MMC Musings

The curious case of Oxfordshire’s recycled road safety press release

Back in May, Oxfordshire County Council announced that the county was bucking the regional trend on road safety. It based that on provisional Department for Transport figures and, to its credit, said so. The word provisional appeared six times. The Leader said Oxfordshire “appears to be bucking” the trend. The Cabinet Member said the figures “provide further evidence.”

On 3 August the DfT published the confirmed figures. They show a smaller fall than the provisional ones did – 16.2% rather than 18.

The same day, the council reissued May’s press release, removing the hedging and claiming that Vision Zero and its blanket 20mph policy were responsible.

Not some of the hedging. All of it.

• “That strongly suggests the approach we have taken is working” became “This shows the approach we have taken is working.”
• “Oxfordshire appears to be bucking the wider Thames Valley trend” became “Oxfordshire is bucking the wider Thames Valley trend.”
• “While these figures are still provisional, the early signs are extremely encouraging” became “This is extremely encouraging.”

And the one passage rewritten from scratch – Cllr Gareth Epps’s quote – went from “although these figures are provisional, they provide further evidence that our safety measures … are having a positive effect” to “the patterns are good across all indicators.”

The evidence got weaker. The claims got stronger.

The bit they missed

Nearly 80% of the August release is word-for-word identical to May’s, including the two headline statistics, which are still the superseded provisional ones: 18% and 20%. The confirmed figures are 16.2% and 14.5%.

Perhaps someone had gone through the original with a ‘find and replace’ for word provisional. They caught five of the six. One escaped. The fourth paragraph of a release announcing confirmed results still describes KSI as “the most robust indicator within the provisional dataset.”

Mind you, I’m hardly in a position to throw stones. I once managed to misspell “Pensions” in a full-page advert in the national weekend press. For a pension company.

A stray word is forgivable (I hope). Publishing superseded statistics under a headline saying “confirmed” is less so. Deleting every caveat while doing it is remarkable.

In fairness, one of those revisions went the council’s way. On the basis it used, the Thames Valley increase has been revised upward, from 19.7% to 21.7%. Had anyone actually checked the confirmed figures, they could have made a bigger noise about it. They didn’t check. They just deleted the caveats.

The row beneath

Here is where it gets interesting, and where the choice of figure starts to matter more than the wording.

For every area, the DfT publishes four rows: killed, KSI unadjusted, KSI adjusted, and all casualties. The adjusted row exists for a specific reason. Since 2016 many police forces have moved from officer judgement to injury-based severity recording, which reclassifies large numbers of injuries from slight to serious without anybody being hurt any worse than before. The adjustment is the DfT’s attempt to strip that artefact out so areas can be compared.

The May release took the unadjusted row for both halves of its comparison:

Provisional, May 2026	2024	2025	Change
Oxfordshire, KSI unadjusted	260	213	−18.1%
Oxfordshire, KSI adjusted	281.5	232.5	−17.4%
Thames Valley, KSI unadjusted	604	723	+19.7%
Thames Valley, KSI adjusted	679.0	771.7	+13.7%

To be fair to the council, that is at least internally consistent. Both figures come from the same row type, so there is no sleight of hand between the two halves.

The problem is which row. Unadjusted is precisely the measure the recording change contaminates, and adjusted was sitting immediately beneath it. Picking unadjusted turned a 31-point gap into a 38-point one.

And the DfT flags it on the very rows the numbers came from. Thames Valley’s entries carry note 8: injury-based reporting adopted recently by this police force, severity adjustments based on a limited volume of data and potentially subject to greater revision. The severity column itself carries note 5: where forces have changed reporting systems, serious injury figures are not comparable with earlier years.

That is the whole argument in a footnote. The council built its headline on a comparison the DfT explicitly warns against, on a table where the warning is printed alongside the number.

Thames Valley is not a control group

There is a more basic objection. Thames Valley includes Oxfordshire, which makes it an odd benchmark to start with. Strip Oxfordshire out and the remaining eight authorities recorded a 36 % increase on adjusted figures.

Before anyone concludes that Berkshire’s roads became lethal overnight, they didn’t. Here is every South East authority ranked by adjusted KSI change in 2025:

Every authority that rose by more than 13% is policed by Thames Valley:

Every authority in the region not policed by Thames Valley sits between −17% and +13%.

The six authorities with the fastest-rising crash rates in the entire South East share one police force. That is not a road safety collapse. What it is, is a force part-way through changing how it records injury severity. DfT names the forces that have recently moved to injury-based reporting, and Thames Valley is on the list, alongside Wiltshire, Avon and Somerset, Cheshire, Cleveland, Derbyshire, Dorset, Dyfed-Powys, Hampshire, Leicestershire and Northamptonshire. DfT’s own warning is that severity adjustments for these forces rest on limited data and may be revised substantially.

Some of it is small numbers. Wokingham’s adjusted total went from 25 to 62, Bracknell’s from 26 to 37, so modest absolute changes produce dramatic percentages. But Buckinghamshire went from 132 to 192, and that is not a small-numbers story.

You can see it happening in the tables. The severity adjustment factor applied to the Berkshire authorities collapsed in 2025 – Wokingham from 1.24 to 1.03, Reading from 1.25 to 1.06, West Berkshire from 1.14 to 1.03. Oxfordshire’s barely moved, from 1.10 to 1.11. When the adjustment stops adjusting, it has stopped doing the job it exists to do.

None of that proves Oxfordshire’s reduction wasn’t genuine. It simply makes “Oxfordshire versus Thames Valley” a very shaky thing on which to hang a claim that a particular council policy is working.

What the figures actually show

Let me give the council its due, because the honest version of this is more interesting than my natural tendency towards the sceptical one.

Oxfordshire’s 2025 adjusted KSI total of 239 is the lowest in the ten years the DfT table covers. And since 2019, Oxfordshire is down 8.9% while Great Britain is up 0.6% and the South East is up 3.3%. Oxfordshire really has diverged from the national picture.

The question is why, and the release offers nothing on that at all.

Start with the timing. Between 2016 and 2019, well before blanket 20mph was council policy, Oxfordshire’s adjusted KSIs fell from 400 to 262 – down 34.5%, against 4.4% for Great Britain. The steepest improvement in Oxfordshire’s road safety record happened in the four years before the policy the council is now trying to credit for it.

Then look at what happened next. Adjusted KSIs were 262, 248 and 264 in the three years to 2021. After February 2022 they went 333, 255, 285 and 239. The year the 20mph policy and Vision Zero were adopted was the county’s worst since 2016.

Nor is Oxfordshire alone. Of the 101 local authorities recording more than 100 KSIs in 2024, forty-eight recorded falls in 2025, against a median change across the group of +0.7%. Ten fell further than Oxfordshire, including Devon, down 20.6%, and Cornwall, down 16.5% – large rural counties with comparable roads and no county-backed 20mph rollout. Oxfordshire’s fall is a good one. It is not one that requires 20mph to explain it.

The evaluation nobody has done

Did 20mph limits cause Oxfordshire’s improvement, as the council claims?

The press release certainly implies they did. The evidence it presents doesn’t. It’s our old friend in road safety again: correlation dressed up in causation’s party frock.

There is no breakdown of casualties on 20mph roads. No before-and-after analysis. No comparison with similar roads left at 30. No speed data. No collision causation analysis.

Yet more than 275 communities have introduced schemes, at different times, across a county with a decade of published casualty data. If ever there were a natural experiment sitting ready-made for a proper evaluation, this is it. Instead we are asked to accept the political conclusion and skip the working.

Finally, the release’s own opening line says serious and fatal collisions are “falling sharply,” and Cllr Epps says “the patterns are good across all indicators.” The confirmed figures don’t quite justify either. Fatal casualties fell from 20 to 19. Fatal collisions fell from 20 to 18. Casualties of all severities fell by 3%. Real improvements, all of them. Sharp, they are not. And “all indicators” is doing enough heavy lifting to induce a hernia.

Why this matters

If 20mph limits are delivering these results, that is genuinely good news. They are relatively cheap, relatively easy to introduce and parish councils like them. I would rather be wrong about this than right.

But this press release doesn’t demonstrate it. It assumes it, and hopes we will too. What we have been given is a warmed-over version of May’s announcement, carrying May’s superseded numbers, with every expression of doubt surgically removed.

If we are serious about reducing deaths and serious injuries on Oxfordshire’s roads, we need something rather more rigorous than recycled press releases and optimistic conclusions. We need a genuinely independent (i.e. not conducted by OCC or influenced by them) evaluation of the 20mph programme.

The council has the data. It has 275 schemes and ten years of casualty records. What it needs now is the appetite to find out.

---

Sources: OCC press releases, 29 May and 3 August 2026; DfT, Reported road casualties Great Britain, final results: 2025, published 30 July 2026; DfT tables RAS0401 and RAS0403. I’ve used severity-adjusted figures for percentage changes..