Today’s the day we’ve all been waiting for. The Boston Athletic Association announced the cutoff time for the 2026 Boston Marathon: 4:34.
If you made it in, congrats! The outcome was slightly lower than I predicted, so I’m sure there are some happy and relieved runners out there who made it through by a hair.
While you’ve all been waiting with bated breath to find out if you made it in to Boston, I’ve been anxious for another reason. I’ve spent the better part of the year very publicly tracking this data and projecting the outcome – and every once in a while, there’s that little voice in the back of my mind that rears its ugly head. “What if you’ve been wrong all along?”
So I was excited for this morning’s announcement. It’s a mixed bag. This is validation that my methods are fairly accurate and the general thrust of my projection was true. But still, my final prediction was a little higher than the outcome. So I wanted to take a quick look at the data that BAA released today to see if there are any clues as to what may have gone wrong.
So let’s dig in.
How Accurate Was the Prediction?
We’ll start with the basic question: how accurate was my prediction?
Last week, after the number of applicants was released, I adjusted my final prediction to be 5:16 +/- 20 seconds. This prediction was based on data from the last decade and the fairly stable linear relationship between the number of rejected applicants and the cutoff time.
So on its face that final prediction overshot things modestly. The lower bounds of that range would have been 4:56 – which is still 22 seconds above the actual cutoff time (4:34). I’d call this outcome close, but definitely not perfect.
If you rewind the clock to before the number of applicants was released, my final prediction was between 5:30 and 6:00. That prediction was based on a projection of ~34,000 applicants, which was slightly high. And compared to the final outcome, it was off by about a minute. Less close, but definitely in the ballpark.
On the one hand, this highlights the difficulty of this process. A certain amount of data is publicly available, but much of it depends on the private decisions and unknown factors influencing tens of thousands of runners. It’s very difficult to nail down with precision what the final outcome is going to be.
On the other hand, if you zoom out across the last year, the message all along has been that the cutoff will likely land somewhere around 5:00. And from that 10,000 foot view, this was 100% accurate. The fundamentals of what’s changing – more runners and more qualifiers – is indisputable.
Projecting the ballpark is a lot easier (and more accurate) than trying to project the specific outcome.
Did the Distribution of Buffers Change?
Now let’s dig into the data a little a bit. One of the key things that BAA releases along with the cutoff time is the number of applicants that fall into certain buckets. Usually, they group these runners by 20+ minute buffers, 10-20 minute buffers, accepted applicants under 10 minute buffers, and rejected applicants.
For the visual below, I combined the two bottom buckets into 0-10 minutes. I’ve included the last three years to see how things have changed.
Among runners with the deepest buffers (20+ minutes), the number of applicants decreased this year. This was an expected outcome, given the fact that the new qualifying times reduced the number of runners who beat their qualifying times by this much.
The difference from 2024-25 is 8.5% – similar to the overall 10.1% increase in applicants. But the difference from 2025-26 is a decrease of 22.6% – much larger than the overall 8.6% drop in applicants.
The next group, runners with 10-20 minute buffers, had a similar amount of applicants this year compared to 2024. Last year was a huge outlier, with a 26% increase over 2024. That group accounted for most of the growth in applicants that year. And this year, it saw a 20.0% decline – again much more than the overall decline.
The final group, runners with 0-10 minute buffers, grew slightly each year. This year was slightly higher than 2024 (4.8%). But what’s strange here is why 2025 was so low. Despite a 10% bump in the number of applicants that year, the number of runners in this bucket remained virtually the same.
There’s something happening here that’s worth digging into more deeply later. But that sharp decline in applicants with 10-20 minute buffers is one explanation for why the projection overshot things.
And if you think that what happened here is the new qualifying times pushed a lot of those 10-20 minute runners up into the 0-10 minute range: they didn’t.
The number of qualifiers with 0-10 minute buffers increased slightly (1.7%) which is slightly less than the increase in the number of applicants in that range (3.8%). For runners with 10-20 minute buffers, the number of qualifiers only declined 6.4% – which is far less than the 20.0% decline in applicants.
Did the Field Size Make a Difference?
One of the other key variables in the prediction game is how many runners BAA actually accepts. Over the years, it’s been between 22,000 and 24,000. Last year was a relative high watermark for both time qualifiers and all runners.
My final prediction was based on an anticipated field size of 24,000. If you moved that assumption up to 26,000, the projected cutoff time would have been slightly under 4:00. And if you moved that assumption down to 22,000, the projection would have been almost 6:30.
How many runners did they actually accept? 24,362.
For context, that is the largest number of accepted time qualifiers in the last ten years. In 2020, they accepted 24,127 and last year they accepted 24,069.
Did this change move the needle on the final cutoff time? Slightly.
My final prediction (5:16) was calculated based off a linear regression of the percentage of applicants that were rejected in each year with the actual cutoff time. The higher the rejection rate, the higher the cutoff. Increasing the number of accepted applicants will have the opposite effect.
By increasing the number of accepted applicants to 24,362, this reduces the number of rejected applicants to 8,887 – 26.7% of the total applicants. If you plug that into the regression equation, you get a new projection: 5:02.
In other words, that change shifted the cutoff time by about 10-15 seconds. The projection is still high, but if you assume a possible variance of 20-30 seconds, the final cutoff of 4:34 is right around the lower bound of what’s expected.
Here’s an updated version of the regression line with this year’s cutoff incorporated into the calculation. The blue dot along the dotted line is just below the blue line – which predicts the cutoff time in seconds based on the percentage of applicants who were rejected.
After recalculating the regression with this new piece of data, this year would have had a predicted outcome of 4:58. How much does this regression vary from the actual outcome?
- This year, 2025, 2021, 2019, and 2017 all have predicted outcomes within 20-30 seconds of the actual outcome.
- 2020, 2018, and 2016 have predicted outcomes within 10-20 seconds of the actual outcome.
- 2015 had a variance of less than 2 seconds.
- 2024 is the only major outlier, with a variance of 51 seconds.
Again, once you know the number of applicants it’s fairly simple to come within 30 seconds of the actual cutoff time. This year’s data fits well in the model, and it offers a slightly refined regression to use next year.
Re-Evaluating Previous Hypotheses
When BAA announced the number of applicants, I offered two hypothesis for why the number was lower than projected. Now that we have a little more evidence, we can re-evaluate those hypotheses.
Previously, I hypothesized that there might be fewer than expected applicants in the 0-5 minute buffer range. This was based on behavior from last year’s qualifying period and it would have pushed the cutoff time up. This turned out not to be the case. A larger percentage of those runners applied than they did last year.
There’s been some heated conjecture around whether the number of international applicants is down, and in a previous analysis I suggested that an across the board drop in applicants could support this hypothesis. But based on the data BAA released today, there’s a concentrated drop in runners with 10-20 minute buffers which explains why the number of applicants was lower than expected. It’s unclear why this would be, but there’s nothing that suggests this would be related to international applicants.
On a related note, there are 120 countries represented by the accepted runners – compared to 118 from last year. This doesn’t tell you anything about the number of runners from some common countries (like Canada, Great Britain, China), but it does suggest that there isn’t a massive international boycott. There are lots of countries with only a handful of runners represented, and if there was a large decline in international applicants you’d expect some of those countries to disappear.
The real question now is why the number of applicants from the 10-20 minute bucket was so low – and whether that will continue to be the case next year.
Other Interesting Observations
I saw a few people comment that they were impressed with the number of first time Boston athletes: 10,710. This is actually lower than last year (11,326) and 2024 (11,391). This is also evidence, for the people who doubted whether qualifiers from Boston will run Boston again. 56% of this year’s accepted applicants have previously run the Boston Marathon.
The top qualifying races were similar, with one exception: London is missing. This is no doubt a result of the warm weather there this year. Boston is the number one qualifying race again, with 2,981 entrants, and Berlin had more entrants this year (1,127) than last year (809). The new race in the top five this year is New York City, which leapfrogged CIM to take the #4 spot (832 entrants).
There are 110 non-binary runners – an increase over the last two years. The share of runners who are women is pretty consistent – 42.81%, compared to 42.62% (2025) and 42.87% (2026).
The number of qualifying para athletes increased from 26 to 195.
The number of legacy runners continues to climb slowly. It was 660 two years ago, 709 last year, and 719 this year. This reminds me that I’ve been meaning to look at the results and see if I could identify this group (or a large sample of it) for some analysis.
And finally, despite what you may have read elsewhere, this is not the lowest cutoff time since 2018. The two COVID years (2022-23) are obvious outliers, and those should be set aside.
But before the 2020 race was canceled as a result of COVID, they held a registration period in September 2019. 24,127 applicants were accepted out of a total of 27,288 qualified applicants – resulting in a 1:39 cutoff. That was also the last time that they implemented new qualifying times.
What’s Next for 2027?
If you made it into Boston, you’re probably focused on travel planning and training right now. But if you’re one of the 9,000 runners who got rejected – or the many thousands of other runners who didn’t quite make their BQ – you’re probably looking ahead to 2027.
In the next couple of weeks, I plan to:
- Write a more thorough analysis of this data to improve the projection methodology for next year.
- Evaluate the impact of the downhill results index and the elimination of the double dip period.
- Outline the likely scenarios for the cutoff time in 2027.
- Launch a new version of the cutoff time tracker.
If you want to keep up with the latest developments, use the form to the right to subscribe to my weekly newsletter. You can also follow me on Threads for weekly updates on race results and the cutoff time.
But for now, I would anticipate a similar or larger cutoff next year. I’m running Twin Cities in two weeks, and I’ll be aiming for BQ-7 just to be on the safe side. Plan accordingly.