Countering Bracketology Groupthink
If we all agree, we're all wrong.
Bucket Listener
While consuming an episode of the Eye on College Basketball podcast in late December, I heard something that deeply resonated. Gary Parrish commented on the NCAA Tournament seeding process, and how it was both consistent and predictable, Parrish stated, “We put 68 teams into the tournament, and we all agree on 65 or 66 of them.” I saw this as problematic, and that the level of groupthink in the bracketology community leaves an unseen number of blind spots.
This is, of course, how the system is designed. We spend three months of the college basketball regular season seeding teams based on how we think the selection committee will seed them. When presented with the same numbers, historical references, and a community echoing ideas, we’re going to see similarities. This is not an indictment on the bracketology community as a whole, but instead a critique of the seeding process. What school of thought are we favoring at the opportunity cost of another?
Dazed and Confused
When filling out our final bracket to be posted on the Matrix, we aren’t asking ourselves of the most deserving teams. The well-known nature of the Selection Committee's lack of consideration of the final weekend of the season beyond auto-bids is just a single example. The 68 teams in the tournament should be the 32 most deserving from each conference, in addition to the 36 at large teams. This should consider the entirety of the season, not simply conference play. In an ideal world, the conference tournament wouldn’t exist.
Regardless of what we’ve been taught to think, I’d love you to consider the scenario below, and which team is more deserving of representing their single-bid conference for the NCAA Tournament.
Team A
22-9 (12-6)
Top-seeded team in the conference tournament
Lose conference tournament finals
Star point guard caught mono
Starting wing injured shoulder
Team B
17-14 (9-9)
Sixth-seeded team in the conference tournament
Beat the heavily injured top-seeded team in the finals
Though injuries and unexpected losses, especially in games that matter more than others, are an important part of the game, are the 32 most deserving teams from each conference not supposed to be the auto-bid representatives? This school of thought may make conference tournaments completely obsolete, but the weight of three months of regular-season basketball should hold much more weight than just determining the seeding for conference tournaments. There are two options worth discussing which take unique looks at the conference tournament.
Stick to the Models
The Ivy League has long been a trailblazer, or more accurately, a footprint retracer, in their commitment to the past. The first conference tournament for the conference was not until 2017, and they have yet to commit to the most popular model, though their conference of eight teams is perfect for an option like the one below.
Image per the Southwest Athletic Conference
The Ivy League doesn’t grant the ability for every team in their conference to compete for a National Championship come the first day of their conference tournament, as they instead grant their top four teams admittance.
Image per ViSN
This model treats the conference season more importantly than any other, but the restrictive nature of the tournament isn’t as appealing for half of the teams in their conference. If this were used universally, instead of about 350 teams having the opportunity to win the National Championship, there would be fewer than 200, and it is clear why that solution might not be best painted with a broad brush.
Instead of making a conference tournament restrictive, the WCC just makes it more difficult for lower-seeded teams to earn the auto-bid, and rewards higher-seeded teams.
Image per the West Coast Conference
Though this model is the most effective in rewarding well-performing teams while not completely omitting bottomfeeders, it has a flaw, which makes the non-conference schedule obsolete in single-bid leagues. The games played before January hold no weight in determining this seeding, and in turn how easy it is to qualify for the tournament. There is a conference, similarly in the western part of the conference, that has an effective solution, to place more value on these games.
The WAC adopted a seeding algorithm the previous season, which evaluates games played throughout the season, valuing not only performing in conference, but the strength of that schedule.
Conference record is not directly that team’s seed, as is most evident in Sam Houston claiming the top seed over Utah Valley, who had the 24th toughest non-con schedule, compared to Utah Valley’s 48 (out of 363). Though it isn’t a dramatic difference, it could be clearer in high major conferences. In 2024, TCU has a non-con SOS in the 330s, compared to Kansas, with one in the 60s. Though this brings us full circle to the point stated before, high major conference tournaments don’t matter, outside of bid stealers.
Mission Impossible III (My Own Team Sheet)
Currently, team sheets are used by the selection committee to fill out, and subsequently seed the field. Here’s top-seeded Houston’s team sheet per Warren Nolan in 2023.
A combination of results and predictive-based metrics, combined with the team’s performance against all four quadrants, their overall strength of schedule, and average NET wins and losses are all evaluated. What matters here?
The most prevalent feature of team sheets is the outcome of games, though not all wins, even in the same quadrant, are treated equally. Though winning close games matters, there comes a point in time that the amount of luck involved would prove considering the game negligible. An example coming to mind is Alabama against North Carolina in November of 2022. The win probability graph below shows just how easily the game could have gone one way or the other. How much could truly be revealed about a team in a game such as this one?
Image per KenPom
Alabama was comfortably a one-seed, while North Carolina was not near the field, but games such as this should not have an implication on seeding, at least to the same extent as blowouts. Keeping the quad the same, wins and losses need to be evaluated separately from the current state.
Instead of focusing merely on wins and losses based on quadrants, the margin of victory or defeat is of tantamount importance. Separating quadrants based on close margins (within four points or OT), and further segmenting moderate margins and losses against blowouts (more than 14 points).
In a project last year, I wrote an algorithm for bracketology, gaining an understanding of what the selection committee valued concerning the team sheet. Outside of the top three in their round of 64 matchups, there aren’t going to be any Q3 or Q4 opportunities in the tournament. Why does beating up on these teams matter?
This provides a good segway to another metric worth considering, the NET, which is a results-based metric, based on who a game was played against, where the game was played, and the margin of outcome. A topic of discussion ahead of conference play was the pliability of the NET, with Iowa State being a topic of conversation.
Image per Dave Ommen
The Cyclones forged a top 15 NET not through a difficult strength of schedule, but instead by beating up on bad teams, which was just dismissed as not mattering at all in the conversation of the Tournament.
Image per KenPom
Regardless of how the Cyclone’s season finishes, the NET is not validated through a strong conference performance, as 40-point victories in Q4 games don’t indicate that a team is going to thrive in the Big 12.
A reworking of the NET would alter the quads that each victory falls into, and while this isn’t going to be accomplished through a think piece, relying this heavily on what is a flawed and easily manipulated metric is an obvious flaw in the system.
With little reason to alter the otherwise undiscussed computer metrics, our team sheet is formed.
Our new team sheet provides an easier way to view the important games played by a team and provides insight into how each team dealt with their opponents. When presented with an opportunity, Houston didn’t play with their food, with six of their seven quad-one victories being won by at least five points. Houston was also not proven to be dominated in their mere three losses, with two of those results being moderate margin outcomes (one of which was on a neutral), and their worst loss being a close margin of defeat. Finally, we’re able to easily ignore almost half of Houston’s schedule, with 15 of their 17 Q3+Q4 opportunities ending in a blowout.
Looking Backwards
A model is nothing if it cannot be tested. Alongside my reverse engineered, this is one of two two brackets I’ll review come Selection Sunday. There are several unique aspects I’ll be valuing when I put our current to the test.
I may individually select conference victors, looking towards the methodologies I previously laid of the current model not valuing the regular season enough. If a team wins their single bid conference tournament in a format similar to the SWAC’s in 2022 (the conference has since expanded), that team isn’t guaranteed a bid in this model. Regardless of the outcome, games played to a close outcome are going to be valued much less than decisive victories.
I think it might be possible that our model doesn’t come close to matching the 68 teams that play in the Tournament.
Have you ever shot that shot?
-BM