CoachCox

Kona Statistics 2012: The Relationship of Swim, Bike and Run

I’ve not quite finished. A couple of months back I examined the results from Challenge Roth – retrieved far more easily, complete with splits – to look at the relationship between bike and run performance across the field. I was interested to see if that data backed up the notion that age groupers who bike too hard pay a price on the run. My initial conclusion at the time was shockingly that faster athletes are faster. Scatter graphs from Roth results (below) show the distribution between splits in swim, bike and run with overall times and gave a hint towards the degree of correlation.

Comparison of Swim, Bike, Run and Overall Times for Athletes at Challenge Roth 2012

Each comparison has a trend line in red giving an indication of the kind of relationship; in each case faster finish times come from faster splits. Faster athletes go faster. The degree of scatter around the best fit line gives some indication as to the strength of the correlation; for example, in Roth, any correlation between swim performance and finish time is weaker than any correlation between run performance and finish time. Not that correlation is necessarily cause. Unsurprisingly repeating the process for Kona shows similar results.

Comparison of Swim, Bike, Run and Overall Times for Athletes at the Ironman World Championship 2012

Faster athletes are faster at the World Championship too. The relationship seems a little tighter too – less dispersal when comparing swim times and finish times, or run times against bike times. Perhaps the result of the selective entry criteria for the majority of athletes: qualification removes the weakest swimmers, cyclists and runners. Having to qualify means that you cannot afford to have a genuinely weak discipline, there is variance, but in relative terms these athletes are quite balanced.

If we only consider the fastest athletes, which I’ll take to mean those going under 10 hours, then the relationships between the disciplines are remarkably similar for both races. Roth is faster, it is easier to break 10 hours there, but both have a high standard of fast athletes. As the chart below shows, in Kona the sub-10 athletes are again balanced. Notably the swim distribution is more tightly packed than for the equivalent athletes at Roth, selection at play once more or perhaps a side effect of the mass start. Perhaps it’s worth reflecting that the mass start may improve drafting opportunities during the swim and unfortunately increase the impact of drafting on the bike, we might expect both to lead to a tighter distribution of splits.

Comparison of Swim, Bike, Run and Overall Times for Sub-10 Athletes at the Ironman World Championship 2012

In my previous examination of the bike-run relationship I didn’t stop here, I attempted to find better evidence that a hard bike often leads to a slow run. Bear with me. I divided finishers into groups by percentage – 10% brackets from fastest to slowest, and then calculated the average bike splits for the fastest runners (top 25% of run splits in the group), slowest runners (bottom 25%) and the whole group. The idea being that if a hard bike leads to slower run we would expect to fine that the slowest runners tended to ride faster than the group average and the fastest runner tended to ride slower. Which is exactly what the data showed for Roth. Would I call it proof? No, but it offered a little support for a long held belief.

For Kona I’ve slightly altered my approach, dividing the field into fixed 30 minute blocks (with the exception of those going sub-9 who are grouped together). While the percentage approach ensured each group of finishers was of equal size, fixed finish times perhaps ensures groups contain a more comparable set of athletes. The results are in the graph below (you can also see a table of the data here)

Average Bike Splits for Fastest, Slowest and All Athletes in Each 30 Minute Finishing Period of the Ironman World Champs 2012

As was the case in Roth the fastest runners have slower bike splits and the slower runners have faster bike splits; and as with Roth the question remains as to whether this is indicative of a relationship. It is worth noting that in the Sub-9 group, mostly male pros, there is little difference between the group average bike and the fastest runners’ average bike split, but as we spread back in the field the gap opens up. Also the faster an athletes finish time the smaller the difference between bike performance for different runners, a reflection of overall fitness and possibly better pace awareness. Of course swim and transitions being equal an athlete who runs 3:15 to finish in 9:45 must have biked more slowly than an athlete who runs 3:20 to finish in 9:45. As I said – this is not proof, more a potential indication. But if it is the case that those slower runners are biking too hard, the data shows that what they typically gain on the bike they more than lose during the run; their bike may be 5 minutes faster than the group average, but their run is 10 minutes slower.

Conclusions? The patterns I previously saw in the Roth results seem to be present in the Kona data, if there are differences then Kona is – questionably – more tightly bunched. Perhaps a subtle consequence of a mass start versus waves and the greater drafting and bunching opportunities that followed it, but equally potentially a reflection of the selective qualification process. Needless to say it remains the case that the fastest athletes are all round faster and that working too hard on the bike may really make you pay on the run.

Ironman Training Library

From nutrition to pacing - a collection of CoachCox blog posts focused specifically on Ironman training and racing.

Comments

  • Rob Knell

    Hi Russ

    Your stats need some work: just about all of your data show strong “heteroscedasticity” which means that the variance increases with the mean – hence the data clouds on your scatterplots are fan-shaped rather than oval. This means that trendlines fitted with standard linear regression techniques (which is what Excel does if you used that) are not going to work too well. See, for example, the way that the lines often fail to predict the faster times well – the trendlines are above the actual data. Just eyeballing the data I think there’s probably some non-linearity (curved rather than straight relationships – see the run time vs bike time and run time vs overall time plots) in quite a few of these relationships as well. If you want me to have a crack at fitting some better models to the data email me the spreadsheet and I’ll see what I can do.

    PS I’m not a real statistician but I play one on the TV.

    Rob Knell

  • Fair to say it’s been a long time since I’ve had to do real stats (roughly 15 years I guess)!

    Agreed linear regressions (which is what I opted for) doesn’t fit the data well (especially as I was lazy in the Kona graphs and left in a few non-finishers), which I guess isn’t surprising – pros are a distinct group from age groupers and the impact of age must lead to some segregation in performances and their relationships. Actually quickly playing myself, removing all non finishers improves things as does opting for a polynomial regression, looks better and a higher R2 value.

    Welcome to play with the data, I’ll mail it over to you. I’ll admit part statistical rustiness (and to be honest I hated stats!) and part laziness on my part. Be good to see what you make of them.

    Russ

CoachCox

Coaching

Ironman Training

Ironman Analysis