Statistical Issues in the Doping Case of Andrus Veerpalu

• Activity Number: 567
• Type: Other
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: ASA
• Author(s): Krista Fischer*+ and Donald Arthur Berry
• Companies: University of Tartu and M.D. Anderson Cancer Center

Abstract

We address statistical issues regarding WADA's hGH test in the Veerpalu case, for which we served as consultants, and more generally. Doping tests are similar to diagnostic tests. Concluding substance abuse based on a doping test alone requires high specificity to minimize false positives. WADA claims their hGH test has specificity 99.99%. They do not advertise sensitivity, which is less important in a case of alleged doping. The decision limits of WADA's test are based on a convenience sample of about 300 athletes, about half of which are Caucasian males. Because large values seemed more likely to come from athletes who were dopers, they excluded them from their analysis! To enable a conclusion as strong as 99.99% on the basis of a small sample requires strong distributional assumptions. Namely, WADA assumed the non-doping hGH data had a gamma distribution. In sum, they concluded that fewer than 1 in 10,000 athletes would have a result as large as Veerpalu's based on a sample size of about 100. In addition, they cut off the distribution's tail and concluded it had a tiny tail! We will describe our counter arguments to WADA's statistical approach in this regard and more generally.