Wednesday, July 17, 2013

The Elusive Dopeometer

Readers of this blog no doubt have seen a number of examples I give about the (in)accuracy of estimating power from climbing speed, and in particular the confounding impact of wind on such estimates. Any climb with wind is going to be subject to error, and it doesn't take much wind at all to introduce quite a sizeable error.

There was my 2010 post about Alpe D'Huez ascent times, timely given the dual ascent this year, and my item yesterday about the significant error introduced by that great unknown, the wind.

I consider such W/kg estimates to be fine for a bit of pub chat fodder, but as a serious means to detect doping, really? I'll get back to this in a bit.

What has been more amusing than climb power estimates was more twitter/blogger/forum sphere musings on the power differential estimates from the Stage 11 individual time trial. Seriously. People are actually thinking they can reliably estimate the difference in power output of riders based solely on their time or speed in an individual time trial.

What complete and utter nonsense.

Aside from that affront to physics by Gazzetta dello Sport's Claudio Ghisalberti, there was also a post by the somewhat infamous Dr Ferrari about speed differentials from the same ITT and what that meant for power differences. Now Ferrari did say that this assumes all riders have the same "aerodynamic efficiency" as he calls it, which he then points out they don't.

So, given that they don't, why would you then proceed to produce and publish numbers as if they were all the same? More nonsense.

All this does is misinform the debate and feed Internet trolls.

OK, so people get the concept of comparing climbing performances (or attempting to) using power to body mass ratios, the now ubiquitous W/kg numbers. But can you do something similar with time trials over flatter terrain?

If you want to normalise flatter terrain ITT data, then you can make a reasonable stab at the differential of each rider's power to aerodynamic drag ratios i.e. the rider's time trial wattage output divided by the rider's coefficient of drag area (CdA), measured in units W/m^2. Of course this assumes the same wind conditions apply for the riders being compared which can be problematic in itself when the riders being compared are on course several hours apart (as was the case for instance with Tony Martin and Chris Froome).

On flatter ground, the higher your W/m^2, the faster you will go. This neat chart courtesy of Robert Chung shows the close relationship between flat road speed and power to aero drag ratio:


Hence we can reasonably say that the faster rider on the day has a higher W/m^2 (putting wind differences to one side of course). This is the flat land equivalent of the hill climber's W/kg.

What we can't say however is how much of that speed difference is due to higher power output and how much is due to a lower CdA. Indeed, it's possible for a rider to produce less power than another yet go faster if their aerodynamics is superior.

I’ve dealt with many riders of similar morphology who have significantly different CdA. I have a former team mate who was same height and weight as me (actually he was a bit heavier) and we have similar power output as well (mine was a bit better), yet his CdA is ~ 20% less than mine on our respective pursuit bikes. Our equipment was very similar. His natural body shape on the bike means he is just far more aerodynamically gifted. That’s why he medals at worlds and I don’t even make state finals.

Unless you know each individual rider’s CdA, attempting to derive power differentials from ITT speed is just pissing in the wind.

In this 2011 post I provided a short snapshot into this, with data provided by Andy Coggan who devised a draft variation to his Power Profiling tables, this time creating an Aero Profiling table:



The top of that list is a good indicator of what's required to set/break Boardman's hour record of 56.375 km (35.030 miles), now classified by the UCI as "best human effort".

Now if people think getting pro riders or their team management to release their power data is proving difficult, try getting their wind tunnel or aerodynamic field test data.

The Dopeometer

OK, so let's get back to obtaining riders' power meter data and using it as a dopeometer.

This seems to be a popular request. Greg Lemond wants the data released, as do the Bike Pure people. All over the net on forums and blogs and twitter and so on people are calling for the data to be released. The point being such transparency is a good thing, and I have no argument with that. But will it actually help? Or will it just be a public relations exercise and not really provide any additional insight into the issue?

Let us for a moment imagine that tomorrow morning we all wake up to find Froome and his professional riding colleagues and competitors release their power meter data.

Then what?


Will having more certainty over the accuracy of power data help us confirm or deny doping status?
A: No. All it will do is re-emphasise confirmation bias for those with an opinion one way or another.

What power output will confirm doping and what won't?
A: Nobody actually knows. All we have is differing opinions on the subject of what constitutes possibly suspicious performances. Yet people already have their suspicions. An SRM file isn't going to change that.

Just where is the doping power plausibility line? Can we really assign such a line? Is 6.2W/kg for an hour proof? 6.3? 6.4? 6.41?
A: In reality we simply can't put a clean line in the sand. The line for each rider may be different, and the line may vary depending on context. How long was the effort? When did it occur? What were the environmental conditions? How steep was the climb? Was it solo or with others? Was it a consistent effort or variable? Who responds better to doping?

Will it change which riders should be placed under scrutiny?
A: No, we already know who they are. They ride bikes professionally and at the elite level, win races and/or go up hills faster than the rest of us mere mortals.

Will it make doping detection easier?
A: Hardly, since proof of doping requires a positive test, a confession or reliable testimony and evidence, and we already know who should be scrutinised.

Will it prove riders aren't doping?
A: Of course not. Since it assumes there is an arbitrary upper power limit for doping to be confirmed, it does nothing to pick up any doping by riders who are below whatever that arbitrary limit is. No green jersey contender for instance is going to out ride the GC contenders on major cols. Hence such data only serves to tell us what we already know, i.e. a handful of riders finished ahead of their competition on the mountain top finishes.

Can power data be manipulated?
A: Yes, of course it can. Accidentally, inadvertently or deliberately. So then we'll have those on the conspiracy trail of a new doping detection avoidance technique of "data doping". Since we already know the amount of slop in power estimates from other methods, then fiddling with the numbers means no-one can really know if numbers are fiddled or not. There are of course forensic data analysis techniques that can identify some examples of that, but only if crude data manipulation methods are used. If riders and their support people are clever enough to manipulate blood to avoid detection, I'm pretty sure they'll be able to work out how to manipulate data to avoid detection.

Has "data doping" happened before? 
A: Sadly, yes as this example shows when a rider attempted to use doctored power data to prove a performance benefit from using a particular type of cycling equipment. Fortunately in this case the fraud was detected - but it took a Professor from Berkeley to point it out.

What would it cost to run such data collection in an independent manner, and free from possible manipulation?
A: Millions of dollars. Think about the number of bikes in the ProTour, the need to carefully calibrate say 1,000 SRMs, to have non-tamperable data loggers, to ensure all riders correctly perform zero-offset checks before and during races. The data collection process. Staff to manage this. Millions of dollars that perhaps would be better directed at improving doping control processes, technology, reducing testing costs, and simply performing more tests and more frequently testing in and out of competition.


I get that people want to see the data, and hope it's a short cut way to provide certainty around establishing whether a particular rider is doping. I get that release of such data may appear to increase transparency. But at the end of the day we'll just be back to where we started before all this data becomes available: i.e. none the wiser about riders' doping status.

SRM make a fine power meter, but it's not a dopeometer.

4 comments:

CLB said...

Good post Alex. Maybe too much accuracy for many people, but covers all the bases I think.

big jonny said...

More information is always preferable to less.

Alex Simmons said...

Only if the information adds value. Else it's just more noise to filter out.

djconnel said...

Great stuff, as always, Alex! I think the argument is that if you see anomalous changes in power production, in excess of what is expected from physiological adaptation accounting for sickness, fatigue, and motivation, then that's a flag similar to the flag which would be raised by biological passport scores on blood values. So it's not a CP or AWC value per se, but the rate of change of these parameters. The limiting factor, however, is how do you quantify the effect of fatigue and sickness and, most difficult, motivation? If a rider goes from uncompetitive in Dauphine to winning the Tour de France is that because of blood manipulation (donating before Dauphine, infusing during the Tour) or because the rider is using Dauphine for measured training and the Tour is going 100%? In Froome's case all that can be claimed is that he's produced power and recovery in the past consistent with his power and recovery in this race, and his body mass has been fairly stable (no AICAR-induced anorexia in Tenerife). But it's all terribly grey-scale.