Why near-miss data matters more than crash data
Collisions are counted. Near misses are what riders actually experience, what changes their behaviour, and what almost nobody records.
Almost every decision about cycling infrastructure is made using collision data: police-reported incidents involving injury. It is the only dataset that exists consistently, it is collected to a standard, and it goes back decades. It is also a poor description of what riding a bike is actually like.
The counting problem
Collision statistics have two well-documented weaknesses. The first is under-reporting: a substantial share of cycling incidents never reach a police record, particularly those without a motor vehicle or without hospital treatment. Studies that link hospital admissions against police data consistently find casualties in the health records that never appear in the road-safety ones.
The second is subtler and more damaging. A junction that frightens riders so reliably that they stop using it will, over time, generate excellent safety statistics. No riders, no collisions. Measured on collisions alone, successful deterrence is indistinguishable from a safety improvement.
What near misses capture instead
Work on cycling near misses (Rachel Aldred's Near Miss Project in the UK is the best-known example) established two things fairly clearly. Non-injury frightening incidents happen at a rate orders of magnitude above collisions: frequently enough that a regular commuter can expect them on a routine basis rather than a yearly one. And they have real consequences, shaping route choice, riding frequency, and whether people continue cycling at all.
That last part is the argument. If the goal of a road authority is more people cycling, the thing suppressing that number is not primarily the collision rate. It is the daily experience of being passed too closely, which is invisible to every dataset the authority holds.
The obvious objection
The classic move here is to invoke the safety triangle: many minor events sit beneath each serious one, so reducing the minor events reduces the serious ones. That model deserves the criticism it receives. The causal chain from near miss to fatality is not a simple ratio, and treating it as one has led safety programmes astray in other industries.
So we would put it more carefully. Near-miss data is not valuable because it predicts deaths. It is valuable because it measures something collision data structurally cannot: the conditions riders are actually experiencing, at a sample size that supports conclusions about specific stretches of road rather than whole cities.
Why it has historically been hard to collect
Near-miss data has always depended on self-reporting, which brings the problems you would expect. People report when annoyed and forget when busy. Estimated distances are unreliable. Whoever is motivated enough to fill in a form is not a random sample of riders. The result is data that establishes near misses are common, but struggles to support claims about a particular corridor at a particular hour.
Passive, instrumented collection changes the shape of that problem. If every pass on every ride is measured, not just the ones that frightened someone enough to file a report, you get a distribution rather than a complaint log. You can compare the same road at 7 am and 7 pm. You can compare a corridor before and after a bike lane. And crucially, you count the passes that were fine, which is what makes the close ones interpretable as a rate.
What we intend to do with it
Every Rydiq ride measures every overtake, not only the ones that breach a threshold. Our intention is to publish aggregate, de-identified distributions from the beta cohort once the sample is large enough to be honest about, broken down by road class and time of day, with the methodology stated so it can be argued with.
Individual riders get a per-ride distribution in the analytics drawer today. The aggregate picture is the more interesting one, and it is the one road authorities have never had.