Originally Posted by
Science Goy
Then as someone who purports to understand statistics, certainly you're aware that the 0.0169% sample rate is irrelevant. It's possible to have a statistically robust subsample of ten flights, or a massively biased subsample of 20,000 flights.
OK, let's get deeper then. I'm probably incorrect as I no longer use this on a daily basis, but to get a delay rate that's within 1 percentage point of the overall 53,200 flights, and still only have a 99% confidence that you got it right within the 1 percentage point,
Dataplumber would have had to fly on over 12,500
random flights (and not biased by route, as someone correctly pointed out).
Relax the requirements of precision by a whole lot, and let's say that we only want a 95% confidence that the sampled cancellation rate is within 4 percentage point of the overall AA operation's one, and the number of
random flights to sample would have had to be close to 600.