Anyone still defending Net rankings should just stop. Mich st is 25 in Net despite being 12-8, 4-5 in conference. It’s a joke
2023-24 NET ratings thread
- Thread starter Styre
- Start date
Michigan State’s losses include #7 Duke, two to #6 Wisconsin, #11 Arizona, at #14 Illinois, at Nebraska, and at Northwestern. That’s five top 15 teams and two tourney bound teams who are very strong at home. They also blew out #18 Baylor.
KenPom has them at #18, and while I’m a little surprised they’re quite that high given they’re 1-5 vs the top 20, their schedule is a great example of all losses being far from equal.
Any system they values “good losses” more than wins is completely flawed. MSU is 34 spots higher than NU despite NU having more wins, better wins, fewer losses and beating them by double digits. I think we try to make things more complex than they Need to be.
If only Northwestern had beaten Chicago State as badly as they beat Michigan State.
If one bad game can have that big an effect, point stands. Spectacularly flawed.
You can whine about it all you want, the reality it is HIGHLY HIGHLY successfully predictive, which is what people are using it for. People, by the way, like Vegas line setters and larger gamblers who have many many millions of dollars riding on the line of knowing what they're doing on this.
NU also isn't being pushed down by just one game, it's really how the team pattycaked through the entire extremely squishy nonconference schedule with a bunch of way too close wins over very bad teams.
Two issues with this comment.
First, curdog is not "whining." He is offering completely valid criticisms and you are re-characterizing them (incorrectly) as "whining."
Secondly, the Ken Pom #'s are not "HIGHLY HIGHLY successfully predictive." They are used to set betting lines and over/unders. Predictive of the betting lines? Sure. Predictive of the results of the games? NO WAY. The standard error for (Actual margin - expected margin) is on the order of 10 points per game. That is a huge error.
I know you understand enough about statistics to understand that any individual game has a massive bell curve of possible outcomes and thus a highly accurate prediction is merely seeking to find the midpoint.
We are WAY past the point in the sports analytics debates where burying your head in the sand and just getting mad about MATH is anything more that being obstinate.
Again, a ranking system that has msu 34 spots higher than NU is absurd. Defend it all you want. Doesn’t make it any less absurd.
I'm not suggesting that MSU deserves to be 34 spots ahead, but:
MSU
Top 100 Wins - #16 by 24, #37 by 12, #52 by 2, #54 by 20, #84 by 10 and #95 by 18.
Other wins - #115 by 31, #150 by 17, #208 by 44, #301 by 31, #310 by 32, #335 by 23
Losses - #5 by 6, #10 by 13 and 15, #11 by 3, #13 by 9, #47 by 14, #51 by 7 and #75 by 3 in OT
NU
Top 100 Wins - #2 by 4 in OT, #11 by 5 in OT, #18 by 14, #25 by 5, #52 by 3, #60 by 25
Other wins - #111 by 19, #115 by 4, #196 by 11, #267 by 11, #282 by 10, #284 by 4, #288 by 22, #300 by 11 and #358 by 32
Losses - #10 by 8, #11 by 30, #32 by 9, #51 by 6, #291 by 2
They've played a harder schedule than us, beaten the crap out of every team ranked 95 or lower, all but one of their 12 wins has been by 10 points or more, and didn't lose to Chicago State at home.
You keep trying to defend the indefensible. In no universe should MSU be ahead of NU (as evidenced by both the AP poll and coaches poll where NU has received votes while MSU has not). So MSU beat bad teams by a lot. Great. They have one (!) win over a top 25 team. NU has 3 and beat sparty head to head.
Curdog...I'm on your side on this, believe me. I believe that NU is better than MSU for all of the reasons you mentioned.
However, the data suggests that MSU is "better" than NU over the course of the whole season, independent of the actual results of the games. We play to win the game, and NU is better at winning, at least so far, but one way to look at it is NU is playing much better than the data suggests (because more wins) and MSU is playing much worse than the data suggests (because fewer wins). Doesn't mean NU is lucky and MSU is unlucky, just that whatever the data is trying to measure (such as the predictive value of all games against all teams over an infinite timeline), MSU is "better."
If NU continues to beat MSU in W/L percentage and head-to-head, it could mean any numbers of things, including "garbage-in, garbage-out," luck, and just that NU is "better," etc.
However, the data suggests that MSU is "better" than NU over the course of the whole season, independent of the actual results of the games. We play to win the game, and NU is better at winning, at least so far, but one way to look at it is NU is playing much better than the data suggests (because more wins) and MSU is playing much worse than the data suggests (because fewer wins). Doesn't mean NU is lucky and MSU is unlucky, just that whatever the data is trying to measure (such as the predictive value of all games against all teams over an infinite timeline), MSU is "better."
If NU continues to beat MSU in W/L percentage and head-to-head, it could mean any numbers of things, including "garbage-in, garbage-out," luck, and just that NU is "better," etc.
I hear you smelly. It’s clearly gsrbage in garbage out. The model is placing way too much emphasis on margin of victory (does it matter if you win or lose by 10 or 20). The BCS dropped margin of victory for a reason. Maybe Net will do the same.
It doesn't use margin of victory, it uses net efficiency. It just works out that generally if you have a very efficient game on offense and on defense you tend to win big.
So it uses something highly correlated with margin of victory. Got it.
I’m sure you also understand that outliers happen. Simply saying well “MATH” is lazy. Especially when the criticism is valid.
Better way to do this would be to eliminate the two outliers (best and worst wins).
If the point of the game is margin of victory then fine. But if the point is actually winning then what you laid out is inane.
But nobody is doing that (burying your head in the sand and just getting mad about math).
We're trying to discuss the strengths and flaws of the NET and KenPom and Torvik.
I looked into KenPom. It ain't that complicated. I can approximate his results.
Its an algorithm. The algebra will perform flawlessly, but the algorithm can be badly flawed. KenPom has some glaring weaknesses, like weighting all games the same.
But the estimated game scores (predictions) are not particularly good. If I say NU is going to lose to MSU by 3 and NU wins by 7 and that sort of thing happens over and over, how can I claim that my predictions are good?
I'm quite confident that the big (savvy) bettors use KenPom as a starting point and look to bet against.
I think what people need to realize is that most of these rankings are backward-looking error-minimization routines. They correct themselves as the new scores inform the system where it was previously wrong.
There are many ways to do that. It is good to talk about the possibilities.
Don’t even think of it as a model. In reality the math isn’t that complicated at all, you can calc the impact of each game pretty easily. Also don’t think of it as a ranking system, it isn’t really (though unfortunately the NCAA uses it that way).
Think of it as a statistic that describes your average margin of victory (or defeat), adjusted for quality of opponent, home or away adjustment, and scaled up to 100 possessions (most games are in the range of 65-70 so this is a multiplier of roughly 1.5). It ignores W or L, and yes it separates out offense and defense for the sub scores. But the top score is basically just your average margin, with a +/- for H or Away, adjusted for opponent quality, multiplied by roughly 1.5.
The model simply runs a massive linear algebra equation to minimize the errors across NCAA every game played. It’s not that complicated.
And yes I agree NU should be ranked ahead of MSU, but if you think of as a descriptive (and maybe predictive) statistic rather than a “ranking”, maybe that helps contextualize KenPom better. PS I also don’t think KenPom should be a direct input into the NET, but it is, so (i) we should schedule tougher OOC with more road games, (ii) unfortunately we kinda need to blow out the bad teams, it might be poor sportsmanship but the incentives are clear.
Its a bit of smoke and mirrors.
Efficiency (Ken Pom's term) is Points Scored per Possession.
I'm not sure how significant the difference is between Points Scored per possession and Points scored per minute when it comes to estimating if a team is good at scoring points.
My guess is that its primary appeal is that if you count possessions you can create another new stat called "Pace of Play." Which is a nice stat, but does it have predictive power? Not so sure about that.
Yes, this is all correct. I have no doubt the math checks out but just going off how teams performed at the possession level without factoring in the actual game results (you know, the whole point of playing the games) seems ridiculous.
If Net was just used as a predictor of future results, fine. But it is being used as a way to assess how good a team is relative to other teams by the committee. And it is clearly wildly flawed.
Btw the thing that surprises me more than KenPom is that our NET ranking is even lower than KenPom (59 vs 47 last I checked, also 40 for Torvik). So we are leaking value even further beyond what the predictive efficiency metrics are saying.
Given our quality wins and 15-5 record, I wouldn’t have thought it’s the Team Value Index component that is doing that, although the weak OOC schedule is certainly weighing things down. But I guess it must be. And I think the lack of road wins for the adjusted winning % metric is hurting us too.
I also think the lack of transparency around the NET rankings absolutely sucks. I have looked into finding the details of it. This is the most recent thing I can find from the NCAA. The headline laughably claims that it “explains” the NET rankings. All it does is tell you the types of things that are incorporated. It lists 4 components but doesn’t tell you the weights of those components, or details behind most of the calculations. I don’t believe (let me know if I’m wrong?) that they publish any of the ranking by those 4 sub components either. What is NU ranked by their so called “Team Value Index”, by their Efficiency measure (which might be KenPom, Torvik, a combo of those, or something similar?), and by Adjusted Winning % (at least that one is fairly clear though and should be relatively easy to calculate)? They should publish the rankings of those sub components every time they update the NET so that teams can be better informed about what is driving their ranking and adjust scheduling accordingly (or try to blow teams out more at end of game situations). The lack of transparency stinks IMO.
Back in the days of the RPI it was a flawed metric for sure (more flawed than KenPom IMO) but at least it was simple and everyone knew how it worked.
College basketball's NET rankings, explained
Here's everything you need to know about college basketball's NET ranking system.
www.ncaa.com
I agree with you here. Except I don’t think KenPom is flawed really, it’s more the way the NCAA is using it is flawed.
It isn't "flawed" in that it is a baseline algorithm that does what the author told it to do.
It has some logic to it.
But the logic itself is almost rudimentary.
I love how dorky we all are, and I say that out of love.
The real question is that are the things it is using as inputs to its formula the right things or given the right weights? I'm sure a lot of data over many years is fed into making its baseline formula "right," but what that question presupposes is...what if it isn't?
The real question is that are the things it is using as inputs to its formula the right things or given the right weights? I'm sure a lot of data over many years is fed into making its baseline formula "right," but what that question presupposes is...what if it isn't?
This was the thing that stunned me when I read it...
It looked like the NCAA backtested on something crazy like 2 years of tournament games when they first came up with their system.
Not the regular season. Just 100 - 150 postseason games.
Because they wanted to see what criteria from the regular season produced the best results in the tournament.
After a few years, they chucked some of their initial factors entirely.
Wow...so when there's just one day of rest sometimes, which almost never happens in the regular season.
I wonder if "days of rest" is factored into the numbers in general, even if just a little bit. Like NU get one fewer day of rest than Ohio State last week, so if they'd been the same, NU would've won by 40 instead of 25.
I wonder if "days of rest" is factored into the numbers in general, even if just a little bit. Like NU get one fewer day of rest than Ohio State last week, so if they'd been the same, NU would've won by 40 instead of 25.
There is some semi-revisionist history going on with how NU handled its early season schedule. They were not good. Yes, they played great against Purdue and well against Dayton but they were awful against bad teams.
Look at the Western Michigan game - a 63-59 win over a team that is 8-12 - Kempom 285 - that game was wretched to watch. NU shot 45% from the field and 23% (6/26) from three. Defense was a mystery.
This was a pattern that continued at home against Chicago State and others. I remember thinking as NU headed to Illinois that they had played about 5 halves of good basketball - against Purdue and Dayton and then the second half against NIU. Other than that, it was ugly. NU ended up getting smacked upside the head by the Illinois.
Not sure exactly when things started to change but when you compare the early season games to what has happened recently, I think its fair to give the computers a break on judging NU. This is a team playing much better basketball but they deservedly must own the crap hoops they played in November and December with only a few exceptions. That said, if NU keeps winning and playing like they are, the rest will take care of itself.
I agree with your comments, but there's no "semi-revisionist history" going on.
I think most fans who understand the ratings also understand our relatively low rating.
But it is like a professor trying to reach a final grade for a student.
We had 10 tests and you got a 95 on 9 of them, but this other one you got a 60.
95? 91.5?
Its up to the professor.
Common sense tells you that a straight average of all results is a little simplistic and overly punitive.
Kids get sick, basketball players get injured, etc. People are not robots.
Toss the outliers? Weight them lower?
Winning twice by 1 is a lot better than winning by 6 and losing by 4. But not to the NET.
Depends on the team. There are outliers like Virginia in recent years who are often highly efficient but low-scoring thanks to a very slow tempo. But generally it looks like the difference isn't that significant.
There's nothing revisionist. Many of us just believe a win is a win is a win. And while I agree that strength of schedule should matter, I outright reject the idea that a 30 point win should count significantly more than a 5-point win. It's not that we're forgetting the start of the season or that we don't understand our relatively low rating. We just reject the criteria for the rating.
Disagree completely. Sign of a good team is winning ugly. These are literally all wins you are complaining about in the non-con (except obviously Chicago State). The whole point of this deal is to win and the ‘Cats have been doing it better than their computer rankings indicate.
Don’t you just end up with RPI then? That was considerably worse.
The good news is that these metrics are discussion points now but don’t particularly matter until all the games are played. Plus, as important as they are to put teams in ballpark tiers, the committee still watches games and uses their eyes too.
Boise State was higher ranked than us in the metrics last year but the committee put us 3 seed lines higher and they were right.
I mean. The RPI factored in Who won the actual game! Seems like a good place to start.
Who cares about any of this? Barring a collapse we are dancing.
Leave it to a NU board to have the longest active thread be about math and not the basketball team. 😂😂😂
Similar threads
Latest posts
-
-
-
At ILL result not surprising but the how is worrisome- Latest: Purple Pile Driver
-
