If I'm doing the math right, there were only 3 fatalities on a 777 if you don't count MH370 or the Ukraine/Russian shoot-down. That would change the FLE from 2.01 to 0.01.
I understand not wanting to count a shoot-down but why not count MH370? For all we know there was a safety issue with that aircraft that caused the crash.
Isn't that because suicide is a taboo subject in many countries? This was at least the case for EgyptAir Flight 990 and SilkAir Flight 185, and IIRC people expressed the same thing about Malaysian culture.
You need a larger dataset if you're dealing with very small probabilities. If you measured millions of events and only saw 1 anomalous event, you have enough data to say that the probability of such an anomaly is extremely rare, but you don't have enough data to compare it to other very rare events.
More concretely: if you flip an unfair coin 1 million times and get a single heads, you know that the odds of getting that heads are extremely low, but you can't yet say what the odds are. It's possible that the odds are one in a million, but it's also possible that you got very lucky or very unlucky relative to the actual odds. You have to have a lot of data in both buckets before you can distinguish.
Millions of flights, but less than ten crashes for most, and it was pointing out that even some of those are incorrectly categorized. When you're dividing A by B to get ratio C, you cannot ignore that A is bad data because B is good, it still makes for a C that is useless.
All this dataset provides is that with small enough samples it is too noisy to draw reliable conclusions.
For example, 777 has had very few incidents, but its figure is impacted by MH370 and MH17.
747's number is also skewed (by about 10%) due to 3 incidents involving bombing and missile shootdown.
By and large though, most plane crashes are due to human error, mostly by pilot. Terrorism, as well as maintenance or engineering defects are pretty rare.
