People misunderstand NET's use as much as they misunderstood RIP's use.
Edit: "RPI"
Original text remains for entertainment.
Edit: "RPI"
Original text remains for entertainment.
Last edited:
NET Rankings are Flawed
- Thread starter BillBrasky4Cy
- Start date
But several games ago ISU and Iowa had very close net efficiencies. ISU had 7 q1 wins, Iowa had 0. ISU had a better record and SOR. Yet Iowa was still several spots better in the NET.
ISU was as good or better in every way, including the all critical wins vs. Quad 1, EXCEPT for offensive efficiency. I don't really need to see inside the black box to very safely make the assumption that offensive efficiency, or offensive performance is weighted VERY heavily.
It will be interesting to see how it shakes out. And I get that NET (or KP for that matter) are not going to be followed explicitly to seed teams. However, I have a tough time believing that Iowa's lofty NET and KP rankings are not going to be a boost for them despite actual w/l.
These ranking systems generally give the same weights to the cupcake games as they do against peers.
Imagine two teams. Let's call one State and the other Uof.
State has some squeakers against some bad cupcake teams on its schedule, but ultimately wins them all. We completely forget about those games -- W is a W -- and they don't indicate much.
Uof trounces the cupcakes. Even their fans don't try to argue that proved much of anything, though.
But the computer rankings look at that and say... well, there were 5-6 games back in the fall when Uof played better than State, so maybe Uof is a better team based on the indications of those results.
I think all of us can see the problem here.
There is no more weight to offensive efficiency, all points are created equal, it doesn't matter if it's one more point scored or one less point given up.
The same argument for why Iowa football should not have been #2 in the country (squeaked by opponents, forced a lot of turnovers, terrible offense, but hey they got the W) are being used by people to say Iowa State basketball should be ranked higher in bracketology.
Trying to compare these two is...not possible when Iowa State has actually beaten good teams. Nice try though.
You're totally missing the point. How can you compare FB and BB when we're talking about metrics like NET? It's a fair attempt...but not working for you.
What about the game where Uof played State and State beat their breaks off?
Let me try again.
Iowa was a good team by metrics and a great team by record in football early in the season.
Iowa State is a good team by metrics and a great team by record in basketball.
Here is what SP+ looked like when Iowa was ranked #2 in football.
What about the game where Uof played State and State beat their breaks off?
As demonstrative as that game was it only made up 1/6th or so of the possessions State had against the JV team of the Indiana University Purdue University-Shipshewana Barnbuilders.
So it only counts for 1/6th the weight.
I could be wrong, but I don't think ISU has been much above 40 in KenPom or related metrics all season. We've had bad efficiency metrics all season and Iowa's have been quite good.
There is a lot of consternation about NET on this website but quite frankly I think it's working exactly as it is intended to work. It's clearly an efficiency metric. I like Warren Nolan's team sheets because they give you 6 ratings and break them down by the type (predictive vs. results). I don't think there is much more to it than that. Again, look at Iowa - SOR and KPI (results) are 44 and 50. BPI, KenPom, and Sagarin are 19, 21, and 20. NET Ranking? 21. It's basically perfectly aligned with the efficiency metrics and completely out of sync with the results metrics. That should really tell you all you need to know about it.
Now, if you want to know my opinion, having a ranking system that HEAVILY favors efficiency over results, for a tool designed to put together the NCAA field, is absolutely absurd but to pretend like it's broken or something is incorrect.
Is this explicitly stated by NET? I am interested to have an explanation how a couple weeks ago ISU and Iowa’s offensive and defensive efficiency rankings were essentially inverse one another, ISU had 7 Q1 wins while Iowa had 0, and Iowa could be higher in NET.
Intended or not, offensive efficiency is being weighted heavily. If it is a pure ppp difference, then it is certainly weighing offense more heavily. The same applies to a ppp basis, but it is easier to demonstrate on a total score basis, and it would correlate extremely well.
Think about it. A team that is great on D and bad on O wins 56-44. That’s a dominant win. Margin of victory is like 30% of the losing teams point total. That is domination.
Conversely, a team that is great on O and bad on D might win 98-86. That’s not nearly as dominant a win. Yet since net efficiency on a ppp basis and margin correlate almost perfectly, the two games would look the same in net efficiency. Or put another way, high scoring games tend to have higher margins of victory simply because you are dealing with larger gross numbers.
If you really wanted to eliminate bias toward offensive teams, you would use net efficiency on the basis of relative ranking, not the raw ppp numbers.
NET Efficiency is just points per possession minus points allowed per possession. That portion of things doesn't appear to take into account opponent.
The problem I have with criteria's 2 and 5 is that you are rewarding teams for running it up against sh!t competition. When you look at Iowa's non con it's pretty clear Fran had this figured out because outside of the Cyhawk and ACC challenge Iowa played a pathetic schedule. Notice both of these games were out of his control... I don't care which way you cut it, at some point a win is a win and the efficiency jazz goes out the window. Not all schedules are created equally and that needs to carry some weight.
A models purpose is to predict outcomes when insufficient data is present. In the terms of KP or NET they exist to tell us who is best/better. Otherwise it serves zero purpose. Efficiency for efficiency’s sake is meaningless and worthless. We have thousands of CBB results data points. The predictive tools that use efficiencies are no longer needed by this point in the season.
Moreso, in every other walk of life, whenever available, real outcomes are used to validate models. If the models aren’t very good at reflecting the actual outcome, then the model is in fact considered broken.
Analytics dorks want to try to pump up the use of these models, and when they churn out nonsensical results they like to throw out the “that’s not their purpose.” A model that churns out nonsense is either a broken model, not relevant, or possibly both. They are interesting early in the season when there are limited games between good teams. But nearly 30 games these serve limited purpose.
Broadly, you're right that Iowa State's efficiency metrics have lagged behind actual rankings. But just to note, Iowa State did get as high as 24 in Kenpom, which was before the Kansas home game. Even before the game in Morgantown, Kenpom had them 29, then that game sent them tumbling to 41. Barttorvik has been more favorable and was in the teens for quite a while, as good as 12 after the Texas home game.
