That's... not true. The question is "how likely is there to be an electrical fire in the Tesla, given that it has crashed (or is crashing)" not "how likely is there to be an electrical fire".

It's easy enough for things not to catch flame when they're being used properly; the real test is how things behave when other things are going wrong.

That's not to say that this means all Tesla cars are fiery death-traps of doom. Just that the sample size really is the number of crashed cars.

Following your logic, we can ignore Tesla cars that catch fire without crashing, because we can assume they're not going to catch flame when they're used properly; We can assume cars that don't crash aren't going to catch fire anyway, so, even if they may happen, let's throw that data away, because we know our assumption holds true.

That's ridiculous.

That is besides the point. We are talking about what happens when your Tesla S does crash. There is on "if" in the certainty that Teslas will be involved in accidents. The questions are "will they catch fire when they crash?" and also "how bad will that fire be for persons and property involved?"

There's also "Under what conditions when Tesla cars crash will they catch fire", and then you only look at your 12 crashes, and that's when everything goes wrong.

    The Space Shuttle Challenger exploded shortly after take-off in January 
    1986. Subsequent investigation determined that the cause was failure of 
    the O-ring seals used to isolate the fuel supply from burning gases.....

    NASA staff had analysed the data on the relation between ambient 
    temperature and number of O-ring failures (out of 6), but they had 
    excluded observations where no O-rings failed, believing that they were 
    uninformative. Unfortunately, those observations had occurred when the 
    launch temperature was relatively warm (65-80 degF).

http://www.datavis.ca/gallery/missed.php

Let's look at this question "how bad will that fire be for persons and property involved?"

You look at your 12 crashes where fire occurs, so (e.g.) 6 crashes. Of 6 crashes 3 were accompanied with injuries. All 3 crashes that were accompanied by electrical fires and injuries occurred was when the car's speed was between (e.g.) 40mph and 50mph. Does that mean when the speed is not between 40mph and 50mph, people won't be injured when there's a crash and a fire? It means nothing like that but it's too easy to make a wrong conclusion; you're skipping all the other data where injuries and/or fires happen independently of crashes.

Sure, you can look at all the crashes, but you'll still run into exactly the same problems with insufficient sample size because now you're dealing with a rare event, which means the required sample size balloons massively.