We’ve gotten some clicks from an Illinois forum in the last 24 hours, and we had one guy mad-online in our emails about this, so it seems like a good topic for today. Why isn’t Illinois a 1-seed in our model’s bracketology?
Illinois is, in this morning’s update, the topmost 2-seed. Their median seed is the same as Alabama’s. Their mean seed is slightly poorer than Alabama’s. Our model puts Alabama (and Michigan, and Baylor, and Gonzaga) ahead of Illinois. It’s close. It’s very close. To our model, it’s just past a tossup. But our model does, narrowly, expect Alabama to end up ahead of Illinois in the seed list.
What’s holding Illinois back?
Some of this is what’s expected to happen. Our model relies on median simulations for each team, and in Illinois’s median simulation, they beat Ohio State on Saturday but lose in the Big Ten semifinals (a close-to-median simulation has them losing to Ohio State on Saturday but making the Big Ten championship, then losing there). One more good win, one more loss, while Alabama is projected to make the SEC Championship in their median simulation, thereby picking up more notches in the belt. Our model doesn’t measure things where they currently stand. If they did, it’s possible Illinois would be ahead of Alabama right now. It’s that close.
Why is it as close as it is, though, when Bracket Matrix almost uniformly has Brad Underwood’s team ahead of the Tide? Well, our model might be wrong. It’s a statistical model, and it’s growing every year, and it’s going to be wrong sometimes. But the reason it sees Illinois the way it does is the KPI gap between Champaign and Tuscaloosa.
Our model relies heavily on SOR, NET, and KPI, the three “reflective” metrics on the selection committee’s team sheets (NET is designed in a way that’s more predictive than reflective, except in the case in which a pandemic neuters the nonconference portion of the season for some and Colgate exists in a bubble nearly entirely cut off from the rest of Division-I and…etc.etc.etc.NET has some issues with this season in particular). It doesn’t rely on these metrics because we think the committee looks at them and nothing else. It relies on these metrics because, in our research, they’re indicative of how the committee thinks. They capture everything: nonconference strength of schedule, Q1 wins, Q3/Q4 losses, and all the rest. They don’t perfectly reflect the committee’s thinking, historically, but they do a decent enough job for us to use them as the base for our model’s seeding process. So when KPI doesn’t like Illinois, our model thinks the committee will have similar qualms, not because the committee’s looking at KPI, necessarily (though that is, to an extent, the case), but because the committee and KPI are looking at the same thing.
And why doesn’t KPI like Illinois? If I had to venture a guess, it has to do with margin. To my knowledge, KPI doesn’t consider margin of victory or defeat. So the fact that Illinois beat Michigan by 23 doesn’t matter to KPI any more than Illinois hypothetically beating Michigan by a free throw, and the fact that Illinois lost to Mizzou, Rutgers, Maryland, and Ohio State by an average of four points doesn’t discount those losses for KPI. In NET, those things matter immensely. In SOR, those things—this is my impression because I don’t know the guts of ESPN’s SOR, which is the one the committee uses—matter indirectly, because BPI evaluates them and the ESPN SOR is built off of BPI.
There’s a point at which public opinion becomes more predictive of the committee’s behavior than historic trends, but with the AP Poll not constructed in a way to reflect this, we don’t have a great way to measure it, which is a problem for our model in a few specific cases. This is likely one of them. If pressed, I’d say Illinois is more likely to get a 1-seed than Alabama, and we’re going to keep working on our model every offseason to better account for things like this. In the meantime, if you’re here wondering why we are directing such outlandish disrespect towards the fine basketball program in Champaign (we do like Illinois hoops here—got a big Route 47 contingent on the masthead), this is what’s going on. We’ll have a post up later this week on how much to trust our model, but consider this a prequel: The short answer is that you can trust it as a baseline, but that you should look for oddities the model won’t capture because we haven’t coded in the hundreds of weird little one-offs that can happen, due largely to a lack of historic data.
Today’s biggest movement:
Moving Up: UConn
Is this happening again? UConn up to a projected 8-seed, in Michigan’s region currently (real crapshoot there, so don’t read too much into it at this point), James Bouknight taking over the Twitterverse…
Would be pretty funny if they did it three times in eleven years.
Moving Down: Florida, Maryland
Florida’s loss to Mizzou wasn’t bad so much as unexpected. As we’ve probably made clear, our model relies on KenPom for the bones of the individual game simulations, so when KenPom says Mizzou isn’t that great, our model believes it and projects things like Florida being fairly likely to beat Mizzou at home. When it doesn’t happen, Florida dips. Cause, effect.
Maryland’s loss, meanwhile, which came at Northwestern, was both bad and unexpected. Welcome back to the bubble, Terps. A good side of it, but close enough to be very scared of what Penn State might do this weekend.
Moving In: Weber State (auto-bid)
Eastern Washington lost to Idaho State, and that really shook up the Big Sky picture. Yesterday, the Eagles were 38% likely to win the Big Sky Tournament in our model’s eyes, with the Beekeepers (come on, Weber State, do it—change the name) 30% likely. This morning, Weber State’s 32% likely and Eastern Washington’s 31% likely. Extremely close. Just outside day-to-day simulation variation. But a change nonetheless.
Moving Out: Eastern Washington (auto-bid)
And, in a corresponding move, we bid EWU at least a temporary farewell.
***
We got the Big Sky, Big East, MWC, and SWAC to almost-finalized in our model last night, which means there are only eight more conferences for which we haven’t eliminated un-rescheduled postponed games and locked in the conference tournament format: C-USA, Southland, Pac-12, SEC, Big Ten, MAC, WAC, AAC. Getting close. As we’ve been saying, as we make these changes within conferences, we’ve yet to see them impact our bracketology, but we do want to be transparent about where we’re at in the process, which is unique to this year.
Conference Tournament probabilities for those in action today and/or yesterday:
Atlantic 10 Conference
VCU: 27.2%
Saint Louis: 23.4%
St. Bonaventure: 23.2%
Davidson: 11.6%
Richmond: 6.9%
Dayton: 3.5%
Rhode Island: 1.6%
George Mason: 1.2%
Duquesne: 0.5%
Saint Joseph’s: 0.1%
The model, to be clear, does not incorporate injury information.
Atlantic Sun Conference
Liberty: 71.4%
Bellarmine: 13.4%
Lipscomb: 5.9%
North Florida: 3.6%
Stetson: 2.6%
Florida Gulf Coast: 1.6%
North Alabama: 1.5%
Kennesaw State: 0.2%
If you know what the A-Sun is doing this year for their auto-bid if Bellarmine or North Alabama wins, please let me know.
Big South Conference
Winthrop: 75.9%
Radford: 11.3%
Campbell: 8.4%
Longwood: 4.5%
Still very much Winthrop’s to lose.
Missouri Valley Conference
Loyola: 62.0%
Drake: 23.5%
Missouri State: 7.8%
Indiana State: 4.3%
Bradley: 0.9%
Northern Iowa: 0.6%
Valparaiso: 0.5%
Evansville: 0.3%
Southern Illinois: 0.1%
Illinois State: 0.1%
There’s a probability lesson to be had, for the select few who follow the A-10 and the MVC with some attention, in Richmond’s conference championship probability being lower than Missouri State’s.
Ohio Valley Conference
Belmont: 49.9%
Morehead State: 25.2%
Eastern Kentucky: 8.5%
Murray State: 6.9%
Jacksonville State: 6.7%
Austin Peay: 2.8%
What’s in a name? I saw Murray State close to Morehead State in a futures market yesterday. The Ja Morant effect, perhaps.
Patriot League
Colgate: 58.1%
Navy: 29.4%
Army: 5.3%
Lafayette: 2.9%
American: 1.3%
Loyola (Maryland): 1.3%
Bucknell: 0.9%
Boston University: 0.7%
The Boston University victory over Lehigh did not, in fact, shake this up.
West Coast Conference
Gonzaga: 82.6%
BYU: 15.6%
Saint Mary’s: 1.0%
Loyola Marymount: 0.2%
Pepperdine: 0.2%
Pacific: 0.1%
Santa Clara: 0.1%
San Francisco: 0.1%
San Diego: 0.01%
Portland: 0.00%
I’m curious how many simulations we’d have to run to get Portland to win this once.
Note: An earlier version of this stated that Illinois loses twice more in our median simulation. This is incorrect. In our median simulation, the Illini lose one more time as of the morning of this being published.
