Part of the problem is it just gets shoved into a model without enough data support. So I hit the gas hard merging onto the freeway? Of course; your brakes are much more powerful than your throttle if you need to take evasive action, and merging onto the freeway at 50mph is dangerous. The model just sees that you like keeping your foot on the gas from time to time. To keep your rates low you have to drive more dangerously, and that's not the sort of incentive structure I like to participate in.
You generally don't need to hit the gas hard to merge onto a freeway, you need to accelerate evenly on the ramp and be at 65-70 by the time you get to the merge point, being aware of traffic and where the gap is that you're going to merge into, and hitting it.
And occasionally having to hit the gas or brake hard is not going to raise your rate. They look for more of a pattern than that. If you're always braking hard and then accellerating hard, surging and changing lanes, consistently speeding, etc. that's going to be the trigger and what do you know, that's also the kind of driving that causes crashes.
South Bay Area freeway onramps are notoriously short. My amount of "aggressive" acceleration decreased a lot just from moving. I used to also have to maintain huge gaps in case the driver in front stops on the onramp, which happened every other day.
The thing about these models is they do increase accuracy, just not in a fair way.
Do you think car manufacturers hire people who can program or even care to program this stuff into the tracking? Knowing some of the people who those tasks get outsourced to, the answer is no.
User drives through grass which the sensors detect as hard collision. Gettem, NexusLexus!
Do you think insurers care to differentiate between those things? Looking at the equivalent behavior from UnitedHealth and their auto claim rejection behavior, my guess is also no.
Any data collected about you can be and will be used against you.
