Which team races go to form?
Orderliness is a real property of a team race regatta (split-half r = +0.34, true spread 0.30) and nothing known in advance predicts it: field R² +0.02, host effects negative, wind r = -0.07.
Some team race regattas go to form and some are a lottery, and the difference is real: it shows up in the first half of a round robin and again in the second. But nothing known before the first start predicts it. Not the field, not the host, not the wind.
Measuring it
For every match the forecast gives a log-odds that the stronger side wins. An event's orderliness is the scale that best fits those log-odds to what happened at that event: 0 means the forecasts were right-sized, above 0 the favourites won even more than stated, below 0 the event was upset-heavy. It is the same quantity as the temperature on the Weather tab of fleet regattas, computed from matches instead of races. 603 ICSA team race regattas since 2013, 29,889 matches; the typical standard error on one event is 0.30.
It is a property of the event
If every event had the same orderliness, the spread of the estimates would be explained by their standard errors; it is 1.76 times that, which implies a true event-to-event spread of about 0.30. Split each event's matches into odd and even and measure orderliness twice: the two halves agree at r = +0.34; first half against second half, r = +0.24. 55 of 603 events sit more than two standard errors from zero where about 28 would by chance.
| Most upset-heavy | Orderliness |
|---|---|
| St. Mary's Quad Team Race (spring 2018) | -2.27 ± 0.26 |
| Washington College Team Race (spring 2022) | -1.91 ± 0.31 |
| Big 10 Team Race (spring 2026) | -1.68 ± 0.33 |
| Seahawk Team Race Invite (spring 2018) | -1.31 ± 0.23 |
| Most orderly | Orderliness |
|---|---|
| Harpswell Sound Team Race (Formerly Mystic L. TR) (spring 2017) | +1.06 ± 0.47 |
| MAISA Team Race Challenge (spring 2017) | +0.63 ± 0.26 |
| Duplin Women's Team Race (spring 2017) | +0.58 ± 0.25 |
| Jen Harris Women's Team Race (spring 2025) | +0.44 ± 0.19 |
Nothing predicts it
- The field. Number of teams, matches, mean and spread of strength, newcomers, championship status, women's event, how much history the forecast had: a ridge model on all of them, cross-validated by event, explains R² = +0.02.
- The host. A per-host effect on what is left over makes held-out prediction worse at every shrinkage tried (best R² -0.04, 61 hosts). The host's fleet race orderliness, which is a real and persistent venue property, correlates with its team race orderliness at r = +0.00 over 589 events.
- The same regatta next year. r = +0.05 over 284 consecutive editions.
- A fleet regatta at the same venue the same weekend. 56 cases, r = +0.10.
Not the wind either
In fleet racing, some venues get orderly in a breeze. Apply each host's fitted fleet surface at the team race event's wind and the slope is 0.14 of the fleet effect, not distinguishable from zero. A two-boat-per-side format with mark traps and pass-backs apparently erases what the wind does to a fleet.
High school
ISSA: 350 regattas, true spread 0.52, split-half r = +0.36, the same regatta next year r = +0.24. One thing does predict a little: championships are more orderly (R² = +0.04 on its own), which fits a format where the field has been qualified down to teams whose strength is well measured.
What this means on the site
Each team race regatta's Ratings & Forecast tab shows its measured orderliness and the share of events it was more orderly than, as a description of what happened. There is no prediction, because there is nothing to predict it with. If you think a particular venue or a particular breeze makes team racing a lottery, this is the test that says otherwise, at least so far.
Every number on this page is produced by analysis/team_race/orderliness.py; numbers via analysis/blog/team_race_post_data.py in the project repository, from the same database that powers the rest of the site. Corrections welcome: quinnbrighton2005@gmail.com.