How every number here is produced
Every dataset on this site is built from named official files by automated pipelines whose rules are documented below. If you can't trace a figure of ours back to the official record, that's a bug — tell us.
Named source files
- • Road collisions: DfT STATS19 —
dft-road-casualty-statistics-collision-last-5-years.csv, plus the matching casualty and vehicle files, from data.gov.uk (Road Safety Data). - • Traffic volumes: DfT AADF local-authority traffic table from roadtraffic.dft.gov.uk.
- • Vehicle crime: police-published street-level records via the data.police.uk API (category: vehicle crime).
- • MOT results: DVSA anonymised test results —
dft_test_result_2019/2020/2022/2023.zipfrom data.gov.uk. - • Workplace injuries: HSE RIDDOR tables —
ridind.xlsx, ridkind.xlsx, ridreg.xlsxfrom hse.gov.uk.
All under the Open Government Licence v3.0. Our pipelines run automatically when sources publish; every API response and page states its data vintage.
The rules, exactly as coded
Collision clusters (hotspots)
Collisions are grouped by rounding coordinates to 3 decimal places (≈111m cells). A location is a “persistent” hotspot if it recorded 2+ collisions in at least 4 of the last 5 years. Clusters are labelled with the nearest named road via OpenStreetMap reverse geocoding and may span a junction of several roads.
Fatal collisions vs fatalities
We count fatal collisions (incidents in which someone died). DfT headline releases count fatalities (people who died) — a fatal collision can involve more than one death, so our figure is slightly lower than DfT’s casualty count by design. Casualty counts on this site are people; collision counts are incidents.
Area definitions
14 cities use their ONS council-district boundary; London is all E09* boroughs; Wiltshire is the unitary authority; Sheffield uses an 8km radius of the centre (its district code changed in the 2025 data); other towns use a 6km radius (Chippenham 5km). Each area page states its definition. Per-100k rates use mid-2023 ONS population estimates for the named town/district and are indicative where the area is radius-defined.
Risk per vehicle-mile (districts)
District KSI (killed or seriously injured collisions) divided by billions of vehicle miles from the DfT AADF local-authority table. The traffic vintage in use is stated in the API meta; where traffic data is older than collision data, treat the normalised rate as indicative.
Vehicle crime
Street-level records within a 1-mile radius of each town centre, all police “vehicle crime” categories, summed monthly. Locations are police-anonymised snap points — street-level, deliberately not addresses. Scotland is absent (Police Scotland doesn’t publish) and Greater Manchester is excluded for known feed gaps. Year-on-year compares the latest 3 months to the same 3 months a year earlier.
Car safety outcome grades
For each car model (STATS19 vehicle type 9), we take every occupant casualty (driver or passenger) over 5 years and compute the share killed or seriously injured. Models need 500+ occupant casualties to appear. Bands: A = lowest ~20% of serious-outcome shares, B to the 45th percentile, C to the 75th, D above. This is a real-world outcome measure shaped by driver demographics and mileage as well as vehicle design — it is not a crash-test rating; see Euro NCAP for those.
MOT pass rates
Class 4 vehicles, normal tests only (no retests). Pass = P; fail = F and PRS (pass after rectification counts as an initial failure, matching DVSA’s initial-failure definition). Models need 10,000+ tests. 2021 is absent from our series: the published 2021 file parsed to zero valid class-4 normal tests and we exclude rather than estimate. 2024+ bulk files are not yet published by DVSA.
Workplace injuries
Employer-reported RIDDOR injuries, latest published year. Industries are SIC sections; rate rankings require 100+ reported injuries, and the “households as employers” category is excluded as a small-denominator artifact. RIDDOR under-reports true injury levels — HSE says so itself.
Projections
Any figure labelled “modelled” or “outlook” is a least-squares trend over the yearly series shown, floored at zero. Projections are estimates, clearly distinct from measured figures, and never mixed into tables of actuals.
Corrections policy
If you believe any figure is wrong, tell us with the page URL and the official source you checked against. Verified errors are corrected in the data pipeline (not hand-edited), so the fix propagates to pages, CSVs and API automatically, and this page is updated where a rule changes.
Been in an accident? Let's sort it.
One call starts your claim. Our line is open 24/7, 365 days a year.