Method · v2.1
Method for the development curve model v2.1
How swimmers of a given age, sex category and event have historically improved, expressed as a spread rather than a single path — and why that spread does not become a forecast for one person.
What it does
For a defined cohort — an event, a sex category as recorded, an age band, a period and the federations covered — the model fits the relationship between age and adjusted time across every swimmer in that cohort, and publishes it as an interval at each age rather than a single line. Fitted on 118 402 swimmer-seasons, 31 federations, 2009–2026.
The output is a curve with a 95% interval band at each age. The band narrows where the cohort has the most races per age and widens at the edges of the age range, where fewer swimmers contribute.
What it does not do
It does not describe an individual. A swimmer’s adjusted time can be plotted against the curve for their cohort — Thomas Ceccon’s development curve shows this for one swimmer — but the curve itself is a statement about the cohort’s history, not a prediction for that swimmer’s next season or their time at 18. No page on this site draws the curve continued forward for a swimmer under 18.
What it assumes
- The cohort is fixed before the curve is fitted. Event, sex category as recorded, age band and period are set first; the curve is not re-fitted to flatter a particular swimmer’s position in it.
- Age is measured at the date of the race, not the season. Two races six months apart at ages 15.4 and 15.9 are treated as different points on the curve.
- Below three races at a given age, the model does not publish that age. The excluded ages are stated on every chart that uses the curve, rather than interpolated silently.
Where it is weak
The curve is calibrated the same way the adjustment is (method: how the intervals are calibrated), and it inherits that calibration’s undercoverage below five races in a window. It is also weakest at the top and bottom of the age range it is fitted on — 13 and under, 24 and over — for the same reason any regression is least certain at the edges of its data: 6 104 of the 118 402 fitted swimmer-seasons fall in that range.
Version history
- v2.1, 2026-08-20 — fitted per event rather than pooled across strokes. Moved published curve figures by a median of 0.11 s and by 0.94 s at the largest, concentrated in the backstroke events, where the four distances had previously shared one curve.
- v2.0, 2026-03-02 — first version to publish the interval band at each age rather than a single fitted line.