I'm not questioning the official stats, I am questioning the interpretation. It is not clear which category road debris would fall.
Last edited:
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
I'm not questioning the official stats, I am questioning the interpretation. It is not clear which category road debris would fall.
SCW-Greg
Active Member
As SwedishAdvocate so notably brought back, and I will bring back again to the top - Mario Kadastik's real assessment...
"Please everyone stop [drawing erroneous conclusions from (My edit.)] the statistics. The ICE fires are in the thousands and their distribution is governed by the normal distribution that most of your statistics are based on. The Model S statistics are so low that it's governed by Poisson statistics and that has completely different characteristics. I deal with low probability events daily (Higgs search at LHC) and have had to handle the differences and you can't believe how much difference there is. Your math here has error bars that are so huge that you cannot draw any conclusions really. In Poisson statistics 0-2 events are statistically inseparable so even if you expect 0 events and observe 2 you cannot claim disparity between the two measurements. With three you start to get somewhere, but only if you really expected 0 in the first place. If you expect even one (or worse ca 3), then one to about six events are fully compatible (or one to ten). You can start using your normal statistics when the number of incidents expected is largish i.e. my statistics teacher used to say that 30 and infinity are about the same, it's not quite that simple, but around that region the Poisson starts to converge towards the normal distribution..."
Ugh. Another Model S fire - 2013-11-06 - Page 40
What do you do for work? Just passing time waiting for updates from Tesla
- Page 36
"Please everyone stop [drawing erroneous conclusions from (My edit.)] the statistics. The ICE fires are in the thousands and their distribution is governed by the normal distribution that most of your statistics are based on. The Model S statistics are so low that it's governed by Poisson statistics and that has completely different characteristics. I deal with low probability events daily (Higgs search at LHC) and have had to handle the differences and you can't believe how much difference there is. Your math here has error bars that are so huge that you cannot draw any conclusions really. In Poisson statistics 0-2 events are statistically inseparable so even if you expect 0 events and observe 2 you cannot claim disparity between the two measurements. With three you start to get somewhere, but only if you really expected 0 in the first place. If you expect even one (or worse ca 3), then one to about six events are fully compatible (or one to ten). You can start using your normal statistics when the number of incidents expected is largish i.e. my statistics teacher used to say that 30 and infinity are about the same, it's not quite that simple, but around that region the Poisson starts to converge towards the normal distribution..."
Ugh. Another Model S fire - 2013-11-06 - Page 40
What do you do for work? Just passing time waiting for updates from Tesla
- Page 36
Last edited:
Krugerrand
Meow
Absolutely not. I can not support that in any way, shape or form. You've changed the entire emphasis to focusing the craziness strictly on the collision when everyone knows it was the tequila that was responsible and therefore the emphasis must be placed on Mexico. The crazy Mexican collision is far more accurate. I won't have it any other way. We're done talking about it.
I have but one question in response:
Do you have heels on when you put your foot down like this?
lolachampcar
Well-Known Member
NigelM
Recovering Member
lolachampcar
Well-Known Member
Zzzz...
Member
can't mix USA road safety with Mexican road safety, even if the cars are the same Driving in Mexico: is it safe relative to other countries? | Geo-Mexico, the geography of Mexico
another point, very rough principle - Tesla is akin to a vehicle whose fuel tank is divided into 16 pieces and spread accross the base of the vehicle.
likihood of fire is 16 times higher than a single location energy storage
severity of fire is 16 times lower than a single location energy storage
If the outcome of the fire are below a threshold, then perhaps it is safer to have increased frequecny of 'mild' fires than to have fewer but of greater severity.
ie falling down one step is alot less dangerous than falling down one storey.
another point, very rough principle - Tesla is akin to a vehicle whose fuel tank is divided into 16 pieces and spread accross the base of the vehicle.
likihood of fire is 16 times higher than a single location energy storage
severity of fire is 16 times lower than a single location energy storage
If the outcome of the fire are below a threshold, then perhaps it is safer to have increased frequecny of 'mild' fires than to have fewer but of greater severity.
ie falling down one step is alot less dangerous than falling down one storey.
NigelM
Recovering Member
Sorry, that doesn't fly. There's massive differences in build, protection, combustibility etc. never mind the cars response to any given accident.
lolachampcar
Well-Known Member
renim,
I kinda agree in some respects. The large flat plane battery exposure increases the risk of damage over ICE. If damage occurs, 1/16 of the available energy is likely exposed. It is not a direct comparison to ICE but I see the connections in your point.
I kinda agree in some respects. The large flat plane battery exposure increases the risk of damage over ICE. If damage occurs, 1/16 of the available energy is likely exposed. It is not a direct comparison to ICE but I see the connections in your point.
Agree, its more the principle that Tesla appears to chosen a design that trades increased frequency + reduced severity, compared to a reduced frequecy + greater severity.
Its a fundamental difference compared to single tank/battery, its a fundamently effective way to improve safety, even though it goes against the intuition gathered from 100 years of petrol vehicle use.
ken,
i believe i have a much better answer for your question today. yesterday i cited the definition of mechanical failure that was in the nfpa document, which to me clearly indicates that hitting a trailer hitch is not a mechanical failure.
last night i found this article which was a research paper prepared for the nfpa titled "motor-vehicle collision-fire analysis..."
http://www.mvfri.org/Library/GM-DOT Reports/B02/NHTSA-1998-3588-77.pdf
the author states his intention in selecting samples for case study was: "the intention was to find examples of incidents over a broad range of field experience." (pg 4)
i refer you to table 1 on pages 8 and 9 of the pdf.
sample 8 is a front impact with a tree. that means hitting an inanimate object is a collision.
sample 13 is an impact with a "tow dolly". that's as close to a trailer hitch as i think you'll get.
here's the description on page 10 of the pdf:
"In the case of Sample 13, the vehicle was driven over a tow dolly left on thehighway. A cylindrical rod on the dolly punctured the gasoline tank below the driver’s door.
Scratches on the dolly showed that it had been dragged against the pavement while lodged under
the van until rest. No other fuel or ignition sources were available."
for me, a case study of "collision-fires" prepared specifically for the nfpa which has nearly identical cases to the tesla accidents pretty much seals that hitting a trailer hitch, tree, or large metal object is considered a collision.
thanks for your question, because it sharpened the research from this angle.
one note: the article is dated 1998, so i suppose some will wonder "did they change the definition?" i'm satisfied with this aspect of my research now, so i'll defer to another poster to answer.
- - - Updated - - -
i'm not referring to the specifics of the honda problem. i'm referring to how the investigation was approached, and what level of focused specific data was enough to cause a legitimate concern.
for the crv, i think roughly the numbers were you had 60 fires in 280,000 vehicles over 2 years - all in a specific area after an oil change.
tesla has 3 fires in 19,000 vehicles operating for 13,300 "car-years" vehicles all in the battery after a collision. the ratios are very similar - multiply by 20 and the numbers get very close.
also i think the case study is interesting (there are many other articles besides this nyt summary) as it tells you a bit about how these things progress.
- - - Updated - - -
i don't know if you're serious or asking this question in frustration.
you're asking here how the model s does vs. ice in the situation of hitting a concrete wall at 30/40/50/60 mph. in this case, i can tell you will not get a statistically significant answer.
even though we don't know the statistics for surviving such an event in an ice, we could use a low arbitrary figure and see if we can draw valid conclusions as an exercise.
so let's say the chance of surviving such a crash in an ice is 10%. let's say we know this from a large sample of ice crashes over many years and many crashes, so that we can essentially treat it as a point estimate.
now we can ask this question: given that an ice passenger has only a 10% chance of surviving a high speed crash with a concrete wall, and the observation of one tesla passenger surviving one such crash, can we reject the null hypothesis at a 5% significance level?
null hypothesis: the tesla model s offers a passenger a lesser or equal chance of surviving when impacting a concrete wall at high speed vs. an ice.
a very similar analysis to what i used before can be applied -
p = 0.1 (the chance of surviving)
n = 1 (number of concrete wall crashes)
x = 1 (concrete wall crashes where victim survived)
under these conditions, the answer is very simple. the odds of observing this outcome are 10%. so we fail to reject the null hypothesis at a 5% significance level because we cannot be 95% sure that the model s is safer.
you could also frame it the other way, using the model s passenger has a greater chance of surviving as your null hypothesis. in this case you would fail to reject the null hypothesis as well.
meaning, even assuming very very low rates of survival for an ice crash of this nature, you couldn't draw a conclusion that's statistically significant at the 5% level.
Babylonfive
Power12
I think that the fire risk is still much lower in the EV and existing data does not reflect some very important points.
One thing that past statistics do not reveal in a quick compilation of existing data is the source of fire. The Tesla has one particular situation that makes it vulnerable to fire while an ICE has multiple potential sources and vulnerabilities. That particular situation in the Tesla has been taken into consideration and there are engineered systems in place to control a fire and protect the occupants, reducing injury risk.
Tesla's vulnerabilies are the underside being hit by very large projectiles that can act like a missle or a severe impact that would obliterate most ICE vehicles and end up with the driver having an engine in their lap. The Tesla undercarriage is not likely exposed in a multi-vehicle accident due to the low center of gravity. At the same, time an ICE's multiple vulnerable systems are significantly impacted in many types of collisions and there more many more potential point sources for spontaneous ignition in an ICE.
One thing that past statistics do not reveal in a quick compilation of existing data is the source of fire. The Tesla has one particular situation that makes it vulnerable to fire while an ICE has multiple potential sources and vulnerabilities. That particular situation in the Tesla has been taken into consideration and there are engineered systems in place to control a fire and protect the occupants, reducing injury risk.
Tesla's vulnerabilies are the underside being hit by very large projectiles that can act like a missle or a severe impact that would obliterate most ICE vehicles and end up with the driver having an engine in their lap. The Tesla undercarriage is not likely exposed in a multi-vehicle accident due to the low center of gravity. At the same, time an ICE's multiple vulnerable systems are significantly impacted in many types of collisions and there more many more potential point sources for spontaneous ignition in an ICE.
In addition to the points Nigel made, must not forget the fuel line that runs the length of the car. That points out another variable- for ICE, even a small amount of the flammable exposes the entire vehicle (including the passenger compartment) and extremely quickly. It's a good point denim, and i would agree it mitigates to some degree, but I'm inclined to the 'different animal' side of things for the moment
- - - Updated - - -
yep- that put it over for me as well. Given that publication description detail, I would put it very likely to be in the Collision category - I relent the point; Thanks for following up on that;
Based on the model as it stands, when would it predict (under the same certainty and based on current car-years) we would see the next fire- should be a high certainty right?
I think on Bonnie's question, without more examples, the best data would be the official destructive crash test data coupled with this incident
My entry (should count as a US accident because):
"Jose can you see"
clearly the tequila
You are looking at a test examples and ignoring other critical information from the report.
The report cherry picked things that they could determine a cause:
"While it was understood that numerous vehicle, collision, and environmental factors are involved in
the incidence of fire, it was also clear that it was not practical to control for all in the selection process
for at least two reasons: 1) it is very difficult if not impossible to perform detailed investigations of
enough observations to differentiate numerous variables, and 2) collision-fire events are rare and
difficult to learn of in a timely manner for investigation. Therefore, samples are drawn from a limited
pool of candidates."
and later they say this:
"Extent of Vehicle Damage: It is difficult to determine specific fire causation and propagation factors
in vehicles with extensive damage from both impact and fire. To maximize the information learned
from each incident selected, samples were taken from three categories:
1) Vehicles with minor fire damage and any degree of impact damage.
2) Vehicles with minor collision damage and any degree of fire damage.
3) Incidents in which fatalities or burn injuries are involved with any degree of collision
and fire damage.
For those incidents with either minor impact or fire damage, fire causation factors may be identified
with a higher degree of confidence. For incidents with both extensive impact and fire damage,
investigation still provided insight into causation of injuries, propagation times and entrapment issues."
I spoke to a fireman and asked him about car fires and how much effort goes into investigating the cause.
He told me that if there is no reason to think it is arson and nobody is hurt there is no significant investigation ( the Fire Marshall is not called ).
He also told me that the cars are frequently reduced to slag and there is very little recognizable material left.
I will reiterate: When the first responders get to the car fire and they don't see the obvious thing that the car hit, it's not recorded as a collision event.
Nobody is going to scrape all the pieces up and put them in a white room CSI style. Nobody is going to walk down the highway with a swab and look for pieces of evidence.
It just doesn't happen.
for reference my original null hypothesis:
the risk of a collision related fire in a tesla model s is no greater than the risk of a collision in the average ice automobile.
the null hypothesis you propose:
the risk of fire in Tesla Model S collisions is no greater than the risk of fire in the average ICE car collision.
first off, i want to compliment you on one of the most thoughtful responses i've received on this thread. you are correct that the null hypothesis you are presenting is a better choice. the difference is so subtle i'm not sure many of the readers will see it, so i'll rephrase your null as the question: "once a model s or an ice get into a collision, which is more likely to catch fire?"
and your thought here is correct, we don't have access to data on model s collisions to directly answer the question.
i had thought about a few of these issues as i did my research.
i think the first thing i realized is that if i could know that the odds of a model s collision were less than or equal to the odds of an average ice collision, then my test would err in favor of tesla.
what i found is that there is a very powerful effect in auto insurance underwriting from education and occupation. that is either highly educated or blue-collar employed individuals have 15-20% fewer claims than the average. there was a whole controversy about this because geico was using these factors to adjust insurance premiums, and there was a big stink about it being a proxy for race. for the data table, refer to page 22 of this study (page 23 of the pdf):
http://www.state.nj.us/dobi/division_insurance/pdfs/ed_occ_april2008.pdf
"This information demonstrates that Occupation Groups 1 and 2 have better loss experience than the others, and that drivers with a Bachelor’s or Master’s degree
are similarly less risky than the population generally. The differences are
statistically significant and thus sufficient under current insurance statutes to be
reflected in the rates charged to these driver groupings.
Based on this data, for example, individuals in Occupation Group 1 generate
about 15% less claims than average drivers, while individuals in Occupation
Group 5 generate greater than 25% more claims than average drivers. Similar
results are documented in the loss ratios for groups with various levels of
education."
i think it's reasonable to assume that the model s drivers are likely to be either more educated or have a white collar job vs. the average ice driver. for example, it's probably safe to say a lot fewer teenagers are driving teslas than ices.
unfortunately that's all i could find that would relate to the likelihood of a model s collision vs a regular ice collision.
however, another thing i learned along the way is that the nfpa statistics cannot distinguish between a pre-collision fire and post-collision fire in most cases. that is, did the car catch fire because of an electrical malfunction, then the driver panicked and had a collision? or did the driver have a collision, and then the car caught fire? because mechanical and electrical failures are the primary causes of ice fires, there's likely a significant pre-collision fire effect in the ice collsion-fire data. this was discussed in some of the source documents i read.
the national data is such that it simply cannot distinguish the scenario of whether the caught fire before or after the crash. so, it's not possible to directly address the question you posed.
a further consequence is that the nfpa collision-fire data is for sure higher than the actual frequency of post-collision fires. in the 3 model s fires, they were all post-collision fires. so the way i compared these statistics, they are skewed in favor of tesla because what we're interested in, and what i should be using is the rate of post-collision ice fires.
it gets back to what another poster said, we do the best with what we have.
- - - Updated - - -
you've raised a valid concern here. i had thought about this too from a few different angles, and i'll share with you why i included the mexico crash. i can also share with you what happens if it is excluded.
first, reliable data is not available for mexico, so let's just forget about trying to adjust the analysis for mexico.
so the only question is, do we include the mexico crash in the count, or do we leave it out.
i decide to include it for these reasons:
(a) the car must have been built to usa safety specifications. it must have been an "american" tesla for lack of a better word.
(b) the driver did not collide with another vehicle, and that would certainly complicate the situation.
(c) the driver hit inanimate objects like barriers and trees, which would be relatively similar in different countries.
(d) i think the question of interest is "are all tesla model s less likely to have fires in a collision than the average ice in the united states?" if that's the question, i believe it's ok to use all tesla model s regardless of geography.
however, if you wanted to throw it out, you could easily adjust my analysis and do so. at that point the null hypothesis is still rejected but only at a 10% significance level, as there is just a little over a 90% chance that you shouldn't even see 2 crashes (using my collision-fire probability).
i don't think a blanket recall is going to be issued unless the investigation reveals a meaningful problem with the undercarriage or battery placement. all of the data supports elon's claims that overall the model s is safer than ices. it's just this nuisance of the post-collision battery fires where they seem to be having a problem.
well as i stated in my original post, all the data supports elon's overall safety conclusions (although he used exaggerated claims, his conclusions were correct). so there's no need for us to spend time on model s overall safety questions, we are in agreement there.
i am looking at just collision fires.
the problem with using the millions of miles traveled is that the nfpa data on collision-fires is all based on time, not mileage. so to meaningfully address collision fires, the only way i see is to use time on the road ("car-years"). if i were to add in the mileage analysis based on the federal highway administration's estimates of miles traveled, it would introduce more error into the analysis. not every car maker has the luxury of tesla, where tesla can know at every instant how many miles each of their cars on the road has traveled.
That is just one of the problems with statistical analysis of this situation. We do not know how the data was qualified or validated for the studies regarding accidents and fires. We also do not know if it is a valid sample rate because there isn't consistent controls in place regarding how or when data is reported and compiled. The existing data can be unintentionally skewed in any direction depending on the source of the report and the subjective nature of interpretation of the events.
Financial data can be rather straight forward when it comes to statistical analysis.
Scientific data collection regarding of real world events that occur with human variables is an entirely different animal.
Similar threads
