U.S. warned Tesla to stop giving "misleading statements" about the Model 3's safety

[orenwolf]

@navarro

if you read the correspondence provided at the link above, the nhtsa is saying that there is no validity to comparing vehicles results if the vehicles differ in weight by more than 250 pounds. tesla might have a leg to stand on if they claimed they were the safest vehicle in their weight class but to haul off and say they are safest over all vehicles is wrong and dangerous.

So changing the blog-post to “the lowest probability of any car the safety agency has ever tested in this weight class” would have somehow avoided all this drama and alleviated the “danger”?

Or are you suggesting that people went out and bought Teslas instead of a ford Expedition or whever the heavy trucks of the world are, and the danger is there? Because I question whether or not there is a venn diagram of those two classes that actually overlap.

[navarro]

i think the danger the nhtsa perceives is in making a false comparison based on the metric of the safety rating by using it as an absolute scale rather than as a relative scale. i don’t know if that would necessarily have alleviated the danger, such as it is, of using the star rating as an absolute scale but it would have handles the stated objection of the nhtsa in their letter.

[KatjaKat]

This is a nice explanation of power in science and statistics. Kudos! Makes me wonder how many collision tests are actually done though. With something as expensive as a Tesla and possibly using cadavers, even five per car model seems like a high number. So their statistical power really can’t be very good.

[DukeTrout]

It has to be at least 5 per model, since they do 5 crash tests: front, rear, side barrier and pole, and rollover.

[orenwolf]

Please remember that Assume good faith is one of our guidelines here. Someone disagreeing with your position is not automatically a shill or fanboi (two terms thrown around exceedingly often when it comes to cars, computers, or gaming).

You can argue against a position without needing to throw shade at the poster.

Thanks.

[KatjaKat]

So they probably actually just do one or maybe two of each, add up the points and that’s it. Putting it closer to a diagnostic test. Wonder if they have stats on sensitivity/specificity.

[DukeTrout]

I don’t think they could even calculate sensitivity/specificity for individual tests per vehicle, since those stats require a sampling of data points. They could probably do a meta analysis using their full database and normalizing to the mean/median and pulling in the real-world crash data they have access to.