This should be looked into IMO. The Model S (and X, I think) were celebrated as getting the highest safety marks by a big margin. Either the safety ratings are garbage, Tesla’s manufacturing is substandard, or something else fishy is going on.
Maybe the study is flawed by incorrect statistical analysis when combining two disparate data sources (FARS and iSeeCars mileage estimation). I'd really love to see where the non-hybrid Honda CR-V ranks compared to the hybrid CR-V. Same for the Hybrid vs. non-Hybrid Corolla. Also a rough sanity check would be the fatalities for each vehicle model divided by the total number sold of that model.
This reminds me of better doctors having higher fatality rates because the most critically ill patients seek them out. Since Teslas are at the top of every crash test safety study, I suspect that this is a spurious outcome of Simpson's paradox and that proper controls would make the effect disappear.
Even more fun would be if alcoholics were more likely to buy Tesla's because of the self-driving aids. And it turned out this cohort was less likely to be involved in fatal accidents than the non-Tesla-driving alcoholics. So that even after controlling for more alcoholics purchasing Tesla, there would be a reduction in the overall fatality rate.
But my gut feel is that there is a systematic bias in the iSeeCars mileage data that they aren't properly accounting for (say prevalence of leasing EVs skewing the mileage estimates).
With all of the telemetry, am I correct to presume that Tesla probably knows the current odometer reading for pretty much every car they've ever sold? So they could easily give out the denominator for the fatality rate in the original reporting? The total miles driven seems like it could be the weak link in the iSeeCars data, since I'm assuming they have to infer the total miles driven off of their used car sales data.
> Therefore, he added, their elevated accident rates likely “reflect a combination of driver behavior and driving conditions.”
In other words, this Tesla person is confirming the no-brainer that the behavior of buying a Muskmobile is not statistically independent of the behavior of being a dickhead on the road.
"Our cars are incredibly safe. It's just that some of the idiots who comprise our customer base are more dangerous."
I don't buy the part about driving conditions, because when you're looking at national level statistics from everywhere, everyone drives the same damn conditions. Conditions can only explain one particular accident. Or a string of accidents at a particular intersection.
I think no because the idea that most accidents happen close to home is a fallacy. It's only true an absolute numbers, because most driving happens close to home. This story is about fatal accidents, not just accident rates.
This should be looked into IMO. The Model S (and X, I think) were celebrated as getting the highest safety marks by a big margin. Either the safety ratings are garbage, Tesla’s manufacturing is substandard, or something else fishy is going on.
>something else fishy is going on.
Maybe the study is flawed by incorrect statistical analysis when combining two disparate data sources (FARS and iSeeCars mileage estimation). I'd really love to see where the non-hybrid Honda CR-V ranks compared to the hybrid CR-V. Same for the Hybrid vs. non-Hybrid Corolla. Also a rough sanity check would be the fatalities for each vehicle model divided by the total number sold of that model.
This reminds me of better doctors having higher fatality rates because the most critically ill patients seek them out. Since Teslas are at the top of every crash test safety study, I suspect that this is a spurious outcome of Simpson's paradox and that proper controls would make the effect disappear.
Even more fun would be if alcoholics were more likely to buy Tesla's because of the self-driving aids. And it turned out this cohort was less likely to be involved in fatal accidents than the non-Tesla-driving alcoholics. So that even after controlling for more alcoholics purchasing Tesla, there would be a reduction in the overall fatality rate.
But my gut feel is that there is a systematic bias in the iSeeCars mileage data that they aren't properly accounting for (say prevalence of leasing EVs skewing the mileage estimates).
With all of the telemetry, am I correct to presume that Tesla probably knows the current odometer reading for pretty much every car they've ever sold? So they could easily give out the denominator for the fatality rate in the original reporting? The total miles driven seems like it could be the weak link in the iSeeCars data, since I'm assuming they have to infer the total miles driven off of their used car sales data.
Discussion (296 points, 7 days ago, 442 comments) https://news.ycombinator.com/item?id=42150443
Possibly related but anecdotally tesla drivers drive more like BMW drivers than any other car brand except BMW itself.
> Therefore, he added, their elevated accident rates likely “reflect a combination of driver behavior and driving conditions.”
In other words, this Tesla person is confirming the no-brainer that the behavior of buying a Muskmobile is not statistically independent of the behavior of being a dickhead on the road.
"Our cars are incredibly safe. It's just that some of the idiots who comprise our customer base are more dangerous."
I don't buy the part about driving conditions, because when you're looking at national level statistics from everywhere, everyone drives the same damn conditions. Conditions can only explain one particular accident. Or a string of accidents at a particular intersection.
I wonder if the difference could come down to electric vehicle owners being more likely to drive locally than gas/hybrid cars?
I think no because the idea that most accidents happen close to home is a fallacy. It's only true an absolute numbers, because most driving happens close to home. This story is about fatal accidents, not just accident rates.