Monday, July 15, 2013

Froome: zestfully clean, but only until n at T+1?

Yesterday, Froome landed on the podium of the all-time greats.  He ascended Mont Ventoux only two seconds slower than Lance Armstrong in 2002.

Here's the all-time context:

Ventoux: Last 15 kilometers
1. Lance Armstrong: 48:33 (2002)
2. Chris Froome: 48:35 (2013)
3. Andy Schleck: 48:57 (2009)
4. Alberto Contador: 48:57 (2009)
MONT VENTOUX (last 15.65 km [from St. Esteve], 8.74 %, 1368 m) 

What's the big deal?  Why's everyone freaking out?

Because it feels like the second coming of Lance and US Postal!

Of course, it isn't.  It's a different era and Froome is a very different personality than Lance.  Froome has a different face and a different attitude than Lance.  He asserts his innocence far more forthrightly.  Of the categories of new clean riders, he falls into the zestfully clean.


"I just think it’s quite sad that we’re sitting here the day after the biggest victory of my life, quite an historic win, talking about doping." (source)

His messaging is the same as Tyson Gay's, another zestfully clean athlete, in terms of messaging.  Gay faced questions with the same zest Froome does.  I'm not sure the stimulant Gay tested positive for yesterday qualifies as hardcore doping, so I'm not yet willing to label him a fraud, but zestfulness only goes so far.  OK, you're passionate.  That does little in the question at hand:

Doped or not?

When it comes to doping and anyone racing in the Tour now, we can't answer that question now.  But that doesn't mean we can't estimate W/kg, and that, over time, we won't be able to figure out whether Froome is doping.

That's because of two things:  (1) the power of aggregation and (2) market forces.

Aggregation and estimation

Cycling Power and Statistical Power

Froome's earlier win on stage 8 meant very little.  It was a 20-minute effort on the first mountain stage of the Tour.  His chief competitors--notably, Contador, struggled uncharacteristically.  It was by objective measures (i.e., time and VAM) impressive, but even those numbers are subject to exogenous factors--wind, drafting (Porte's shepherding Froome up the middle portion of the climb certainly factored), and so on.

And most notably, it was a one-off.  It was only one climb.  It, like the Karate Kid catching a fly in his chopsticks, can be shrugged away.  It lacked statistical power.

Bias

If there is a damning objection against Vayer's and Science of Sport's model for estimating W/kg, the lack of statistical power is it.  While critics have argued persuasively that estimation is inevitably subject to error, no one that I'm aware of has persuasively argued that it is biased (i.e., that it consistently over- or under-estimates power output.

There's a difference between bias and error, and it's important and relevant here.

Error: randomly different from the actual.
Bias: consistently different from actual.  I often underestimate how long it will take me to complete mowing the lawn.  My model here is biased in one direction.  (Geeks may want to check out homoscedasticity).

Estimating W/kg using speed, time, and slope is subject to error, but it is not, as far as I've seen, subject to bias.  A wind picks up (as it did on Ventoux yesterday), and it may underestimate.  A rider sits in the shelter of the pack the entire climb, and it may overestimate.

But the point is that it does both--it over- and underestimates power.  And it does so equally.

That's because the sample size behind the model is large enough that biases were corrected.

Error

Statisticians deal with measurement error by increasing sample size.  The larger the error, the larger the sample size required.  As long as the measurement error is not correlated with anything (that is, it is unbiased), given a large enough sample size, the aggregate estimation will come close to the actual.

Think about darts.  If you throw three darts at a board, you may or may not hit the bullseye. And even if you aggregate the three spots on the board, the average location of the darts will probably not be the bullseye.  But if you throw a thousand darts, the aggregated location of all the darts will probably match the bullseye.

The takeaway:  Vayer's model may be wrong about a particular climb, but if its predictions from several hundred climbs are aggregated, it is correct.

Market forces

Bike racing, like everything from music to soybeans, is subject to market forces.  It responds to incentives.  It changes as the rules and demands change.

The EPO era was driven by market forces.  Cyclists and teams didn't set out to dope, but they were driven to succeed, to make money and earn glory, and EPO was a precondition for success.  Equilibrium was established at a point somewhere near the point where clean cyclists could even hang on the back of the peleton.

Early adopters were the big winners.  Dr. Conconi and his disciples Ferrari and Cecchini ensured Italian domination in the 1990s.  

Lance's relationship with Ferrari, of course, came later.  Lance paid Ferrari extraordinary sums of money to corner the kind of expertise superior to the services Ferrari provided other cyclists.  

In the same way, market forces drove sponsors toward "clean" teams such as Jonathan Vaughters' Garmin squad and Sky.  

Let's be clear--the information that came out during the USADA investigation was nothing new.  It was merely a confirmation of what we had already heard, except this time there was money behind it.

So what market forces are at work now?

Sky and Garmin have pioneered an approach to cycling performance that shuns dope.  They've hinted at techniques, particularly altitude training, but they keep a lot of secrets.    

They won't be able to hold these secrets--and we're hoping they're clean secrets--forever.  Other teams will adopt them.  That's how market forces work.  

And if that's the case, then down the road we'll see others pulling off rides like Froome's, and we'll see Braillesford's statement about "clean performances [surpassing] the doped performances of the past" occuring regularly, and not just on his team.

Conclusion

As Froome continues to race, we'll get more and more estimations.  And as our estimations increase, we can be more confident that the aggregate of estimates of Froome's W/kg are closer to the actual W/kg of Froome.  We will also, hopefully, see more parity down the road as Sky's training secrets come out.

Of course, if we had Froome's weight and power files, this would eliminate the need for our reliance on estimation and aggregation altogether.  Wouldn't that be nice?

Where precisely we can draw and line based on performance and use a W/kg to mean certain doping, I'm no authority to declare.  All I can assert is that until critics address bias, the model not only stands, it grows stronger and more powerful over time.

Even more convincing would be training logs and the models behind Tim Kerrison's coaching--assuming it's something other than a needle in the vein.

Certainly, Sky will guard this information.  It's business, after all, and Sky has a corner on the market for the near future.

For those of us who want to learn something about our bodies and how they function, to know what Froome means when he says he earned his triumph with hard work, waiting for this information, hoping it's clean--well, we can be excused for not rejoicing at the triumphs of a question mark.

1 comment:

CoCo Rico said...

Here's another athlete who was Zest-fully clean...at least he was in the commercials.

http://www.youtube.com/watch?v=s5iqpb6L7pY