124: MOV
Let's talk about margin of victory, plus some random notes.
The start of conference play in basketball has brought a lot of blowouts.
Kansas beat UCF by 51 points on the road last week. It was the largest margin of victory in a road game in Big 12 history.
In the Big East last week, Villanova beat DePaul by 44 points. And earlier this week in the Big Ten, Illinois beat Penn State by 39 points.
The almighty SEC has blowouts too. Tennessee beat Arkansas by 24 points last week and lost by 30 points at Florida a few days ago.
Margin of victory is a hot topic. Last month, San Diego State head coach Brian Dutcher explained how he thinks about scoring margin:
“Point margins matter now,” Dutcher said. “You don’t go out there and run up and down and not care at the end, because every point matters metrically. Winning is not enough. Winning by a lot of points is what you’re trying to do at some level. Our first goal is to win the game. And then, obviously, a lot of teams are trying to win by larger margins because it affects the metrics.”
San Diego Union Tribune article by Mark Ziegler, December 20, 2024
So does margin of victory matter?
First, when we refer to metrics, we’re referencing everyone’s favorite college basketball sorting tool, the NET.
The exact formula of NET is a secret, however, I’ve recently found the closest public attempt to reverse engineer it. Brian Hare shares his take on the formula and explains the two factors of adjusted net efficiency and team value index.
Let’s focus on adjusted net efficiency as it relates to scoring margin. Hare points out that the net efficiency targets the difference in the score per 100 possessions.
My assumption is we all might be misunderstanding scoring margin as it relates to pace. Games are not played at 100 possessions, but when converting to 100 possessions, we can imagine how a 14 point win in 65 possessions is different than a 14 point win in 75 possessions.
Here is a sample of games with an identical scoring margin of 20 points that are all played at a different pace. Notice how that pace impacts raw net efficiency per 100 possessions.
Houston’s 20 point win over Troy featured around 55 possessions. The raw net efficiency is around +37.7 when calculating that number per 100 possessions or if the game extended to another 45 possessions, we would expect Houston to win by around 38 points.
While Ohio State beat Indiana State by 20 points, but the game featured around 79 possessions. The Buckeyes would only be expected to win by ~24 points if the game was extended to 100 possessions.
Of course, the raw net efficiency is not the only variable that requires an adjustment. The location of the game and the strength of the opponent matter too.
Hare runs a ridge regression to adjust for home court advantage and a method from Alok Pattani to adjust for strength of opponent. For example, Louisville’s 20-point win over Virginia was on the road, which means the Cardinals get an adjustment because it’s viewed as a tougher to win on the road than at home.
In his estimations, Hare weights the adjusted net efficiency at 80% and the team value index at 20%. The estimations are quite accurate from my findings, for example, for January 10 NET rankings:
157 teams have identical rankings (43.1%)
119 teams are within one spot of the actual NET ranking (32.7%)
50 teams are within two spots of the actual NET ranking (13.7%)
The largest delta is an eight spot ranking difference for Grambling State. Hare’s estimations show a 324 ranking, while the actual NET for today has the Tigers pegged at 310th.
As the season goes on, I plan to continue to tinker with these NET estimations. A few rough thoughts:
spending time on daily NET ranking changes is about as useful as weighing yourself every hour (it’s a waste of time)
expect the estimations to get more accurate with more data, especially as teams ramp up their games on the road
does the NCAA not share the actual rating of a team because there are two numbers (adjusted net efficiency and team value index)?
Anyhow, when it comes to the metrics or the NET, scoring margin does matter a lot. Dutcher is right, winning by a large margin matters, but that raw margin can be misleading if we don’t account for pace.
A team that slows the pace down and limits possessions with a large lead might be a sound strategy because of the raw net efficiency. There is also diminishing returns with that strategy, plus it’s terrible to watch.
As conference play continues this weekend, it will be fascinating to follow the average margins of victory. Over the past three seasons, the highest average margin of victory in league play was 16.3 points in the WCC. That margin was buoyed by Saint Mary’s +312 and Gonzaga’s +305 point differentials in league play last season.
So far this season, two conferences show an average margin of victory of about 16 points in league play (SEC 15.9 and Big 12 15.8).
Random notes and recommendations
Debugging the NET
If you’re curious in reverse engineering the NET, you can find Hare’s work in this GitHub repo. The code is in Python, but I’ve transformed it to R using a lot of trial and error. You can find the R code in this gist.
Modeling referees
Last summer, Connor (@cobrastats) and I put together a tiny app that tracked college basketball referee logs. One of the things we tried to tease out was the tendencies of different referees. We didn’t have a ton of success, but earlier this week, Ken Pomeroy shared a model to do just that.
Read more about Pomeroy’s tracking of fouls above average (FAA) here. For example, Paul Szelc has FAA of 1.7 or Szelc’s whistles on average ~2 more fouls than other referees. This kind of analysis is fascinating, and I hope to experiment with it more in the future.
State and UNC
Speaking of fascinating, did you know NC State and North Carolina have nearly identical records since the last time they played?
NC State beat North Carolina 84-76 in the 2024 ACC Tournament Championship game. After that game, the Wolfpack won four more games en route to the 2024 Final Four. Carolina lost in the Sweet 16 to Alabama.
Since that ACC Tournament title game, State is 13-7 and Carolina is 12-7.
UNC is 38-7 against NC State since the 2003-04 season, however, Carolina is just 3-2 over the last two seasons against the Wolfpack.
Curious if that trend continues when the two teams meet tomorrow in Raleigh.
Thanks for reading this far, and please subscribe if you so choose. You can find ode for the charts in the post here.
If you’re looking for more college hoops data, check out this tiny app. For example, did you know the ACC played 34 games against the SEC in non-conference play? Tennessee and Florida account for ~27% or 8 games of the 30 SEC wins against the ACC.
This is really interesting, because in isolation point differential is one of the most important stats out there to gauge the quality of a team. However, if a team knows that they're being graded by point differential and begins to play accordingly, it becomes a lot less meaningful statistic. Being an NFL guy, my mind drifts to the 1989 Cincinnati Bengals, who if you just look at top level numbers are one of the biggest outliers in the history of the league, finishing with a .500 record despite a point differential of positive 119 (7.44 points per game). That is, until you see that the team scored at different points in the season a 42-7 win, a 41-10 win, and a 61-7 win. Adding these three games together you get a point differential of 120.
Exempting these three games, the Bengals are still a team that underperforms their peripherals (a 5-8 record with a -1 point differential is not typical), but not one of the biggest outliers in NFL history, like they look on paper. Part of me wonders if college sport formulas that bake in point differential, given long enough, would incentivize every team to make themselves look like the 1989 Cincinnati Bengals. Every team would want to make themselves look like such big outliers, when in reality they aren't.