2022 → 2025
-9.4%3,316 → 3,005
Retrieving and verifying the requested analytical release. Result data loads before optional map boundaries and source checks.
Your filters and release stay in the address. Navigation remains available.
20MPH EVIDENCE LAB · WALES
Recorded counts fell over the longer period, then rose in 2025. The policy’s contribution remains uncertain.
A collision is an event with at least one injury. A casualty is a person killed or injured; one collision can involve several casualties. Neither measure counts damage-only events.
OBSERVED CHANGE · ALL WELSH ROADS
3,316 → 3,005
2,919 → 3,005
These are observed counts, not a policy effect or a risk per journey. StatsWales, 29 October 2026 release; 2025 revised. All roads include roads whose speed limit did not change.
MODEL ESTIMATE · injury collisions
Compare Wales with two English benchmarks over the same periods. Negative estimates mean a lower fitted Welsh trajectory relative to that benchmark.
Conditional 95% interval: -14.8% to -4.4%
The interval excludes no relative change within this model. It does not capture all uncertainty about causation.
Conditional 95% interval: -10.1% to +4.5%
The interval includes no relative change. It does not capture all uncertainty about causation.
Policy effect: not established. The areas were already following different trends, and the comparisons are not verified free of their own 20mph policies. Tests using fake policy dates also find apparent changes.
The link restores this selection. The text retains the period, scope and qualification.
Observed counts
Police-recorded injury collisions on all Welsh roads fell from 3,316 in 2022 to 3,005 in 2025 (-9.4%). The 2025 count was higher than 2024’s 2,919.
Limit: Counts are not exposure-adjusted risk. The decline predates the national policy; this comparison does not identify its effect.
Lab data file · machine-readableStatsWales — police-recorded road collisions and casualties · 29 October 2026 release
Read the source guide before using the raw data files.
Model associations; not causal
The weighted English comparison gives a -9.7% collision association (95% calendar-HAC interval -14.8% to -4.4%). Pooling England gives -3.1% (-10.1% to +4.5%).
Limit: The intervals condition on the model. Comparison areas are not verified untreated; pre-trends and fake-date tests weaken identification. These are not bounds on a causal effect.
Lab data file · machine-readableDepartment for Transport — STATS19 road safety open data · model data through 2025
Read the source guide before using the raw data files.
Full controlled models, trend charts and sensitivity checks →
The earlier release retains full charts, exact tables, specifications and downloads. Its phase numbering describes research development.