The Chase for Championship Weekend
Midseason predictions for each division
In all sports it goes without saying that championships define outward success. However, the journey on the climb to the top differs in difficulty depending on the sport and postseason structure. Take the NBA, for example: including the play-in games, more than half of the teams in the league every year qualify for the postseason. Speaking harshly, getting to the postseason means your team didn't tank to improve your lottery odds for the upcoming draft. However, getting through the playoffs of a sport and tournament that has the most enjoyable deterministic properties (my opinion) and to the finals is a feat. In the NFL, your team is as good as trivia history if you didn't at least get to the Super Bowl, the true pinnacle of the most popular league in the country. NCAA football's current Mount Everest is being one of the four teams chosen for the College Football Playoff. Progressing through the 68-team gauntlet of March Madness requires taking exponentially more difficult steps at each round.
I've exhausted the point, but certain milestones are of more interest depending on the sport and postseason structure. As it stands currently in the AUDL, three teams from each of the four divisions qualify for the playoffs. In each division, the three seed plays at the two seed for a chance to play at the one seed's home stadium for the division title. The four division champs converge at Championship Weekend, where the semis and AUDL final will be played on back-to-back days. Qualifying for the playoffs means you are in the better half of the teams in your division, but getting to Championship Weekend is an exponentially higher jump from the playoffs than getting into playoff contention.
Looking to the past
So, if Championship Weekend is so lofty, let's take a look at the teams that have gotten there since the 2012 inaugural season.
The recent success of New York is obvious and most notable with Champ Weekend appearances in each of the last four seasons (and winning two of them). However, Toronto, Madison, and Dallas seem to appear quite a bit as well. The bar chart below shows the cumulative number of Championship Weekend appearances for each franchise since the beginning of the AUDL.
The Radicals and Empire are tied at the top with six appearances, with the Rush right behind at five and the Legion at four (it sure doesn’t feel right to say that out loud about the Legion — even though it’s the same franchise, it definitely seems disrespectful to the Roughnecks to attribute their success to the current logo and outfit).
Predicting the future?
Looking to the past is fun and all, but what about this year? We are now essentially half way through the 2023 season and are nearing the end of the way-too-early playoff prediction season, so there’s no better time to start to track who might make Championship Weekend. We could opine with the AUDL talking heads (shoutout Adam and Daniel on the Swing Pass podcast and voice-of-ultimate’s Evan Lepler’s consistent commentary at the league level), but that’s not what I’m here for. Instead, I am introducing a model that has some potential to predict what we might see this year. Before I do, let me quickly explain what goes into the model and what the historical data indicates.
Model details
Put simply, this model predicts how likely a team is to make Championship Weekend given four basic season level statistics for each team, two offensive focused metrics and two defensive focused metrics:
O Line Conversion Percentage — frequency that the offensive line scores over all possessions they have
Completions Per Game — number of total team completions on average over the course of the season
Break Percentage — frequency that the defensive line scores over all of the points they play
Blocks Per Game — number of total team blocks on average over the course of the season
The more specific details of why each of those were chosen aren’t necessary here (if you want them feel free to ask), but these seem to be good explainers of how good a team is in a specific season. It accounts for how stingy and successful the offense is with the disc and for how consistently a defense can take the disc away and get breaks. The model also accounts for how efficiently the team plays and moves with completions per game.
The remainder of the model details can be skipped for the reader who is just interested in the midseason 2023 Championship Weekend predictions, but those who are interested in a little bit more information may continue on…
Data from the 2012 and 2013 seasons were excluded because O line conversion percentages and break percentages weren’t recorded until 2014. The Bayesian logistic regression model used was first trained on the 2014 through 2021 seasons in order to test how good the model was at predicting the 2022 Championship Weekend attendees. The model predicted that the Empire, Summit, Union, and Hustle were the most likely teams from each division (with the Flyers as the second most likely from the South division). Without the knowledge of what happened in the 2022 season, the model performed quite well since Carolina was the only team that made Championship Weekend that the model didn’t predict as being the most likely to make it out of the South, so I went forward and used all the data from 2014 through 2022 to refit the model.
Below are the results from the posterior draws for each parameter in the model. The basic takeaways are actually quite insightful, even if they’re not completely surprising:
O line conversion percentage is quite important in explaining whether or not a team makes Champ Weekend with higher conversion percentages improving a team’s chances quite drastically.
While the effect is small, the more completions a team has per game the less likely a team is to make Championship Weekend (looking at give-n-go obsessed 2022 Oakland Spiders…).
Being better at getting breaks seems to help increase a team’s chances, but definitely isn’t as obvious or significant as offensive efficiency is.
Getting more blocks surely improves a team’s probability of making Championship Weekend, but how great an effect of an extra block per game is much more uncertain.
In summary, the model performs well and makes sense, so let’s go on and see what it has to say about the future!
Model results
Before showing the results in the table below, I’d like to give a reminder that these probabilities aren’t at all based on what is actually technically possible given the specific team’s win-loss record or their remaining games, but instead how likely the model determines the teams are to make Championship Weekend given the team’s summary stats so far this season. The model is outputting probabilities based solely on those four team-season statistics, agnostic of how many games the team has played, won, lost, or what their overall point differential is. That said, below are the model’s predicted probabilities:
The immediately most interesting thing is that the results fall about in line with what the aforementioned AUDL pundits are saying these days. Most out-of-line, however, is the relative certainty with which the model says Atlanta will be the team advancing out of the South. Despite their most recent loss to Carolina, the model still predicts they are 65% more likely to make Championship Weekend than the Flyers. While the model says nothing about the playoff race specifically, it’s hard to not see those probabilities and think almost every division has at most two teams actually in the hunt to be in the final four.
Concluding thoughts
I’m curious to see how these probabilities change as we approach the back half of the season, but it is remarkable that the model almost perfectly falls in line with the standings, even with some movement for adhoc strength of schedule adjustments. The graph below further explores this by showing each team graphed by their team’s win percentage on the x-axis and the team’s probability of making Championship Weekend as of the midseason point on the y-axis.
The general quadratic trend is relatively convincing, showing further evidence that the four team level metrics used in the model are not only quite good at predicting not only Championship Weekend attendees, but also explaining team win percentage. This makes sense broadly, so it is a good sanity check on the model’s general usefulness.
So, as of right now it’s the Wind Chill, Empire, Hustle, and Shred as the favorites to make Championship Weekend in Minnesota. We’ll know in less than two months whether or not that pans out.
감사합니다
What do you think? How will things shake out come the end of August?
Feel free to leave questions or feedback. You can find me on Twitter @Comet_Miller. Thanks for reading!
— 밀혜성