| # | School | Team | Rec. | |
|---|---|---|---|---|
| 1 | Tufts University | Jumbos | 9–2 | |
| 1 | University of Vermont | Catamounts | 9–2 | |
| 3 | Fordham University | Rams | 7–4 | |
| 4 | University of Rhode Island | Rams | 5–6 | |
| 5 | Harvard University | Crimson | 4–4 | |
| 6 | Massachusetts Institute of Technology | Engineers | 3–5 | |
| 7 | University of Pennsylvania | Quakers | 3–5 | |
| 8 | Northeastern University | Huskies | 2–6 | |
| 9 | Princeton University | Tigers | 0–8 |
Each side is the sum of its three boats (the skipper's fleet rating times 0.72 plus a team race offset learned from their past matches, plus the crew's offset and a small experience term) plus the school's team race strength and a home-water edge for the host. The win probability is a logistic function of the difference. The forecast was made before the event: the model was refitted on every match sailed before the first day, with fleet ratings as of the week before.
Fitted on the 12,301 team race matches sailed before 2019-03-09, using the line-ups that actually raced. It called 30 of 42 matches (71%) with a log-loss of 0.491 (a coin flip scores 0.693), so the event went about as expected for the forecast. Orderliness −0.25 ± 0.33: a typical event, more orderly than 21% of college team race events (temperature of the forecast log-odds that best fits the results, relative to the league average; nothing about the field, venue or wind predicts it, see the Weather tab). vs forecast is a team's wins minus the wins the model expected from the matches it sailed: positive means it beat its rating.
For each side, the performance strength is the value that would have made its record least surprising: the strength at which the model's expected wins over exactly these opponents equal the actual wins, with every opponent fixed at its pre-event strength. For a skipper, the same shift is applied only to their boat in the matches they sailed, with team-mates and opponents fixed; team-mates who sailed every match together therefore get the same number. Surprise is in bits: -log2 of the probability of the record at the pre-event strength minus the same at the performance strength. P(this good or better) treats matches as independent coin flips at the model's pre-event probabilities, so it ignores form within the day; read it as a rough guide.
Pre-event strength against the strength that would have made each record least surprising, everything else fixed. Δ is in the model's logit units, the same scale as the boat values above (a side is three boats plus the school term); green is a clear over-performance, red a clear under-performance. Skippers need three or more matches.
| Place | School | Strength | Perf. | Record | Exp. wins | Δ | Surprise (bits) | P(this good or better) |
|---|---|---|---|---|---|---|---|---|
| 1 | University of Vermont | 8.29 | 10.13 | 9–2 | 5.8 | +1.83 | 3.9 | 2% |
| 3 | Fordham University | 8.20 | 8.87 | 7–4 | 5.8 | +0.67 | 0.6 | 30% |
| 5 | Harvard University | 7.47 | 7.76 | 4–4 | 3.6 | +0.28 | 0.1 | 53% |
| 7 | University of Pennsylvania | 7.01 | 7.02 | 3–5 | 3.0 | +0.01 | 0.0 | 65% |
| 6 | Massachusetts Institute of Technology | 7.17 | 7.03 | 3–5 | 3.2 | −0.13 | 0.0 | 71% |
| 1 | Tufts University | 10.09 | 9.48 | 9–2 | 9.8 | −0.61 | 0.4 | 89% |
| 8 | Northeastern University | 7.13 | 6.05 | 2–6 | 3.1 | −1.08 | 0.8 | 94% |
| 4 | University of Rhode Island | 9.02 | 7.64 | 5–6 | 7.7 | −1.38 | 2.8 | 99% |
| 9 | Princeton University | 2.16 | -0.84 | 0–8 | 0.0 | ≤−3.00 | 0.0 | >99% |
| Skipper | School | Value | Perf. | Record | Exp. wins | Δ | Surprise (bits) | P(this good or better) |
|---|---|---|---|---|---|---|---|---|
| Eduardo Mintzias | Fordham University | 2.94 | 3.61 | 7–4 | 5.8 | +0.67 | 0.6 | 30% |
| Francis Doyle | Harvard University | 2.22 | 2.50 | 4–4 | 3.6 | +0.28 | 0.1 | 53% |
| Eric Hansen | Harvard University | 2.01 | 2.30 | 4–4 | 3.6 | +0.28 | 0.1 | 53% |
| Amanda Majernik | University of Pennsylvania | 2.41 | 2.61 | 2–2 | 1.9 | +0.19 | 0.0 | 69% |
| Ava Esquier | University of Pennsylvania | 2.66 | 2.54 | 1–3 | 1.1 | −0.13 | 0.0 | 73% |
| James Peraire-Bueno | Massachusetts Institute of Technology | 2.36 | 2.23 | 3–5 | 3.2 | −0.13 | 0.0 | 71% |
| Christopher Keller | Tufts University | 3.24 | 2.62 | 9–2 | 9.8 | −0.61 | 0.4 | 89% |
| Clayton Snyder | Fordham University | 2.12 | 1.03 | 2–3 | 3.3 | −1.09 | 1.0 | 95% |
| Connor Mraz | Princeton University | 1.61 | -1.39 | 0–8 | 0.0 | ≤−3.00 | 0.0 | >99% |
| William Mandelbaum | Princeton University | -0.44 | -3.44 | 0–8 | 0.0 | ≤−3.00 | 0.0 | >99% |
| Connor Larson | Princeton University | 0.42 | -2.58 | 0–8 | 0.0 | ≤−3.00 | 0.0 | >99% |
Probability the model gave the side that went on to win. Red bars are upsets (the winner was given under 50%).
| Team | Rec. | Jumbos | Catamounts | Rams | Rams | Crimson | Engineers | Quakers | Huskies | Tigers | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Tufts University Jumbos | 9–2 | ✕ | L | W | W | W | W | W | W | W | |
| University of Vermont Catamounts | 9–2 | W | ✕ | W | W | W | W | W | W | W | |
| Fordham University Rams | 7–4 | L | L | ✕ | W | L | W | W | W | W | |
| University of Rhode Island Rams | 5–6 | L | L | L | ✕ | W | L | W | W | W | |
| Harvard University Crimson | 4–4 | L | L | W | L | ✕ | W | L | W | W | |
| Massachusetts Institute of Technology Engineers | 3–5 | L | L | L | W | L | ✕ | W | L | W | |
| University of Pennsylvania Quakers | 3–5 | L | L | L | L | W | L | ✕ | W | W | |
| Northeastern University Huskies | 2–6 | L | L | L | L | L | W | L | ✕ | W | |
| Princeton University Tigers | 0–8 | L | L | L | L | L | L | L | L | ✕ |
| # | Winner | Loser | Pos. | ||||
|---|---|---|---|---|---|---|---|
| 1 | Huskies | 1-2-3 | def. | Tigers | 4-5-6 | ||
| 2 | Jumbos | 1-2-3 | def. | Engineers | 4-5-6 | ||
| 3 | Quakers | 1-3-5 | def. | Crimson | 2-4-6 | ||
| 4 | Rams | 1-3-6 | def. | Rams | 2-4-5 | ||
| 5 | Catamounts | 1-2-6 | def. | Huskies | 3-4-5 | ||
| 6 | Engineers | 1-2-3 | def. | Tigers | 4-5-6 | ||
| 7 | Jumbos | 1-2-3 | def. | Crimson | 4-5-6 | ||
| 8 | Rams | 1-2-3 | def. | Quakers | 4-5-6 | ||
| 9 | Rams | 1-2-3 | def. | Catamounts | 4-5-6 | ||
| 10 | Huskies | 1-3-6 | def. | Engineers | 2-4-5 | ||
| 11 | Crimson | 1-2-3 | def. | Tigers | 4-5-6 | ||
| 12 | Rams | 1-2-3 | def. | Jumbos | 4-5-6 | ||
| 13 | Rams | 1-2-3 | def. | Quakers | 4-5-6 | ||
| 14 | Catamounts | 1-2-6 | def. | Engineers | 3-4-5 | ||
| 15 | Crimson | 1-2-4 | def. | Huskies | 3-5-6 | ||
| 16 | Rams | 1-2-5 | def. | Tigers | 3-4-6 | ||
| 17 | Jumbos | 1-3-4 | def. | Rams | 2-5-6 | ||
| 18 | Catamounts | 1-2-6 | def. | Quakers | 3-4-5 | ||
| 19 | Crimson | 1-2-6 | def. | Engineers | 3-4-5 | ||
| 20 | Rams | 1-2-5 | def. | Huskies | 3-4-6 | ||
| 21 | Rams | 1-2-3 | def. | Tigers | 4-5-6 | ||
| 22 | Jumbos | 1-2-3 | def. | Quakers | 4-5-6 | ||
| 23 | Catamounts | 1-3-5 | def. | Crimson | 2-4-6 | ||
| 24 | Rams | 1-2-4 | def. | Engineers | 3-5-6 | ||
| 25 | Rams | 1-2-3 | def. | Huskies | 4-5-6 | ||
| 26 | Quakers | 1-2-6 | def. | Tigers | 3-4-5 | ||
| 27 | Jumbos | 1-2-5 | def. | Catamounts | 3-4-6 | ||
| 28 | Crimson | 1-2-4 | def. | Rams | 3-5-6 | ||
| 29 | Engineers | 2-3-4 | def. | Rams | 1-5-6 | ||
| 30 | Quakers | 1-3-6 | def. | Huskies | 2-4-5 | ||
| 31 | Jumbos | 1-2-5 | def. | Tigers | 3-4-6 | ||
| 32 | Catamounts | 1-4-5 | def. | Rams | 2-3-6 | ||
| 33 | Rams | 1-2-5 | def. | Crimson | 3-4-6 | ||
| 34 | Engineers | 1-2-3 | def. | Quakers | 4-5-6 | ||
| 35 | Jumbos | 1-2-5 | def. | Huskies | 3-4-6 | ||
| 36 | Catamounts | 1-2-3 | def. | Tigers | 4-5-6 | ||
| 37 | Catamounts | 1-2-4 | def. | Rams | 3-5-6 | ||
| 38 | Jumbos | 2-3-4 | def. | Rams | 1-5-6 | ||
| 39 | Catamounts | 1-2-5 | def. | Rams | 3-4-6 | ||
| 40 | Jumbos | 1-2-6 | def. | Rams | 3-4-5 | ||
| 41 | Rams | 1-2-3 | def. | Rams | 4-5-6 | ||
| 42 | Catamounts | 1-2-5 | def. | Jumbos | 3-4-6 | ||
Orderliness is how tightly a fleet race’s finishing order follows the pre-regatta ratings, relative to a typical event (0): positive is the favourites’ day, negative the underdogs’. A value of ±0.02 means a rating gap counts about 2% more than on a typical day. This event is not a fleet race the rating model scores, so there is no actual value for it: this is what this venue’s fleet racing history says about the breeze on the day, read off the chart at the star. How this is estimated · compare venues.
Short write-ups of what the ratings and results show, with the numbers and how they were computed.