More understanding.
Every mile.

Why it exists

Runners make decisions about pace, targets and courses with incomplete information. The Marathon Pacing Study, created by Andrew Kam, makes large collections of race results understandable and useful to explore.

Its ten main analyses follow the questions a runner might ask before or after a race: how people pace, what follows different starts, how conditions compare, and where improvement appears.

What the data contains

The essential ten analyses use 3,328,159 finishes with a complete set of usable splits, across 28 cities. A finish is a performance, so one runner may appear more than once.

The weather comparisons examine 3,324,934 eligible finishes across 177 race editions in 27 cities. Each race edition contributes equally.

We evaluated humidity, warming and wind together. 2 of the 3 questions met the publication rule. Results for humidity did not support a clear enough takeaway for a separate page.

Times at successive checkpoints show average pace for each section. Supplied records also include age, recorded gender and candidate links between runners’ races. Weather estimates and course profiles add context where available.

Marathons and years

The database contains results from 34 marathons and 240 city-and-year editions. The years below are the years recorded, including partial editions. They do not imply complete fields or usable splits for every runner. Each analysis applies its timing and source-quality checks.

Race results

Recorded names, marathon and year, finish times, checkpoint splits, and age and recorded gender where supplied. Candidate links help find a runner’s other results; visitors confirm which races belong to them.

Weather data

Usable weather context is available for 233 of 240 editions. Fields include air temperature, feels-like temperature, dew point, relative humidity, wind speed and direction, precipitation, cloud cover, sea-level pressure and weather-condition codes. Hourly readings also show how conditions change after the scheduled start; some fields may be missing.

These are Open-Meteo historical weather estimates near the race location. They describe modeled conditions around the supplied scheduled start, not a runner’s actual wave or exposure along the course. Precipitation is the preceding hour’s total, not a race total.

Elevation data

Supplied course profiles provide context for 238 of 240 editions. Recorded fields include route distance, elevation along the route, minimum and maximum height, climb, descent and net elevation change, including summaries for the timed sections. The unit switch shows elevation in feet or metres.

The profiles come from supplied routes and elevation models. The route used in each historical year has not been verified. Whole-route and section totals can differ, so the site identifies their source and treats terrain as context rather than an adjustment to finish times.

What an answer means

These are observed patterns. A fast opening and a fast finish can go together because a runner is fitter; that does not mean a faster start caused the improvement.

Each analysis shows the actual comparison group. If a narrow selection has too few records, any broader comparison is labeled. Missing ages, splits and route history stay missing.

The study is a way to investigate your race, not an individual training plan or a promise of a result.

Open to everyone

The complete exports, source code and methods are publicly available. No account or decryption key is needed. Newer exports can be downloadable before they have passed the checks required for website calculations.

Begin with one question

Start with the pacing pattern, then follow the next question or choose another of the ten.