OFFICIAL NET Thread - 2022/23

[PSAL_Hoops]

Sure, which is the stated goal. A one point loss to Purdue is better than a one point win over IUPUI. Nobody wants to use the NET as the sole (or even main) criteria for picking the field. It's a useful sorting tool.

The NET tells you who is better (usually) but it doesn't tell you who has the better resume.

Better for what, exactly? IMO, a win can never be worse than a loss for field selection purposes. The two can be neutral for sure. It’s different to say, you don’t penalize a team for a close loss to Purdue relative to a win over Columbia. But at the end of the day, in a vacuum - which resume is better?

A) 27-3 midmajor resume with 25 Q4 wins, and wins over Maryland and Rutgers. Blowout losses to Kansas and Purdue. Close loss in conference tourney finals on neutral floor.

Or

B) B-10 team with 16-15 record - those same wins over Maryland and Rutgers - no other wins over field teams (but obviously better other wins than team A). Losses to Kansas and Purdue by 1 point.

Team A gets an at large bid. Team B does not.

 

[fluoxetine]

Better for what, exactly?

Better as in more likely to win tonight on a neutral court.

 

[MiloTalon13]

Better as in more likely to win tonight on a neutral court.

Is "tonight" right? or is it better to win, on average over the course of the season so far?

 

[kcg88]

Better for what, exactly? IMO, a win can never be worse than a loss for field selection purposes. The two can be neutral for sure. It’s different to say, you don’t penalize a team for a close loss to Purdue relative to a win over Columbia. But at the end of the day, in a vacuum - which resume is better?

A) 27-3 midmajor resume with 25 Q4 wins, and wins over Maryland and Rutgers. Blowout losses to Kansas and Purdue. Close loss in conference tourney finals on neutral floor.

Or

B) B-10 team with 16-15 record - those same wins over Maryland and Rutgers - no other wins over field teams (but obviously better other wins than team A). Losses to Kansas and Purdue by 1 point.

Team A gets an at large bid. Team B does not.

Yes, Team A has the better resume but Team B very well could be the better team. That's what NET is trying to calculate and this is my point: better team and better resume are not always the same.

 

[bac2therac]

Got bored and scrolled through some of the bacatology threads from last year. Kinda fun to go back through the highs and lows (not as fun to swipe past pages and pages of arguing about our OOC schedule last year). Also a reminder of how quickly things can change even as it gets to the final games. Early in conference tournament week I felt great about being above the Dayton line, by the end I was praying we'd make it to Dayton.

In the middle of February we thought Iowa and UNC were frauds. Virginia Tech had great computer numbers but a bad resume -- and then won the ACC tournament.

thats why I dont like early bracketology. And even early February bracketology is too soon. Yeah last year Iowa had great net rank but not many quality wins until the end of the season. UNC was out until they beat Duke. Wake was in safely until they werent. Looks like RU does have the same horrible ooc sos rank from last year.

With respect to Va Tech I thought I read somewhere that they would not have been in if they didnt win the NCAA tourney. Add in the Texas A&M snub even as they reached the SEC finals and had some good wins and I dont think conference tourneys really move many needles

 

[fluoxetine]

Is "tonight" right? or is it better to win, on average over the course of the season so far?

Most of these models assume that a team has constant skill throughout the season so these would be equivalent.

 

[PSAL_Hoops]

Yes, Team A has the better resume but Team B very well could be the better team. That's what NET is trying to calculate and this is my point: better team and better resume are not always the same.

Fair enough. That makes sense.

 

[bac2therac]

It really feels like NET is valuing the efficiency numbers more than the W/L numbers. Play well consistently across all of your possessions, and the model gives you a lot of credit. Which leads to things like valuing "good" losses over "bad" wins.

I think wins and losses need to be weighted more, they overdid it in a way to reward close losses to good teams

 

[PSAL_Hoops]

Wow we are #9 in ESPN BPI

In my limited analysis - BPI seems to value defensive efficiency more than offensive (seeing how Marquette is down in the 20s).

RU by the way, moved up to number 2 in Kenpom defense.

 

[bac2therac]

its very flawed when a team loses 6 of 8 and is ranked 7th....its more than flawed, its absolute garbage.

Note they lost by double digits to a very mediocre St Johns on their home court. That should move them out of the top 10 alone. its not like they are playing the 70s Celtics every night in the Big East. its not the Big 12.

 

[fluoxetine]

its very flawed when a team loses 6 of 8 and is ranked 7th....its more than flawed, its absolute garbage.

Note they lost by double digits to a very mediocre St Johns on their home court. That should move them out of the top 10 alone. its not like they are playing the 70s Celtics every night in the Big East. its not the Big 12.

Are you saying NET is flawed or "team value index" is flawed? Because "team value index" is just a small part of the NET so it's hard to tell if or how it is flawed. The things you are complaining about are mainly happening because the adjusted net efficiency is weighted so highly, but this is separate from "team value index" which is what your original post said.

 

[fluoxetine]

Like the team value index could be 100% perfect or it could be garbage.. it's really impossible to tell from looking at NET because NET is only giving it like 15% weight or something (I made that number up but it's very clearly small compared to the other component)

 

[pmvon]

Net efficiency is basically just scoring margin. The only difference is that it is per possession as opposed to per game.

Adjusted net efficiency is just that adjusted for opponent and home court advantage. So if I play against an opponent on a neutral court and their adj net efficiency is +5 points per 100 possessions and I beat them by 3 points per 100 possessions then my adjusted net efficiency for that game is +8 points per 100 possessions (+3 raw, +5 for the opponent).

I probably wouldn't say that to a five year old.. the real five year old explanation is just "efficiency = scoring margin with some tweaks that don't matter very much"

 

[PSAL_Hoops]

its very flawed when a team loses 6 of 8 and is ranked 7th....its more than flawed, its absolute garbage.

Note they lost by double digits to a very mediocre St Johns on their home court. That should move them out of the top 10 alone. its not like they are playing the 70s Celtics every night in the Big East. its not the Big 12.

OSU has lost 6 of their last 7 and still remain in Kenpom top 20. So odd.

 

[bac2therac]

OSU has lost 6 of their last 7 and still remain in Kenpom top 20. So odd.

this too and they lost to a horrific Minnesota team on their home court.

I firmly believe non conference is being weighed too heavily and its more so because wins are not valued

 

[bac2therac]

Are you saying NET is flawed or "team value index" is flawed? Because "team value index" is just a small part of the NET so it's hard to tell if or how it is flawed. The things you are complaining about are mainly happening because the adjusted net efficiency is weighted so highly, but this is separate from "team value index" which is what your original post said.

both

 

[fluoxetine]

OSU has lost 6 of their last 7 and still remain in Kenpom top 20. So odd.

Remeber kenpom doesn't care if you win or lose the game beyond the points you score.

Loss (#5) purdue at home by 2 (roughly +17.96 performance)
Loss (#33) maryland away by 7 (roughly +10.33 performance)
Loss (#188) minnesota home by 3 (roughly -9.90 performance)
Loss (#17) rutgers away by 4 (roughly +18.86 performance)
Loss (#99) nebraska away by 3 (roughly +7.06 performance)
Win (#37) iowa home by 16 (roughly +33.67 performance)
Loss (#28) illinois away by 9 (roughly +8.53 performance)

So an average of roughly +12.36 for those 7 games which would be like #62.

But they were up pretty high (#12) for the first 13 games. I don't know exactly what their rating was but the current #12 is +21.02

So 13 games of +21.02 and 7 games of +12.36 averages to +17.99 which would be #24. Their actual rating is +18.77 / #20 not that far off

 

[PSAL_Hoops]

this too and they lost to a horrific Minnesota team on their home court.

I firmly believe non conference is being weighed too heavily and its more so because wins are not valued

Then again - you’d think the metrics would like Illinois a lot more than they do. Can’t get much better than that pair of neutral wins against UCLA and Texas OOC.

 

[fluoxetine]

They're just doing a simple average of all the games/possessions. There's no special weighting of OOC games.

 

[PSAL_Hoops]

Remeber kenpom doesn't care if you win or lose the game beyond the points you score.

Loss (#5) purdue at home by 2 (roughly +17.96 performance)
Loss (#33) maryland away by 7 (roughly +10.33 performance)
Loss (#188) minnesota home by 3 (roughly -9.90 performance)
Loss (#17) rutgers away by 4 (roughly +18.86 performance)
Loss (#99) nebraska away by 3 (roughly +7.06 performance)
Win (#37) iowa home by 16 (roughly +33.67 performance)
Loss (#28) illinois away by 9 (roughly +8.53 performance)

So an average of roughly +12.36 for those 7 games which would be like #62.

But they were up pretty high (#12) for the first 13 games. I don't know exactly what their rating was but the current #12 is +21.02

So 13 games of +21.02 and 7 games of +12.36 averages to +17.99 which would be #24. Their actual rating is +18.77 / #20 not that far off

I get how the math could’ve worked out that way. They had one super efficient performance with Iowa mixed in there.

Top 20 teams should not go 0-2 vs. Minny and Nebraska though.

 

[Loyal_2RU]

I still think efficiency ought to count 0. The is one summative matter of efficiency and out is whether to win or lose. The game direct have indicated scoring based on a bunch of factors of. It is simply did you purchase your opponent at the end.

Think about a race. You can lead the whole face and rub it off fuel because someone else was dragging behind you. Who was more efficient?

 

[bac2therac]

Remeber kenpom doesn't care if you win or lose the game beyond the points you score.

Loss (#5) purdue at home by 2 (roughly +17.96 performance)
Loss (#33) maryland away by 7 (roughly +10.33 performance)
Loss (#188) minnesota home by 3 (roughly -9.90 performance)
Loss (#17) rutgers away by 4 (roughly +18.86 performance)
Loss (#99) nebraska away by 3 (roughly +7.06 performance)
Win (#37) iowa home by 16 (roughly +33.67 performance)
Loss (#28) illinois away by 9 (roughly +8.53 performance)

So an average of roughly +12.36 for those 7 games which would be like #62.

But they were up pretty high (#12) for the first 13 games. I don't know exactly what their rating was but the current #12 is +21.02

So 13 games of +21.02 and 7 games of +12.36 averages to +17.99 which would be #24. Their actual rating is +18.77 / #20 not that far off

well thats a problem then...wins and losses matter alot

 

[fluoxetine]

well thats a problem then...wins and losses matter alot

Matter a lot for what? They don't matter very much for predicting who will win and lose in the future (which is all that kenpom is trying to do).

 

[PSAL_Hoops]

They're just doing a simple average of all the games/possessions. There's no special weighting of OOC games.

Right - I’m not crazy about the simple average part. Sometimes it’s just not your day, and I’m not sure being able to cut the lead down to a more respectable MOV at the end of a poor performance or even get within quasi striking range is a great indicator of future performance. Our 6 point loss to Temple (injuries not withstanding was not really that close of a game). Illinois lost to NW by 13 but that game was just as close in reality. NW went on a timely run. (Simple example).

 

[fluoxetine]

I still think efficiency ought to count 0. The is one summative matter of efficiency and out is whether to win or lose. The game direct have indicated scoring based on a bunch of factors of. It is simply did you purchase your opponent at the end.

It depends what you are trying to do. I would be in favor of win/loss only metrics when it comes to tournament selection.

Think about a race. You can lead the whole face and rub it off fuel because someone else was dragging behind you. Who was more efficient?

Well, they were. But the basketball analogy is a game you lead the whole way and then blew at the end.. kenpom would say they were more efficient there too. You are always more efficient in a win than a loss (a scoring margin >0 is higher than one <0).

In order to complete this analogy you need, like, a series of races. Your opponent wins 5 races by 1 second each and you win 1 race by 5 minutes. Who is faster / more likely to win in the future? (I don't know, racing obviously includes a pacing element that is not nearly as present in a basketball game.)

 

[fluoxetine]

being able to cut the lead down to a more respectable MOV at the end of a poor performance or even get within quasi striking range is a great indicator of future performance.

There is a lot of noise in scoring margins for sure. If you have the ability to parse out / properly de-weight garbage time scoring you will do better than if you don't. Obviously the couple minutes we have Aiden Terry and Chol in at the end of some blowout are pretty much pure noise in terms of predicting our future performances.

However, models that know nothing about wins/losses and only know points will still do way better at predicting the future than models that only know about wins/losses and don't care about points. It's not close. And if you make a hybrid model that gives some bonus for the actual win it will be non-zero (there is skill to winning beyond just pure points scored) but it will be pretty small (it's mostly just luck whether you win your really close games and you mostly just want to look at points scored)

This is all about predictions though and not about selecting teams for a postseason. Kenpom only cares about predictions, so I think it is silly to criticize his metrics in terms of whether they properly account for resumes. That's not what he's trying to do.

If you are making a model for some other purpose (e.g. tournament selection) then that's a completely different story and becomes more of a subjective exercise.

 

[bac2therac]

Matter a lot for what? They don't matter very much for predicting who will win and lose in the future (which is all that kenpom is trying to do).

I dont care about predicting and it shouldnt be used for selecting ncaa bids and thankful kenpom and all of the others arent..just appearing a number

Lets fix the Net

 

[PSAL_Hoops]

Matter a lot for what? They don't matter very much for predicting who will win and lose in the future (which is all that kenpom is trying to do).

I disagree. It’s flawed to call all close scoring margin games 50/50 games.

Teams with good FT shooting guards should (on average) be able to close out games where they have a lead at the end. It’s not “luck” that our dependable FT shooting guards went 6-6, 4-4 and 3-4 from charity to close out the UMass Lowell, NW and OSU games we won. It took a string of bad officiating luck on top of an untimely missed FT by a +78% FT shooter for us to lose that other OSU game. My point is teams with poor FT shooters statistically blow close leads at a higher rate.

Purdue was our only true 50/50 game which I would define only as a game that goes back and forth until the last posession with no intentional fouling.

 

[fluoxetine]

I disagree. It’s flawed to call all close scoring margin games 50/50 games.

Teams with good FT shooting guards should (on average) be able to close out games where they have a lead at the end. It’s not “luck” that our dependable FT shooting guards went 6-6, 4-4 and 3-4 from charity to close out the UMass Lowell, NW and OSU games we won. It took a string of bad officiating luck on top of an untimely missed FT by a +78% FT shooter for us to lose that other OSU game. My point is teams with poor FT shooters statistically blow close leads at a higher rate.

Purdue was our only true 50/50 game which I would define only as a game that goes back and forth until the last posession with no intentional fouling.

Like I said, there is some skill to winning close games but it's not that much. This is demonstrated by actually fitting models on large amounts of data.. wins/losses provide a little but not much predictive value when added to pure points. Regardless of what logic you put forth as to why this should or should not be true, it's demonstrably true.

Your example (good FT shooting guards) is probably the best example of where it actually happens. Having a 90% guy is obviously great at the end of games if you can get the ball to him.

Also,
It’s not “luck” that our dependable FT shooting guards went 6-6, 4-4 and 3-4 from charity to close out the UMass Lowell, NW and OSU games we won

(1) those points still count in the efficiency metrics
(2) it is a little bit luck, that's 93% FT shooting and obviously a weighted average of our guards does not shoot 93% long term. Even just Cam is almost certainly not 93% long term.

Just think of anecdotal evidence from this season alone. In the first OSU game, if that guy misses that heave (or if the ref makes the right call), would that significantly alter your opinion of our team going forward?

In the second OSU game, if the guy banked in his three at the end and we lost in regulation, would that significantly alter your opinion of our team going forward?

In the SHU game, if Hyatt puts up the 3 pointer on the last play and makes it, does that significantly alter your opinion of our team going forward?

In the Northwestern game, if Cam bricks his three on our last possession, or if someone commits a foul diving for that loose ball and Northwestern wins the game at the line, does that significantly alter your opinion of our team going forward?

 

[kcg88]

Why are we trying to "fix" the NET? I don't want any computer picking the field.

 

[PSAL_Hoops]

Like I said, there is some skill to winning close games but it's not that much. This is demonstrated by actually fitting models on large amounts of data.. wins/losses provide a little but not much predictive value when added to pure points. Regardless of what logic you put forth as to why this should or should not be true, it's demonstrably true.

Your example (good FT shooting guards) is probably the best example of where it actually happens. Having a 90% guy is obviously great at the end of games if you can get the ball to him.

Also,
It’s not “luck” that our dependable FT shooting guards went 6-6, 4-4 and 3-4 from charity to close out the UMass Lowell, NW and OSU games we won

(1) those points still count in the efficiency metrics
(2) it is a little bit luck, that's 93% FT shooting and obviously a weighted average of our guards does not shoot 93% long term. Even just Cam is almost certainly not 93% long term.

Just think of anecdotal evidence from this season alone. In the first OSU game, if that guy misses that heave (or if the ref makes the right call), would that significantly alter your opinion of our team going forward?

In the second OSU game, if the guy banked in his three at the end and we lost in regulation, would that significantly alter your opinion of our team going forward?

In the SHU game, if Hyatt puts up the 3 pointer on the last play and makes it, does that significantly alter your opinion of our team going forward?

In the Northwestern game, if Cam bricks his three on our last possession, or if someone commits a foul diving for that loose ball and Northwestern wins the game at the line, does that significantly alter your opinion of our team going forward?

No to those examples. Because those things would fall into the same “luck” category as the OSU game we lost. Bad luck happens, it’s just not inherently binary for all close games. When your PG shoots 61% from charity (Pike’s first team) it’s going to be statistically less likely that you can hang onto a late lead.

Some teams are just built to win close games easier but not necessarily to blow teams out based on their style of play. Winning and losing (adjusting accordingly for SOS) are much more important than how you win IMO.

 

[fluoxetine]

No to those examples. Because those things would fall into the same “luck” category as the OSU game we lost. Bad luck happens, it’s just not inherently binary for all close games. When your PG shoots 61% from charity (Pike’s first team) it’s going to be statistically less likely that you can hang onto a late lead.

This is true, but you're also going to be worse all the time in general which will show up in all the efficiency metrics. For sure the end of some close games start to emphasize different things, so if you are good at doing something you have to do a disproportionate amount of at the end of close games (free throws?) and bad at some other thing that doesn't matter as much at the end of close games as it does at other times (unclear what this would be honestly) then maybe you will be better or worse in that situation than in other situations. But, again, the effect is just not that large.

Some teams are just built to win close games easier but not necessarily to blow teams out based on their style of play. Winning and losing (adjusting accordingly for SOS) are much more important than how you win IMO.

You can say this all you want "Winning and losing (adjusting accordingly for SOS) are much more important than how you win IMO" but, assuming we are still talking about prediction here, that's not actually an opinion. It's just a factually untrue statement. You can actually build models using different factors and fit them and see how well they predict and how much weight they put on something like a binary win/loss factor if they already have the points information. There is a reason kenpom and Bart are not using a binary win/loss factor in their models and it isn't because it's too complicated or something. It's also the reason NET is not weighting it very highly.

If we are NOT talking about prediction then your opinion statement is perfectly valid (and I agree with it). But creating a model for selecting a tournament field is subjective because there is not a defined correct answer. If you are predicting wins/losses or point spreads you can measure the results.

 

[PSAL_Hoops]

To the first part - when you have a lead at the end of the game the usage shifts to a high confidence interval that your next possession will put you on the FT line relative to the rest of the game. So FT efficiency has more relative impact in that situation because it happens more.

I also think predictive outcomes are more match up driven than they are tied to efficiency metrics. There are plenty of more statistically efficient teams I’d rather us play than Michigan.

 

[bac2therac]

Why are we trying to "fix" the NET? I don't want any computer picking the field.

because its not sorting them right either

 

[kcg88]

because its not sorting them right either

It does a good enough job to be useful as a sorting tool.

 

[bac2therac]

It does a good enough job to be useful as a sorting tool.

Not when it has uconn 7

 

[fluoxetine]

UConn is #actually good.

 

[bac2therac]

Any school losing 6 of 8 shouldnt be ranked 7

 

[bac2therac]

6th in their own conference 7th in the country yes I apologize for being wrong all along

 

[Joey Bags]

this too and they lost to a horrific Minnesota team on their home court.

I firmly believe non conference is being weighed too heavily and its more so because wins are not valued

Why is there a weighting at all? Why not just value the OOC and conference slate the same and base it on the opponent regardless of whether or not it’s a conference game.