*I run, therefore I'm injured.*

It was Puffing Billy that did it. My win against the train came at a cost - a nagging groin injury which I'm only just starting to see the back end of six weeks later. That means I have about six hours a week to try to fill a running-shaped hole in my schedule. I could use those six hours to spend more time with family, or donate my time to a charity. But I’m not going to do that. Instead, I’m going to spend it looking at data. Running data. Ultra running data.

Why? Because I’m a data guy. That’s my job – to look at data and turn it into information. So, what information can I find in the world of running data? Specifically, I’m interested in data from ultra marathons and what it can tell us about those who do well and those who do less well. (Yes, yes, everyone who completes an ultra has done well, but the cold hard fact is, some do better than others. They are races, after all.)

At the outset, I’m going to confess that I offer no guarantee of the statistical validity of my findings. Maybe if I had enough data, I could offer this. But I don’t. Or at least, I can’t be bothered finding enough. Instead, I’ve looked at two of Australia’s biggest 100km races (which couldn’t really be more different in terms of terrain) – UTA 100 (nee: The North Face 100) and the Surf Coast Century. I picked these two events because of (1) the size of the start list and (2) the checkpoint data that’s available for these races.

I wanted to focus on pacing and specifically, which groups did it well and which did it not so well. Did the front runners go out hard and barely hold on? Did the backmarkers take it easy at the start, knowing they had a long day ahead of them? Or was it the opposite? Were women better than men? Old better than young? Did certain parts of the course slow down the back-of-the-packers more, relative to the guys and girls at the front? Or, did it not matter – did fast runners slow down at the same rate as slow runners?

Why the focus on pacing? It just so happens that it’s an area that I’ve worked very hard on in recent years. My main goal in a race is now to perform better in the second half of the race relative to the average runner and runners around my final position – if I don’t do that, I’m disappointed.

Onto the results.

Firstly, UTA 100.

I used the checkpoint data currently available for the 2015 edition of the race to split the race into eight segments, as well as a rough first “half” (0-46km) and second “half” (46-100km). I was first interested in how much longer the second “half” of the race took, compared to the first (remember, the second “half” is actually 17% longer than the first). On average, across all competitors with the relevant data, the average “slow down” was 60% - that is, the 46-100km segment took 60% longer than the 0-46km segment.

It turns out that, on average, there was no difference in the proportional slow down between men (607 runners) and women (171 runners) – both had an average slow down of 60%.

More differences emerged when you looked at the finishing position of the runner. And it’s important here to remember that I’m not comparing overall times – just the rate at which different runners slow down over the race. There’s no rule that says a slow person will slow down faster than a fast person – it’s all about how you judge your own level of effort and endurance. Think about a 10km race – a 34 minute runner and a 60 minute runner can have an identical ‘slow down’ rate if they’re both good at pacing to their ability.

The below chart shows the relationship between finishing position and the rate of slow down. You can see there’s a small correlation between the two.

I think the uptick towards the end of the series (i.e. the final 10% of finishers) is probably down to something more than just bad pacing. These may be people who have injured themselves, or are completely new to ultras and are just doing whatever it takes to finish. Still, it does appear that those at the front of the field are better judges of what is a sustainable pace than those in the middle to the back of the pack.

What’s striking though, is that there are many runners, at whatever position in the rankings, who mix it with the best in terms of percentage slow down. This suggests to me that pacing can be learned and applied to ultras – just because you finish 600th, doesn’t mean the second half of the race is going to be a nightmare compared to the first.

What about age? It turned out that the older you were, the more you slowed down. On first glance this might not sound surprising – but remember, we’re not talking about overall speed here, we are talking about how much runners slowed down relative to their own early splits. And even in the “super masters” category (50-60 years old), there was a healthy proportion of runners who outperformed the average slow down – even those who finished in the bottom quarter of the field. Let’s take another look at that scatterplot, with the different age categories visible.

Remember, the average slow down is 60%. Each age category has plenty below the average and plenty above it. So whilst on average, the older runners slowed down a bit more, at an individual level, it didn’t mean much.

So, where were the different groups slowing down? Was there a particular part of the course where, say, the backmarkers started to slow down more than the leaders, or did the difference just gradually emerge? A little of both. The below chart looks at a few groups of runners sorted into finishing position. It takes their time into checkpoint 1 (10.5km) as a base time and then shows how their race progressed as a multiple of that first split.

For example, if you got to the first checkpoint in one hour and finished in 10 hours, the line would end at 10 on the y-axis. The numbers on the x-axis correspond to the distance at each timing point. If everyone slowed down at the same rate, the lines would all be identical. And the flatter the line, the less you slowed down.

The chart is a little busy, but it’s clear to me that for those at the back of the pack, they really started to slow down after 66km – the last third of the race. The section with the largest spread between the front and back of the pack was between 78-99km. This also happens to include a very long descent followed by a very long climb. And the bottom 50 (that is, the bottom 50 with timing data for all timing points – this was a bit hit and miss) slowed down a LOT in the last kilometre, which I’ve heard (I’ve never done UTA) is a bastard.

Finally, what about the overall distribution of slow down rates? There's a long tail out to the right of the average - if you're in that tail, there could be some improvement to make.

So, what about the comparatively flat and fast Surf Coast Century?

It turns out that most of the conclusions from UTA applied to its flatter Victorian counterpart. Although, in terms of sample size, SCC is much lower (~175 runners with data on all timing points, instead of ~650 for UTA), so the results are even less robust.

The average second half (and this time, it’s pretty much even in terms of halves – 49km/51km) slow down was 24%, with males (122 runners) slowing down by 25% and females (53 runners) 23%. Not much difference there. The second half of SCC is also hillier than the first, so it’s not just fatigue that led to the slow down.

Again, there was a small correlation between finishing position and pacing. But like UTA, lots of variation irrespective of finishing position. Some people in the middle to the back of the pack even went close to a negative split.

There was also the same general trend of the older runners slowing down more, but not by much. The 20-39 year olds slowed down by 23%, 40-49 by 24% and 50-59 by 28%. What’s interesting is that three out of the four runners in the 60+ category averaged 17% or less – putting them in the top quartile of the field in terms of slow down.

Similar to UTA, you see a good mix of younger and older runners finishing the second half strongly relative to the average field slow down (24%). By the way, I hope my Dandenongs Trail Runner comrade Mathieu DorĂ© doesn’t mind me pointing out that he’s the outlier at the front of the field – 13th place, but the biggest slow down in the field! (If I know Mathieu, he’ll have a good laugh about that and then (1) go out and destroy some Strava segments and (2) win a whole bunch of races.)

What about where the gaps started to open up? Here’s the same index chart as UTA. I’ve taken out the top 20 category and changed the bottom 50 to the bottom 20, because of the smaller field.

You can see the front and back of the field really start to diverge after the half way point. As I noted earlier, this is also the hillier section of the course. And in the last 15km (which is pretty much flat along the coast), the back end of the field was struggling. Maybe this was affected by injuries, or maybe the final 15km of a 100km ultra was biting people who went out a bit too fast.

And to round things out, here's the slow down distribution. Again, a longer tail to the right, but for whatever reason, it's a bimodal distribution (two peaks). This is probably just a symptom of the relatively small sample size.

Two different races, very different terrain, but similar conclusions.

So I think what I learned from this exercise is this – for individuals in the race, sex, age and speed don’t matter a great deal when it comes to pacing well in an ultra. On average, sure, it’s better to be in your 20s and fast. But you can be over 50, placed near the back of the pack and finish as strong (relative to your own ability) as the young and fleet of foot.

One final thought – I’m tipping a few of you have read this far and you’re now thinking “So, his conclusion is that older people slow down more and the elite runners know how to pace themselves. But everyone’s different. Whoa! This guy deserves a Pulitzer!” But hey – that’s the thing about data. Sometimes it surprises you and sometimes it doesn’t. But at least the next time you talk to someone about pacing in ultras, you might have some facts to back you up. And plus, I got to play with some data. Win-win.

## No comments:

## Post a Comment