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

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.