WADA Doping Study Suppressed

agip wrote:

...

3) I don't understand the 'statistical analysis' that determines the birthday answer to the drugs answer

...

The survey technique is described pretty well in the wiki article linked below. Which of the questions is to be answered needs to be deterministic from the interviewee's perspective and the randomizing question statistics need to be known by the analyst. The Times article didn't adequately convey that in their description of the survey procedure. One would hope that the researchers are sufficiently competent to do that correctly.

What did the "Caracas" 1983 study show again?

That sounds like a complete Rube Goldberg method of coming up with the 29% figure. You could have had 99% of the athletes answer the birthday question and then you extrapolate from the 1% who answered the drug question. How could you ever have confidence that you can extrapolate the 1% who answered the doping question to the 99% who answered the other question and to all of the athletes who didn't participate in the study?

That is exactly what I was thinking. The error bars on this survey must have been huge. I would not be surprised if the range is 29% +/- 25%. This was a survey done for one reason, to shock the public into forcing WADA to conduct more tests, and, therefore, pad the pockets of the testers who probably conducted this survey.

important fact: the success rate of anti-doping tests is about 0,1% max. (germany), in many countries even lower.

Then you are showing your statistical colors; rather lacking. This is not the sort of thing you can understand with an opinion, you have to understand it from the point of some sophisticated statistics. So look at the research on AIDS and statistical analysis. This approach allowed investigators to obtain surprisingly (compared to prior analysis) results in the aggregate without knowing ANYTHING at the individual level. Unless you can pick apart the protocol, your comments are really useless musing. This is a creative and effective use of statistical analysis. This is brought to you by some very bright PhDs in statistics and throwing in an opinion without adequate background is essentially besides the point. I have a PhD in a related field and I am familiar with the protocol from my wife's work (and she is a lot smarter than me with an area much closer to the analysis). If you do not think that this is science then you really do not understand science.

1) what incentive is there to lie; the answers on the questions cannot be linked with the individual by way of the random element introduced that hides individual information while retaining aggregate information. Prove, using some analysis that this is not the case; opinions are useless, only analysis counts and if you cannot do the analysis then you cannot understand the protocol and the implications.2) As in 1, why lie. It was strongly established in the AIDS research that very good, pretty much unbiased results were obtained. Again, their is no link to the individual's random draw, as it is done by the individual and not related to anything that anyone else can know.3) What the heck are you critiquing this when you cannot understand the protocol. Go try a big experiment on your own if you want to, but you have to be very careful about following the protocol, and it is likely that if you cannot understand it that you will make mistakes implementing it.4) Well, this seems to be the most far fetched guess, but one that seems to go counter to the bias it seems that you think exists.5) What has this got to do with it? Carbon testing of what under what circumstances and would it detect all cheating that has taken place over the last 12 months? Note that the question was not have you cheated in the last few days.

Cacao Starfish wrote:

They should have taken DNA samples from all the surveys and busted the 29% etc.

Why don't you try to explain exactly how this would work. Only the person flipping the coin knows whether they are answering the doping question or the non-doping question. No protocol where you would have at most a 50% chance of getting an appropriate answer is any good outside of, say, situations like Nazi Germany, and where they would not care about any of the 'false positives'.

You are guessing at the error bars? Why don't you try to calculate what they are. This technique takes pretty substantial sample sizes because you have to subtract out the random component that will supply half the +/- answers, but if n is very large, the standard deviation on that 50% can be pretty small.

I did a few 'back-of-the-envelope' calculations on the standard deviation. If the sample size is only 10 (and randomly drawn without bias from the population), then the standard deviation is quite large, approximately 20%. However, increase it to 100 and it drops to a bit over 6%, while increasing to 1000 drops it to 2%.

You would be a rather inept investigator if you use the 99% figure; I do not think that they are that stupid. There are subtleties here but your comments are simply in the 'no idea' category.

26mi235 wrote:

You would be a rather inept investigator if you use the 99% figure; I do not think that they are that stupid. There are subtleties here but your comments are simply in the 'no idea' category.

The article said:

Then, depending on the date of the birthday, they were instructed to answer one of two questions that appeared on the same screen: one asked if the birthday fell sometime between January and June, and the other asked, “Have you knowingly violated anti-doping regulations by using a prohibited substance or method in the past 12 months?”

So all responses to the birthday question would be a yes? 5 months between june & jan or 7?

if it's inclusive, I mean. man I havent touched stats since high school, I'd love to see the methodology/full study. my head hurts...

concur espically about other sports there is no way Adrain Peterson did not have help

There are numerous problems with scientific polling. The question here is whether the kind of error that the poster charged is valid. It is not. Half were randomly selected for the drug question, nothing wrong with that protocol in itself. The primary issues here would be that some chose not to participate at all--maybe that was a different group vis a vis drug testing--and it is done by self-reporting: some lied. It's surprising that 29% would admit doping within the last year. So, you'd think the real # would be higher.

If I remember it right, some 2000 athletes answered the survey. The sample looks quite substantial and representative to me.

jjjjjjjjj wrote:

There are numerous problems with scientific polling. The question here is whether the kind of error that the poster charged is valid. It is not. Half were randomly selected for the drug question, nothing wrong with that protocol in itself. The primary issues here would be that some chose not to participate at all--maybe that was a different group vis a vis drug testing--and it is done by self-reporting: some lied. It's surprising that 29% would admit doping within the last year. So, you'd think the real # would be higher.

Yes, there are some issues that are not clear from the modest-length coverage of a big study. The relatively large number of athletes that participated, about 2000, is a good sign, but out of how many. I suspect that the paper itself provides much of what we want, but it will be hard to pickup and fully understand without the right background. There are places to get things wrong, which the referring process should catch, so this is as-yet un-refereed paper and caution is necessary.

29% with a +/- of 20 still yields a minimum number(9)at least twice as high as the testing shows.

___

You are right on some points re; my ignorance but, but there are few people less defensive than a social scientist when challenged about the validity of social science methods.

Defensive enough to be willing to overlook the problems with a study using a self selecting group of subjects. It's like a website poll - it is interesting but not valid.

the consensus is, yes, the actual number of doping should be higher given that people tend to lie more about their unethical behavior than ethical conduct.

oy vey ismir

Talk about the suppression of the study not the damn tested to be validd methodology!

http://www.nytimes.com/2013/08/23/sports/research-finds-wide-doping-study-withheld.html?pagewanted=1&_r=0

[/quote]

What am I being defensive about; critiquing a low-content, not fully coherent complaint is not particularly defensive.

As to the study, I think I have figured out the confusion at the beginning. I think that they do the follow:

1) get a birthday theirs or someone close (so they will know the date). The use the date, e.g., 17th. If it is odd, then you get one question, even the other.

Comment, known to the respondent but not the analyst, but the analyst knows the frequency of each which is almost .5 (slightly more for odd).

2) One question ask if the birthday is from January to June

comment: (chances are (182/365*3+182/366)/(365*3+366), expected value almost .5

3) the other question, is taken PEDs of some sort in the last 12 months.

comment; they cannot know if you have been directed to the Jan-June or drugs question, but they know the frequency of each. Go look at the links cited early on (e.g., Wikipedia

).