"My rule was I wouldn't recruit a kid if he had grass in front of his house.
That's not my world. My world was a cracked sidewalk." —Al McGuire

Monday, January 22, 2024

Bracketology: Forget About NET

As rankings, seeding, and brackets are discussed, we come to the part of the calendar where the NET is argued over and complained about. The NET replaced RPI in 2018-19 and has been misunderstood ever since. Pundits repeatedly respond to the NET rankings as being flawed. They talk about how any system that ranks <Insert Team Here> at <Insert Rank Number> must be broken. But every year when Selection Sunday comes around, there are vast discrepancies between the NET Rankings and the official seed list. Last year, NET #9 Kansas was #3 on the Selection Committee's S-Curve. NET #13 FAU was down at #33 on the S-Curve while NET #18 Utah State barely made the field as #40. In the other direction, #24 Kansas State was #11 on the S-Curve while #67 Pittsburgh was the worst NET but earned #44 on the S-Curve.

What does that mean? Yes, the NET is flawed, as is every other ranking system that tries to evaluate college basketball teams. But the point of the NET is not to be perfect, it is to put data in context and make it easier for human eyes to evaluate.

This is the case every year and has been going back to the RPI. The reason for seeding discrepancies is because while the NET does order the teams in the list, it is not used like a ranking system by the NCAA. If it were, we could just punch the tickets of the 32 league champions, then take the top 36 non-champs from the NET rankings, sort them in NET order, and we'd have a Tournament Field. This whole Bracketology project would be much easier, that's for sure. But that's not what the NET is for.

The NCAA has said it repeatedly: the NET is a sorting tool. But what they don't say often enough (and should) is they actually use it for sorting opponents. NET doesn't get you in or out, it tells you where your opponents fall in Quadrants (the boxes on the team sheets where teams are sorted). When I build out a bracket, I only use NET when I am putting together my candidate list because Team Sheets are listed in NET order. Once I have teams on the spreadsheet, a team's individual NET loses all meaning. The only way NET is used beyond that point is for what your opponents' NET rankings are, which is how teams are sorted into Quadrants.

Looking at Marquette, their NET of 17 doesn't matter nearly as much as the NET rankings of teams they play. When the NCAA calls NET a sorting tool, it is sorting teams into the four Quadrants at the BOTTOM of the team sheet, and subdividing Quadrants 1 and 2. More important to Marquette is the ranking of opponents like Illinois, Creighton, and Kansas, who are their best wins, because it is the rankings of those teams that give Marquette three wins in Quadrant 1A, which are the best wins a team can get. Similarly, Butler's #66 NET is a negative for Marquette and if it continues to fall, having a Q3 loss would be more damaging to Marquette's resume than whether their NET #14 or #24, particularly as last year the team ranked #14 was 5-seed San Diego State while #24 was 3-seed Kansas State. Why was that? San Diego State had just 5 Q1 wins while K-State had 9. Their individual NET rank didn't matter.

NET is also primarily an efficiency model. So, like kenpom, its rankings are based primarily on how efficient a team is on offense and defense. So if you score a lot and don't allow many per possession, you will be rewarded, especially if you do that against quality opponents. It does also reward teams that run up scores on lesser opponents. I know some people have a problem with this, but because NET is opponent-focused, all it needs to tell you is when you beat that team in that Quadrant, what does that team look like on the average night? It weights the efficiency of performances, averages them out, and then ranks teams on that basis.You might get the occasional outlier like an Alabama whose NET is higher than their record would indicate, but as that doesn't give them a seed-line boost (#8 NET Alabama is #22 on today's S-Curve) so the only people it actually benefits are Alabama's opponents. The teams that beat Alabama did beat a team that on the average night plays really well. A team like Creighton might benefit from that win over Alabama, but as that single result only accounts for 1/18th (so far) of Alabama's overall season and NET ranking, it really isn't statistically significant that Creighton benefits from Alabama's overall performance against a top-5 schedule.

Another complaint about NET is how much teams move. The reason teams move like they do is because NET has a team score, just like kenpom uses Adjusted Efficiency Margin to order his teams. But unlike kenpom, we don't see what those team scores are. Let's look at the current top-20 in kenpom as of January 21:

Let's say that Marquette had a massive victory that moved their Adjusted Efficiency Margin up 3.0 points. Looking at the AdjEM column, that would push them to 24.22, which would take them from #17 to #9, an 8-spot jump (tied with Illinois). However, if #2 Purdue were to do the same thing and move their AdjEM from 30.25 to 33.25, they would still be ranked #2 because the gap at the top is that much wider. If you go further down the rankings, that 3.0 difference makes for even wider gaps. If Brian Wardle's #75 Bradley (10.99 AdjEM) bumped up to 13.99, they would jump 25-spots. It's even more stark in the middle of the rankings, where #195 Northern Colorado (-2.44 AdjEM) would skyrocket 42 spots to #153 if they saw a 3.0 bump to 0.56.

With kenpom, we see those differences. But with NET, while those numbers exist, they are hidden so we only see the ranking. That's how this past weekend you could see Creighton win by 3 at #56 Seton Hall and move up 5 spots from 16 to 11 while Marquette won by 1 at #36 St. John's and stay the same at #17. While we don't see a rank change, behind the scenes, the odds are both teams moved a similar amount in terms of their hidden rating, but because of gaps like we see above, the same movement that pushed Creighton up 5 spots kept Marquette in the same place. In addition, NET is influenced by the teams around you, so if Creighton moved past 5 teams but 3 of them lost that day while 2 didn't play, it was easier for the team that started a little bit ahead to make a big move while a comparable move didn't have a similar impact.

What are the takeaways? First, NET does not correlate to the S-Curve or Seeding, so don't overstate its importance. Second, the NET of your opponents is far more important than your own because it will determine what quadrants your wins and losses are, which actually does impact seeding. Third, don't pay close attention to the movements because while you might see a rank change, that doesn't give you context of the associate rating change or what happened to the teams around you.

Enough about the NET, let's get on with where we stand today:

Multibid Leagues

Big 12: 10

SEC: 8

Big East: 6

Big 10: 6

Mountain West: 5

ACC: 4

Pac-12: 3

American: 2

No comments: