Bracketology: How Are They Going to Seed Memphis?

Our NIT Bracketology, NIT/NCAAT probabilities, and NCAA Tournament Seed List are all updated heading into today’s games.

There isn’t much change in today’s NIT projection. Nobody moved in or out, and only Boise State and Wake Forest swapped seed lines, with our model now effectively projecting an NCAA Tournament bid thief when it effectively wasn’t projecting one before.

(1. The median NCAAT/NIT cut line moved from 46/47 to 45/46 after Gonzaga beat San Francisco, but it still sits very close to 45.5/46.5.)

(2. This affects Wake Forest because Wake does well in NCAA Tournament selection metrics and poorly in NIT seeding metrics. If they’re one of the NCAA Tournament committee’s First Four Out, they’ll receive a 1-seed. Otherwise, it’s probably a 2-seed, at least going by what we know now.)

That shook up some first round matchups, which in turn shook up second round matchups and quarterfinal matchups. But overall? The picture didn’t change much. Our bracketology will be most accurate this week at telling you who’s in the picture. It will be least accurate at telling you the specific matchups to expect. Whether to pair Akron with Texas, SMU, or Dayton in the first round is a judgment call.

So, instead of dwelling on the NIT bracket, let’s check one of our model’s more volatile outputs: NCAA Tournament Seedings.

As we get closer to Selection Sunday, we might move some teams around on our Seed List (though our NCAA Tournament Bracketology, which won’t be updated ‘til tomorrow, will always include our model’s true outputs). We’ll be transparent about who we moved and why, and each piece of movement will tie in with something we want to improve about our model next year. The goal with these changes is two-parted: First, it helps us contribute a more accurate projection to Bracket Matrix, which pulls from that seed list. Second, it helps us find things about our model we want to reexamine. Our model isn’t good enough to blindly trust. We know that and readily admit it.

Here, then, is everywhere our model deviates by two seed lines or more from yesterday’s Bracket Matrix projection, excluding teams where our model picks a different auto-bid than the matrix:

Before we dive into these five, a note about Drake from yesterday:

Yesterday, our model had Drake an 8-seed. The reason for this is that our model’s AP Poll approximation treats all wins and all losses as equal in the eyes of AP voters. Most of the time, this works disappointingly well. With mid-majors, though, especially those verging on low-majordom like Drake, it overestimates how much a team will move when it wins three games in a week. With the pre-Champ Week AP Poll finalized yesterday, we no longer have to worry about how our model treats future changes. This, of course, is something we want to improve next year. We want to stop systematically overseeding teams who are a lot better than the rest of their conference.

With that settled.

Let’s start with Memphis.

The short story on Memphis is that they’re a mediocre team with a good résumé. In KPI, SOR, and WAB, they average an 18^th-place ranking. In BPI, kenpom, and Torvik, they average 50^th. They’re 48^th in the NET, which is basically blurry kenpom.

If you count NET as a predictive metric (which it more closely resembles), our model’s preliminary NCAAT selection score is 97% résumé and 3% predictive. Its preliminary seeding score is 60% résumé and 40% predictive. The seeding score also includes a mid-major adjustment (which narrowly hurts Memphis) and the AP Poll adjustment (which helps, because Memphis is ranked 16^th). It’s said that selection is about résumé and seeding’s about how good you are, and we agree…to an extent. The question here is whether we’re overweighting résumé in seeding, and whether we’re specifically overweighting KPI, the largest single contributor to our seeding score based on its correlation to the 2021, 2022, and 2023 brackets.

We won’t make any changes today, but if Memphis is still an 8-seed in tomorrow’s Bracket Matrix, we’ll move them to a 6-seed. Their average Bracket Matrix seed is closer to 7 than 8 anyway, sitting at 7.35 (there’s a decent gap right now between the top 29 and teams 30^th and beyond). So we won’t take them all the way down to a 7, but the matrix has a much better track record than we do when it comes to seedings. Things to check here would be whether our model’s overweighting KPI relative to the other résumé ratings and whether our model’s overweighting the AP Poll, or overweighting it specifically with mid-majors.

With Louisville, we again have a team with better résumé numbers and a better AP Poll ranking than their predictive scores. The Cardinals are 13^th in the AP Poll. Our model thinks that matters. I’m curious, though, whether part of Louisville’s position is our model expecting the Cardinals to win another game comfortably. Maybe it’ll be proven right, but it’s possible it’s overestimating how much KPI, SOR, and WAB will change. If it’s also overestimating the importance of those ratings to committee seedings, then we’ve got an exponential miss.

If Louisville’s still a 6-seed in the matrix tomorrow, we’ll bump them down to a 5.

Arizona fits this trend as well, but in the other direction. The Wildcats are unranked, and their résumé ratings are worse than their predictive ones. If it sounds like our model is definitely overrating those résumé ratings and the AP Poll, consider the number of teams on which our model matches the matrix. It’s possible there’s something more nuanced than this at play. (But we’d be thrilled to either be proven right or to find an improvement this simple.)

If Arizona’s still a 5-seed in the matrix tomorrow, we’ll bump them to a 6 in our seed list, especially because they’re closer to a 4 on the matrix than they are to a 6.

BYU is tougher to read. Their résumé average and predictive average are closer to one another than is true of the other non-UCLA teams on this list. They’re ranked 17^th, which is probably part of it, but I’m still surprised the matrix isn’t higher on the Cougars. Our model only gives Torvik a 5% weight, which I’d worried was too small. (Torvik is new to the team sheets this year.) Torvik’s higher on BYU than anybody.

Our model doesn’t expect BYU to beat Iowa State in the Big 12 Tournament quarterfinals, for what that’s worth. So frankly, I’m a little confused here. It’s easier to come up with theories about why the matrix might be wrong than it is to find reasons our model is missing on BYU. Maybe this is because BYU’s won eight straight? Maybe the matrix is slow to react, with the more subjective bracketologists operating from a prior formed in January or February?

We will continue to follow BYU.

Last: UCLA. The Bruins sit between 23^rd and 27^th on every single ranking on the team sheet. They’re effectively 28^th in the AP Poll (our model cuts off the benefits at 25). But the Bruins have a winning Q1A record and a full nine Q1 wins. If I had to guess, this is the piece others like, whereas it’s a piece that our model doesn’t even consider. So, if we’re still seeing the gap tomorrow, we’ll move UCLA up to a 7, and we’ll make a note to spend time with some Q1A and Q1 data before next year.

**