• Welcome to OGBoards 10.0, keep in mind that we will be making LOTS of changes to smooth out the experience here and make it as close as possible functionally to the old software, but feel free to drop suggestions or requests in the Tech Support subforum!

2023-24 Wake Forest Basketball Season - 21-14 (11-9) - KP#29 / NET#43

KidA23 said:
My issue with the quads is it seems like the emphasis on Q1 wins is resulting in some double-counting. I’m a lawyer so math ain’t my strong suit, but it seems if our overall NET is so high, it shouldn’t be a sub-component of the NET calculation (Q1 wins) that then gets held against us. It’s your effing formula and it shows we’re good.
Click to expand...

Yes that’s exactly the problem. They’re cherry picking an input from that formula and arbitrarily assigning cutoff points to group larger batches of teams. Which is bad enough when it gets all the airtime but it’s worse when you know the idiot committee is going to be talking quads too and ignoring their own better tool as a result.
 
There are lots of dumb parts / flaws of the criteria (the arbitrary quad designation is an obvious one, the lack of clearly articulated criteria for selection/seeding, the significance of the short non-conference season is another as has been mentioned), but I think some of the complaints are off-base.

The NET, KenPom, Torvik, BPI, etc. are all power ratings that are optimized to be predictive based on efficiencies. The typically have gotten some criticism for ignoring W-L record, overvaluing margin of victory, ignoring head-to-head, etc. But they are designed to be predictive, and do a good job of that.

The tournament doesn't and never has been selected based on power ratings, but rather on resume, or results. (I do think picking the tournament based on power ratings would be pretty absurd in practice). Winning vs. losing is pretty much all that matters. The RPI used to do this, since W-L was literally all RPI cared about (along with location of game).

There were a bunch of flaws with the RPI and other similar systems, but the big one was that a team could play a game and go down in the rankings, no matter what, just because they were playing a bad team, so that was obviously dumb.

Using the NET to develop quads theoretically still rewards winning and losing but theoretically does a better job of evaluating strength of schedule, although the arbitrary cut-offs on the quad designations is a new, but seemingly obviously dumb, flaw in the system.

ESPN calculates "strength of record" (it's more commonly used in college football) that I think does a good job of combines the two and solves the quad problem (https://www.espn.com/mens-college-basketball/bpi/_/view/resume).

The conference strength issue is a bit complicated, since it's not "biased," it's just noisy. Kenpom calculates a metric called "luck" which is basically the difference between your predictive power rating and expected W-L rating based on that power rating. Teams with more wins against their schedule than you would expect are "lucky," teams with fewer wins than expected are "unlucky." We rank as the 352nd most lucky team out of 362. Theoretically, luck explains the difference between how good teams are and how likely they are to get a tournament bid.

One of the issues is because the non-conference play is both so impactful, but also has a small sample size, is that you see a lot of correlation between the luck factor of teams in the same conference. That is, if a conference is, on aggregate, unlucky in their non-conference games, that flows through into all the teams for the whole season. I don't think it will shock anyone to see that it looks like overall, the ACC has been rated as pretty unlucky by kenpom (on top of just not being that good). FWIW, I'm not sure this is a new problem.
 
Last edited:
IAppreciateIt said:
There are lots of dumb parts / flaws of the criteria (the arbitrary quad designation is an obvious one, the lack of clearly articulated criteria for selection/seeding, the significance of the short non-conference season is another as has been mentioned), but I think some of the complaints are off-base.

The NET, KenPom, Torvik, BPI, etc. are all power ratings that are optimized to be predictive based on efficiencies. The typically have gotten some criticism for ignoring W-L record, overvaluing margin of victory, ignoring head-to-head, etc. But they are designed to be predictive, and do a good job of that.

The tournament doesn't and never has been selected based on power ratings, but rather on resume, or results. (I do think picking the tournament based on power ratings would be pretty absurd in practice). Winning vs. losing is pretty much all that matters. The RPI used to do this, since W-L was literally all RPI cared about (along with location of game).

There were a bunch of flaws with the RPI and other similar systems, but the big one was that a team could play a game and go down in the rankings, no matter what, just because they were playing a bad team, so that was obviously dumb.

Using the NET to develop quads theoretically still rewards winning and losing but theoretically does a better job of evaluating strength of schedule, although the arbitrary cut-offs on the quad designations is a new, but seemingly obviously dumb, flaw in the system.

ESPN calculates "strength of record" (it's more commonly used in college football) that I think does a good job of combines the two and solves the quad problem (https://www.espn.com/mens-college-basketball/bpi/_/view/resume).

The conference strength issue is a bit complicated, since it's not "biased," it's just noisy. Kenpom calculates a metric called "luck" which is basically the difference between your predictive power rating and expected W-L rating based on that power rating. Teams with more wins against their schedule than you would expect are "lucky," teams with fewer wins than expected are "unlucky." We rank as the 352nd most lucky team out of 362. Theoretically, luck explains the difference between how good teams are and how likely they are to get a tournament bid.

One of the issues is because the non-conference play is both so impactful, but also has a small sample size, is that you see a lot of correlation between the luck factor of teams in the same conference. That is, if a conference is, on aggregate, unlucky in their non-conference games, that flows through into all the teams for the whole season. I don't think it will shock anyone to see that it looks like overall, the ACC has been rated as pretty unlucky by kenpom (on top of just not being that good). FWIW, I'm not sure this is a new problem.
Click to expand...
you say, "The NET is optimized to be predictive" and "Using the NET to develop quads theoretically still rewards winning and losing" but there is a component of the NET, separate and distinct from the adjusted net efficiency rating, intended to reward winning, already baked into the NET:

www.ncaa.com

College basketball's NET rankings, explained

Here's everything you need to know about college basketball's NET ranking system.
www.ncaa.com www.ncaa.com

The remaining factors include the Team Value Index (TVI), which is a result-based feature that rewards teams for beating quality opponents, particularly away from home, as well as an adjusted net efficiency rating. The adjusted efficiency is a team’s net efficiency, adjusted for strength of opponent and location (home/away/neutral) across all games played.
Click to expand...

The TVI would explain why WF currently has a lower NET rating than KP/Torvik, and why the reverse is true for Virginia.
 
Great post Appreciate. I can't speak for BPI because I have no idea how that's calculated but I've concluded that the NET is intending to serve as a hybrid predictive/resume-based ratings system. To what degree it actually does that is up for debate. It seems pretty straightforward that the adjusted efficiency component comprises the bulk of the system (as it probably should - as Torvik and KP both use this as well, and it seems to be the best indicator of future success - but not necessarily wins and losses in a backward looking way).

KP does not use any resume aspect at all - it's just adjusted efficiency (I'd imagine each of these metrics "adjusts" in a slightly different manner). Torvik uses mostly adjusted efficiency but also factors in some game control type aspect based on the average scoring margin throughout individual games which is why Wake was struggling earlier in the year it seemed. BPI no idea. NET uses the nebulous TVI as discussed above.

Long way of saying, I think the NET itself does a pretty good job and has for the last few years since the last tweak. I think strength of record would probably be a smoother way to handle the resume portion rather than quadrant wins. Perhaps the strength of record measure is generally what the NET uses for TVI though.

Regardless, the quad system is arbitrary overall and does, to me, seem to be disproportionately impactful in the way it appears on the team sheet
 
thatguy2016 said:
you say, "The NET is optimized to be predictive" and "Using the NET to develop quads theoretically still rewards winning and losing" but there is a component of the NET, separate and distinct from the adjusted net efficiency rating, intended to reward winning, already baked into the NET:

www.ncaa.com

College basketball's NET rankings, explained

Here's everything you need to know about college basketball's NET ranking system.
www.ncaa.com www.ncaa.com



The TVI would explain why WF currently has a lower NET rating than KP/Torvik, and why the reverse is true for Virginia.
Click to expand...

Good call -- curious what the correlation between NET and the other three main power ratings are (too lazy to do it myself).

We are:

KP: 20
Torvik: 24
BPI: 21
NET: 27

(51st in ESPN SOR)
 
It seems like the efficiency rating has one big negative inthe Net because a possession taking place in the last minute of a 20 point game is weighed the same as a potential game winning possession. With all the data on hand I would think some sort of weight could be applied to "unimportant possessions" so they are given less of an impact. This is is helping us right now since we seem to be a pretty awesome front running team, but I think everyone would take the rating more seriously if you didn't have teams keeping their starters in until the last minute of a 25 point game.

For all the other variables in the formula, I'd think a possession weight could be added in.
 
TooManyWeapons said:
It seems like the efficiency rating has one big negative inthe Net because a possession taking place in the last minute of a 20 point game is weighed the same as a potential game winning possession. With all the data on hand I would think some sort of weight could be applied to "unimportant possessions" so they are given less of an impact. This is is helping us right now since we seem to be a pretty awesome front running team, but I think everyone would take the rating more seriously if you didn't have teams keeping their starters in until the last minute of a 25 point game.

For all the other variables in the formula, I'd think a possession weight could be added in.
Click to expand...
Except it's been shown that those possessions that you are calling meaningless do have predictive value, which is why they are included.

Teams didn't pull their starters all that much earlier 20 years ago than they do now. It's not like walk-ons were regularly playing 5 minutes in blowout games.
 
IAppreciateIt said:
Good call -- curious what the correlation between NET and the other three main power ratings are (too lazy to do it myself).

We are:

KP: 20
Torvik: 24
BPI: 21
NET: 27

(51st in ESPN SOR)
Click to expand...

I don't know - I do know that New Mexico, whose wins I highlighted earlier, has the following rankings - so they are getting a nice boost from their quality wins:

KP: 26
Torvik: 23
BPI: 47 (!)
NET: 20

So New Mexico already gets a boost in the NET, and WF already gets dinged in the NET, and then we turn around and say "New Mexico has more Q1 wins !!"
 
If they want to make the tournament a little larger without total disruption, make all the "eleven seeds" be play-in games. So the eight teams are "eleven seeds," go to Dayton, play a game and if they win, they get to play in thevfield of 64.

I wouldn't increase the play-in games for the conference winner AQ's. They did all they could by winning their respective leagues. If your an 11 seed and get beat and don't like it? Play better and win more games next year.
 

NIT Bracketology

www.nitbracketology.com www.nitbracketology.com

Most recent N.I.T. bracketology has as a #1 seed. If we win tomorrow, that clearly takes us off their current 1 seed and out of the tournament , but does a loss also take us off the #1 seed? Wow, heady stuff.
 
deacon14 said:
Except it's been shown that those possessions that you are calling meaningless do have predictive value, which is why they are included.

Teams didn't pull their starters all that much earlier 20 years ago than they do now. It's not like walk-ons were regularly playing 5 minutes in blowout games.
Click to expand...
Yes, not to be an asshole about this but I do find it interesting that folks keep piling on about this when KenPom obviously did his research on whether or not the data he was using was predictive (throughout the entire game - not just on balance). He considered all of this, I'd imagine far far more than we have combined on here, and concluded this was the best predictive way to go.
 
Does NET or Kenpom have a better record predicting NCAA Tourney games?
 
PhDeac said:
Does NET or Kenpom have a better record predicting NCAA Tourney games?
Click to expand...
No idea...predictive metrics probably don't have much of a leg up in a single elimination format. Likely better than wisdom of the crowds, but on a one off (or even three off at this stage given the current NET format since 2020 with a COVID year) I think you're just getting a lot of noise.
 
Deaconblue said:
If they want to make the tournament a little larger without total disruption, make all the "eleven seeds" be play-in games. So the eight teams are "eleven seeds," go to Dayton, play a game and if they win, they get to play in thevfield of 64.

I wouldn't increase the play-in games for the conference winner AQ's. They did all they could by winning their respective leagues. If your an 11 seed and get beat and don't like it? Play better and win more games next year.
Click to expand...

Love this 8x11 seed idea and love this discussion. Learning a ton about these rankings and great debate. I do know this related to eye test:

Pre-Efton it was clear Carr was overmatched vs other true 5’s. Marsh can dunk but no real post moves so any foul trouble (or injury) with either and we were screwed. Reid gave us immediate depth (provided he stays on the F’ing floor).

Maturation of PFred, re-emergence of Marsh, Cam finding a new ‘D-dog’ role, have made us a tough out anywhere vs. anyone. We still have games to work out some kinks and roles like Boopie and DMon. Drives me crazy we don’t run more set plays for PFred. Don’t reinvent the wheel, just watch Duke/ JJ Redick game tape - would be sweet to burn them with their own plays, right? Marsh can play Boozer role. We’re trending up at the right time with a chance to finish Feb 6-2.

If the committee is doing their job, they know these things too and see us as a dangerous team with just some work to do and tons of upside. Fair to say we haven’t earned the benefit of the doubt yet.

Duke is just the beginning.

LFGo Deacs!


Sent from my iPhone using Tapatalk
 
BPI is doing something differently. The MWC has two teams in the top 50 on BPI. Under resume there are five MWC teams in the top 50 on SOR. The SOR is probably a result of better w-l records. I'm guessing BPI is the result of the weak schedules.
 
Wakeforest22890 said:
Yes, not to be an asshole about this but I do find it interesting that folks keep piling on about this when KenPom obviously did his research on whether or not the data he was using was predictive (throughout the entire game - not just on balance). He considered all of this, I'd imagine far far more than we have combined on here, and concluded this was the best predictive way to go.
Click to expand...
Kenpom all knowing, all considering, the not to be questioned judge jury and executioner of college basketball.

All ye who disrespect him have yet to consider as much as lord KenPom considered considering.
 
Wakeforest22890 said:
No idea...predictive metrics probably don't have much of a leg up in a single elimination format. Likely better than wisdom of the crowds, but on a one off (or even three off at this stage given the current NET format since 2020 with a COVID year) I think you're just getting a lot of noise.
Click to expand...
The fact that Virginia is considered in the tournament because their home win against us was a quad 1, but our fucking beat down of them at home was a quad two makes me want to die inside.
 
You must log in or register to reply here.
Back
Top