All the data from this page was collected from this database of finishing statistics for the Toronto Waterfront Marathon. I omitted all participants who did not finish or were disqualified.
Finishing Time
The histogram below shows the finishing times of participants of the Toronto Waterfront Marathon from 2002 to 2016. You can filter the results by age, year, and gender.
Notice the big bumps below 3:00, 3:30, and 4:00 (especially in 2014) -- I suppose these are due to runners meeting their goal times. I ran the 2015 and 2016 marathons and felt that 2016 was the harder year. I feel vindicated by the data, which shows that 2016 was the slowest race (by median finish time) of the last fifteen years. Let's hope 2017 goes better!
Splits
The violin plots below compare the runners' pace during a segment of the race to their average pace over the entire race. For example, the plot labelled '0-10km' measures:
(average pace over the first 10k (sec/km)) minus (average pace over the entire race (sec/km))
The big bump around -8 means that many people ran their first 10k roughly 8 seconds/km faster than their overall race pace. You can filter these results by age, gender, and finishing time. You can also change which segment you're comparing against by clicking on the title of the plot.
It's fun to see how the faster runners have more even splits -- but even they still tend to start out fast and then slow down, just like the midpack runners. Also, look at how well the 21.1-30km pace predicts your overall race pace. So next time you're 30km into a marathon, maybe the best way to predict your final time is to multiply your pace from the past 10k (in min/km) by 42.2/10. At least this gives you some mental math to keep your mind off the pain of the last 12k of the race.
Usually, amateur runners like me are told to have as even splits as possible. Since fast runners have pretty even splits, this seems like reasonable advice. However, maybe a better strategy is to try and pace yourself using the same pacing strategy that the pros use -- almost even, but a bit slower towards the end of the race as towards the beginning. For example, suppose your goal is to run a 4-hour marathon, but you want to pace yourself so that your proportion of the time spent in each segment of the race is the same as the proportion of time that a super-fast person (sub 2:50) spends on each segment. The calculator below tells you what splits to follow if you want to pace yourself ``like the subelites'' in this way.
5k: 0
10k: 0
15k: 0
20k: 0
21.1k: 0
25k: 0
30k: 0
35k: 0
40k: 0
A few warnings about this visualization:
The 40-42.2km segment is much shorter than the other segments. Because of this, for purely statistical reasons we'd expect there to be more variability in runners' paces over 40-42.2km than in the other segments. This is true even if we ignore all the other psychological and physiological reasons that people go abnormally fast or slow at the end of a marathon. So don't overanalyze the 40-42.2km segment. It's just there for fun!
If you're running a 4:00 min/km pace, then a 20 sec/km deviation from this pace represents a dramatic difference in speed; if you're running a 7:00 min/km pace, then a 20 sec/km deviation doesn't change your speed much. For this reason, the raw pace difference is a flawed measure of how ``even'' splits are. If I spend more time on this visualization, I might include the option to measure the pace deviations as percentages of average pace, rather than absolute pace differences.
There are problems with the 2012 data -- many runners are recorded as finishing the race before they hit the 40km checkpoint. I guess the 40km checkpoint was just buggy that year, so the plot above does not use the 2012 data in its calculation of any statistic that uses the 35-40km or 40km-42.2km paces.