A statistics project on college sailing results, 2008 to 2026
College sailing ratings
I scrape results from the ICSA and ISSA scoring sites, fit a rating to every sailor, and
test how well it predicts races it hasn't seen. 1.5 million boat-race results since 2008.
Quinn Brighton · Applied Math–CS, Brown University · Class of 2027EmailLinkedInGitHub
100K+
Training races
24K+
Sailors modeled
13.7M
Parameters
70%
Held‑out accuracy
Predictions
Win probability from 5,000 simulated regattas on the predicted lineups. One event per weekend, the one with the strongest field.
Best sailor in each of the last five graduating classes, weekly rating over their career
ICSA. The sailor with the highest peak rating in each class; the line is their rating every week they were rated.
Program rank: average rating of each school's two best active skippers
Ranked among all ICSA programs at the end of each season. Schools shown were top three at least once; a line that reaches the bottom row is outside the top ten.
Method
Every sailor gets a skill rating for each week they race, θ[sailor, week].
I fit the ratings so the observed finishing orders are as likely as possible under a
Plackett–Luce model, which handles a full finishing order rather than just win or loss.
Penalties on week-to-week change keep a rating from jumping on one regatta. About 13.7 million
parameters, trained in PyTorch.
The Boat Speed Analyzer uses the same data. It compares a school's fleet-race
results across two time windows and tests whether the difference is larger than race-to-race noise.
Accuracy
I held out a random 1 in 20 regattas before fitting. On those races the model picks the
faster of two boats 70% of the time. Forecasts a full season ahead, and the comparison with
Elo and simpler baselines, are on the methodology page.
70%
of boat pairs ordered correctly on held-out races
Why
Most college sailing rankings are win totals or opinion. I wanted ratings I could check against results.
I'm Club Captain of Brown Sailing and use these numbers for lineups and scheduling.
It is the project that made me want to build statistical models for a living.