Where London’s roads are most dangerous
People killed or seriously injured on London’s roads since 2010, ward by ward, measured against the people there, the people who live there and the traffic, with a test of whether 20mph limits helped.
- Year
- 2026
- My role
- Solo rebuild and extension of the geographic analysis I owned in a 2024 LSE group project
- Tools
- Python, pandas, GeoPandas and Leaflet
- Data
- DfT STATS19 collision and casualty records, DfT road traffic counts, DfT casualty values (RAS4001), 2011 Census workday and resident population and London Datastore ward and borough boundaries
The question
Raw casualty counts make central London look like the most dangerous place to travel, partly because so many people work and travel there. Dividing by residents makes it worse: the City of London has about 7,400 residents but about 360,000 people there on a working day. So I measured casualties against the workday population, the people actually present on a typical weekday, using the 2011 Census, because the 2021 Census took place during lockdown. Then I added the traffic itself as a third measure.
What I did
- Took every collision recorded by the Metropolitan Police and City of London Police from 2010 to 2025 from the Department for Transport’s STATS19 data, and matched each of the 382,744 with a location to the 2011 ward it happened in.
- Counted people killed or seriously injured (KSI), the measure behind London’s Vision Zero goal, using DfT’s adjusted severity estimates. These correct for the Metropolitan Police’s switch to a new injury-recording system in late 2016, which otherwise makes serious injuries jump by about 57% from 2016 to 2017.
- Measured exposure three ways: workday population, residents, and traffic from DfT’s road traffic counts. For boroughs the traffic covers all roads; for wards only major roads have counted lengths, so the ward traffic view compares casualties on major roads with the traffic on them.
- Valued casualties with DfT’s official figures for the cost to society of each one: about £2.6 million per death, £294,000 per serious injury and £22,640 per slight injury, at 2025 prices.
Explore the map
Highest in this view
| Area | Rate a year | People a year |
|---|
Data: DfT STATS19 (2010 to 2025) and road traffic counts, DfT casualty values (RAS4001, 2025 prices), 2011 Census via Nomis, London Datastore boundaries. Base map © OpenStreetMap contributors. Borough logos belong to the councils.
What I found
The way you measure exposure changes the story. Per resident, the City of London, St James’s and the West End are London’s most dangerous wards. Per workday population, the City falls from 1st to 610th of 625 wards, and St James’s and the West End drop below 500th. Per vehicle-km, the order flips back: the City has the highest rate of any borough, about 49 people killed or seriously injured per 100 million vehicle-km from 2022 to 2025, against about 5 in Havering and Hillingdon. Central streets carry little motor traffic for the number of people walking and cycling on them, so each measure tells part of the story.
On the workday measure, risk concentrates in inner south and west London. Lambeth, Wandsworth, Kensington and Chelsea, Southwark and Richmond upon Thames have the highest borough rates, at about 6 to 7 people killed or seriously injured a year for every 10,000 people there on a workday. The highest ward, Cremorne in Kensington and Chelsea, averages about 13 a year among roughly 8,000 people.
Most of the people hurt weren’t in cars. From 2022 to 2025, pedestrians made up 32% of those killed or seriously injured, cyclists 27% and motorcyclists 21%: four in five were walking, cycling or riding a motorbike.
Some hotspots are through-routes. In two of the top five wards, East Sheen and Ham, Petersham and Richmond Riverside, about three-quarters of those killed or seriously injured were cyclists, which may reflect the popular cycling routes in and around Richmond Park. Many of the people hurt there are passing through, so a population measure overstates the risk to locals.
The cost to society is large. At DfT’s 2025 values, the casualties reported in London cost about £1.9 billion a year from 2022 to 2025, three-quarters of it from deaths and serious injuries.
London’s roads have become safer, but slowly. Killed or seriously injured casualties averaged about 4,660 a year from 2010 to 2012 and about 3,850 from 2022 to 2025, a fall of roughly 17%, while deaths fell from about 140 a year to about 103. That leaves London a long way from Vision Zero’s goal of no deaths or serious injuries by 2041, and 2025 was the worst year since 2018.
Did 20mph limits help?
Many London boroughs have lowered their default speed limit to 20mph. To test whether that cut serious casualties, I compared boroughs that switched with boroughs that didn’t, before and after the switch: a staggered difference-in-differences design.
Council records of when each borough switched are patchy, so I dated each switch from the data itself: the year in which the share of a borough’s collisions on 20mph roads jumped by at least 20 percentage points to above 40%. That only works from 2017, because the Metropolitan Police’s 2016 recording change also changed how speed limits were recorded. It picks out Ealing, Merton and Richmond upon Thames in 2020, and Kensington and Chelsea, Wandsworth and Westminster in 2021. The comparison group is the nine outer boroughs where 20mph roads stayed below a quarter of collisions.
I found no evidence of a fall. Before the switch the two groups moved together, which supports the comparison. After it, killed or seriously injured casualties in the switching boroughs were about 12% higher relative to the comparison group, but the 95% interval runs from 2% lower to 31% higher, and a placebo test that gives the same dates to random boroughs produces a gap that large 18% of the time. Per vehicle-km the gap is a little larger, about 16% (1% to 37%), and a placebo gap that large appears only 4% of the time, so if anything serious casualties rose relative to the comparison boroughs.
This design can’t separate the speed limit from everything else that changed at the time. The switches happened during the pandemic, when travel changed sharply and unevenly across London, for example through the growth in cycling and temporary street schemes, and every comparison borough is in outer London. A road-level comparison of streets that changed limit with similar streets that didn’t would be the stronger test.
Caveats
- The workday population comes from 2011. London’s daytime population has shifted since, not least with hybrid working.
- Ward figures rest on small numbers: even the highest wards see 10 to 14 people killed or seriously injured a year, so a few collisions can move a ward a long way up or down.
- STATS19 only covers collisions reported to the police, and many slight injuries, especially to cyclists, are never reported.
- The 20mph dates are inferred from the data rather than taken from council records.
What I’d do next
- Check the inferred 20mph dates against council records, and run the road-level comparison.
- Use DfT’s cycle counts to measure where people cycle, since motor traffic understates cyclists’ exposure.
- Build an ArcGIS StoryMap version, and compare London with Toronto’s open collision data.
This rebuilds and extends the geographic analysis I did for a 2024 LSE group project, which used TfL’s collision feed up to 2019.