jeff tallon wrote:
i read somewhere(i cant remember where) that the figure of elite athletes doping is 44-57% and it was supposedly based on surveys.I think the pan arab games,and pan american games were mentioned,and world championships.Its from a few years back.Im also not sure what passes as an elite athlete.Perhaps olympic B qualifier leval.I can believe about half at B leval or above are using drugs,and literally everyone at top leval is using.
It was discussed here many times. It was the 2011 IAAF World Championships and the 2011 Pan Arab games (28 sports, including athletics). There was no distinction of "elite" athletes, just participants at these events.
For those who haven't read the study - here is "rekrunner's" notes:
It was anonymized by offering two questions, a birthday question, or a doping question, with a random selection of which question to answer, only known to the survey respondent.
The survey results, mathematically corrected to factor out the expected percentage of birthday questions, was 44% and 57%, respectively.
The study then hypothesized many forms of non-compliance, and modeled each form with a mathematical equation, as a function of number of athletes not complying.
This is important -- they did not measure any form of non-compliance, but just built equations taking the non-compliance factor as an input. In their analysis, they assumed a high degree of non-compliance for each model, not as an estimate of likelihood, but as a tool to help indicate which forms lead to small bias, and can be ignored, and which forms lead to more significant bias.
This is also important -- independently of these models, they also measured response times, noting that those who respond quickly, likely did not read the instructions and responded automatically. This would lead to many non-dopers responding that they doped. When checking the responses with the response times, they observed a significant positive bias with the fastest response times. The top 10% fastest responses indicated a 96% doping rate. It would seem strange that so many dopers would want to confess so fast.
The large bias was linked to response times, and not to any of their models, although "automatic yes" from non-dopers would likely be over-represented in the fastest responses. They recommended discarding the fastest 30% of the responses, to eliminate the observed bias, which would contain a spectrum of responses, including "automatic yes", "birthday switching" responses, as well as truthful doper and non-doper responses, etc.
The recommendation to eliminate this large observed positive bias, and only consider the slowest 70% of responses, leads to a lower bound estimate of 31% and 45%, respectively. Note that the remaining 70% can still contain all forms of non-compliance.
This is important -- "fastest response" is NOT one of the non-compliance models, and eliminating the fastest responses is not the same as assuming a factor of non-compliance in a model.
The conclusions of the paper did not decide which number was "real", but concluded that more doping surveys should be conducted to get better estimates than biological estimates.
Outside of this study, there is plenty of research on the accuracy of survey methods. One paper used the same survey technique in British athletes, cross-checking the results with other techniques. They found that for sensitive questions like doping, the survey result was 60%, compared to three other methods of 20%. When the question was a non-sensitive question about herbal supplements, the results were consistent with each other. They speculated that many dopers and non-dopers switch to the birthday question, as even non-dopers fear to answer a doping question, even anonymously when they have nothing to fear, which inflates the doping estimate.
Survey methods can potentially bring better estimates, but it is still a work in progress.