Perfect Pacing at the London Marathon
How to run like an elite and achieve a personal-best at London.
- An analysis of more than 215,000 London Marathon runners during the period 2011–2016;
- How do runners pace the London Marathon? Where are the fastest and slowest sections of the race?
- How does the pacing of elites compare to recreational runners and those running a PB?
- Putting what we learn into practice we create an optimal pacing chart for recreational runners that is tailored for London.
With the London Marathon just a few weeks away, participants will be planning their pacing strategy for race-day. Many will look to pacing charts to take some the guesswork out of race-day pacing, but the one-size-fits-all approach of traditional charts limits their usefulness in practice. In this article we describe a new type of pacing chart that is optimised for a particular marathon course — in this case the London Marathon — to provide participants with fine-grained, tailored pacing advice, based on the pacing patterns of the best of the best.
The data for this study is part of a larger dataset of 1.7m race records from more than 60 city marathons, London among them. The London dataset includes the last 6 marathons (from 2011 to 2016) and includes 215,575 race records, each with 5km split-times.
The Problem with Existing Pacing Charts
Marathon pacing charts aim to take the guesswork out of pacing, by calculating various split-times for a specified target finish-time. For example, the Runner’s World Pace Charts list the times for various points in the race (5km, 10km, halfway etc.) for a wide range of paces and finish-times, from 3 mins/km (just over 2:06 finish) to 11 mins/km (a 7:44 finish).
Convenient as they might be, these charts suffer from an important shortcoming because they assume even pacing throughout the marathon. In practice, even pacing is extremely rare for a variety of reasons, not least because of athlete fatigue and course conditions. Consequently race-day finish-times typically deviate from chart-times for all but the most disciplined of runners.
More sophisticated pacing charts have been developed, which go some way to modelling more common pacing patterns. For example, pacing charts, such as those at Marathon Basics, calculate split-times based on even pacing and a 47/53 positive split; that is the runner spends 47% of their time running the first-half and 53% running the second-half of the race. While this might be a step in the right direction it remains limited by assuming a fixed split-type. Even when other split-types are offered (e.g. see Running For Fitness) the resulting charts still assume evenly paced splits within each individual half of the race.
We need more fine-grained pacing plans, which account for ability and fatigue of runners and different course conditions. In what follows we will analyse the pacing patterns of London marathoners to better understand how a typical runner completes the course. We will see how, despite the large variation in runners (gender, age, experience, ability), most runners follow a remarkably similar relative pacing profile, which provides a starting point for a more tailored pacing chart.
Pacing the London Marathon
The chart below shows the pacing (mins/km) for men and women during each of the 5k segments of the London Marathon, including the final 2.2k segment. As expected men are faster than women and, for both men and women, pacing tends to slow as the race progresses. The start of the race (the first 5k segment) is the fastest (5.47 mins/km for men and 6.17 mins/km for women, on average) and the penultimate segment (35k to 40k) is the slowest (6.92 mins/km for men and 7.92 mins/km for women); both men and women manage to speed up during the final 2.2k section of the race as they ‘sprint’ to the finish. Thus, the average London participant starts outs fast but then gradually slows, completing a positive split with the second half of the race approximately 15% slower than then first.
A word of caution here, this average pacing profile includes people who hit the dreaded wall in the second half of the race. This means that some of the slow down shown will be exagerated, at least somewhat, by these poor souls.
Rather than looking at the actual pacing values in mins/km we can look at the degree to which a runner runs each segment slower or faster than their own average race-pace. This is shown in the next graph below where the y-axis stands for this relative or percentage pace; we refer to this as a pacing profile. Now, a percentage pace of 90% means that the runner is running 10% faster than their average race-pace, whereas a percentage pace of 110% means that they are running 10% slower, and a percentage pace of 100% means they are running at their average race pace.
We can see how, on average, men and women both tend to start their race about 10% faster (90% percentage pace) than their average race-pace. Then, as the race unfolds, the percentage pace gradually increases, indicating a gradual slowing-down. After this fast start women slow more quickly than men. For example, women reach their average race-pace (100%) just after then 20k segment, but it takes men until after the 25k segment before their pace slows to this level. As a consequence of this, men tend to slow more than women in the second half of the race, reaching almost 113% (13% slower than average race-pace) by the 40k segment. Women, on the other hand, only slow to 109% (9% slower than average race-pace), during the 35k segment. Women also speedup more than men in the final sprint to the finish, recovering to within 2% of their average race-pace during the final 2.2k segment; men run this final segment 7% slower than their average race-pace.
The lesson here seems to be that, on average, runners start fast, probably too fast, but women tend to adjust their pace to more effectively than men to enjoy a faster finish, relatively speaking.
Perfect Pacing for London
Thus, there is a fairly consistent pattern when it comes to how recreational runners pace London. Relative pace varies predictably across a broad field of runners, as it is influenced by growing fatigue, the undulations of the course, the noise of the crowd, the excitement of the start etc. It is a consistent pattern but, presumably, it is not a very optimal one. For example, it includes the pacing deterioration experienced by the many runners who hit the wall later in the race. Certainly the range of pace variation is significant, from 10% faster than average race-pace to 13% slower (men), for a total range of 23%.
This begs the question: what is the optimal pacing profile for London? To answer this we need to isolate the pacing profiles of a subset of the best runners and to do this we will consider two groups of runners: those running a personal best (PB) and elites.
Pacing for a Personal Best
Our dataset includes runners who have run more than one marathon and for these runners we can identify PB times from their set of times; essentially we treat their fastest time among their set of races as their PB. For the purpose of this study we will consider those runners who have run at least 3 marathons. There are 33,440 such runners out of our total set of 215,575 London participants. Among these 33,440 repeat runners, 8,908 have run a personal best at one of the London races in our dataset; that is, these 8,908 runners all have their quickest times in London. Of course this might not be their true personal best, since it depends entirely on the window of data that we have access too, but it at least counts as a good time for these runners.
Next, we can compute and compare the pacing profiles for these PB runners to the average recreational runners as shown above. When we do, we see a more even pacing profile, ranging from 95% — 108%. These PB runners still start out faster than their average race-pace, but not as fast as the average runner, and they finish slower than their average race-pace, but not as slow as the average runner. Men continue to run the first half of the race relatively faster than women, but the difference is less than was found for the average recreational runner. And men and women finish faster when they PB.
The Optimal Pacing of Elites
What about elite runners? How does their pacing profile compare? Our London dataset contains 1,453 elites runners, defined here, somewhat loosely, as men and women finishing within 150 and 185 mins, respectively. The graph above also shows the pacing profile for these runners. This time we see a more even pacing profile again, with elites operating within +/- 4% of their average race-pace. As with recreational and PB runners, elites start out faster than their average race-pace, but only 4% faster, and they finish slower than their average race-pace, but only 4% slower. Throughout the race elites run closer to their average race-pace than either recreational or PB runners. It is also interesting to notice how there is virtually no difference between the pacing profiles of elite men and elite women, despite their differing speeds. This perhaps indicates that both male and female elites have converged onan optimal pacing strategy for London.
An Optimal Pacing Chart for London
As we have compared the pacing profiles of recreational, PB, and elite runners we have found key similarities and differences. In all three cases fast starts are followed by slower finishes but the pacing profiles of more able runners (those achieving a PB and elites) become more even with less difference due to gender. If we adopt the view that the pacing profile of elites is near optimal, for London in this case, then we can use it as the basis for a pacing chart that is tailored for perfectly pacing the London marathon. All we need to do is map recreational running times to this profile.
To do this for a given target finish-time of say 4 hours, we first calculate the average race-pace, in this case about 5 minutes and 41 seconds per km. Next we map the elite pacing profile to this target time and pace. Elites run the first 5k about 4% faster than their average pace and so we adjust this 4-hour pacing to recommend a start pace of 5 minutes and 28 seconds (4% faster than the 5 minute 41 second pacing). Doing this for each of the race segments gives a tailored pacing plan for a 240-minute finish that matches the relative pacing of elite runners. It is not an even pacing plan because even elites typically don’t run in this way. They adapt and adjust their pacing to their course conditions, for example.
The result, for target times ranging from 180 – 420 mins (3 – 7 hours), is presented in the chart above. For each target time we show the corresponding split-times for each of the 5km race segments based on the pacing profile of elites at London. Obviously these split-times are progressively slower than the elite times but the point is that they follow the same pacing profile. For completeness, the next table shows the per-kilometre pacing information of each of the race segments.
Of course the notion that elites pace perfectly in every race is probably too strong an assumption to make. In any given race there will be many factors that influence their pacing: course conditions, the competition, how they feel on the day etc. However, on average, when pacing is analysed over many runners and a number of years, a strong pattern emerges and it is this pattern that might be important when it coms to informing the pacing of other runners; if not optimal then it’s certainly not a poor pacing profile.
Moreover, it is a pattern that is closely connected to the course in question. The elite pacing profile for Boston is different from London, because the Boston mnarathon is different from the London marathon. And so we can reasonably conclude that the elite pacing profile for a given race likely corresponds to a highly optimised profile that is tailored to a particular course. Then the pacing tables produced from this pattern may allow recreational runners to replicate this ‘perfect’ pacing profile in their own races and for their running ability and target times. Doing so may help to produce better performance on the day.
If these pacing charts help you during your race please let me know!