The Silliness of Bracketology

We’re less than one month from Selection Sunday, which means the burgeoning field often called Bracketology is in full swing. Bracketology has taken on some broader meanings over the years, but it most often refers to predicting the selection and seeding of teams in the NCAA Tournament bracket. ESPN’s Joe Lunardi (aka “Joey Brackets”) has made a name and a living on his projections and there are now so many bracketologists that there is a site called The Bracket Matrix that collects all of them (dozens and dozens), displays them in a matrix, and grades them when the final bracket is released.

As a March Madness lover, I am a fan of most things involving the tournament and endorse almost anything that brings interest and discussion to the event. While predicting the NCAA Tournament field certainly falls into that category–and I myself have dabbled in my version of it–there are some aspects of the current state of Bracketology that range from misguided to downright silly.

Error #1: Bracketologists follow all the bracketing procedures

Fix: Ignore bracketing procedures and just put teams on their “true” seed line (or better yet, produce the full ordered 1-68 s-curve)

The Bracket Matrix states in its 2016 welcome post that aspiring bracketologists should read the NCAA’s procedures and principles for selecting, seeding and building the bracket and adhere to them when building their projected brackets. The most applicable parts of the document for bracketologists concern how to place teams in the bracket. For instance, there are rules to help avoid teams from the same conference being placed too closely together within the bracket (example: the first 4 teams from each conference must be placed in separate regions), minimize regular season rematches, and maximize the proximity of games for the top seeds (i.e. place higher seeds in geographically-favorable venues) while preventing any home-court disadvantages in the early rounds. The rules are somewhat convoluted and can be conflicting, but they strive to result in benefits for the overall product and more fairness for the teams involved.

The problem with adhering to them in a projected bracket is that we don’t really care about the little tweaks and adjustments that serve secondary purposes. What we primarily care about is the strength of each team’s resume and where we project that to place them in the bracket. Whether or not a team happened to fall victim to a technical bracketing procedure in one of the billions of potential bracket layouts doesn’t help us in our assessment of a team’s current NCAA Tournament case. It is simply noise and randomness injected into the bracket for no gain. It’s a large waste of time for a bracketologist to decide what region or site Baylor should be a 10-seed in. We really just care that they are a 10-seed.

In fact, it’s even worse than that! Teams can be adjusted up or down one, or sometimes even two (!), seed lines from their “true” seed line to satisfy bracketing rules. So, we start with a “true” ordering of the 68 teams, which is what we really care about. Then we spend an inordinate amount of time satisfying inconsequential bracketing procedures which just inject randomness into the result that we really want (and already have!) and end up with something not just the-same-but-different, but actually worse. That makes no sense, especially the further out from Selection Sunday we are when the bracketing procedures applied have an infinitesimal likelihood to be anything more than purely random.

Take a quick example. Let’s say that Iowa State is projected to be the 4th-highest Big 12 team and we project them to be a solid 4-seed. We place Kansas, Oklahoma, and West Virginia in their respective regions. Now, we are confined to just one region that Iowa State can be put in. However, we already have a 4-seed in that region, so we now must bump them up to a 3-seed or down to a 5-seed. Let’s say we put them at the 5-seed. After we fill out the bracket some more, we notice that they would be in a 1st round matchup that would be a regular season rematch, so we swap them with the 6-seed. So now instead of just making Iowa State the 4-seed in one of the other 3 regions and saying, “hey, in the real bracket they won’t actually be in the same region as West Virginia, but who cares”, we have spent more time to somehow end up with Iowa State as a 6-seed. Or they could have just as easily been a 3-seed. How does that add to our understanding of where Iowa State actually currently stands in (our projection of) the eyes of the committee?!

Which brings me to my next point…

Error #2: Bracketologists are predicting where teams currently stand

Fix: Predict where teams will likely finish, or at least what scenarios will lead to which results

While it’s certainly not useless to assess where teams stand in the present, we’re really trying to predict where they’ll end up on Selection Sunday. Purdue started off the season 11-0 and on December 12th probably warranted a #1 seed. But it was probably unlikely that a team barely ranked in the preseason Top 25 would end up there, and the Boilermakers have fallen off going 10-7 since and are likely headed for something like a 5-seed.

This error actually is perhaps more apparent towards the end of the season. For example, say that Gonzaga and Saint Mary’s meet in the WCC final, and a bracketologist projects that right now (before the final has been played) that Saint Mary’s is the auto bid and Gonzaga gets the final at-large bid. In reality, that Gonzaga projection is worthless. Either they beat Saint Mary’s and receive the auto bid, or they lose and fall out of the tournament. So Gonzaga is projected for an at-large bid, when in reality we predict they have no chance at all to receive an at-large bid!

More relevant and helpful would be a prediction of a team’s likelihood of receiving an automatic bid, an at-large bid, or no bid at all. This allows us to account for the quality of the team (like in the Purdue example above) and the effects of the team’s remaining schedule (like in the Gonzaga example) on their selection and seeding. The way I have chosen to present this in the past is to show a team’s (1) Auto Bid %, (2) At-Large Bid % when they don’t get an auto bid, and (3) the total In/Out %. Those pieces seem the most important–how likely is the team to make (or miss) the tournament, how often do they receive an automatic bid, and how likely–when they need it–are they to get an at-large bid. We can also project teams’ likely seed, based on all their potential outcomes weighted by the likelihood of each. These changes give a much more complete and accurate picture of a team’s true outlook for the big dance.

Error #3: Bracketologists are graded based on how many of the 68 tournament teams they correctly predict

Fix: Grade each bracketologist on a curve adjusting for each team’s selection chances, and add in grades on projected seeding

Imagine you’re in high school you sit down for a final exam for your math class. There are 68 questions. Before the test, your teacher hands you the correct answer to the first 31 questions. Another 30 questions or so are Kindergarten-level (2 + 2 = ?). The final 7 or so questions vary in difficulty from 2nd grade math to high-school level algebra. Oh yeah, and each question is multiple choice with just two options, A or B. You miss 2 questions but get to score a 66 out of 68 as your grade.

That’s exactly the environment bracketologists live in. They are trying to predict the 68 tournament teams. Thirty-one of them are automatic bids, and so are literally impossible to get wrong. By the time Selection Sunday hits, another large slew of teams will be “locks” even without auto bids–whoever doesn’t win the Big 12 out of Kansas and Oklahoma, for example, is definitely getting in. That leaves just a handful of teams for the final few spots and even among those, a few of them will be more likely than a coin flip. Wouldn’t you like your job performance to be graded like that?

It goes even further. Bracketologists are mostly graded on whether they put the team in or out of the bracket. In theory, they get credit even if they predict a team for a 3-seed but they get in as a 7-seed!

The fixes for this are simple. First, if we’re going to grade bracketologists on the percentage of teams they correctly predict to get in, at least take out the auto bids…and we’re probably safe to remove teams from at least the top 4 seed lines (and probably more) from the equation as well, as those are gimmes. But there are other relatively simple adjustments that can lead to better grades and let the truly strong bracketologists shine through. We can, for instance, take all the bracketologists from the Bracket Matrix and find how often each team was predicted to get in the tournament. If everyone correctly predicted a team to get in, no credit is given. Correctly predicting a team that was predicted by only half of the other bracketologists gets more credit, and more credit still is given to correct predictions with fewer bracketologists on the correct side.

But the best grading would be for each bracketologist to produce their final true s-curve list from 1-68 (and beyond for the teams they project to miss out on the tournament). The committee now releases their final 1-68 s-curve list, so we can easily compare bracketologists projections to the committee’s choices. This would yield the best possible evaluation of bracketologists’ true quality.

***

While I enjoy the role that Bracketology plays in the March Madness community and commend the hard work of bracketologists around the web, I hope the community can implement some of these changes so that bracketologists can spend their time more efficiently, giving fans more pertinent and relevant information and helping them gain a better understanding of each team’s true NCAA Tournament outlook, while also allowing the best of the Bracketology world to rise to the top and receive the recognition they deserve.

In fact, there is one site out there that has already implemented many of my suggestions–TeamRankings’ Bracketology. On their site, they project where teams will end up (not where they currently are) and display their selection chances–broken down by auto/at-large–along with other interesting information like seed breakdowns, tournament odds by # of wins, and even advancement odds for when the team does make it in the tournament. Check out their Bracketology team page for Butler as an example.

With just a few weeks left until Selection Sunday, Bracketology will be in the spotlight. Hopefully it will continue to evolve and add even further to the interest and excitement leading up to the March Madness season.

Category: College Basketball, descriptive, March Madness, review | Tags: , , , , , Comment »


Leave a Reply



Back to top