THE SHORT BRIEF · WALES · EVIDENCE FROZEN 4 NOVEMBER 2026

Did fewer people get hurt?

On all Welsh roads, recorded casualties fell from 4,448 in 2022 to 4,079 in 2025, but rose from 4,008 in 2024. Injury collisions fell 9.4% from 2022 and rose 2.9% from 2024. These observed changes do not isolate the effect of 20mph limits.

How much did the policy cause?

casualties (people killed or injured): -6.7% against weighted English regions; +0.2% against pooled England. The comparisons disagree on the direction of change. These are model estimates. A causal policy effect is not established.

Did journeys take longer?

Across 60 sampled route, direction and peak combinations, 53 had longer mean journeys from spring 2023 to spring 2024. The median change was +37.5 seconds. Each combination has equal weight; this is not average delay for a Welsh traveller or an isolated policy effect.

Who and which places were affected?

Broad subgroup patterns vary. They cannot tell you the effect on an individual street. Neighbourhood income describes the collision location, not an injured person’s income; changes on faster roads do not prove traffic diversion.

Was it worth the cost?

A national net benefit is not established. Reported spending has differing scope and price bases. The optional calculator uses hypothetical trips, delay and casualty changes; missing impacts are not zero.

Confidence: the displayed arithmetic is reproducible. The causal policy effect and national net benefit are not established.

Share this finding with its limits

The link restores this selection. The text retains the period, scope and qualification.

Open this exact view

Read the full article and source notes

Safety: a fall in counts is only the start

The public debate often asks a single question: did 20mph work? That bundles together several different questions. Were speeds lower? Were fewer people injured? How much of any change was caused by the policy? And were those benefits worth the travel time and other costs? Each needs its own evidence.

Wales is the focus of this lab. The national change took effect in September 2023, but road safety had already been changing. Choosing a favourable before-year can make a simple comparison look decisive. Showing both the longer comparison and the latest annual movement is more informative.

The controlled comparison asks what the Welsh trajectory might have looked like without the change. It is more demanding than a before-and-after calculation, but the counterfactual still fails important checks: English areas had their own 20mph programmes, trends were not parallel, and the model detects breaks at fake dates before the policy. The smaller pooled estimate also has an interval spanning zero. We therefore publish associations, not a count of injuries proved to have been prevented.

Evidence and sources for this section

Observed counts

Fewer collisions than 2022, but an increase in 2025

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.

2022–2025; StatsWales release 29 October 2026

Limit: Counts are not exposure-adjusted risk. The decline predates the national policy; this comparison does not identify its effect.

Model associations; not causal

The estimated reduction depends on the comparator

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%).

Post: October 2023–December 2025; pre: 2015–2019 and January 2022–August 2023

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.

Drivers faced a trade-off in the measured sample

Lower spot speeds do not translate directly into the same percentage increase in whole-journey time. Routes include different limits, junctions and congestion. The lab follows the same route, direction and peak combination across periods and makes spring-to-spring the main comparison.

The route sample shows that many journeys became longer and less variable at the same time. That is useful evidence for the debate, but it does not isolate the policy’s contribution. Traffic volumes, roadworks and other changes may contribute; no validated route weights convert these observations into a national delay. The autumn comparison remains available as a separate sensitivity because it changes the season.

Evidence and sources for this section

Observed sample

Measured speeds fell at retained sites

The weighted mean at 38 retained monitoring sites fell from 28.5mph in summer 2023 to 25.2mph in summer 2025.

Summer 2023–summer 2025

Limit: These selected main through roads are not representative of every 20mph road. Five sites reverted to 30mph and are excluded throughout.

Observed sample

Most sampled journeys took longer

Of 60 route/direction/peak combinations, 53 had longer mean journeys, 1 was unchanged and 6 were shorter. The median change in the mean was +37.5 seconds.

Spring 2023–spring 2024; same season

Limit: Combinations have equal weight, overlap and are not independent experiments. This is not a national traveller-weighted delay or a policy effect.

Observed sample

Longer and less variable can occur together

The observed 95th–5th percentile time spread narrowed in 50 of 60 combinations; 44 became both longer on average and narrower in spread.

Spring 2023–spring 2024

Limit: The percentile spread is neither a confidence interval nor a standard deviation. It cannot alone price reliability benefits.

Where should the limit apply?

That is a decision about particular roads and people, not simply a national average. The lab exposes differences across road-user groups, age, road settings and neighbourhoods. Those patterns can help prioritise further investigation; they cannot yet tell a council which road would become safer if its limit changed.

A credible local evaluation needs dated road boundaries and legal changes, comparison roads with verified histories, traffic exposure, and enough before-and-after observations. Roads with no collisions must remain in the denominator. A rise on roads recorded at higher limits is not, by itself, proof that traffic or harm was displaced there.

Evidence and sources for this section

Exploratory associations

Local patterns do not establish where the policy works

Subgroup associations differ across road users and settings. Collision-neighbourhood deprivation describes the place of the event, not the injured person’s income. Recorded speed-category differences do not demonstrate traffic diversion.

2015–2025; same main pre/post windows as the comparison

Limit: No causal adjacent-road diversion estimate or road-specific reversal casualty effect is identified. Multiple comparisons and nonparallel pre-trends remain relevant.

What would the safety gains need to be worth?

The economic question cannot be answered by attaching a pound value to the lab’s aggregate casualty association. A defensible appraisal needs a causal change by severity, policy-induced travel time, matching trip volumes and a compatible cost history. Those inputs are not yet jointly established.

The calculator instead lets readers see the assumptions needed to reach a break-even threshold. The illustration is deliberately labelled as hypothetical. Costs set to zero are excluded components, not evidence that implementation was free. Casualty prevention values describe social welfare; they are not equivalent to insurance claim reductions, lower premiums or cash returned to the NHS.

For a decision to retain or reverse a limit now, already-spent implementation money is sunk. The relevant future comparison concerns the costs and outcomes still changeable. An evaluation of the original decision must instead place historical costs on their actual dates. These are different appraisal questions.

Evidence and sources for this section

Provisional reported expenditure

Spending disclosures have different scopes

A 19 May 2025 Welsh Government disclosure reported approximately £39.504m spent to date. The annual outturn was still being finalised and the total included the separate Operation Ugain grant for 2024/25.

Disclosure: 19 May 2025

Limit: This is not a final lifetime cost. Forecasts, awards and spending disclosures are not additive, and do not share a validated payment schedule and price basis.

Hypothetical; frozen May 2026 TAG v2.03

A break-even illustration is not a verdict

Assume 100,000 car trips per day, 250 days per year, 1.5 people per car, 30 seconds of policy-induced delay, commuting values and ten years. With implementation and recurring costs excluded, the included time cost has a £44.39m present value. Preventing 16.03 serious casualties each year would be one alternative equivalent threshold.

£2023 market prices; 2026 present-value reference; annual flows 2027–2036

Limit: Every traffic and effect input is assumed. No casualty effect is selected by default; benefit and balance stay null. Missing impacts are not zero. This is not a Wales appraisal or a forecast of insurer or NHS cash savings.

Dated guidance notice

Valuation guidance is changing

The scenarios retain May 2026 TAG v2.03 inputs for reproducibility. On 5 November 2026, DfT announced revised discounting and economic parameters for its expected December release, following Treasury’s October changes. Those revisions are not included in this version; updated calculations will require a separate version.

Checked 5 November 2026

Limit: Frozen rates must not be described as the latest Green Book recommendations. No silent replacement of historical calculations.

Do not turn missing environmental evidence into a zero

Air quality, emissions, active travel and access may matter to the decision even when this release cannot value them. The absence of an identified effect is not evidence that the effect is absent. They remain explicit gaps rather than hidden zeros in a national total.

Evidence and sources for this section

Coverage gate not met

The air-quality question remains open

The pilot did not collect a pre-implementation air-quality baseline. Sensor changes, gaps and incomplete traffic comparisons limit its post-only observations. A new causal model was not advanced.

Pilot report: May 2024; feasibility review: 4 November 2026

Limit: This establishes neither an effect nor no effect. Ambient nitrogen dioxide and particles are not measurements of carbon emissions or fuel use.

What the evidence supports now

The defensible conclusion is bounded: speeds fell at the retained monitoring sites; most measured peak-route combinations took longer; and the controlled safety result depends materially on the comparison. This lab does not establish a national causal safety effect or a national net benefit.

The next evidence investment should target those decision gaps: verified treatment and comparison-road histories, matched exposure and journey weights, compatible severity records, and dated costs. Until those are available, publish the observations and their uncertainty together. That gives the public a basis for judging claims without pretending the remaining questions have been settled.

Evidence and sources for this section

Model associations; not causal

The estimated reduction depends on the comparator

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%).

Post: October 2023–December 2025; pre: 2015–2019 and January 2022–August 2023

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.

Observed sample

Most sampled journeys took longer

Of 60 route/direction/peak combinations, 53 had longer mean journeys, 1 was unchanged and 6 were shorter. The median change in the mean was +37.5 seconds.

Spring 2023–spring 2024; same season

Limit: Combinations have equal weight, overlap and are not independent experiments. This is not a national traveller-weighted delay or a policy effect.

Hypothetical; frozen May 2026 TAG v2.03

A break-even illustration is not a verdict

Assume 100,000 car trips per day, 250 days per year, 1.5 people per car, 30 seconds of policy-induced delay, commuting values and ten years. With implementation and recurring costs excluded, the included time cost has a £44.39m present value. Preventing 16.03 serious casualties each year would be one alternative equivalent threshold.

£2023 market prices; 2026 present-value reference; annual flows 2027–2036

Limit: Every traffic and effect input is assumed. No casualty effect is selected by default; benefit and balance stay null. Missing impacts are not zero. This is not a Wales appraisal or a forecast of insurer or NHS cash savings.

Evidence frozen 4 November 2026; the DfT guidance notice was checked 5 November. This is a descriptive and analytical publication with an unresolved causal gate, not a recommendation to retain or reverse every limit.