"All of that" is not encompassed in the human driver's home zip code. OP is talking about more than just the zip code, they're also talking about restrictions on the times of day Waymo operates, the types of vehicles they use, whether Waymo shuts down in adverse weather, etc. I'm not sure what all the current and past restrictions are, but those have to be controlled for.

The study you linked only discusses one uncontrolled variable (which is a bad sign given how many others there are that they don't talk about) but it's one I hadn't thought of:

> A limitation of the selected human baseline is that the location of crashes that generate claims is not known, which limits the ability to filter claims based on Waymo's Operational Design Domain (ODD). As a result, whereas the Waymo ODD largely does not include freeway driving, the human database includes miles driven and claims which occur on freeways. Due to variations in collision frequency per million miles between freeways and non-freeways, this may have led to a baseline which may be more conservative than a roadway- matched baseline.

Fair enough. Maybe not “all of it”, but most of it is covered.

Waymo has always operated 24/7 since the beginning and had 99.4% fleet uptime [1] during the past winter season (SF had record rains). They used to shut down during inclement weather in Phoenix years ago, but their SF rain operations have scaled up with the mileage. Waymo vehicles use its own set of sensors and don’t rely on the vehicle’s safety features, so I’m not sure how types of vehicles is relevant for the comparison.

This is pretty good as far as “apples-to-apples” comparisons go.

[1] https://waymo.com/blog/2023/08/the-waymo-drivers-rapid-learn...

I don't live in SF, but I was under the impression that Waymo's 24/7 operation is a relatively new thing [0].

EDIT: On further research, it looks like for Waymo this change only applied to when they could charge fares, so OP is correct that they were authorized to operate 24/7 [1].

[0] Article from August 10 of this year: https://www.theverge.com/2023/8/10/23827790/waymo-cruise-cpu...

[1] https://www.cpuc.ca.gov/news-and-updates/all-news/cpuc-appro...

Nope. It was Cruise who gave rides only at night. Waymo has operated 24/7 since day 1. The only restriction has been who is approved to ride (they have a long waitlist).

>so I’m not sure how types of vehicles is relevant for the comparison.

Surely "car with ABS and good tires" vs "car without ABS and with bald tires" makes a difference. And presumably waymo operates exclusively the former while some people drive the latter.

Sure, but that's driver discretion, so it's a factor that we should be considering.

That depends on exactly what you're trying to measure.

Abs has been mandatory in the us since 2012, and was very common long before that (my base model truck from 1999 has it) we can consider cars without abs statistical noise by now.

>Abs has been mandatory in the us since 2012

The average US vehicle's age is more than 12 years.

>and was very common long before that (my base model truck from 1999 has it)

My non-base-model car from 1995 didn't.

>we can consider cars without abs statistical noise by now.

I'm not so sure that's true. But even if it is, that doesn't mean that there are no meaningful differences in safety between an exclusively new (or at least very recent) and maintained vehicle fleet and the general population.

And for cars that are geofenced. Not accurate comparison.

If you're comparing to other accidents in the same (geofenced) zipcode, then why is it not an accurate comparison?

The linked study even calls this out: the zip codes they controlled for are the residence zip code, which means accidents that happen on the freeway (where Waymo typically doesn't drive) or even in other zip codes are included in the human baseline.

Precisely.

In Phoenix, the waymo drive on the freeway around the city, and given the number of crashes I see in that city I have few doubt that Waymo cars are safer.