The WIAA Boys Basketball Tournament came to a close on Saturday by naming state champions in each of the WIAA's five divisions. This post serves as a before-and-after, comparing what the models predicted before the tournament to what actually transpired on the hardwood.
First up, the state championship teams. The chart below shows the pre-tournament probability that each of the teams would win the state title. In many cases, a low probability reflects how difficult a team's tournament run was expected to be rather than expectations of poor performance. Cochrane-Fountain City earns the title of the least expected champion with only a 7.9% chance of winning.
Diving deeper, the chart below shows the least likely tournament runs. To qualify for this list, a team needed to win at least a regional championship and have one of the two least likely tournament runs in their division. Similar to the above list, these probabilities reflect the pre-tournament rating of the team, the expected difficulty of their path, and the length of their tournament run. The least likely tournament run goes to Sectional Runner-Up, Cedar Grove-Belgium, which had only a 0.02% of becoming the sectional runner-up (or better) before the tournament began.
Another view of similar data is shown below. This chart shows the 10 largest rating increases for teams during the tournament. The rating change measures similar results as the prior chart (unexpected wins) but also includes margin of victory as a main driver of the model's ratings. Wausau East received the largest rating increase.
No review of tournament surprises would be complete without looking at the individual games. The following exhibit shows the least likely upsets that occurred during the tournament. Sheboygan South takes the top spot here after their defeat of Hartford which had a pre-game probability of 3.3%.
And finally, it is time to review how the model performed during the tournament. The following exhibit only uses games between teams that were seeded in the same group (through sectional finals for division 1 and through sectional semifinals for divisions 2-5). The higher seed in these matchups only won 78.2% of the time. The model predicted the winner correctly in 82.0% of games which was better than the seedings and was close to the model's own expectations (82.1%). The main driver of the model's superior performance was the picking of upsets. In the 77 games where the model disagreed with the seedings, the model was correct 47 times.
