I can't really check this, but I wouldn't be surprised if that's the worst loss for a top 25 kenpom team ever.
OFFICIAL NET Thread - 2022/23
- Thread starter fatsam98
- Start date
I know it would never happen but if I was on the bench and a coach was going to put me in just to dribble out the clock I'd love to see a " no thanks". Like when they put in the backup QB just to take a knee. Insulting.
Iowa letting Eastern Illinois shoot 60% 🤣
People love to say how we can't score and it will cost us games, but teams that don't play D can cost them games too.
Eastern Illinois 92 points against a B1G team is embarrassingggg
People love to say how we can't score and it will cost us games, but teams that don't play D can cost them games too.
Eastern Illinois 92 points against a B1G team is embarrassingggg
Splitting hairs at this point but this is worse than the Lafayette loss.
Lafayette beat us at the buzzer. Iowa lost their game by double digits. Ouch.
Rider has been the mid-major version of us. Outside of our game, lots of heart breaking close losses.
A good week would go a long way towards bumping them to Q3. They have a road game at Georgia coming up.
A good week would go a long way towards bumping them to Q3. They have a road game at Georgia coming up.
Rutgers up 2 spots to 38
Iowa to 60 in the NET (from 27).
Ken Pomeroy discussed the somewhat counterintuitive effect on his ratings of pre-season data (and other good stuff) as the guest on yesterday’s Solving Basketball podcast.
Ohio St 25 ▶ 14, 95-61 Maine 236▶248
VA Tech 14 ▶ 29, 65-70 @ BC 247▶229
Auburn 55 ▶ 30, 84-71 @ Wash 94▶117
Ariz St 22 ▶ 50, 60-97 @ SanFran 119▶92
Iowa 27 ▶ 60, 83-92 vs E.Illinois 350▶333
How far down do you have to be to only gain 17 places with this win?
JMU 54 ▶ 69, 100-107 @ Coppin St 249▶234
Fla-GC 75 ▶ 77, 84-81 vs Canisius(1-9) 283▶261 (Up 21, close loss)
Ohio 169 ▶ 119, 95-76 @ Delaware 165▶191
Richmond 152 ▶ 147, 81-71 vs Bucknell 209▶224
Lafayette(2-11) 311▶271, 90-65 @ La Salle 270▶307
California(1-12) 341▶319, 73-51 vs UT Arlington 248▶284
VA Tech 14 ▶ 29, 65-70 @ BC 247▶229
Auburn 55 ▶ 30, 84-71 @ Wash 94▶117
Ariz St 22 ▶ 50, 60-97 @ SanFran 119▶92
Iowa 27 ▶ 60, 83-92 vs E.Illinois 350▶333
How far down do you have to be to only gain 17 places with this win?
JMU 54 ▶ 69, 100-107 @ Coppin St 249▶234
Fla-GC 75 ▶ 77, 84-81 vs Canisius(1-9) 283▶261 (Up 21, close loss)
Ohio 169 ▶ 119, 95-76 @ Delaware 165▶191
Richmond 152 ▶ 147, 81-71 vs Bucknell 209▶224
Lafayette(2-11) 311▶271, 90-65 @ La Salle 270▶307
California(1-12) 341▶319, 73-51 vs UT Arlington 248▶284
Yup. I'm not going to get over the refs taking that W from us all season.
Rutgers up to 24 in KenPom
I think he draws an important distinction between his approach and that of NET. Pomeroy is trying to be predictive - and including priors helps to do that (and he's apparently weighting them a bit less earlier in the season this year, but including them further into the season - even into February). NET, though, is meant to be descriptive of performance to date - not predictive of how they'll do in their next game - so priors are data that wasn't generated by the current team.
Bigtime upset for Cal over UT Arlington moves them up 22 spots 🤣
Including prior year into February is dumb, especially in the portal / 1 and done era
Up to #38 but only 2 teams ahead of us with same or less wins - Kent State with 7-3 record, Sam Houston with 6-2 record. RU 7-4
Not really. You want it to fade out over time but prior expectations are ALWAYS relevant. If Duke and Morgan St. have the same adjusted efficiency over 30 games, who are you betting on when they face off in game 31?
The correct answer is Duke. I'm not sure whether their win prob should be 51% or 60% or what, but it's definitely >50%
It’s simply not true if you are trying to make predictions. data from previous years, if used properly, will improve your predictions
I work for a company that produces predictive analytic software and the above statement is true.
From the podcast, he said that the day he removed priors in January each year always saw a spike in the data that made it less accurate as a predictive model. He also said he was going to look in the offseason about changing the amount that priors are weighted on a team-by-team basis, based on things like coaching changes.
He's not trying to build a model that shows how well teams have performed in a ranked list - he's trying to build a model that's predictive of future game results. There's no way to do that early in the season without prior data, and he's finding that even in late January his predictive model was more accurate with priors than without.
NET is trying to say whether Team A or Team B has performed better to date. Kenpom is trying to say by how much Team A would beat Team B if they were to play tomorrow. I think that's an important distinction.
Good discussion. It seems obvious that looking at prior years is useful for prediction. While players can turn over from year to year, I think we all agree that the quality of the coach and program matter. New players get coached up by good coaches as the season progresses. Good programs are more likely to have good coaches and good players that are more likely to improve over the course of the season than weaker programs.
I don't care about predictions and I care even less about Kenpom. Its not part of the selection process
How is this relevant to my comment? I didn't say prior expectations aren't relevant. Prior SEASON data with a different team are not a good predictor of what a different team the next season will due in February and beyond.
For example, how is Murray State's stats with Ja relevant to the next year's team in February when he's gone?
How you are performing with the current team and current roster in the current season is much more relevant
What about teams with an entirely new coaching staff?
The new coaching staff is something Pomeroy talked about on the podcast, that he's considering adjusting how much prior data should be used in such situations.
Even if you lose a star player like Ja, there is still consistency year to year with returning players and overall system. In Murray State's case, they were definitely worse without Ja in the lineup, but they still played tough defense and went 23-9, tying for 1st in their conference.
Overall, though, he's found that prior season data, even in late January, made his predictive model more accurate. On its own it's not a good predictor, but as a small (and ever decreasing) factor, it seemingly has some benefit if your goal is making a more accurate predictive model.
Makes sense. I personally don't care about predictive models which are more useful for gambling. I want to see rankings based on the team's performance this season. Merit based.
I see value in both - but I prefer descriptive rather than predictive, personally. NET has a lot of rough edges, but it's much more focused on representing a team's "resume to date" rather than worrying about how they'll perform tomorrow.
Looking at Rutgers' and Bucknell's relative strengths for tomorrow's game, kenpom is of more use. Once the game is over, though, kenpom becomes useless for that game and the resultant NET changes are then the focus.
The prior expectations are based in part on how the team performed in the past.
Definitely much MORE relevant, once you have a decent number of games. But that doesn’t make everyone else irrelevant.
There will be more noise and uncertainty. If you are trying hard to build a preseason model the coach is probably a component, as well as player turnover etc. but the base for all of that is still going to be “how good were they last year?”
This is all about predictions. If you don’t care about that, it’s perfectly fair, but that is what Kenpom is trying to do. Something like NET is being used for a different purpose, and using prior data wouldn’t be appropriate.
Sacred Heart and Rider both won tonight.
CC lost again, but I’m thinking that doesn’t matter because losing by 17 at St Joes might actually improve CC’s NET they are so bad.
CC lost again, but I’m thinking that doesn’t matter because losing by 17 at St Joes might actually improve CC’s NET they are so bad.
I still think you can make an accurate ranking without prior year results by just going further into W-L transitive property. To use the above example, of Morgan State and Duke having the same efficiently after 10 games and who would you pick - if you look at the records of the teams Duke has beaten and the records of the teams Morgan State has beaten - and look at the records of the teams they have beaten etc. and keep going, you’ll find that Morgan State is weaker than Duke because Morgan State’s teams they beat will have beaten worse teams or lost to more “better” teams etc. because they generally play in a weaker conference (ie small teams that lose a lot of early OOC games to big teams) .. basically I’m saying there’s a mathematical numbers-only way for justify your “gut” feeling why Duke would be ahead of Morgan State given the same efficiency/record
