A couple of weeks ago, a new study was released that analyzed pacing data from the 2015, 2016, and 2017 Boston Marathons. The title, “Pacing strategy patterns and performance outcomes in marathon runners,” is interesting. But when you dig into the methodology, I’m not sure they actually measured and analyzed what they expected to examine.
This study suffered from some of the same defects as the recent Berlin Marathon study on splits. Chief among them is the idea that pacing outcomes – as observed by the positive or negative change in pacing from the first half to the second half of the race – is synonymous with strategy. But we don’t know that someone who ran positive splits planned to run positive splits. To the contrary, I’m sure many people have had the experience of executing the first half a race thinking everything was going swimmingly … just to have it fall apart somewhere in the later stages of the race.
I included some thoughts on that study in my newsletter a few weeks ago, but that’s not really the focus for today. Instead, this made me think back to the analysis I did last year of the splits at the 2025 Boston Marathon. And I wanted to take another look at the data from this year’s race to see how pacing strategy relates to outcomes.
From a high level, the question we’ll be exploring is – how does a runner’s pace in the downhill segment of the Boston Marathon relate to their final performance? In other words, how aggressively or conservatively should you take that first part of the race?
Let’s dig in.
Overview of Methodology
In a perfect world, we could interview or survey every runner to understand their strategy and goals. Then, it would be much easier to compare different strategies to different outcomes and determine which are the most effective.
Unfortunately, we don’t have that data. But the Boston Marathon’s course is unique and it’s elevation profile makes it possible to make some inferences about a runner’s strategy from how their pace varies in the first half of the race.
For the most part, the first 10k of the race is downhill. It’s steeper in the beginning, and then it there’s a bump and a more gradual decline. The next 5k, however, is relatively flat. By the time you get to the 15k split, you should have settled into your race pace. You’re also not deep enough into the race that you would be suffering the negative effects of fatigue.
For our purposes, we’re going to assume that the pace a runner runs in the third 5k of the race is their benchmark pace. It’s a proxy for their goal pace. Unless something goes wrong, they would expect to run this pace – or close to it – for the rest of the race.
Since the previous two 5k segments are downhill, runners tend to take them faster. There’s a risk and reward here. Cruise downhill, and you can potentially run faster. But get too aggressive, and you may burn out on the later hills.
So the question is – how does the aggressiveness of those early miles relate to a runner’s final outcome?
You need to operationalize both of those variables – the aggressiveness and the outcome. And we’re going to use that split at the 15k mark to do so.
If you assume that third 5k of the race is a runner’s goal pace, you can measure how much faster or slower they were during the first 5k, the second 5k, and across the entire race.
The first two variables are measures of aggressiveness – where faster paces indicate more aggressiveness. The final variable is a measure of success. The closer a runner’s overall pace is to their benchmark pace, the more successful they were in the race.
Examination of the Data
Next, let’s explore the data from the 2026 Boston Marathon to get a sense of how runners are distributed across these strategies and outcomes.
This first chart shows the distribution of how aggressive runners are. The left graph shows how their 5k split compares with their 15k benchmark split. The right graph shows how their 10k split compares with that benchmark split. In both cases, negative numbers and blue bars indicate that they were faster in this segment than they were in the benchmark segment.
The numbers on the x-axis represent the difference in paces as a percent – i.e. 1% faster or slower. This helps normalize things across runners of different paces.
The first 5k is relatively evenly distributed. The two sides of the graph are similar, with slightly more runners going a little slower in that first 5k. This may be by choice (a deliberate conservative strategy). It also may be a result of congestion in the beginning of the race, with runners hemmed in and unable to be aggressive. In either case, it’s a measure of how aggressive their actual pace was compared to later in the race.
In the second 5k, the graph becomes a little less evenly distributed. Runners are clustered more tightly towards the center of the graph, and more runners are on the aggressive side of things. Just under 14,000 runners (of 29,000 finishers) run 0 to 3% faster than their benchmark time. The other side of the coin, where runners are 0 to 3% slower, represents between 10,000 and 11,000 runners.
The next graph disaggregates this data into four groups by gender (men vs women) and age (under 40 vs over 40). The drop down filter toggles between the data for the 5k split and the 10k split.
When you look at the 5k split data, the four graphs look similar. Runners under 40 are slightly more likely to be on the aggressive side. This is true of both men and women. But the difference is small.
Flip to the 10k split, and the data looks different. For both young men and young women, the aggressive side of things (0 to 2% faster) is much taller than the conservative side. For both men and women above 40, the two sides are still relatively evenly balanced.
In all cases, there’s also a greater clustering towards the center. When you put it all together, it’s likely that the first 5k offers less room for runners to maneuver, and their paces are more randomly distributed based on crowd conditions. But in the second 5k, runners have a little more room to maneuver – with everyone settling into their pace and younger runners tending to be more aggressive.
The next graph includes the distribution of runners’ overall pace vs their benchmark pace. Again, blue bars and negative numbers mean that their overall pace was faster than their 15k split. In other words, they sped up later in the race.
The vast majority of runners were slower, overall, than their benchmark pace. This makes sense, given the hilly nature of second half of the course. Some people (like John Korir, who won the race) can power through the hills and speed up. Most people satisfied just to survive and slow down as little as possible.
When we evaluate the overall outcome, we can also expect most runners to be slower overall than their benchmark. So the question of a successful outcome is how close they are to zero.
This last graph shows the overall pace data broken out by the same four demographic groups. Younger runners appear to more likely to have a quicker overall pace (although it’s still a minority of them). Older runners are more likely to slow by several percent (or more).
There doesn’t appear to be much difference here between men and women.
How Do Pacing Decisions Relate to Outcomes?
Now we can take our key variables – how aggressively runners were in the first two segments of the race – and see how the outcome varies.
As a baseline, the baseline outcome for all runners is a median of 2.9%. Their overall pace across the entire race was 2.9% slower than their pace during the 10k to 15k benchmark segment. As an example, if they running 6:30/mi on the benchmark segment their overall pace would be about 6:41/mi.
This first graph divides runners into bins based on how their 5k split pace compared to their benchmark (15k) split pace. This is the x-axis.
The y-axis represents how much slower their overall pace was compared to that benchmark. A lower number is good (a more consistent race effort) while a higher number is bad (greater slowing). The number on the graph is the median value for the entire group.
The four lines break the runners into their demographic categories by age and gender.
There’s a difference here based on age, but across all four groups there’s a consistent pattern. The best results come from runners who maintain a relatively even pace in the first 5k. The optimal range is from -1% to 2%. The number (1) on the graph represents the bin of runners who ran 1.00% to 1.99% slower than their benchmark time.
Runners who were particularly aggressive – more than 2% faster than their benchmark time – performed more poorly overall. And this impact was greater for runners at the most aggressive end. But runners who started out more conservatively also suffered. Perhaps in these cases they weren’t able to make up as much time on the back end.
The next graph here shows the same data based on the runners’ pace at the 10k split. There’s still a (more mild) difference between younger and older runners. And there’s still a similar pattern across all four groups.
For all four groups, the optimal outcome occurs with relatively even pacing – either slightly aggressive (~1% faster) or slightly conservative (~1-2% slower). Runners at each of the extremes saw increasingly poorer outcomes.
How Does This Relate to Finish Time?
In the previous graphs, I used the percent of slowing as the outcome or independent variable because it helps normalize things across age, gender, and performance levels. But we can examine the same data based on the actual finish times, as well.
This first graph shows the data for the four groups, split out by how aggressive or conservative their 5k split was. The y-axis in this case is the median finish time for that group.
For women over 40 and men under 40, the best times come from running 1.00 to 1.99% slower in the first 5k. For men over 40 and women under 40, the fastest group was between 0.00 and 0.99% slower. In other words, just slightly conservative.
On the more aggressive side of things, runners are slightly slower at the 2-3% range. But the biggest jump in finish times comes from runners who were 4% faster in the initial segment.
The difference here is so drastic that it could also be a distinction between qualifiers and non-qualifiers – with qualifiers more likely to be more consistent and non-qualifiers more likely to be overly aggressive. This is a flaw in using just the finish time – it’s hard to know if that 3:30 for a man under 40 is horrible (if he qualified with a 2:50) or great (because he was a charity runner hoping for 3:30).
Here’s the same graph with the data at the 10k split. Here, the fastest runners were between -1.00% and 1.99%. More aggressive runners (3% or faster) and more conservative runners (2% or slower) were more likely to have slower finish times.
What’s the Bottom Line on Pacing?
This analysis took a look at one very specific and particular question when it comes to pacing a marathon, and the Boston Marathon in particular. How does aggressive or conservative pacing early on impact a runner’s overall pace.
Without being able to ask runners about their actual pacing decisions, we used one specific split (10k to 15k) as a benchmark. Then, we analyzed outcomes based on how early splits compared to that benchmark.
In all cases, the optimal outcome – the smallest increase in overall pace – resulted in a consistent pacing strategy. Running slightly faster than the benchmark was fine, but runners who were overly aggressive suffered. On the flip side, running slightly slower was also fine. But runners who were overly conservative also suffered.
When you compare these to actual finish times, you also find that runners who finish faster tend to run more evenly in the early miles. This isn’t necessarily a causal relationship – because it could also just be that better runners are more consistent. So it doesn’t necessarily mean that you will have a good race because you adopt this strategy. But it gives you some insight into how the faster runners approach the race.
One final tidbit: a group of about 5,000 runners started out slow in the first 5k (maybe because they were hemmed in) and then sped up in the next 5k. This group performed slightly worse (across age and gender) than runners who were consistently on the conservative side. And in every case, the runners who were consistently on the faster side also performed slightly worse.
In other words: you’re probably better off erring on the side of caution – and taking that first segment slow – than either starting off aggressive or hitting the gas once the course starts to clear. As long as you’re not overly conservative, you’ll probably fare better in the later part of the race and suffer less from fatigue.