This analysis claims someone that detonated a bunch of fireworks and then shot himself in the head as a "fire fatality" and acknowledges it.
"NOTES: 2001 figures exclude two events—the September 11th Attacks and the Flight 587 crash."
(which is especially weird, because the flight 587 crash was pilot error)
We can, and especially in this small sample size, we can exclude it if we want to. The question is if a cybertruck is a dangerous vehicle, not can I have mental illness and shoot myself inside of my vehicle, which happens to be a cybertruck. That danger is present for any vehicle, there's nothing particular about the design of the cybertruck that helps or hinders self-inflicted gunshot wound, unlike a gas tank at the back of the vehicle that causes fire when it gets into an accident.
Compensating for dataset bias in your preferred manner, we get 4 fire fatalities over 34,438 vehicles giving a Cybertruck fire fatality rate of ~11.6 per 100,000 units in comparison to a Ford Pinto fire fatality rate of ~0.85 per 100,000 units; ~13.6x the fatal fire rate of the Ford Pinto. Even counting mere incidents we get 2 over 34,438 vehicles which still results in a fire fatality rate of ~5.8 per 100,000 units; ~6.8x the fatal fire rate of the Ford Pinto.
Of course, this is all actually a massive underestimate as can be seen from my other post: https://news.ycombinator.com/item?id=43039925
Where the comparable analysis on vehicle-years, a much better comparable, actually places the fatal fire rate at ~43.5x if we discount the suicide and ~21.7x if we only count incidents. And in comparison to current vehicles would be ~41.4x and ~20.7x, respectively.
Frankly, it would have been better to just cite the inverse of 1 fire fatality per ~8609.5 Cybertrucks (I already discounted the suicide) versus the 1 fire fatality per ~117,536 Pintos. Does that help you understand the usage and comparison of rates?
If a self-driving car drives one mile with zero accidents, would we extrapolate from there and say they have a perfect driving record? Because thats just how the number work out? No! We'd want to see how this hypothetical self-driving car does over a million or a billion miles, over a wide range of conditions, before drawing any conclusions.
You appear to be under the mistaken assumption that your sample size needs to be in proportion to the failure rate you hope to achieve to draw any conclusions. That is untrue. Your sample size only needs to be in proportion to the actual failure rate to draw conclusions.
If your self-driving car drives one mile with zero crashes, that provides almost no evidence to support a claim that your self-driving car has a crash rate of 1 per billion miles. However, if your self-driving car drives one mile and crashes 10 times, that provides tremendous amounts of evidence against the claim that your self-driving car has a crash rate of 1 per billion miles. This is despite the fact that both instances only have a sample of a single mile.
Proving, failure to prove, and disproving are not symmetrical at all. But do not worry, that is a common mistake made by people with no training in scientific or statistical analysis; just take this as a learning experience.
I never said a sample size needs to match the target failure rate—only that proving an extremely low failure rate requires significant data. Yes, a high failure rate can be detected quickly, but that doesn’t mean a small sample is enough to confirm an ultra-low failure rate. Just because a car doesn’t crash in one mile doesn’t mean it won’t in the next billion. You’re oversimplifying the problem while assuming I don’t understand statistical inference.
Lighting a truck on fire is fundamentally different than a truck starting on fire. The fact that they are willing to count that shows me they are disingenuous.
Link for the COVID reference: https://www.freedomfoundation.com/washington/washington-heal...