2.5% of the field ran the second half faster

Every coach tells marathoners to negative split. I cut every finisher's 2014 Boston race in two at the halfway mat. Of 31,912 runners, 803 came home faster than they went out, and the median runner's second half was 14% slower than the first.

One square = 1% of finishers

The field

Nearly everyone slows down

The whole distribution of second-half slowdowns sits on the slow side of zero. Its peak is around 12 to 15%, and the median is 14%. That is the typical runner, so a handful of blow-ups can't explain it.

Hover the chart, or focus it and use the arrow keys, to read a bar.

Each bar is one percentage point of second-half slowdown, as a share of all 31,912 finishers. Rust bars ran the second half faster. The dashed line is the median. From tables/web_splits.json.

By finish time

The faster the runner, the smaller the fade

Runners who finished under 3 hours slowed by a median of 5.3% in the second half. Runners slower than 4 hours slowed by 21.7%, about four times as much. Fitter runners are further from their limit at halfway, so the pattern is an association and can't show that even pacing makes anyone faster.

2:00–2:30+4.5%
2:30–3:00+5.2%
3:00–3:30+8.6%
3:30–4:00+12.2%
4:00–4:30+17.3%
4:30–5:00+22.0%
5:00–6:00+26.8%

Median second-half slowdown by finish band. The write-up compares your own split with your band.

Women slowed a little less than men: a median of 13.4% against 14.8%.

Other years

2014 was a hard year to try it

Run the same count on every Boston year with public splits and the share of negative splits goes from 2.4% (2024) to 10.8% (2023), while the median fade goes from 5.8% to 14.0%. 2014 had the second-lowest share and the biggest fade of the eight. The 2014 numbers above are unchanged; the 2013 race, stopped at about 4:09 on the race clock, is left out.

20142.5%
20157.7%
20162.6%
20173.1%
20214.7%
20226.9%
202310.8%
20242.4%

Share of finishers whose second half was faster than their first, per year, wheelchair and handcycle entrants removed. From tables/years_splits.json (multiyear.py); the write-up has the full table with median fade.

The companion study now has a live split tool that predicts your finish band, wall risk and pacing type as you enter splits, trained on these years and tested on 2024.

Method and limits

Official 2014 Boston results from the public llimllib/bostonmarathon mirror, 31,912 finishers with a valid half-marathon split. First half is the halfway mat time; second half is finish time minus that. Python with pandas and matplotlib; fetch_data.py then analyze.py reproduce every number.

The headline is one race on one course. Boston's early downhill and its hills between miles 16 and 21 make an even split harder than on a flat course, and its qualified field is faster than a typical marathon's. The halfway split is the only checkpoint used, and weather is not modeled.

The useful reading: almost nobody in this field started too slowly. The full write-up has the split medians for every finish band and a tool that places your own split against the field.