Saturday 7 November 2015

The problem with using race predictors for choosing marathon race pace.

Is there a problem with using race predictors for marathon pace setting?

Often marathon runners use race predictors to help set their marathon pace - but, there is a real problem with that approach. I thought it might be interesting to give a few real world examples from people at opposite ends of the performance spectrum to illustrate the problem (and one in the middle). This is not a full scientific analysis - and you could accuse me of cherry picking my data. But, this is real data from runners who I know were making serious attempts to produce their best performance. These are runners I respect and have given their all in pursuit of a good marathon time. Of course there are plenty of examples of where race pace predictors work reasonably well - and I will try and do a statistical analysis of just how well they work for different types of runners at a later date. But, for now I want to show that care needs to be taken when extrapolating from shorter distance race performance to longer distances. What limits performance at 10km is not necessarily what limits performance in a marathon. This is not just a minor problem.



My first example comes from Michael Salt. Let me first tell you about him. He really is an awesome runner - he is way above my ability. Here is a link to his Powerof10 profile http://www.thepowerof10.info/athletes/profile.aspx?athleteid=45117 At 44 years old he is a very experienced runner. Last year he did a 16:31 5K (84.5% age-grading) and in 2011 he did a 1:11:37 half marathon (86.8% age-grading). He is a seriously tough runner and knows how to push through the pain barrier. In April this year he ran the Virgin Money London Marathon (VMLM) for the second time, it was his third marathon. Just over a month before he had raced two half marathons, the one listed on thepowerof10 was in 1:15:48 (83.35%) which predicts a 2:37:26 marathon (flat pace 3:43 per km). He finished the VMLM in 3:16:09 - nearly 40 minutes slower than the predictor. Clearly the most obvious explanation for this dramatic failure of the predictor is that he got his pacing wrong. But, his splits don't support this. He went out on the first 5km at 3:41 per km - only two seconds too fast for a flat pace strategy. This looks like pretty good pacing. He then slowed to 3:43 per km for the next 5km - perfect pacing apparently. However, his pace continued to slide from 10km to halfway he did 3:50 mins per km which then slowed further to 4:23 per km by 30km before ending the last 10km at 6:40 per km. The plan of racing at exactly the scaled age-graded performance ended miserably for an experienced, tough runner on his third marathon. Clearly what limits Mike's performance at a half marathon is not the same as what limits him at the marathon. The predictors did not work for him.

My second example is Calvin Sambrook (age 52). He is my brother-in-law. He does not have a powerof10 profile (he is not a member of an athletics club, nor is he particularly fast) but, here are links to his Strava and parkrun profiles. In September this year he ran a PB 19:57 parkrun (age-grading 75.1%) which predicts a 3:09 marathon. He ran the Frankfurt Marathon just over a month later, I don't know for sure, but it must have been about his 10th marathon - may be more. He finished in 3:34:54 which is about 25 mins slower than the prediction. Again, the most obvious explanation is that he went out too fast. But, his first three 5km splits were all within 2 seconds of 4:40 per km (which is flat pace for a 3:17 marathon time - he wanted Good For Age for VMLM) - a very conservative pace for someone who the race predictors said should have been in 3:09 shape. Calvin held close to that pace up until 30 km before slowing. He did not give-up, my nephew was with him right up until he was taken away on a stretcher at the finish. He was desperate for a good performance and he did everything he could during the race to achieve it. But, the race predictors let him down too.

For the final example I will use myself. I think I must have done over 25 marathons - most, but not all, are listed on my powerof10 profile. For those who don't know, I am not an athlete with much history. I took up running in 2009 and did my first marathon in 2011 with a PB of 3:29:33 (link here). Six months earllier I had set my parkrun PB of 19:37 (71.7% age-grading) which predicted a marathon time of 3:04:31 - 25 mins faster than I achieved. But, I had set-off at roughly 4:50 per km pace (the splits are only available on the Strava file) which should have seen me finish in about 3:24. I did a slight positive split, but my failure to achieve my age-grading was not that I went too fast at any point. But, what I want to illustrate is how age-grading predictions can fail in the opposite direction - it is entirely possible (but I admit unlikely) that you could run a faster marathon than the predictors would suggest. This year I decided to try and run a 2:45 marathon and in Frankfurt I got pretty close with 2:45:10 (age-grading 83.6%). It was a 6 min faster than my London PB from this year, which was also 6 min faster than my PB from November the year before. My previous highest age-grading was from May this year - a half-marathon where I ran a 1:21:59 (age-grading 80.3%) which predicted that I should have been able to do a 2:50:41 marathon. I did 5 mins better than that. Of course there are reasons for this, some of which I know. But, the point here is that simply using an age-grading or race performance predictor may well produce an unreliable answer. If it predicts a slower time, like for me, it is no big deal - I would have just thrown away a slightly better performance than I could have achieved as a slow start in a marathon allows for a faster finishing pace. But, if the prediction is too fast the results are terrible - there is almost no limit to how slowly you may have to 'run' in the final stages to get to the finish. It is critical to understand why and when the predictors fail and to develop strategies (like a training log) that will enable you to produce a better guess of what a sustainable pace might be. If you want to read more about the predictors, Alexander White wrote a publically available degree project on it, which although rather old, gives an overview of some of the models. Hillrunner has a lot to say on the topic, Fellrnr attempts to take account of other factors (but, I think he makes some mistakes on pacing strategy). There is much discussion on the Net as to what sort of pacing strategy is best - negative or positive split, flat, U or inverted U. My view is that for most runners it is an unnecessary debate - near flat (within a few seconds per km on flat terrain) is best. The important question is; "What is the fastest time I could possibly achieve?".

So, I need to sign-off - and I am aware that none of this has helped you to work out what pace to choose, but if you want a clue as to how I do it then look at some of my past posts.

No comments:

Post a Comment