An even deeper dive into the Cutoff Time for the 2025 Boston Marathon and the behavior of marathoners
Since last December, I have posted to this blog with my predictions for the cutoff time for the 2025 Boston Marathon. My most recent post in June introduced the concept of “Utilization Rate” for a Boston Marathon qualifier race.
Utilization rate (UR) is the percentage of runners achieving the Boston Marathon qualifying standard in a qualifying race who then use that result to apply for entry into the Boston Marathon. Many runners who BQ do not end up using that result to apply for Boston. The desire to use the BQ for Boston varies widely among the various qualifying marathons.
Some very large influential marathons like London have a surprisingly low UR. Others, like those who market themselves as BQ-generators (read: the downhill marathons), have a high UR.
In that June post, I also introduced the BIB Index (“Bring It Boston!” Index) wherein I look at the data from 50 top qualifiers (the BIB50) to predict the Boston Marathon cutoff time. By identifying individual runners in the 2024 Boston Marathon entry list and the qualifying race they used on their Boston application, I am able to assign UR’s to all the BIB50 qualifiers.
Those 50 races represent 62% of all entries to the 2024 Boston Marathon. There are more than 470 additional qualifiers that were used for entry. In aggregate they account for 38% of the entries yet none of them on their own are very influential. Hence, I am comfortable that trends identified in the BIB50 represent the qualifier world as a whole.
Prodded by reader comments and a restless mind, I felt the urge to reconcile the BIB with other trends being discussed. Specifically, there has been a lot of anxiety about the following:
- The total number of BQ’s for all qualifier races has increased year over year.
- Finish times for many marathons are trending faster.
Either of these trends would push the 2025 cutoff time higher. All things being equal, together they represent a serious risk to those who aspire to run Boston next April.
Although these trends are real, and they will influence the 2025 cutoff time, they don’t tell the whole story. The main point of my June post is that not all BQ’s are created equal. What really matters is the finish times of the likely applicants to Boston. The overall trends in BQ numbers and finish times have to be sampled judiciously in order to determine how they ultimately affect the cutoff time.
I have been obsessing about this point all summer.
With the data I have collected, I should be able to arrive at an algorithm that takes into account UR to re-construct the scenario that forced the B.A.A. to establish last year’s 5:29 cutoff time.
An algorithm that matches the 2024 cutoff time to within a few seconds would be a compelling sanity check. I could then apply that same algorithm to the qualifier data of the current qualifying period to predict the 2025 cutoff time with high confidence.
That goal was elusive. That is until today, the eve of the double-dip qualifier weekend.
I have always maintained that in addition to identifying the number of applicants to Boston, a model must be able to describe the mathematical distribution of the cutoff margin times. In the case of re-creating the 2024 cutoff time, the model must be consistent with the distribution data published by the B.A.A.:
During the registration period, the breakdown of accepted Qualifiers was as follows:
- 6,182 Qualifiers met their qualifying time by 20 minutes, 00 seconds or more.
- 8,858 Qualifiers met their qualifying time by between 10 minutes, 00 seconds and 19 minutes, 59 seconds.
- 6,979 Qualifiers met their qualifying time by between 05 minutes, 29 seconds and 9 minutes, 59 seconds.
- 11,039 qualifier applicants were unable to be accepted into the 128th Boston Marathon.
My original idea was to use the UR that I calculated in June to sample the full set of unique BQ’s for each of the BIB50. If, for example, a qualifying race had a BQ of 50%, I would randomly sample 50% of the full set of cutoff margin times for that race. The sample then represents the finish times that would be submitted to the B.A.A. for acceptance to the Boston Marathon. Do this operation across all the BIB50 races to create an aggregated sample set of all applications to the Boston Marathon.
One sample would not be adequate. I would have to perform a Monte Carlo simulation wherein many such random sample sets were taken and the results aggregated to create the cutoff time estimate.
That approach failed. It overloaded the cutoff margin segments longer than 10 minutes thus driving up the predicted cutoff time.
The breakthrough observation was that the UR rate is not a constant across all finish times of a given qualifying race. Rather, it turns out, runners with longer cutoff times are less likely to apply to Boston than runners with shorter cutoff times.
The plots below demonstrate this effect. In each figure I plot the histogram of all of the BQ’s for a specific qualifying race (All BQ) on the same axes as I plot the histogram of cutoff times actually accepted into 2024 Boston (Accepted Entries).
The relative shape of the orange histograms compared to the blue histograms describe the phenomenon.
The table below summarizes the data for some of the larger qualifiers. The 2023 Boston Marathon represents an exception in that the UR is fairly constant across all cutoff margin segments. All the others show a distinct trend with higher utilization rates in the 5:29 to 9:29 range compared to the longer ranges.
My conclusion is that runners with larger cutoff margins are less likely to apply to the Boston Marathon. A higher percentage of applicants come from the ranks of runners who have smaller cutoff margins. These runners are the ones most interested in getting into Boston.
It is a window into the behavior of marathoners. Those runners who are ever so close to getting into Boston are the ones most likely to apply. Many runners who can get into Boston with ease don’t apply, for whatever reason.
This may not sit well with many of you. But it is a strong trend that holds up for all of the BIB50. And face it, if you are reading this, you are likely to be in that higher utilization rate group. That perspective causes bias. Perhaps you see all these faster finish time trends and you fear getting crowded out. That may happen, but the reasons for it are nuanced.
To be clear, I have already eliminated elites from my dataset and have filtered out the slower finish times for runners who have earned multiple BQ’s from more than one qualifier.
I apportion 62% of the available entry spots to BIB50 races and the remainder to the non-BIB50 races. I’ve confirmed that the BIB50 races uniformly represent ~62% of the applications across all BQ margin segments. Hence, those non-BIB qualifiers do not skew the BQ margin distribution. In the Monte Carlo simulation the cutoff time solutions converge rapidly.
My sanity check on the 2024 Boston Marathon dataset arrives at a cutoff time of 325 seconds (5:25), just 4 seconds off from the actual cutoff time.
Using that information as validation, I calculated the expected number of applications for the 2025 Boston Marathon. Despite the concerns raised above, I see a modest 4% increase in the number of applications compared to last year. Specifically, I anticipate 34,402 applicants for the 2025 race compared to 33,058 last year.
I then broke up the full set of BQ data each BIB50 qualifier into segments and randomly sampled them using the appropriate UR’s of each segment. Again, a Monte Carlo simulation was employed to aggregate a number of random samples to create the cutoff estimate.
Using the same number of accepted applicants as the B.A.A. allowed in 2024 for 2025 (22,019), I get a predicted cutoff time for 2025 of 371 seconds (6:11).
This raises an interesting point. The number of accepted applicants in 2024 was historically low. The B.A.A. could choose to raise that number closer to the historical range of 23,000 to 24,000. By my model, increasing the number of accepted applicants to just 23,000 would drop the cutoff time by 32 seconds to 339 seconds (5:39).
I suspect that is partly why the B.A.A. has resisted dropping the qualification standards by 5 minutes as so many of us have anticipated. They have this other tool at their disposal; the eventual standard change can wait.
I ran a few other combinations of the number of applications and the number of accepted applicants. My current best estimate for the 2025 Boston Marathon cutoff time is in the range of 330 to 375 seconds (5:30 to 6:15). Two BIB50 races have yet to occur (Erie Marathon at Presque Isle and The Tunnel Lite Marathon) but they are small races and I don’t believe they will move the needle much.
As always, comments are welcome. However, with the application period opening up next week we are in the “time’s up, game over” phase and we will have the real answer shortly.
