81st Boston Dinghy Challenge Cup — Analysis

Performance rating answers one question for each skipper: holding every other skipper's rating fixed, which rating makes my actual race-by-race finishes least surprising? It is the maximum-likelihood rating under the same Plackett–Luce model that produces the power rankings (how accurate it is), with each opponent fixed at their rating from the last rated week before this event (2016-W16). Substitute skippers are scored only on the races they sailed, and races the boat did not finish (OCS, DNF and similar) are left out.

Percentile places a rating among the 2,454 ICSA skippers who were rated within the year before this event, so it reads as “better than X% of the skippers racing in this league right now”. The arrow shows where the sailor stood going in and where this performance would put them.

Surprise is how many bits more surprising the sailor's finishes were at their pre-event rating than at their performance rating (0 = exactly as expected). P(this good or better) simulates the regatta 3,000 times with the sailor at their pre-event rating against the actual fleet, and reports how often they would score this many points or fewer in the races they skippered. Races in one regatta are not independent, so each simulation also gives every skipper a good or bad regatta, form that drifts from race to race, and the speed of each boat the rotation puts them in (why). Low means an unusually good regatta, high means an unusually poor one. It deliberately ignores how rivals actually sailed, so it agrees with the performance rating rather than with the division finish, which also depends on everyone else's day. Ratings are shown on the rankings scale (model rating × 100). Only skippers are rated; crews are not part of the model.