> it is making the comparison from a tiny pool of autonomous test cars, about 50 of them in California, versus 269 million conventional cars as of 2013 That is

> it is making the comparison from a tiny pool of autonomous test cars, about 50 of them in California, versus 269 million conventional cars as of 2013 That is the flaw: the control group should be some regular cars in California traveling the same areas at the same hours, not the US average. I suspect that comparing any group of 50 cars in California (autonomous or not) against a US average that takes into account many cars travelling along empty areas will result in the former having a lot more fender bender rates.