Hacking the Bracket with strength of schedule
When the NCAA field of 68 was set this past Sunday, perhaps the biggest story[1. Non-Charles Barkley Division] was the exclusion of Monmouth, a plucky team from the Metro Atlantic Athletic Conference who had gone out of its way to load its non-conference schedule with heavy hitters. The argument on Monmouth’s behalf was that small conference teams’ resumes will inevitably take a hit due to subpar competition within their own league, so a team that challenged traditional heavyweights like UCLA, USC, Notre Dame and Georgetown — let alone a team that beat them — deserved to be rewarded for their valor despite failing to secure an auto-bid in their conference tournament.
Ultimately, Monmouth found itself on the outside looking in, thanks to a trio of uninspiring losses to bottom-tier teams. However, despite the committee’s ruling, proponents of strong non-conference scheduling are right to be up in arms. Not just about the principle of failing to encourage aggressive scheduling by not rewarding a team like Monmouth, though that’s a valid argument in its own right. Why fans should be upset is that the committee is standing in the way of a team who, all else equal, is more likely to be a factor in tournament play than any of the similarly rated teams who will be dancing later this week.
You see, had Monmouth qualified for the tournament, it would have had the highest Non-Conference Strength of Schedule Pythagorean Rating in the field of 68. You may recognize NCSOS Pyth as the number all the way to the far right of the page on KenPom[2. Basic explanation here], but what you may be surprised to learn is that it has historically been a strong predictor of success in head-to-head matchups, with little regard for seed.
Looking at all 882 NCAA Tournament games played in the 14 years for which KenPom data is available, a trend starts to emerge. Specifically, teams whose NCSOS Pyth rating was at least .1 higher than that of their opponent tended to vastly outperform the rest of their peers — whereas teams whose NCSOS Pyth rating was at least .1 lower than that of their opponent lagged behind the average win expectancy rate.
Granted, even with nearly 900 data points to work with, the degree to which we’re filtering games doesn’t leave us with the most substantial sample size for any individual matchup. Furthermore, running a regression based on margin of victory only yields a correlation of 0.18. That having been said, the fact that the phenomenon appears consistently across multiple matchups seems significant. There is no clear explanation as to why the results would be binary (wins/losses) without any regard for margin of victory, but an educated guess would be that the NCSOS Pyth number is more a reflection of how tested a team is — that is, how well they have been prepared to face unfamiliar competition in a high stakes environment — which is not really reflective of the team’s talent level but may rather evince itself in how a team is able to close out a tight game in the NCAA Tournament.
Regardless, let’s just agree to couch these findings as less of a proven hypothesis and more in the realm of “things that make you go ‘hmm.’” If the trend continues to manifest itself, the committee may need to reassess what factors it deems most important to the resume evaluation process. But I doubt you’re here to talk about that. So, let’s get into what you really care about: How can this potentially help you build a better bracket?
Rock Chalk for the Jayhawks
If we are to believe in the predictive power of comparative NCSOS, no top-seeded team benefits more than the No. 1 overall seed in the tournament, the Kansas Jayhawks. Assuming Kansas can get past the 16th seeded Governors of Austin Peay[3. And yeah, let’s go ahead and assume that.], they will face the winner of the 8/9 matchup between Colorado and UConn. The consensus seems to be that UConn will emerge victorious between those two, and UConn over Kansas seems to have become somewhat of a trendy second-round upset pick. However, regardless of who the Jayhawks draw, history likes their chances to move on to the Sweet 16. Holding the +.1 NCSOS edge over both potential second-round opponents, Kansas slots into a group — teams with a 7- or 8-seed advantage who also held a +.1 NCSOS advantage — that has gone a combined 40-4 over the past 14 years. More specifically, Kansas would be the 16th top seed to square off against a significantly less tested team in the Round of 32. The previous 15 all moved on to the Sweet 16, while all other 1 seeds went a combined 33-8 in that same span.
The good news continues in the next round is, barring a surprise run from 12-seed South Dakota State, Kansas projects to have another matchup against a team with a far lower NCSOS Pyth rating. Teams who have held the NCSOS advantage over opponents 3 and 4 seeds lower have historically posted a win percentage between 25-30% higher than teams with the same seed advantage but no edge in strength of schedule. Picking chalk may not be much fun, but based on strength of schedule, no team appears to have a clearer path to the Elite Eight than the Jayhawks.
Danger Zona
On the other end of the spectrum, No. 6 seed Arizona seems to be a popular dark horse pick for the bottom half of the South region. And on paper, it’s hard to blame anyone for picking them to play deep into the second weekend. After all, head coach Sean Miller has been largely successful in past tournament stints, and no one really trusts No. 2 seed Villanova to live up to expectations. That having been said, Arizona doesn’t appear to have done itself any favors in non-conference scheduling this year, ranking third-worst among tourney teams in NCSOS Pyth. The Wildcats will be in danger right off the bat, taking on a thoroughly battle-tested Wichita State team fresh off dispatching Vanderbilt in the First Four.[4. Sigh…] Even with a five-seed advantage, Arizona will be in for some tough sledding if history is any indication. Teams with a five-seed advantage who are -.1 in NCSOS are actually below .500, while all other five-seed favorites have posted a .722 win percentage.
If the Wildcats can survive a potential shocker[5. Sorry.], they’ll face another opponent to whom they will cede the NCSOS advantage, meaning they’ll be facing long odds against No. 3 seed Miami (4-28, historically) or a surprisingly tough draw against No. 14 seed Buffalo (8-9). And just in case they manage to win their first two games, there’s a good chance they’ll run into Villanova (1-12), Iowa (11-23) or Temple (7-5). Back the Wildcats at your own risk.
Good Dogs, Bad Dogs and Difficult Decisions
The last thing we’ll want to look at it is what NCSOS Pyth says about the games that will likely determine who gains the upper hand in your March Madness pool: first-round upsets and pick ’em games.
On the upsets side, take a good, long look at 11-seeds Wichita State and Gonzaga, both of whom not only project to have a good shot of knocking off their first-round opponent but both of whom could actually prove troublesome for the No. 3 seed as well. The lone 12/5 matchup that the numbers like in any direction is No. 12 seed Chattanooga over No. 5 seed Indiana. Conversely, history doesn’t bode kindly for upset-minded No. 13 seeds UNC Wilmington and Hawaii, who are hoping to do what only two of 14 teams have done and pull off a none-seed differential upset in a game where the opponent is +.1 in NCSOS.
As far as toss-ups are concerned, there are disappointingly no 8/9 games that the numbers have too much to say about, but there are a few potential matchups later on down the line that are worth consulting the data for. The numbers like No. 4 Duke over No. 5 Baylor, No. 4 Kentucky over No. 5 Indiana and *gulp* No. 1 Virginia over No. 2 Michigan State. Or, if your bracket happens to take some strange turn, consider: No. 7 Dayton over No. 6 Seton Hall; No. 7 Iowa over No. 6 Arizona; No. 11 Northern Iowa over No. 10 VCU; No. 11 Tulsa OR No. 11 Michigan over No. 10 Pitt[6. Regardless of who wins tonight, both would have the advantage in an unexpected 10/11 showdown against Pitt]; No. 11 Gonzaga over No. 10 Syracuse.
Again, none of this is meant to be taken as the gospel. It’s fully well possible that Arizona could make a deep run, much as No. 3 seed Notre Dame did last year despite several historically unfavorable draws. Kansas could buck history and flame out in spectacular fashion. All we have to guide us is what has happened in the past. The goal here is simply to re-evaluate these matchups through the lens of NCSOS Pyth. The more data we are able to gather, the more we can begin to contextualize what is really happening here — and see whether strong non-conference scheduling actually indicates as much of an advantage as the recent data seems to suggest.