Monday, December 13, 2010

Did the ITT Results at Cross Nationals Act as a Valid Call Up Method?

By now, most of us are home and the CX Nationals hangover is in full effect.  All the, "what ifs, could haves, and if onlys" have been recounted way too many times and, as we all return to the real world, or even the off-season, some lingering complaints and questions remain about the 2010 championship event.

Complaints at events of this caliber are typically equal in quantity to excuses but a few things are obvious.  First and foremost, every single one of the winners earned every thread of that stars and stripes jersey.  Congratulations to everyone!  Second is that the crowd was definitely something that American CX racing has yet to see.  Hopefully, Tom Schuler can keep that momentum rolling and Madison January 2012 will be even an even bigger spectacle to behold.

A consistent complaint that was heard throughout the week involved the Cyclocross ITT, the result from which dictated call up order for the national championship race although the ITT was on a different day, with different weather conditions.  Also, the ITT was run on a completely different course, which means different surface properties.  The possibility of different physical demands and handling skill sets affecting the outcome of the race loomed.

We looked at the new call up method and even called upon some statistical expertise with the hope of being able to answer a question that everyone asked at some point over the weekend: Did the ITT results at cross nationals act as a valid call up method?

In statistical terms, "is there a significant correlation between the time trial performance and race performance?"

In order to examine this question, only results from racers who completed both events could be looked at.  We accomplished this using voodoo magic and something called casewise deletion of missing data.  This removed the first row of call ups from the equation.  Also, it got rid of everyone that did not start the ITT as well as anyone who did not finish the championship race.  Remember that we're only looking at the validity of the ITT as a call up method.  If the ITT and the race aren't completed, then a necessary piece of data is missing in order to be considered.

Next we applied this analysis only to the races with more than 25 athletes completing both competitions.  The larger sample size leads to more precision in the calculation and thus better quality of information.

Results:  After using some spreadsheet ninja skills to work some super sweet math skills that a guy named Pearson discovered back in late 1800's we were able to come up with some results. First, all the races that we applied the above standards to showed significant and strong correlations.  The "p-value" of less than 0.05 is widely accepted in academic literature as a significant finding and the < 0.01 is even better. 

Single speed: r = .85   p <  0.05

13-14: r = 0.91           p <  0.01 for all other races
15-16: r = 0.81
17-18: r = 0.90
30-34: r = 0.90             
35-39: r = 0.87
40-44: r = 0.87
45-49: r = 0.90
50-54: r = 0.80

30-34: r = 0.66
40-44: r = 0.88

What does this mean?
Athletes that perform well in the ITT, are also performing well in the race when compared to other athletes who completed both events.

Ultimately it could be said that the different ITT day, weather, surface conditions and course difference did not matter.

Why does this happen?
Simply put, cyclocross is a combination of fitness and bike handling.  Those athletes that are superbly proficient in both of those general traits will excel in the sport and outperform those athletes who are not as proficient in both, or lack one trait all together. 

Wrapping up, answer the questions we posed earlier: Yes, the ITT has been shown to be a valid call up method and yes, ITT results and race performance are strongly correlated.

Of course, a control is needed for starting position in ITT performance. This could address the issue that start position could affect race performance.

And finally, don't forget the number one rule of statistics, correlation does not imply causation.

Special thanks goes out to Alyson D Abel, MS and her mad crazy spreadsheet ninja skills.