On the other hand I think people mostly know the score. I'm sure the driver in this crash knew he should have been looking and autopilot was not 100% reliable. I enjoy riding motorcycles which have a stupid high death rate compared with cars. It's a free world to some extent.
One immediate problem with real world stuff like this is selection. How do we know that Tesla is safer because it is built more safely than normal cars, rather than that perhaps people who are more safety conscious buy a Tesla? After all it's a more expensive car, the kind of thing you might only be able to afford once you're a bit older, the same time that you might have kids to worry about. But older people are also not the type to be racing around fast and furious style. Maybe there's an imbalance in the man/woman distribution?
It's a real can of worms to unpick these kinds of things. You find it all the time in social sciences.
Maybe you can find stats for Teslas without autopilot, or other cars with autopilot, and other cars without auto. And stats for how Tesla drivers are distributed according to age and sex?
Did you mean to ignore the fact that Tesla knows the score?
Tesla knows all of those statistics about its drivers, and knows how those numbers compare to other published data.
And they chose to write up the data in the manner they did.
In which configuration of all possible worlds is this ok?
In light of this new context, has your thinking around this subject changed?
Of course Tesla and any other firm that releases stats to prove a point need to do so honestly. And they would have made the exact same considerations that I listed. So yeah, they should have released more representative stats.
I'm sure the PR people don't know anything about stats though, plus they would want to pick whatever they think makes the company look good.
10/100 is same percentage as 1/10, so groups would have same mean.
Your point is fair though, but it would be more appropriate to make the example 0/10 in the small group when there was a ~1/10 chance of that happening (based on larger groups distribution) so its not super significant.
HN's Contrarian Skeptics Brigade can correct me on this (I repent!), but I think this is not very different than saying:
"10 out of all 100 people get this disease. We tested this hand-picked group of 10 people and found that only 1 of them had it. This small group is ten times more healthy than the general population!"
The number of Tesla cars with Autopilot in circulation must certainly be a small fraction of all vehicles on US roads (including bicycles, etc). It makes sense that, if you look for an event in a smaller population you should expect to see it happening in a different rate than in a larger population - for the same reason that if you toss a coin ten times and count the proportion of heads it may well be different than 0.5, but if you toss it one million times you're much more likely to see it landing heads half of the time. More observations allow you to better estimate the true distribution of events.