Ahh yes we should go from an actual indicator of actual accidents to a computer interpretation of g forces and what an insurance company wants to use to make more money.
Seems totally reasonable to ditch actual accident data for interpolation. Super smart.
Accident data is rare and random. You might be a good driver and have gotten unlucky once. You might be a terrible driver and gotten lucky every single time.
That’s what happens if you have such an extremely tiny sample size.
In the U.S., millions of car accidents occur annually, with around 6.1 million police-reported crashes in 2023, leading to roughly 40,900 fatalities, though total crashes including non-injury ones reach over 13 million.
There are 237.7 million licensed drivers in the US. That means there’s 0.026 police reported crashes per driver per year. (Crashes not reported to the police are usually also not reported to insurance and thus don’t matter in this discussion.)
Or to put it differently, that’s one crash per 39 years of driving per driver or on average 1.6 crashes in a lifetime.
Yes, every crash is one to many and every fatality of course as well. In that regard it’s far too many, but that’s not what we are talking about.
We are talking about insurances estimating the likelyhood of future crashes of a driver. That means, on average, insurance has 1 data point per driver, and for anyone younger than 35 likely 0 data points.
That’s not nearly enough to make any kind of statistically significant guess on how likely someone is to cause a crash.
For any statistically significant result you’d need at least a few dozen data points.
For that crashes are far, far too rare, so it makes sense to try to get better data that actually has some kind of significance.
Fuck the insurance companies, they should all be nationalized, with zero compensation for the capitalists running them now.
But yeah, we really don’t have useful accident data on the individual level. Tracking behavior that makes crashes more likely (and especially severe crashes, which cause a lot of damage) is a much more reasonable way to determine rates.
Ahh yes we should go from an actual indicator of actual accidents to a computer interpretation of g forces and what an insurance company wants to use to make more money.
Seems totally reasonable to ditch actual accident data for interpolation. Super smart.
Accident data is rare and random. You might be a good driver and have gotten unlucky once. You might be a terrible driver and gotten lucky every single time.
That’s what happens if you have such an extremely tiny sample size.
wut.
In the U.S., millions of car accidents occur annually, with around 6.1 million police-reported crashes in 2023, leading to roughly 40,900 fatalities, though total crashes including non-injury ones reach over 13 million.
And its not random, some car brands have much higher crash rates.
And some brands have 3x more drunk drivers.
There are 237.7 million licensed drivers in the US. That means there’s 0.026 police reported crashes per driver per year. (Crashes not reported to the police are usually also not reported to insurance and thus don’t matter in this discussion.)
Or to put it differently, that’s one crash per 39 years of driving per driver or on average 1.6 crashes in a lifetime.
Yes, every crash is one to many and every fatality of course as well. In that regard it’s far too many, but that’s not what we are talking about.
We are talking about insurances estimating the likelyhood of future crashes of a driver. That means, on average, insurance has 1 data point per driver, and for anyone younger than 35 likely 0 data points.
That’s not nearly enough to make any kind of statistically significant guess on how likely someone is to cause a crash.
For any statistically significant result you’d need at least a few dozen data points.
For that crashes are far, far too rare, so it makes sense to try to get better data that actually has some kind of significance.
Fuck the insurance companies, they should all be nationalized, with zero compensation for the capitalists running them now.
But yeah, we really don’t have useful accident data on the individual level. Tracking behavior that makes crashes more likely (and especially severe crashes, which cause a lot of damage) is a much more reasonable way to determine rates.