British highway authorities, ranked on how much harm their roads do

How the score is built, and what it will not tell you

Each authority is percentile-ranked against every other on four measures, which are then weighted into a single 0–100 number. Higher means more dangerous.

Two corrections most published comparisons skip

One year of casualties is too few to rank on. An authority recording sixty KSI moves several places on chance alone, so a table built on a single year is partly measuring luck. The rate here is a three-year average, which is what road safety analysis normally uses and what makes a movement in the table mean something. Single-year counts are still shown on each authority page, because that is the figure people are used to seeing quoted.

Serious injuries are not recorded the same way everywhere. Police forces moved to injury-based reporting — CRASH, and COPA in the Metropolitan Police — on different dates between 2012 and 2026. Under those systems the officer records injuries from a list and severity is derived automatically rather than judged at the scene, which added between 5% and 15% to the national serious-injury total, and a different amount in every force. Comparing raw KSI between authorities across that transition compares two different definitions.

DfT’s own guidance says the adjustment is required when comparing across areas. But since the July 2026 release, DfT publishes adjusted figures only at national level — the local authority table no longer carries them, and no official adjusted KSI by local authority exists anywhere. What DfT does publish is the per-casualty probability weights, inside the open data. So we apply them: adjusted KSI = fatalities counted, plus the summed serious weights. Fatalities are never adjusted.

The town-level score is a different measure

Our town and city pages carry their own Road Risk Score, and it is not the same number as the one above. Towns are not highway authorities and DfT does not publish traffic estimates for them, so that score is normalised by resident population instead of vehicle miles, and it adds vehicle crime — which is published per town but not per authority. Its weights are KSI per 100,000 residents 40%, vehicle crime rate 20%, vulnerable road user share 15%, trend 15%, fatal share 10%.

The two will sometimes disagree about the same place. A town whose per-head score is mild but whose per-mile score is severe is one where the danger is being carried by people passing through. The band letters are not interchangeable either: the town score bands at 80/60/40/20, this one at 70/50/30, so a town “Band C” and an authority “Band C” do not mean the same thing.

Where this measure is weak

• •Vehicle miles only count vehicles. Where most of the people hurt were walking or cycling, the denominator understates how many were exposed, and the rate comes out high for a reason that has nothing to do with how the roads were built. The City of London is the clearest case and is held out below.
• •Involvement counts overlap. A collision involving both a pedestrian and a cyclist is counted in both figures, so the vulnerable road user measure can exceed the true share of collisions and, in very small authorities, can exceed 100%. It is a comparable index across areas, not a proportion.
• •Adjustment is a model, not a measurement. For forces not yet on injury-based reporting, the adjusted figure estimates what would have been recorded if they were. DfT is explicit that these estimates depend on the model and that other approaches would give other numbers.
• •STATS19 records what police attend. Collisions that are never reported never appear, and under-reporting is worst for cyclists and for slight injuries. The seriously injured and killed figures used here are the most consistently recorded, which is why the score leans on them.
• •A high score is not a verdict on a council. Road danger tracks road type, traffic mix and deprivation as much as it tracks policy. A dense urban authority with a trunk road through it starts from a harder position than a rural one.
• •Boundaries are administrative, not behavioural. People are hurt on the road they are travelling on, not the one they live near, and the responsible authority for a given stretch is not always the one it sits in.

The 13 authorities held out of the ranking

These have the data but not enough of it. They keep their pages and every figure on them; what they do not get is a rank, because at their size the derived percentages describe chance more than they describe the place.

Source: DfT STATS19 road safety data and DfT road traffic estimates (vehicle miles, 2025), both Open Government Licence v3. Headline counts are for 2025; the ranked rate uses a three-year average and, where the adjustment data is present, is adjusted for the change in severity recording; the trend component compares 2021–2025. Peer groups are ONS statistical nearest neighbours (OGL v3). Compiled and reviewed by Stephen Hall, Data lead, Swift Assist. Last rebuilt 2026-10-08. Method and known limitations are set out above and on our methodology page.