“Pace it evenly” is the most worn-out piece of marathon advice there is. The problem is that it is usually backed by a coach's personal experience rather than by numbers. In 2026 PLOS ONE published a paper that closed that gap on a sample it is hard to argue with: 78,912 Boston Marathon finishers from 2015–2017, with nine timing checkpoints each.

Four pacing types

The authors (Qizheng Dong and Zaidong Li) did not define categories in advance; they clustered the actual split profiles. Four came out:

Even pacing — 47.9% of runners
Moderate positive split — 31.9% (controlled slowdown in the second half)
Pronounced positive split — 12.2% (an obvious collapse at the end)
Erratic pacing — 8.0% (pace jumps up and down all the way)

The first thing worth noting: fewer than half of the starters run evenly — at a qualifying marathon, where everyone has already run a standard. In a mass-participation race the share will be lower.

The numbers that make this study worth reading

The frequency of the “wall” — a sharp drop in pace in the second half — by group:

Pacing	The “wall”
Even	0.2%
Moderate positive split	9.9%
Pronounced positive split	22.4%
Erratic	85.4%

The spread here is not a matter of a few times over, but of hundreds of times. And finish times diverge accordingly: the even group finished in about 217 minutes on average, the pronounced positive split group in about 276 minutes. Almost an hour of difference. Erratic pacing trailed even pacing by 39.7 minutes.

One more line that explains a lot: among strong runners, 82.0% choose even pacing; among those running for enjoyment, 19.6%.

Careful with the conclusion “run evenly and you will get faster”

This is an observational study. It shows an association, not a cause, and the direction of the arrow is not obvious here.

Even pace is largely a consequence of fitness, not its cause. A runner who has put in enough volume for marathon pace to be genuinely sub-threshold will hold it without effort. A runner for whom the target pace is too fast will slide into a positive split regardless of intent. In that case the “hour of difference” is a difference in readiness, merely reflected in the shape of the graph.

The reverse link exists too, and that very 0.2% figure confirms it: even among well-prepared runners, evenness protects against a collapse. It is more accurate to treat even pacing as a necessary but not sufficient condition.

Boston is a specific course. The first kilometres run downhill, and the Newton hills fall at kilometres 26–33, exactly where your reserve runs out. An “even pace” on that profile means very uneven power: downhill you collect seconds for free, on the hills you pay for them twice over. On a flat marathon the numbers would be different.

The sample is selective. Boston qualification filters out casual entrants. You cannot transfer these percentages to a city marathon with a six-hour cut-off.

How to apply this

Set your target from the second half. Your target pace is the one you will be able to hold at kilometre 35, not the one that looks nice in the first.
The first 5 km — no faster than target pace. Even if it is downhill and even if your legs are carrying you. Seconds taken at the start are repaid with interest.
On hills hold effort, not pace. Trying to hold a number on a climb is the fastest way to join the pronounced positive split group. Slower up, freer down.
Erratic pacing is the worst of the four, and it is self-inflicted. It is made of surges to stay with a group, accelerations into aid stations and attempts to “win back” lost time. The 85.4% collapse rate is the price of those surges.
A moderate positive split is a perfectly good plan B. Slowing by 1–2% in the second half is better than clinging to the schedule and falling apart: this group hits the wall ten times less often than those who held on to the last.
Remember that your threshold at the end of the race is lower than at the start. Speed at the lactate threshold falls after an hour and a half of work — that is a separate story, covered in the piece on durability. A pacing plan calculated from fresh numbers is optimistic by construction.

Key points

PLOS ONE, 2026, Dong and Li: 78,912 finishers of the Boston Marathon 2015–2017, nine checkpoints, clustering of split profiles.
Four pacing types: even 47.9%, moderate positive split 31.9%, pronounced 12.2%, erratic 8.0%.
Frequency of the “wall”: 0.2% / 9.9% / 22.4% / 85.4% respectively.
The even group is faster than the pronounced positive split group by about 59.5 minutes, and faster than the erratic group by 39.7 minutes.
Even pacing is chosen by 82.0% of strong runners and 19.6% of those running for enjoyment.
The study is observational: evenness partly reflects fitness rather than creating it.
Boston's profile (downhill start, hills at km 26–33) and its qualifying standard limit how far these numbers transfer to other races.

Source: Dong Q., Li Z. «Pacing strategy patterns and performance outcomes in marathon running: A large-scale analysis of split time data», PLOS ONE, 2026. https://doi.org/10.1371/journal.pone.0352808