I was surprised to see three math errors (2 orders of magnitude each) in the fifth paragraph of
http://www.ntsb.gov/Dockets/Aviation...021/421357.pdf and repeated verbatim in Continental's report
http://www.ntsb.gov/Dockets/Aviation...021/436811.pdf.
Another interesting tidbit: "57% of [United Airlines] crosswind takeoffs [in the set of operational data analyzed] in excess of 30 knots occurred at DEN." Think about that a moment. 57% from one airport. I wonder what percentage of all UA operations world wide in that data set departed from Denver? I'm betting a lot less than 57%.
Final thought: this is the second accident report in the past year where data mining of huge operational performance data sets has suggested that the conditions that setup the accident in question were statistical holes in one. (I'm thinking of the BA038 enroute temps as the other.) What does that suggest about the difficulties of reducing an already miniscule accident RATE when the causes are things that are so rare that anticipating / designing / testing / training them out is also such a statistical challenge? The existing sims don't train high and gusty crosswinds below 50' AGL in part because there isn't even a winds model to do it with. And what far more likely scenario do you not train to find the sim time to train this one?