Tournament Probabilities: Was the ACC Undervalued?

Our NCAA Tournament and NIT probabilities are updated through this weekend. A few high-level notes, to start us off:

• Before the tournament began, there was a 90.3% chance of the champion coming from the top four seed lines. That’s up to 95.6% now. So yes, the tournament’s been chalky, not only in its results but in their impact.
• Houston was a narrow favorite over UConn in our model’s pre-tournament simulations. That is no longer close to being the case. No team’s championship probability gained more by magnitude than UConn’s, and that’s independent of our model treating Boston (as it did Brooklyn) as a true neutral site.
• Duke has climbed from ninth-likeliest champion to fifth. Iowa State has dropped from sixth-likeliest to eighth.
• The probability of a full chalk Final Four (counting Arizona as the West Region favorite, as markets hold) is only 4.1%. Before the tournament, it was 1.6%. It is still very unlikely, but it is likelier than it was. Both those probabilities are lower if you’re looking at chalk as only 1-seeds.

Here’s how every team in the Sweet Sixteen has seen their championship likelihood change:

Again, UConn sees the biggest change in terms of magnitude. Proportionally, Clemson made the biggest leap. Iowa State lost ground due in part to its path getting much worse. The Cyclones were benefitting in our simulations from the presence of Auburn in their region. There was a good enough chance of Auburn knocking off UConn to significantly affect Iowa State’s title chance. With San Diego State in the Sweet Sixteen rather than the Tigers, that advantage shrinks, and Illinois both winning and playing well hurts the ISU situation as well.

That last piece—Illinois playing well—is relevant too. Here’s how every Sweet Sixteen team’s kenpom rating has changed since the tournament began:

Houston, the pre-tournament favorite and the second team to see its championship chance decrease, is the only team remaining in this tournament who has underperformed expectations, on aggregate, over its first two games.

To some extent, this is natural. As teams win in the NCAA Tournament, their performance ratings almost always improve. Why that happens: Most NCAA Tournament games have narrow spreads. Most games’ spreads are similar to the spread kenpom implies. Generally, covering the spread will be the result of overperforming kenpom’s expectations, something which improves one’s kenpom rating. In every game, either the underdog wins—outperforming expectations and seeing their rating improve—or the favorite wins. In games with narrow spreads, the window in which a favorite can win and not cover the spread is, naturally, narrow. Most of the time, someone outperforms expectations. It is rare to see a team underperform them and still win. So, even if the eventual NCAA Tournament champion has been a favorite in each game, it usually is better than it was or better than we thought it was at the moment the tournament began. We account for this by running our model “lukewarm,” allowing it to adjust its simulated kenpom ratings just a little as its simulated results come in. (Running it fully “hot,” adjusting ratings as much as they adjust in the real world, tends to overrate how much teams can deviate from their ratings in the long run, because Ken Pomeroy is really, really good at what he does, and his ratings have long-term predictive power.)

With Houston a significant favorite in both its games, the situation is different. Their window to underperform but still win was bigger. Failing to cover the spread last night dropped Houston’s kenpom rating. They only failed to cover by a little, and their rating didn’t decrease by much, but while teams like Clemson turned heads this weekend, Houston’s performance was relatively unsurprising, as were Iowa State’s and Alabama’s, and Houston’s two games combined, plus all the games of teams they played and teams those teams played (I believe kenpom is a Bayesian system), were the only ones that decreased the rating of one of the Sweet Sixteen teams. Marquette’s a bit of a different beast because of the Tyler Kolek injury situation (reaction: so happy to see he’s healthy; lost a bit of money because I didn’t think he was healthy), but their results had the same kenpom effect. Kenpom was more “right” about Houston, Iowa State, Marquette, and Alabama than it was about Clemson, Duke, and Gonzaga.

There’s been a lot made about the ACC going undefeated in the tournament’s first week if you conveniently ignore that pro-ACC arguers will soon again claim Virginia as having made the tournament. We’re salty about this, of course. “Manipulating the NET rankings” is the same thing as playing good basketball, and while the NCAA has done a lot to damage its own system’s credibility by offering little transparency and paradoxically releasing the rankings far too early in the year, that doesn’t mean guys like Brad Brownell and Jeff Capel should get away with saying stupid shit without us calling the shit stupid. But Brownell backed up the talk, and the fact is that even including Virginia, the average ACC team in the tournament was undervalued by kenpom by 1.27 points per 100 possessions. It’s only five teams, so it’s a small sample, and kenpom isn’t NET. Also, the tournament is far from over. But, we have an indication that ratings systems may have undervalued the ACC. If it happened across the board by that 1.27 points per 100 possessions number, then Wake Forest should have exited Selection Sunday 23^rd in kenpom, while Pitt should have been 32^nd. That’s eight spots apiece, and while tournament selection isn’t much about kenpom rating or NET (New Mexico was 23^rd in kenpom and was set to miss the tournament if they didn’t win the Mountain West Tournament), it’s possible Q1 and Q2 wins were missed out on because of an undervaluing of the ACC. Maybe the ACC would have gotten a sixth or seventh team. I personally doubt it. More likely, Clemson would have landed a 5-seed, which ironically could have been accompanied by a more difficult draw.

It’s unlikely the miss was as big as 1.27 points per 100 possessions, and to be clear, there’s nothing nefarious going on here, nor was a mistake necessarily made. With such a huge percentage of games played within conference after January 1^st, the last two and a half months of the season normally don’t manifest big changes in a league’s teams’ average rating. Points are handed off between teams within the same conference more than they’re handed off between teams across conference lines. A conference’s “total rating,” if you want to conceptualize it that way, is rather static. What 1.27 points per 100 possessions would mean, if it’s even real, is that the ACC got better over conference play to a greater extent than other leagues, and that ratings systems missed that happening because the only games being played were between conference teams.

What’s funny about this is that the list of teams who got significantly better doesn’t include Pitt, nor does it really include any of the conference’s four and a half NCAA Tournament teams. Duke and UNC each outperformed their New Year’s expectations by a little, and Wake did outperform its own by a lot, but Clemson, NC State, and Pitt all looked like about the same teams at season’s end as they’d appeared to be before conference play began. (Virginia, of course, looked much worse.) Who got significantly better in the ACC? It’s a tiny list. There are really just two teams on it. There’s Wake Forest, who we mentioned. There is also Notre Dame.

Notre Dame saw one of the biggest post-Christmas improvements in the country, if not the single biggest. Micah Shrewsberry’s freshman-packed team jumped from 223^rd in kenpom entering their conference opener to 120^th today, notching home wins over Virginia, Virginia Tech, Wake Forest, and Clemson along the way. 120^th is still bad, of course, and there’s no reason to think it should be higher, because kenpom is adjusting to the results of Notre Dame’s opponents even as Notre Dame sits idle. Wake Forest took a bad loss in South Bend. But if the ACC was undervalued this season, our best explanation for it is that Wake got significantly better after nonconference play, a lot of other teams got marginally better, few teams got worse, and Notre Dame took off. It’s hard to predict how good teams are at basketball. Even if kenpom missed by 1.27 points per 100 possessions on the ACC (and again, that seems like an outlandish, small-sample number), that’s only about a point per game, which is hilariously accurate. Kenpom is good at indicating how good teams are. NET is good at indicating how well teams have played. SOR and WAB are good at grading the difficulty of a team’s wins and losses. RPI is good at being easy to calculate (and little else). KPI is good at assigning a number to a résumé that doesn’t make us laugh it off the stage.

If the ACC was undervalued, it’s not because anyone was out to get the ACC, and it’s not because systems were wrong about the ACC. If a lot of teams in the same league get a little bit better after nonconference play, it’s very hard to tell that that’s what’s happening. The ACC’s issue was and remains its nonconference performance. I know Brownell loves to cite the league’s record against the Big 12, but in nonconference games prior to January 31^st, the ACC was 29–31 against power six competition, fourth-best among those six leagues. The Big 12 was 30–24, best among the six. (Credit: Jim Root.) I don’t think that’s a great way to measure league strength. It doesn’t account for each team’s position within the league (kenpom, indirectly, does exactly that). But if you use Brownell’s logic as your gospel, even then the ACC stacked up poorly. If—if, if, if—the ACC was undervalued, it was a weird bit of luck. Congratulations to the conference for proving so many doubters wrong in three games (Clemson x2, NC State x1) out of the two hundred or however many it’s played against other conferences this year.

There is, as you well know if you read this site, another postseason tournament going on that’s sanctioned by the NCAA. Three ACC teams played in it. Virginia Tech and Boston College narrowly exceeded expectations, while Wake Forest underperformed slightly on aggregate, likely due to the absence of Hunter Sallis yesterday in its home defeat against Georgia. This all impacts kenpom as much as other ACC games (and to be clear, it impacts the ratings of every other team VT, BC, and Wake played this year, as well as the teams those teams played, and so on down the line). It’s interesting that two of the three exceeded expectations, because it increases the sample size to a total of 15 games involving seven ACC teams, but it’s all still a small sample.

With Wake’s elimination, Ohio State takes over as the NIT favorite, narrowly favored over Cincinnati. Georgia has the worst chance of any team left in the field, and that’s Ohio State’s quarterfinal opponent, so don’t expect the Buckeyes to enter Indianapolis the favorite even if they take care of business against the Bulldogs. The winner of Indiana State vs. Cincinnati will most likely be the NIT favorite next week, followed by whichever three of Ohio State, Utah, Seton Hall, UNLV, VCU, and Georgia, probably in that order.

One thought on each of yesterday’s games:

• Jamal Shead is a great college basketball player, and it’s cool to see the offensive leap he’s made this year. It was not his best game last night, but when Houston needed someone, he was that someone.
• Colorado and Marquette played a pretty game, offensively. Tyler Kolek’s presence for Marquette, as we mentioned above, is among the biggest deals. Tough for Colorado, who had finally gotten its whole roster together and was playing like it down the stretch, but to be fair, they could have done a better job of their late-game management both Friday and Sunday.
• All the credit to Purdue—we know Purdue fans are wanting respect, and we’re more than happy to provide it—but also, credit to Utah State for continuing to be nationally competitive even as they churn through coaches. It is insane how many coaches have succeeded lately at Utah State, and how quickly they’ve done it. Three guys in a row, and two of them are doing decently well at their next stop. There’s no reason to expect Danny Sprinkle to not succeed at Washington, if that move is official by the time this is published.
• While Gonzaga’s overperformance of those kenpom expectations had something to do with playing a McCullar-less Kansas, Duke just dominated the first weekend. We’ve known they have this in them, and the question has always been whether they can do it consistently, and the answer to the question so far has been no. But they only need to do it one more time to have really made some noise, and while we still suspect that Houston will beat them up to the degree that they fold, we’re much more skeptical of our own take on that than we were a day ago.
• It was a problem for Baylor to be built entirely around offense and to shoot 25% from three in an elimination game. Credit to Clemson, though!
• It’s a little funny that Alabama and Grand Canyon, sloppy though they were early, looked like they were about to put together a great finish and then totally fell apart. Those final six minutes after Grand Canyon took the lead were as anticlimactic as they come. But for Alabama, like Duke, the consistency question is only two thirds as relevant as it was on Thursday.
• UConn is the best team in the country. We can acknowledge that. But: They haven’t played anyone in these last two NCAA Tournaments who finished the season ranked better than 8^th in kenpom. It’s possible they’ll pull that off again this year, and some of the effect is how badly they beat Gonzaga last year, but it’s impossible for them to play someone as bad in the Final Four as Miami was last year. They might get a good draw, but they won’t get a Miami draw.
• Hard to generate much of a thought about San Diego State. They throttled Yale. That is an observation.

For the Archives