Why would you measure by population? Shouldn't it be by distance driven?

Otherwise cities like NYC would have a really low rate since so many people don't have cars.

Well that depends. If what I care about is not dying then going twice as far but still dying isn't an improvement.

Measure the thing you actually care about. After issuing metal helmets to infantry in WW1 the number of men hospitalised with head injuries climbed enormously. Were helmets a bad idea? Nope. To get hospitalised first you need to survive your injuries.

The fact that most of us on Manhattan don't own cars unfortunately does not mean that we walk through tree lined streets empty of large, heavy and dangerously driven motor vehicles.

For this subject matter per capita (should really be per licensed driver to account for different rates of public transit usage between populations but per capita is close) is appropriate because people care about the likelihood of something over a lifetime.

Miles driven would be useful too but not as directly useful in evaluating odds of any one random person dying in a car crash.

Regardless, I'm be interested in seeing the stats broken down by accident type (which is tracked by the police and insurance companies to some extent).

My suspicion is that there's several accident modes that are more common in North America that make up most of the difference.

I also suspect that just like pretty much ever other thing that's common at scale but statistically unlikely at the individual level socioeconomic status is the best predictor.