes 1.

Before the change No drift detected

Before the change we could not detect a drift between the two series (slope +0.0009 per month, interval -0.0062 to +0.0080, includes 0). The test is weak: across the 26 months between the two periods' midpoints it cannot rule out a drift from a 14.8 % fall to a 23.0 % rise, about as large as the results above.

How severity is recorded

Dyfed-Powys moved to injury-based reporting from May 2023; South Wales moved to injury-based reporting from Dec 2025; North Wales and Gwent did not. Injury-based reporting records more casualties as seriously injured, so a force that switches inside the window moves its count of serious collisions without any change on the road. In England the share of collisions reported this way rose from 52.1 % before to 70.4 % after.

Cross-check against the Welsh Government Agrees

Our counts from the DfT file agree with the Welsh Government's published figures within 2.0 % on every final figure (9 rows).

Tolerance 3.0 % on figures the releases presented as final. The Welsh Government validates recorded speed limits with local authorities; the DfT file may not carry every correction, which is one reason for the small differences.

Checks added after the run Not registered

An independent check of this page found things the registered tests did not look for. None changes a registered figure; each changes how far the headline can be read.

By police force area. Each Welsh force's area against the same matched districts. North Wales built-up collisions went from about 41 a month before to about 22 after, and its 40 mph and above roads, where the limit did not change, read 0.86 (0.75 to 1.01). A change in how collisions are recorded would look like this too. The table after this one narrows that question; only DfT or North Wales Police can close it.

How North Wales collisions reach the record. The Welsh Government's quality report (updated July 2026) says North Wales Police has not moved to injury-based recording, so that change does not explain its fall. The open file shows where the fall sits: each row below is North Wales Police's collisions a month after the change against before, beside the other three Welsh forces, and the ratio of the two with its 95 % interval. Against its own 40 mph and above roads, North Wales's built-up collisions read 0.62 (0.53 to 0.73) and KSI adjusted 0.80 (0.66 to 0.96); for the rest of Wales, 0.84 (0.77 to 0.92) and 0.95 (0.86 to 1.07). Those roads carry different traffic and a different mix of severity, so this narrows the question without answering it. Police recorded road collisions and casualties: quality report

MODELLED from MEASURED counts. Recorded severity is not comparable across these forces (Dyfed-Powys moved to injury-based recording in May 2023 and South Wales from December 2025), so the severity rows show where the fall sits, not a clean contrast. See post-run note 19 of the pre-registration.

The comparison's own records. DfT's known data issues list police forces whose returns are incomplete in some years. Seven of the matched districts are policed by forces on that list: Avon and Somerset (2022) and Staffordshire (2017 to 2023).

MODELLED. Same estimator, seed and slices as the registered rows; recorded in the pre-registration's post-run notes 10 and 11.

Intervals from a regression. A Poisson regression with a term for every district and every calendar month gives the same ratios, as it must with one treated group; what it adds is the intervals. Allowing for correlation within each district over time widens them (KSI adjusted 0.75 to 0.97, all collisions 0.68 to 0.87). Both stay below 1. No interval here allows for correlation within a police force, and with four Welsh forces none could be trusted.

MODELLED. Poisson regression with district and calendar-month terms over the 22 Welsh authorities and 80 matched districts; 95 % intervals with errors clustered by month (50 months) or by district (102). Post-run note 15.

English districts, 2016 to 2020 Too few to estimate

The rule fixed in advance also looked for English districts whose share of built-up collisions recorded at 20 mph rose by 40 points or more in one year, 2016 to 2020, with at least 60 such collisions in each two-year window either side. One district meets it. The rule needs three, so no estimate is shown.

Rule 4.8 of the pre-registration, run on DfT's full collision file from 1979; post-run note 16. The detector reads how collisions were recorded, not when a council changed a limit.

Other work on this question

Senedd Research's "What impact has 20mph had on road safety in Wales?" (2026) reports that collisions on 20 and 30 mph roads in Wales fell 22.5 % from 2023 Q2 to 2024 Q4, against a fall of 4.7 % in England, and a fall of 17.4 % in killed or seriously injured casualties in Wales on adjusted figures. Those are before-and-after counts without intervals.

The strongest method we know of is Kokka et al.'s study of Edinburgh (2024). It compares streets that switched to 20 mph with streets in the same city that stayed at 30 mph, and finds a 10 % greater fall in casualties on the switched streets (95 % interval 0 to 19 %, so it reaches no change). It also links the change in injuries to a measured fall in speed of 1.3 mph. Citywide, casualties fell 17 % (13 to 22 %) allowing for trend. J Epidemiol Community Health 78(7):437-443

This page adds a pre-registered comparison with matched districts, intervals, a placebo, a force-by-force breakdown and per-authority estimates. It lacks traffic and speed data for the roads whose limit changed, a comparison of Welsh roads that stayed at 30 mph, a before-period longer than 23 months and a synthetic-control comparison. The Welsh Government's commissioned evaluation of the policy reports in stages, with a final report planned for 2029 (Senedd Research, same article).

What is not controlled

• Traffic. There is no open traffic count for Welsh minor roads at this resolution, so collisions are not divided by traffic. If traffic changed differently in Wales and in the comparison areas, that difference is in these ratios. Motor traffic on all roads rose 4.5 % in Wales and 4.9 % in England, 2022 to the 2024 and 2025 average (DfT road traffic estimates). Overall traffic grew at much the same rate in both. That says nothing about traffic on the roads whose limit changed.
• The recorded limit. After the change 74.7 % of Welsh collisions on 20 and 30 mph roads were recorded at 20 mph, against 10.5 % before, while the Welsh Government reports about 13,000 km of 20 mph road after the change against 870 km in 2022. It reports quality issues with the recorded limit on the new roads, so 20 and 30 mph are pooled here, as it does.
• The comparison was changing too. The share of English built-up collisions recorded at 20 mph rose from 23.1 % to 26.9 %. London boroughs, which ran their own 20 mph programmes, are left out of the matched comparison.
• The choice of comparison. The monthly rate of built-up collisions in the matched districts rose 2.7 %; in all of England it fell 4.6 %. Most of that gap is the Staffordshire districts (see Checks added after the run). Against all of England the all-collisions ratio is 0.83 (0.78 to 0.88) and the adjusted KSI ratio 0.89 (0.81 to 0.96).
• Anything else that changed at the same time in Wales and not in the comparison areas (enforcement, publicity, road works) is in these ratios as well.

Provenance

The method, the comparison, the intervals, the placebo band and every sentence above were fixed in a pre-registration before the estimators were written. It was not blind to the likely result: the research pilot and a first look at the data were known when the rules were written, including the adjusted KSI point estimate against all of England (0.89, without its interval) when adjusted KSI was made the headline. The checks added after the run and the reworded trends sentence are recorded in its post-run notes.

Input

collision-last-5-years.csv, 97,669,586 bytes, SHA-256 06a03c0e5504175e23ed98e3e0b4767ac61d396c073405a1bddab9339c18f948

Further checks

collision-1979-latest.csv, 1,534,937,928 bytes, SHA-256 7b80a3a8d9df7b54633a4019e73c8512ca052b166ca2afac0fff4d1448565edc

local_authority_traffic.csv, 433,725 bytes, SHA-256 058a61b9f6f7d07a9363c339829c6723fdecc4270f500611d1790eae6da7b9fc

Pre-registration

SHA-256 f3490273836398288c002afdf0eb48a3806f33f54f44f0b08837490dfbf89bfa, commit 81a1290a3

Comparison

5 nearest English districts per Welsh authority on three before-period features, London left out, 80 districts in all; 22 Welsh authorities.

Generated

23 October 2026

Data: Department for Transport STATS19 collision files (2021 to 2025, and 1979 onwards for the English districts), Open Government Licence v3.0. Department for Transport road traffic estimates by local authority, Open Government Licence v3.0. Welsh Government statistical releases, Crown copyright, Open Government Licence v3.0. 6,955 Welsh collisions on 20 and 30 mph roads in the window.