Thanks.
Having read the conversation, I feel like this really highlights how poor Twitter is as a medium for discourse:
-Why is it so difficult for Tucker to focus on the explanation that releasing her values would make it easier for people to get away with doping? Either this is possibly true, or it isn't. Either doping authorities asked Paula to not release her values, or they didn't. This shouldn't be too difficult to conclusively determine after a decade of asking her about it.
-Tucker's discussion of statistics alarms the statistician in me: "When the probability that a hematological value is not due to doping is about 1 in 1000, and only then with multiple flags and data points." I don't know how the biological passport works, and I mean, 1 in 1000 *sounds* pretty stringent, but if there are dozens or hundreds of tests per athlete (dozens of tests times a handful of markers) and thousands of athletes in the system, then false positives are going to happen all the time. Tucker should know this. Looking at multiple flags might help, but only if they're uncorrelated. Ultimately the decisions are apparently left to a human panel of experts, which doesn't really seem transparent to me either. I did some digging and found this (start on page 50):
https://www.wada-ama.org/sites/default/files/resources/files/wada-abp-operating-guidelines-v5.0-en.pdf. The biological passport sounds like it's designed to modulate the testing frequency rather than make an outright determination of positive/negative tests, which seems like it might help improve the use of testing resources.
-Lauren Fleshman's comments seem out of line. Do you believe in the biological passport, or don't you? If you do, then you can't be worried that Rudisha's results might be anomalous because he's the greatest of all time. The fact that Paula is a woman is irrelevant. If you don't believe in the biological passport, then that's a fair stance to take. As a layperson I'm not sure what to believe.
I want to ask further questions about the biological model. It identifies people who are outliers. Does that mean if everyone is doping then the outliers will be clean athletes? Or is the model trained on clean athletes (do they exist?) or ordinary people?
Personally, I don't even care who is doping (because I think everyone is), I just want to see how prevalent they think doping is, and that might be possible to figure out from the statistical distribution of test results. *If* you believe in the biological passport model and the need for transparency, then looking at the entire distribution of p-values should give us an estimate of the fraction of people who are doping (but won't tell us who they are). You should be able to do this by considering the entire distribution of probabilities spat out by the biological passport's adaptive model. Not just the unlikely events, but the likely ones too. All of them. If nobody were doping, then the null hypothesis would be true and the histogram of p-values would be uniform (actually, I wonder how they trained the model in the first place). Since some fraction of people are doping (because duh), if you believe in the adaptive model then you are likely to get lots of people with low probabilities, even if they don't cross the probability threshold for significance. The distribution would look like this:
http://genomicsclass.github.io/book/pages/figure/multiple_testing-pval_hist2-1.pngSome of those tests won't cross the probability threshold for significance action on their own, but the entire distribution will shift to the left in such a way that you should be able to get an estimate of the number of tests that came from dopers overall, even if you don't know who the dopers are. This is related to the "false discovery rate" in statistics. In the figure, the fraction of the area of the curve above the solid horizontal line (at 90ish) would be equal to the fraction of people doping. We just wouldn't be sure who they were.