Home / Notes / Crash counts and exposure
Field note · road safetyCrash counts need context.
A high count tells you where collisions were recorded. It does not, on its own, say how much travel happened there or whether the next collision is likely. The interface should let analysts inspect those different questions separately.
Keep four kinds of evidence distinct
Recorded crashes describe reported events in a defined period. Travel exposure estimates how much movement occurred in the area. Near-miss indicators, such as harsh braking from connected vehicles, are signals from a particular observed population. Modelled risk is an output of a method with its own inputs and uncertainty. None should borrow the label or legend of another.
Let the analyst compare the same place
Keep geography, time period and road-user group visible while switching between views. A useful path is: select a road segment, inspect the recorded collisions, compare exposure and then look at any other signals. Show the source and coverage of each layer near the result.
Do not hide the denominator
A road carrying more people may record more collisions simply because more journeys happen there. If a view presents a rate, name its denominator and period. If exposure is estimated or unavailable, say so before a colour scale implies a precise ranking.
Make room for investigation
A highlighted segment is a reason to look closer, not an automatic intervention order. The screen should make it easy to inspect underlying records, see limitations and take findings into a review process.
The SOTERIA consortium's public account describes its Madrid prototype combining accidents, travel demand, connected-vehicle events and predictive outputs. Our SOTERIA case study describes our front-end contribution; the data and models belong to the wider project.
Building a safety-analysis view?
Start with the decision your users make and the sources you can provide.