Kevin Sullivan’s training

I don’t know how I missed this, but there was a paper in the December issue of the Journal of Strength and Conditioning Research that is pure track-geek heaven. It’s called “Performance Modeling in an Olympic 1500-m Finalist: A Practical Approach,” and in it, researchers from Eastern Michigan University take seven years of Canadian miler Kevin Sullivan’s logs (from 2000 to 2006) and subject them to detailed analysis.

The goal of the research is to see if they can use basic “impulse-response” training theory to predict upward and downward trends in Sullivan’s race performances. To put it simply, they assume that:

Performance = Fitness – Fatigue

Makes sense so far. Every time time you train, you create some fatigue… and then a little while later, your body compensates by increasing your fitness a bit. So at any given moment, your performance ability can be estimated by adding up the contributions of every training session you’ve done toward your fitness and fatigue. Yesterday’s training session will have a big impact on your fatigue, but none on your performance. A session from three weeks ago, on the other hand, will have a performance impact but not much of a fatigue impact.

So how do you model the impact? Without getting too far into the nitty-gritty, the researchers add up every bit of running Sullivan does and calculate its pace as a fraction of the pace he could maintain all-out for an hour (akin to what runners would think of as threshold pace). A day in which he ran all-out for an hour would get a score of 100. As it turns out, over the course of a full year, he tends to average between 50 and 55 of these “points” per day. During base training, he averages over 60, with individual days sometimes exceeding 100.

So they plug this training data into the “impulse-response” formulas to see if there’s any correlation with performance. Previous studies in other sports have found good predictions averaged over whole teams, but it’s trickier with an individual athlete. They’re not trying to predict exactly how fast he’ll run in a given race — rather, it’s a question of looking for trends, to see whether his “performance score” is getting higher or lower. That way, coaches can react by taking extra rest, training harder, or whatever.

As an example, I’ve included one of the figures here. It shows (A) his 2000 season, when he came fifth at the Olympics, and (B) his 2004 season. The dotted line shows his “training score” — basically how hard he’s training — and the solid line shows his predicted performance. The triangles show his race performances, converted to Mercier points. The circled ones are at the Olympics. In their discussion section, they suggest that maybe he peaked a little too early in 2000 — but it seems to be they’re using their 20-20 hindsight vision to make that call, because that’s not what their model predicts. On the contrary, the solid line is highest right after the Olympics, so maybe he peaked a little late… or maybe he peaked just right. There are many debates that track geeks could have about this data — which is why it’s so much fun!

The original reference: J Strength Cond Res. 2009 Dec; 23(9): 2515-23.