3 ms·

(adding some references to the points you made) Government data show gasoline vehicles are up to 100x more prone to fires than EVs: https://electrek.co/2022/01

(adding some references to the points you made) Government data show gasoline vehicles are up to 100x more prone to fires than EVs: https://electrek.co/2022/01/12/government-data-shows-gasoline-vehicles-are-significantly-more-prone-to-fires-than-evs/ https://electrek.co/2022/01/12/government-data-shows-gasolin... Data from the National Transportation Safety Board showed that EVs were involved in approximately 25 fires for every 100,000 sold. Comparatively, approximately 1,530 gasoline-powered vehicles and 3,475 hybrid vehicles were involved in fires for every 100,000 sold: https://www.fairfaxcounty.gov/environment-energy-coordinatio https://www.fairfaxcounty.gov/environment-energy-coordinatio... Statistics from 2015 showed that 174,000 vehicle fires were reported, and almost all of them involved gasoline vehicles. Tesla claims that gasoline cars are 11x more likely to catch fire than a Tesla, and that the best comparison of safety is fires per billion miles driven. If we compare using this method, there are approximately five EV fires for every billion miles traveled, compared to 55 fires per billion miles traveled in gasoline cars: https://driveelectriccolorado.org/myth-buster-evs-fire/ https://driveelectriccolorado.org/myth-buster-evs-fire/ From https://www.evlithium.com/Blog/advanced-safety-features-of-lifepo4-batteries.html https://www.evlithium.com/Blog/advanced-safety-features-of-l... Non-flammable electrolyte: LiFePO4 batteries use a non-flammable electrolyte that does not catch fire even if the battery is punctured or damaged. The electrolyte is a mixture of lithium salts and a solvent that is less volatile and less flammable than the organic electrolytes used in other types of lithium-ion batteries. High safety: LiFePO4 batteries have a lower risk of overheating and catching fire due to their more stable cathode material and lower operating temperature. They also have built-in protection circuits that prevent overcharge, over-discharge, short-circuit, and physical damage.

hansvm 2y ago
That's an interesting set of stats, but it's not decisive. Some alternative explanations: - The stats are just plain wrong, even at a surface-level inspection. They suggest that over 20yrs you'd have 3 fires for every 10 ICE vehicles ever sold. I ought to know 50 people in my current company who have had a car burn down, 30 in my last company, 5 in my extended family who I'm close with, 100 in a slightly wider net of acquaintances, .... I know 0. I highly doubt _anyone_ in the country comes close to having seen the "right" number of ICE fires, outside of firemen and people who otherwise gravitate toward such problems intentionally. - The problem mostly isn't gasoline; it's shoddy manufacturing, especially wiring. How do the stats look if you ignore Hyundai and other particularly low quality manufacturers? Normalized by miles driven (which is closer to what ought to be correct), the delta is 10x rather than 100x. That's roughly the delta between how often a BMW breaks down vs a Toyota, so as a coarse estimate you might expect the data to be entirely explained by having the wrong denominator (cars vs miles) and the fact that you're just comparing different tiers of manufacturer. - Back to the "denominator" issue, which is always a problem with normalized statistics. If your baseline isn't that each car has an equal chance to burn down for each mile driven, but that older cars are more likely to burn down because rust and other degredation allows for water infiltration and electrical fires, the older ICE fleet would naturally generate the reported stats. - Back to the "denominator" issue, per car sold this year is maybe correct, per car ever sold is maybe correct, per currently running car over 20yo is maybe correct, per mile per one of the previous items is maybe correct, .... You really want to see the stats broken down by those categories, and to help avoid fishing for hypotheses to then come up with other testable ideas to examine for each interesting subcategory you find. - It's always worth mentioning with EVs that the distributions aren't directly comparable. You have different populations driving the cars in different places from their ICE alternatives. Similarly to the previous point, you want to see results broken down at least by crude demographics. It matters more for self-driving safety claims, but it might be relevant here.
- hansvm 2y ago
  Elaborating slightly on the denominator and population issues: The reason you care is that these stats aren't just abstract quantities; you're trying to gauge what the impact of a specific intervention might be. Suppose you could magically flip a switch and transition the country to EVs (which I'm not arguing against, just potentially the fire data thing), and you based that decision on this data. You weren't sold before, but since the fire risk is so low you were willing to apply that intervention. Instead of low fire rates, you might see: 1. When EV/EV collisions happen instead of EV/ICE, the higher total energy from heavier vehicles makes short-circuits and fires more likely than your model predicted (this is different from most of the rest of my points since it assumes your model is actually correct and unbiased today in some meaningful way but would still fail to be very helpful). 2. The data was just wrong and thus had no bearing on the real world. 3. You have the same fire rate as shoddy manufacturers enter the mix. 4. You have low rates initially, trending toward the same rate we're at now as the cars degrade, except now those fires are incredibly dangerous and toxic. 5. A mandatory software update while you're driving down the road inadvertently triggers a short circuit, leading to unknown-unknowns potentially affecting large swathes of the population at once (unlikely probably, but you are exposed to new failure modes and don't have a ton of data about the rarer ones yet -- much like (1) this isn't actually a criticism of the stats themselves, just their interpretation). 6. When you get more people driving these things in rural, snowy, mountainous regions you get more collisions and fires. 7. Having more teenagers driving EVs substantially reduces any claimed fire hazard deltas. And so on. Global stats (like fires per crash, per mile, per car, ...) are suggestive of the next place to look for more data when trying to make an informed decision, but when you're talking about something affecting hundreds of millions of people, they're the beginning of the conversation, not the end. That's doubly true when they obviously have some kind of glaring flaw (like the napkin-math observation that they're off by at least an order of magnitude, probably 2 in this case). Separately, once that data is interpreted correctly, you'll probably find other important observations. E.g., if ICE and EV have the same fire rate excluding shoddy manufacturers, and you have more solid evidence backing that up, that gives you very easy followups. The conversation can shift to the severity of those fires, their mitigations, and whether they're worth the cost given the other benefits EVs have. Moreover, you learn that you can greatly reduce car fires by just having better engineering, so (carefully and thoughtfully) toss in a law or tax or something to try to encourage better behavior. Similarly, if you find old vehicles are the root cause then you can encourage recycling or rust-proof coatings. That might even be a point in favor of EVs since the batteries don't last long, so they'll be recycled before they're likely to combust (again, assuming for simplicity that everything else stays constant, which a policy-maker hopefully would not). Just the stats presented though, especially without evidence that the obvious confounders don't apply, would likely not be helpful in making an informed decision. You could do just as well, perhaps better, from a gut instinct, and the presence of those stats in isolation is (charitably) just to help provide to people the data we do have and encourage further discussion, or (uncharitably) to present decontextualized information in a way that biases the populace in an intended direction, despite the fact that it's useless for the stated goal.

hansvm 2y ago

That's an interesting set of stats, but it's not decisive. Some alternative explanations: - The stats are just plain wrong, even at a surface-level inspection. They suggest that over 20yrs you'd have 3 fires for every 10 ICE vehicles ever sold. I ought to know 50 people in my current company who have had a car burn down, 30 in my last company, 5 in my extended family who I'm close with, 100 in a slightly wider net of acquaintances, .... I know 0. I highly doubt _anyone_ in the country comes close to having seen the "right" number of ICE fires, outside of firemen and people who otherwise gravitate toward such problems intentionally. - The problem mostly isn't gasoline; it's shoddy manufacturing, especially wiring. How do the stats look if you ignore Hyundai and other particularly low quality manufacturers? Normalized by miles driven (which is closer to what ought to be correct), the delta is 10x rather than 100x. That's roughly the delta between how often a BMW breaks down vs a Toyota, so as a coarse estimate you might expect the data to be entirely explained by having the wrong denominator (cars vs miles) and the fact that you're just comparing different tiers of manufacturer. - Back to the "denominator" issue, which is always a problem with normalized statistics. If your baseline isn't that each car has an equal chance to burn down for each mile driven, but that older cars are more likely to burn down because rust and other degredation allows for water infiltration and electrical fires, the older ICE fleet would naturally generate the reported stats. - Back to the "denominator" issue, per car sold this year is maybe correct, per car ever sold is maybe correct, per currently running car over 20yo is maybe correct, per mile per one of the previous items is maybe correct, .... You really want to see the stats broken down by those categories, and to help avoid fishing for hypotheses to then come up with other testable ideas to examine for each interesting subcategory you find. - It's always worth mentioning with EVs that the distributions aren't directly comparable. You have different populations driving the cars in different places from their ICE alternatives. Similarly to the previous point, you want to see results broken down at least by crude demographics. It matters more for self-driving safety claims, but it might be relevant here.

hansvm 2y ago
Elaborating slightly on the denominator and population issues: The reason you care is that these stats aren't just abstract quantities; you're trying to gauge what the impact of a specific intervention might be. Suppose you could magically flip a switch and transition the country to EVs (which I'm not arguing against, just potentially the fire data thing), and you based that decision on this data. You weren't sold before, but since the fire risk is so low you were willing to apply that intervention. Instead of low fire rates, you might see: 1. When EV/EV collisions happen instead of EV/ICE, the higher total energy from heavier vehicles makes short-circuits and fires more likely than your model predicted (this is different from most of the rest of my points since it assumes your model is actually correct and unbiased today in some meaningful way but would still fail to be very helpful). 2. The data was just wrong and thus had no bearing on the real world. 3. You have the same fire rate as shoddy manufacturers enter the mix. 4. You have low rates initially, trending toward the same rate we're at now as the cars degrade, except now those fires are incredibly dangerous and toxic. 5. A mandatory software update while you're driving down the road inadvertently triggers a short circuit, leading to unknown-unknowns potentially affecting large swathes of the population at once (unlikely probably, but you are exposed to new failure modes and don't have a ton of data about the rarer ones yet -- much like (1) this isn't actually a criticism of the stats themselves, just their interpretation). 6. When you get more people driving these things in rural, snowy, mountainous regions you get more collisions and fires. 7. Having more teenagers driving EVs substantially reduces any claimed fire hazard deltas. And so on. Global stats (like fires per crash, per mile, per car, ...) are suggestive of the next place to look for more data when trying to make an informed decision, but when you're talking about something affecting hundreds of millions of people, they're the beginning of the conversation, not the end. That's doubly true when they obviously have some kind of glaring flaw (like the napkin-math observation that they're off by at least an order of magnitude, probably 2 in this case). Separately, once that data is interpreted correctly, you'll probably find other important observations. E.g., if ICE and EV have the same fire rate excluding shoddy manufacturers, and you have more solid evidence backing that up, that gives you very easy followups. The conversation can shift to the severity of those fires, their mitigations, and whether they're worth the cost given the other benefits EVs have. Moreover, you learn that you can greatly reduce car fires by just having better engineering, so (carefully and thoughtfully) toss in a law or tax or something to try to encourage better behavior. Similarly, if you find old vehicles are the root cause then you can encourage recycling or rust-proof coatings. That might even be a point in favor of EVs since the batteries don't last long, so they'll be recycled before they're likely to combust (again, assuming for simplicity that everything else stays constant, which a policy-maker hopefully would not). Just the stats presented though, especially without evidence that the obvious confounders don't apply, would likely not be helpful in making an informed decision. You could do just as well, perhaps better, from a gut instinct, and the presence of those stats in isolation is (charitably) just to help provide to people the data we do have and encourage further discussion, or (uncharitably) to present decontextualized information in a way that biases the populace in an intended direction, despite the fact that it's useless for the stated goal.