a shill by any other name wrote:
Word of caution about interpreting statistics (which you've done a very nice job of compiling and explaining)... it's easy to see patterns that aren't real, and assume that the strength of your dataset is adequate to defend your interpretation.
Excellent points - you posted this while I was typing my comments above regarding dividing a group into smaller sub-groups.
Thanks. Except for my choice of cutoff, to select "top time", I used a rather complete dataset. We could do similar analysis, using top-10 average, top-20 average, or even top-20th time. All the numbers would be bigger, and the percentages would change. I'm not sure it would change the general observations though -- there would still be enough basis to ask, for 5 continents, why so few by so little when EPO promises so much?
a shill by any other name wrote:
rekrunner,
thanks for the interesting analysis. Word of caution about interpreting statistics (which you've done a very nice job of compiling and explaining)... it's easy to see patterns that aren't real, and assume that the strength of your dataset is adequate to defend your interpretation.
Interesting story (if perhaps only to me), a colleague recently shared with me a neat paper about linkages between lunar cycles and large earthquakes, which drew the (evidently well-defended) conclusion that very large (M8+) earthquakes show no relationship with lunar cycles (which correspond to small cyclical changes in gravitational pull, thus small changes in crustal stresses...). The paper got me wild-eyed interested in the topic so I did my own exploration, and grabbed about 14000 "almost random" earthquakes (M2.5+) from available records (USGS) to see if I could detect an underlying lunar signal. Those data DID suggest a relatively higher frequency of earthquakes around the new moon relative to the full moon (with a lot of scatter), so I explored further, grabbing over 50,000 earthquakes. This larger dataset seemed to confirm and strengthen the initial hypothesis, so I got really excited and shared the finding with some colleagues. One smart colleague asked me if the interpreted pattern was statistically significant, a fair question; so, working a little more methodically to choose more earthquakes randomly, I increased the dataset to 104,000 earthquakes. At this point, the evident pattern completely disappeared.
The lesson that I (re)learned in this exercise is that it's easy to fall in love with conclusions you've drawn from statistical datasets, but the reality of (particularly complex, chaotic) natural phenomena is that they like to defy probabilistic description in sometimes unexpected ways. And interpreting the presence or absence of a doping effect from a small number of performances is a risky thing to do. Or at least one that ought to be done without a high degree of confidence expressed in the outcome.
Nice work though. Cheers
rekrunner wrote:... I used a rather complete dataset.
Maybe so, but "complete" isn't remotely the same as statistically significant. In my counter example it took > 100,000 data points to disprove a false interpretation derived from > 50,000 data points. Here you are using a few dozen data points and presuming (because they capture "all" the top performances) that they represent the truth of all humanity. Big stretch... VERY big...
(I'm surprised you don't seem to "get" this, given the evident understanding conveyed in your first summary)
a shill by any other name wrote:
(I'm surprised you don't seem to "get" this, given the evident understanding conveyed in your first summary)
Are you really "surprised"?
casual obsever wrote:
rekrunner wrote:
Splitting the improvements in smaller sub-groups highlights the regional nature of large improvements
It also worsens the statistics. A lot. And potentially skews them when the person who wants to make a point chooses the sub-groups, like you did here.
Also, you need to be more consequent. This here is a good start:
rekrunner wrote:
If these claims are accurate, we should be able to observe the large effect from EPO on the best performances, pre-1990, versus post-1990.
But if you consider that the best performances post-1990 are EPO-aided, then you need to consider that the pre-1990 performances are also enhanced by the then basically undetectable (legal and illegal) methods.
Otherwise you'd postulate that the sport was clean pre-1990. We know better.
The question then becomes: how much - possibly gender- and distance-specific - helps (at first unlimited) EPO and controlled but more powerful steroids vs. unlimited blood transfusions and basically unlimited but primitive steroids. And yet, each time somebody brings this up, you argue against it.
On the other hand, if you were to think that runners are too honest to cheat even without noticeable risk of detection (and not even use blood transfusions while allowed), then this method were absolutely pointless. Then we can only look at the athletes with positive tests, and whose doping history was proven by admissions and witnesses. At the top, those are 0, as nobody talks, and everyone pretends they only doped once.
Finally, let's look at the numbers above. Why would the longer distances benefit more? Ok, the marathon was underdeveloped in the 80s, and now there is more money, so we can't really compare that.
But do 5000 m runners earn more than 800 m runners? Doubt it. More likely, the benefit from the steroids decreases with increasing distance, while blood doping methods become more important.
Begin with two ideological and racist assumptions :
1/ Everybody cheats.
2/ Africans are genetically superior.
Use psuedo-scientific pretence of being experts in statistical analysis to show what was already obvious- Africans improved massively as soon as epo became available.
Use assumptions 1 + 2 and psuedo-scientific pretence to prove assumptions 1 + 2 and ignore obvious explanation that athletes from Third World corrupt nations with no testing are more likely to cheat.
rekrunner wrote:
Your EPO champions were not as fast as the steroids champions of the past.
C'mon man...they're not "my" EPO champions & I never said they were as fast as the pre-EPO era times. But there's a problem when you examine the fastest women's 800 times and conclude that's it's only androgens involved & not augmented by blood doping. 5 out of the 6 fastest times are pre-1990 but 3 out of those 5 and 6 out of the top 10 all-time best are runners from Soviet block countries. In these cases, you can't rule out blood doping as it was being used extensively with the Soviet block nations in the 70s & 80s. So, was it androgens only pre-1990? Blood doping & androgens? OTOH, could some of those non-Soviet block runners possibly be clean?
casual obsever wrote:
rekrunner wrote:
Splitting the improvements in smaller sub-groups highlights the regional nature of large improvements
It also worsens the statistics. A lot. And potentially skews them when the person who wants to make a point chooses the sub-groups, like you did here.
Also, you need to be more consequent. This here is a good start:
rekrunner wrote:
If these claims are accurate, we should be able to observe the large effect from EPO on the best performances, pre-1990, versus post-1990.
But if you consider that the best performances post-1990 are EPO-aided, then you need to consider that the pre-1990 performances are also enhanced by the then basically undetectable (legal and illegal) methods.
Otherwise you'd postulate that the sport was clean pre-1990. We know better.
The question then becomes: how much - possibly gender- and distance-specific - helps (at first unlimited) EPO and controlled but more powerful steroids vs. unlimited blood transfusions and basically unlimited but primitive steroids. And yet, each time somebody brings this up, you argue against it.
On the other hand, if you were to think that runners are too honest to cheat even without noticeable risk of detection (and not even use blood transfusions while allowed), then this method were absolutely pointless. Then we can only look at the athletes with positive tests, and whose doping history was proven by admissions and witnesses. At the top, those are 0, as nobody talks, and everyone pretends they only doped once.
Finally, let's look at the numbers above. Why would the longer distances benefit more? Ok, the marathon was underdeveloped in the 80s, and now there is more money, so we can't really compare that.
But do 5000 m runners earn more than 800 m runners? Doubt it. More likely, the benefit from the steroids decreases with increasing distance, while blood doping methods become more important.
Good point on blood doping methods:
https://www.ncbi.nlm.nih.gov/pubmed/3573270"Reinfusion of RBCs caused a reduction of approximately one minute in time to run the
10-km race."
a shill by any other name wrote:
rekrunner wrote:... I used a rather complete dataset.
Maybe so, but "complete" isn't remotely the same as statistically significant. In my counter example it took > 100,000 data points to disprove a false interpretation derived from > 50,000 data points. Here you are using a few dozen data points and presuming (because they capture "all" the top performances) that they represent the truth of all humanity. Big stretch... VERY big...
Nice posts on interpreting statistics.
As for doping:
Problem is there are many ways to skin this cat.
Every coach has their own training regimes.
Every drug-coach has their own drugs regimes.
There's an art to running and to successful coaching.
Unpicking all of this is something that science can't do yet (possibly ever).
The dataset is too small with lots of inconsistencies.
I do remember in the 90s and early 2000s a lot of runners coming from nowhere to spank
everyone in championship finals and then disappearing. I don't think they cared as much
about times and end of year rankings.
I guess there is no problem then, as I did not exclude transfusions: "I'm not seeing a strong case for replacing transfusions with EPO boosting VO2max to new levels of performances, for men or women."
Rocket Fuel Madness wrote:
rekrunner wrote:
Your EPO champions were not as fast as the steroids champions of the past.
C'mon man...they're not "my" EPO champions & I never said they were as fast as the pre-EPO era times. But there's a problem when you examine the fastest women's 800 times and conclude that's it's only androgens involved & not augmented by blood doping. 5 out of the 6 fastest times are pre-1990 but 3 out of those 5 and 6 out of the top 10 all-time best are runners from Soviet block countries. In these cases, you can't rule out blood doping as it was being used extensively with the Soviet block nations in the 70s & 80s. So, was it androgens only pre-1990? Blood doping & androgens? OTOH, could some of those non-Soviet block runners possibly be clean?
Are Gary and Casual Observer actually different people? I somewhat doubt it, but if you are, couldn't you just have this mutal two way jerk off via e-mail?
I get why you got earthquakes wrong. You took a random selection of samples, and your sample size was too small, and your small dataset was not representative. I guess if I had chosen a small subset of athletes at random to measure, and projected these measures onto a larger group, I risk making the same mistake. What I'm not getting: - What's all this about "the truth of all humanity"? The two metrics I made for a group/event are only intended to be interpreted for that group/event. I chose the parameters I chose, specifically to look for performance effects at the top. What happens with the rest of humanity below the top is another story with a different ending. I didn't "(interpret) the presence or absence of a doping effect from a small number of performances" but "failed to observe a large performance improvement for the best performances of one of the groups I looked at", and asked how that can be explained in the EPO-era, given the many assumptions and promises often claimed for EPO.
a shill by any other name wrote:
rekrunner wrote:... I used a rather complete dataset.
Maybe so, but "complete" isn't remotely the same as statistically significant. In my counter example it took > 100,000 data points to disprove a false interpretation derived from > 50,000 data points. Here you are using a few dozen data points and presuming (because they capture "all" the top performances) that they represent the truth of all humanity. Big stretch... VERY big...
(I'm surprised you don't seem to "get" this, given the evident understanding conveyed in your first summary)
Racist? What race is North Africa? "5-continents" is pretty much all races. 1) I thought we fixed that. "Athletes from everywhere cheat, from 29-57% according to a recent Tuebingen survey". 2) I never assumed or concluded "Africans are genetically superior". I also showed "non-Africans failed to improve massively as soon as epo became available". "Athletes from Third World corrupt nations with no testing are more likely to cheat." And you called my assumptions "racist"? Are you suggesting that none of the top athletes from "5 continents" took EPO during the EPO-era? That's the opposite of what we saw in cycling.
Coevett wrote:
Begin with two ideological and racist assumptions :
1/ Everybody cheats.
2/ Africans are genetically superior.
Use psuedo-scientific pretence of being experts in statistical analysis to show what was already obvious- Africans improved massively as soon as epo became available.
Use assumptions 1 + 2 and psuedo-scientific pretence to prove assumptions 1 + 2 and ignore obvious explanation that athletes from Third World corrupt nations with no testing are more likely to cheat.
Why are you asking me?
Not everybody dopes but u2r dopes wrote:
Are Gary and Casual Observer actually different people? I somewhat doubt it, but if you are, couldn't you just have this mutal two way jerk off via e-mail?
Not everybody dopes but u2r dopes wrote:
Are Gary and Casual Observer actually different people? I somewhat doubt it, but if you are, couldn't you just have this mutal two way jerk off via e-mail?
0/10. What a wiseass comment! (imagine that...wiseass comments on LRC ?). There's been good & insightful information discussed here by rekrunner, casual observer, Coevett, amateur & others...and you come in here and throw a stupid wiseass crack like that. Is this subject matter too complicated for you to offer anything of value? Maybe college classses got out early and you have nothing better to do? ?
Rocket Fuel Rick wrote:
Good point on blood doping methods:
https://www.ncbi.nlm.nih.gov/pubmed/3573270"Reinfusion of RBCs caused a reduction of approximately one minute in time to run the
10-km race."[/quote]
Rick, I agree blood reinfusions obviously help 10k times, people who deny this have serious issues, but what is interesting is that 10,000m bests improved by some 3% within a decade after epo's mainstream arrival c1993-1994. Something like HGH in combination with EPO and altitude training must have produced a mutant class of runners when compared the 1980s 27:20 guys who likely were doing infusions.
casual obsever wrote:
It also worsens the statistics. A lot. And potentially skews them when the person who wants to make a point chooses the sub-groups, like you did here.
What do you suggest I skewed? Would you chose different sub-groups?
casual obsever wrote:
But if you consider that the best performances post-1990 are EPO-aided, then you need to consider that the pre-1990 performances are also enhanced by the then basically undetectable (legal and illegal) methods.
Otherwise you'd postulate that the sport was clean pre-1990. We know better.
Did *I* consider the best post-1990 performances EPO-aided? I'll accept that the sport, pre-1990, was not clean.
casual obsever wrote:
The question then becomes: how much - possibly gender- and distance-specific - helps (at first unlimited) EPO and controlled but more powerful steroids vs. unlimited blood transfusions and basically unlimited but primitive steroids. And yet, each time somebody brings this up, you argue against it.
Fine. How do we get answers to the questions, without answering them ourselves with new assumptions? Were the *best* pre-1990 times a result of "unlimited blood transfusions and basically unlimited but primitive steroids"? Are the *best* post-1990 times a result of "(at first unlimited) EPO and controlled but more powerful steroids"?
casual obsever wrote:
On the other hand, if you were to think that runners are too honest to cheat even without noticeable risk of detection (and not even use blood transfusions while allowed), then this method were absolutely pointless. Then we can only look at the athletes with positive tests, and whose doping history was proven by admissions and witnesses. At the top, those are 0, as nobody talks, and everyone pretends they only doped once.
I don't doubt runners cheat. We disagree on how much that matters.
casual obsever wrote:
Finally, let's look at the numbers above. Why would the longer distances benefit more? Ok, the marathon was underdeveloped in the 80s, and now there is more money, so we can't really compare that.
But do 5000 m runners earn more than 800 m runners? Doubt it. More likely, the benefit from the steroids decreases with increasing distance, while blood doping methods become more important.
Benefit more from what?
So long as we are sticking to what the record shows, Ceplak's PB is from 2002, and her EPO bust is from 2007. In 2007, she was running slower at 1:59 and 2:00.
Rocket Fuel Rick wrote:
And "doping busts" are clear cut and allows for the analysis of progression trends, PBs, competition record & performance post-ban for those who returned to competition. For example, Ceplak had a big 3 second jump in one year before she set her sizzling 1:55 PB. And when she came back from her ban, she was crap never running anything near her fastest times and not even coming close to breaking 2 again.
So, show me where Ceplak (was) using androgens? The record shows an EPO positive ....
rekrunner wrote:I didn't "(interpret) the presence or absence of a doping effect from a small number of performances" but "failed to observe a large performance improvement for the best performances of one of the groups I looked at", and asked how that can be explained in the EPO-era, given the many assumptions and promises often claimed for EPO.
OK, I get where you're coming from.
"... how that can be explained..." is simple. Your experiment, or investigation, was inadequate, from a statistical point of view, to produce a confident answer one way or another. It wouldn't matter what your stats showed, it couldn't be taken as an indication of doping, or not, or any other factor. The sample is (by a very large degree) too small to be statistically significant.
I suspect maybe you see where I'm coming from, but maybe you don't, and in any case I'll explain with a simple, if (grossly) imperfect, analogy. I've just taken a 1 euro coin out of my pocket and flipped it 20 times. I (honestly) got the following sequence of results: H T T T T T T T T T H H T T T T T T. (it may seem like I made that up, but it's bob's honest truth; I think the gods smiled on me today to help me hammer home my point... :-)) If you reproduce my experiment, you will almost certainly not get the exact same result (the probability is vanishingly close to 0), and even getting the same statistical result (17 tails out of 20 trials) is very, very unlikely. But it's a real result, it just happened in front of me at my desk while I ate breakfast and delayed starting my work day...
I can take that "complete" sample, and form some very wrong deductions, including:
- my coin produces long series of tails, hence I can expect long series of tails every time I toss the coin
- my coin has an 85% probability of showing tails on any given toss
You have taken your "complete" (but very small) dataset and concluded that the absence of a pattern, being absence of evidence of a doping influence, is the same as evidence of absence of a doping influence ("what other explanation could there be?"), when the other explanation is simple noise overpowering the signal in your data.
I think (but don't know for certain) that a fairly simple Monte Carlo analysis would reveal how weak your experiment is relative to its ability to divine anything about doping effects.
I took a different 1 euro coin, out of the SAME pocket, mind you... and just produced the following series of results: H H T T T T H H T H H H H H H H T H H .
It seems like THIS coin tends to produce long series of heads, and also yields heads 70% of the time. Now I can do one of two things with this newly developed knowledge about my two coins:
(1) I can go into a bar, confident of my two coins' abilities and trick some guy into betting against me, when I *know* my coins are unfair. Or,
(2) I can recognize that both my coins are actually most probably fair, and will yield about 50% of each heads and tails, in the long run, and NOT throw away my money in the bar.
Here, you've walked into the wild west LRC saloon, threw your coin (race stats) down on the table and challenged us all to a bet about doping, believing in your heart your coin contains some hidden wisdom not known to the rest of us. I defy you! Your coin does NOT contain unfair hidden wisdom.
(All in fun of course; I do admire the effort you took to work out the stats and think it all through, I simply don't agree that any conclusions are warranted)
rekrunner wrote:
What do you suggest I skewed? Would you chose different sub-groups?
I expressed it neutrally. You took a small data set (see the comments from a shill is a shill is a shill), and made it smaller by choosing sub-groups based on what you consider important, e.g. 1) North Africa, 2) East Africa, 3) rest (including South Africa). You also decided that heritage (genes) are more important than environment by placing African-born Europeans into the respective African group (again, except for South Africa). There is something to that, but one could also argue the other way. Plus a little extra treatment of Spain and China here or there.
Personally I wouldn't (actually didn't) do this at all, because there are way too many unquantifiable variables here.
Maybe a better approach: make a list of all occurrences that are debated to have an impact, such as beginning of testo doping, beginning of testo testing, HGH, beginning of EPO, 1st EPO test, 2nd EPO test, CERA, beginning of ABP, beginning of WADA, end of the Cold War, $$$$ for marathon wins etc.
Then take all best-times in any given year, and do best/top-10/top-20, to see whether any trends are visible.
Maybe two scenarios: one with annulled results of drug cheats, and one without.
rekrunner wrote:
Did *I* consider the best post-1990 performances EPO-aided?
Looks like it. Because if you didn't, then those times can't help to look at the impact of EPO.
rekrunner wrote:
Fine. How do we get answers to the questions, without answering them ourselves with new assumptions? Were the *best* pre-1990 times a result of "unlimited blood transfusions and basically unlimited but primitive steroids"? Are the *best* post-1990 times a result of "(at first unlimited) EPO and controlled but more powerful steroids"?
See above for a possible approach.
Having said that, the following statements are more than just assumptions:
- there was, and continues to be, a high prevalence of dopers
- performance enhancing drugs enhance performance
- talent is independent of character
Then it turns indeed into simple statistics. A hypothetical scenario:
Let's take the 100 m for an example. Let's estimate that doping without triggering alarms brings you 0.1 s. Let's estimate 50% at the top dope. Let's take a hypothetical scenario of 30 sprinters capable of running 9.8 - 10.0 s, distributed evenly: 10 have a 9.8 potential, 10 a 9.9, and 10 a 10.0. Questions:
a) Is the winner most likely clean?
b) Will the winning time more likely be over or under 9.8 s?
You could actually simulate the outcomes with a simple spread sheet and a random function. Redo the above with a doping benefit with 0.2 s.
Maybe I'll do that tomorrow.