I know braking data is used to identify dangerous road sites / locations, and dangerous prior driving behavior of this car. The dangerous road site versus driver can be disentangled by statistics: if the road site / location results in similar braking behavior in other drivers, its more associated to the site, if the braking behavior is more correlated with the driver, its more attributable to this driver. However most people tend to have relatively regular commutes due to location of their home, their job, and their working hours, their shopping patterns etc. so it still entangled with other drivers, since they will tend to encounter the same subset of drivers, also having their own relatively steady probabilistic patterns.
For example, when a user suddenly brakes with large delta v, is it really due to this driver's aptitude to not predict the results of their driving decisions? Or is it because they frequently encounter the same reckless drivers?
It seems this could also be detected: for each braking event, consider a disc of sufficient radius and similarily downscore other drivers in this disc, use proper Bayesian inference of course, not naive linear score incrementing decrementing...
Simply downrating the driver of the braking vehicle risks taxing the less reckless chickens vis-a-vis the dare's in chicken or dare scenario's, naive calculations risk taxing specifically those parties that decrease the total kinetic energy in potentially dangerous situations, if the reckless drivers don't flinch even if it would have gotten them into trouble if a chicken had been a reckless dare.
The insurance company doesn’t care whether the risk is because of the driver or because of the road that the driver is on. They are on the hook for the risk either way.
This is why insurance companies also use your zip code to rate you. If you live near roads with more losses, you are more likely to incur losses. Doesn’t matter if you’re a great driver or not — someone might hit you.