Reasons why Jakob Ingebrigtsen is drug-free

justanothercuriousdude wrote:

know thy history wrote:

You are not skeptical at all. You always look only one way - to minimize doping, both effect and prevalence. That peer-reviewed survey for example has a raw result of 44%, and when corrected for dopers not trusting the anonymity, over 60%. Of course you quickly discard both, and go with the low "at least 31%" although that is explicitly described as just one scenario explicitly only valid under two conservative assumptions, and the authors explicitly and correctly mention a couple of times that most scenarios discussed lead to higher numbers than the 44% - as you could quickly verify by browsing through the tables of the attachment where all those details are listed.

Naturally you love to ignore the "at least" and round 31% down to 30%. Your mission is quite obvious.

This isn’t correct at all - f.x in my last post in this thread where I mention percentages I write (#136 in this thread):

4. The researchers raw estimates (based on estimates of the answers) were that 43% admitted to doping in the past year. But since some of the athletes seemed to answer to quick (to be serious / trustworthy) the researchers cut down the estimate to “at least 30%”, meaning 1 in 3, not 1 in 2…

So as you can see from this point 4 pasted in from that post that I write: “AT LEAST 30%”!

You don’t seem to understand my agenda at all, because my point isn’t at all to downplay the researchers opinion about high doping prevalence. And as you are saying they clearly stress that the real percentage might be far higher than 30%, and even significantly higher than their other number (43%). But operating with a range from a relative low 30% to a far higher top estimate shows a scale of uncertainty, and I think this acknowledgment gives credit to the researchers here.

So let me spell out clearly my agenda: 1. To show that the surveys are estimates, and not a counting of number of athletes that have admitted to doping without interpretation. And estimates are always subjective and can be very wrong, especially in a sensitive area. Which lead me to point 2:

What are the odds for truthfully answers if you ask an athlete if he intentionally has been doping the last year? Probably a near 100% lie rate… But what if you give him an asserted anonymity? Well, one would expect that some (if they feel assured) would then tell the truth. But how many would lie even with this anonymity? -Probably a lot, given human nature when it comes to admitting wrongdoing, fear of being exposed somehow anyway, not fully understand that the anonymity is secure, being ashamed, repressing an uncomfortable reality, no reward and incentive for being truthful, and so on. So how to overcome the un truthfulness? Well, one can estimate the percentage of lies by using experience from other surveys on questions that have some likeness, for instance asking random people if they have driven drunk one time or more the last year, and then calibrate this up against percentage of busts in real traffic. And these busts are unlike doping tests highly reliable. And thus one can get an opinion of percentage of lies on sensitive matters, and transfer it to the doping area. But nothing is necessarily alike in my opinion when it comes to two different areas, and the researchers estimates might be very uncertain and even wrong…

My mission is to show a huge uncertainty when it comes to estimated doping prevalence, and suspicion against named athletes based on performances. But that doesn’t at all mean it has to be a low prevalence -I have repeatedly stressed that anyone can be doped, and that f.x Norway has a doping list of popped and even doping admitted unpopped athletes, and even has some severe reported flaws in the anti-doping work (despite also being applauded for quite a lot of effort here) and that in my view nothing in a country cultures is a guarantee against doping.

I admit that my gut feeling is that quite a few top athletes are clean. But a gut feeling is far from logical, so I prefer to say to my self: Everyone might be dirty, everyone might be clean -we cannot follow gut feelings in this matter…

know thy history wrote:

Wrong, justanothercuriousdude, I don't equate anything, I just factually point out that you keep ignoring most tables except for the one that suggests to bring the numbers down - which is also based on assumptions, by the way. There is no evidence whatsoever that 10, 20, let alone 30% answered yes because of apathy.

On the contrary, you can see that even the last, slowest 30% have fewer yes answers than the faster 70%. I suppose there were lots of dopers debating whether or not to admit to it - the longer you think, the less likely you admit... So in turn, may I suggest to delete the slowest, not the fastest 30%?

And no, you don't just report the peer-reviewed numbers, you report certain selected numbers because of your extreme bias, and then put your spin to it.

But yes, I also see the bias in that paper. But that is understandable, because the researchers had to downplay the extent of the doping or the IAAF would never have agreed to publishing the work. We do know that the IAAF was blocking (i.e. negotiating the wording and emphasis of) the publication for years.

rekrunner wrote:

Armstronglivs wrote:

This is such nonsense. You wish to introduce a level of uncertainty into the findings of the survey so that you can hold to the view that maybe doping prevalence isn't as much as the survey suggests. It's a play out of the rekrunner book and you have fallen for it. But it isn't just one championship survey that indicates doping is throughout the sport now and certainly at the top.

Not rekrunner or justanothercuriousdude. Did you read the fine print? It was the researchers themselves who wrote that playbook. They introduced and acknowledged and addressed many uncertainties, providing no less than 9 models with 11 unknowns. They also discussed several other sources of uncertainty limitations, but provided no model for these, assuming they were negligible or favorable. Then they proceeded to offer a corrected estimate based on none of the models, but rather empirical observations based on the well-known historically observed habit of "automatic yes" responses to surveys. They had the experience and foresight to measure response times as part of the survey design, knowing from history that the fastest responses would contain a lot of nonsense. They filtered out the fastest responses based on empirical observation, choosing a cutoff where the bias stabilized, and eliminated a 12%-point bias, and a refined estimate of 31.4% (+/- 2.5%). Even discarding the fastest 30%, there is no guarantee that all of the "automatic yes" responses were filtered out.

Any suggestion of "likely more than 44%", an often repeated misconclusion, at best is still pure speculation and rationalization based on assumptions of the unknowns, and should be perceived as subjective personal bias and discarded as propaganda. Most ironic is the assumption about the primary unknown, that anonymity still doesn't ensure many dopers won't lie -- in other words, the survey design fails to achieve its primary purpose, and this approach essentially adds no value or certainty over the simple one question model "did you dope?", and must be refined further or discarded.

The main survey design flaw is that anonymity reduces the incentive to lie, but it doesn't address survey apathy -- what is the incentive for any athlete to spend any time participating in surveys at a world championship? A second design flaw is the inability to detect and estimate non-compliance.

At the same championship, the same respondents completed another survey with a different design, which can estimate non-compliance, resulting in an estimated 21% doping prevelance, correcting for an estimated 30% non-compliance, giving a wide confidence interval of 9%-30%.

Which of the figures between 9% and 47% (range of 95% confidence intervals) is the "true" figure, for the 2011 World Championship survey responders? Hard to say. The reader is still left to debate it because anonymized random response surveys are not yet proven to be reliable estimators.

The unified conclusion from all of the survey studies is that we need better survey designs. This is not my conclusion, but the consensus of the researchers themselves.

rekrunner wrote:

know thy history wrote:

LOL - you just said you wrote "the researchers cut down the estimate to “at least 30%”. 30%, not 31%? Cut down?? Rather, they suggested that IF one removes the fast responders and IF one ignores all the factors, even then it's at least 31%. That's exactly what I meant with your bias.

Note I said you love to ignore, not that you always do.

As for "And as you are saying they clearly stress that the real percentage might be far higher than 30%, and even significantly higher than their other number (43%)" - exactly - when did you ever say that? When did you ever mention the 62%? If you really wanted to show the uncertainty in both directions, you'd say all the time that it could be 31 - 62%, wouldn't you?

You keep making the mistake of equating empirical observations in Table 4 ("yes" responses as a function of measured response time), with the theoretical distortions based on assumed non-compliance in Table 8 (assumed degree of underreporting doping), not to mention all the tables 7-12.

The main reason to disregard large assumptions in Table 8 is that the survey anonymity was chosen by design to minimize "self-protective" underreporting doping. If we still think that 30% of dopers still responded with a self-protective "no", we should consider this design a failure.

We haven't discussed this aspect much, but it's funny you see the bias in those who just repeat the peer-reviewed figures reported in the study, but don't see the bias from the researchers themselves. Much of their discussion is shaped by what they suggest is "plausible" or "likely", based on rationalizations rather than observations and data.

It is not only a "suggestion" to eliminate fast responders. They went out of their way to design the survey to include collecting the responses times in order to perform the exercise of deleting fast-responses based on empirical measurements and observations, to avoid the well-known "serious problem" of "substantial over-estimating" prevalence. We can see why, as performing this exercise suggests a 12.2% over-estimate (and similarly suggests over-estimates for PAG and Supplements).

Note their criteria for deleting the fastest 30% is arbitrary, and they created an alternate "standard error of estimate" (based on what standard?), resulting in their likely representative lower bound of 28.9%, rather than their reported 95% CI of 26.5%. They also suggest (without basis) that this "almost certainly excluded virtually all of the automatic "yes" responders". Had they chosen 50%, the SE lower bound becomes "at least 27.0%" (removing an additional 1.9% overestimation bias, suggesting not "virtually all"), and their 95% CI becomes "at least 24.1%". The table stops at 50%, but why stop at 50%? Could there still be significant "automatic yes" responses in the slowest responses? How can we know?

Their choice of populating tables with assumptions of 10%, 20%, and 30% is indicative, but also arbitrary.

Although we often speak of two published estimates, the unfiltered 43.6%, and the revised 31.4% after deleting fast-responses, the tables 4, and 7-12 provide us with no less than 39 estimates, ranging from 8.9% to 73.1%, limited by the arbritary choice to show a maximum 30% assumed non-compliance for each table. Given we are left at the end of the study with 9 models of non-compliance, and 11 unknowns, not counting the arbitrary criteria needed to choose how many fastest responders should be deleted, there is plenty of wiggle room for everyone to inject personal bias, with no clear guidance as to which values are more likely correct.

I agree with the conclusion of the study, that survey techniques can be valuable, but they are in need of being refined, and designed in a way as to minimize, and/or detect and/or eliminate survey non-compliance that can bias the "true" prevalence.

Furthermore, to paint a broader longitudinal picture, we need more than 1 data point for one event that combines men and women and countries and sprints and fields and distances and roads and walking.