THE ACTUARY LENS · ROAD RISK
Which road risks persist?
Interpret the evidence on injury-collision occurrence, severity and exposure. The detailed analysis lives in the STATS19 Data Lab.
Calculated from this release using the exact same selectors as Data Lab. Eligible-authority totals are not whole-GB totals. A repeated ordering is descriptive evidence, not a causal result or an insurer pricing validation.
Bodily injury: count people as well as collisions
In 2025, 127,883 casualties were recorded. 34.8% were in collisions injuring more than one person. The Lab separates casualty role, age, road conditions and adjusted KSI, with every person counted once per breakdown.
Multi-claimant exposure can matter for bodily-injury accumulation, but police injury counts do not measure compensation or developed insurer losses.
Inspect injury burden and multiplicity →Comparable EV / ICE outcomes
The Lab standardises conditional collision severity across shared vehicle-age, driver-age, road, urban/rural and light strata, separating cars from taxis. It reports how much of each fuel population remains in common support. The frozen 2025 file has no vehicle-age values, so that year’s adjusted comparison is unavailable.
These controls improve descriptive comparability. They do not establish accident frequency per mile, fault or a causal powertrain effect. Sparse taxi support and unmeasured vehicle/driver differences remain important.
Residence is not road exposure
Recorded home-area agreement and published home-to-collision distance bands now sit alongside explicit missingness in the Lab. Collision geography describes where events occurred; pricing geography usually describes where a policyholder lives. Residential mileage or policy exposures are needed before estimating a postcode relativity.
Frequency: separate volume from rate
The comparable panel moves from 42.86 to 30.38 injury collisions per 100 million motor vehicle miles between 2016 and 2025. Changes in how much, when and where vehicles travel can affect the result.
For an insurer, this supports testing road exposure alongside annual mileage. It does not establish a change in insured claim frequency: damage-only events and many unreported injuries are outside this dataset.
Inspect annual rates and coverage →Persistence: useful structure, with competing explanations
The highest-risk discovery group remains above the lowest in 4 of 4 eligible later years. Earlier-versus-later area rank correlation is 0.954. These comparisons retain the same areas and freeze the original groups.
Stable road mix, exposure patterns and reporting differences can also generate persistence. A pricing study should test decomposed geographic features alongside the existing postcode factor, using features available at quote and later claims held out from model construction.
Compare geographic persistence and exceptions →Severity: injury outcomes and financial loss differ
Adjusted serious-injury measures help compare road outcomes through changes in police reporting. They do not estimate compensation, repair costs, credit hire or large-loss development. The Lab separates casualties per mile from the proportion of recorded injury collisions that are fatal or serious.
For UK motor insurance, these are candidate indicators for bodily-injury exposure. Financial severity requires an insurer’s developed claims experience.
Vehicle and weather comparisons need exposure context
The detailed 2021–2025 records support road, time and vehicle comparisons. Licensed-car stock joins provide a fuel-cohort involvement indicator. Raw stock comparisons do not control for vehicle age, mileage, driver selection or road use. The separate matched analysis reports its own controls and remaining limitations.
The weather view joins annual country weather to a fixed geographic road panel. It describes association across years; it does not estimate the effect of rain or ice on an individual journey.
A road-exposure or telematics analysis should use measured driving exposure in each road and time environment before translating these patterns into pricing factors.
Open vehicle and weather evidence →