Bart Torvik's T-Rankings Now Used by Selection Committee
- Thread starter BlueSince92
- Start date
I guess I don’t understand the point you’re trying to make. And I don’t mean that in a condescending way but literally.
I get that not all teams play so direct head to head so a direct comparison cannot be made. That’s certainly an argument for adding weights for more advanced metrics like efficiency.
However, with the case of Kentucky vs Gonzaga and Kentucky vs Duke, Kentucky beat both head to head in neutral floors and has 1 less loss than both. To me that’s a smell of a less than adequate sort algorithm where real world results differ with the sort model.
I don’t like the initial seeding of a lot these sort algorithms because they are guided by initial human rankings. I agree that they lack meaning at this point on the season.
I don't know if either of those are all that accurate to be fair. Like the system didn't have Duke/Gonzaga higher rated than UK because of initial human rankings. They had them higher because it was due in some part to previous seasons data and current information available.
Computer systems should be free of human bias. This is why even tho in Kenpom UK started off 43rd, he admitted he felt in his opinion that was way too low. But you cant go changing the system because of a few outliers.
The fact that they lack meaning at this point I also think is something people say but isn't entirely true. The system balances out very quickly IMO. The average error in Kenpom's model doesn't change much from November to say March. It gets better but not drastically. It's very valid now.
The top two teams right now in Kenpom is Auburn and UT. I think if you asked many people, they would feel the same or at least have them rated toward the top.
But yeah I mean an argument can be made that UK, Duke and Gonzaga are all very close in rating and UK beat them both and on neutral an road courts at that, UK should be ranked higher. Makes sense.
This is why although I do think people complain about bias with the committee in terms of seeding, I feel like people would NOT accept a system where a computer ranks teams 1 through 68.
Rating teams on actual resume just makes more sense to people. Going to a computer model that rates say UK lower than Duke and Gonzaga currently I think would be a hard pill for people to swallow even if it could be accurate in terms of actual skill level.
Rating teams on actual resume just makes more sense to people. Going to a computer model that rates say UK lower than Duke and Gonzaga currently I think would be a hard pill for people to swallow even if it could be accurate in terms of actual skill level.
I guess I should be more specific. The KenPom
Ratings in particular are revealed the first week in the season and they are influenced by human subjectivity. This is why a Duke miraculously moves toward to top of these rankings when all teams are 1-0.
Written by KenPom himself:
“There is one more ingredient and that involves people. If the best basketball players are consistently choosing certain programs that says something about the stature of the program. So I add an adjustment using a recruiting rating based on the final RSCI rankings for each of the past ten seasons, giving more weight to recent seasons.”
This is the part I disagree with or more accurately why KenPom is prone to inaccuracies at the beginning of the season. I’d like to learn more about eh NET to understand its initial seeding, but again, I’d argue that Gonzaga and Duke being ahead of Kentucky at this point of the season is an indicator of its lack of accuracy at this point in the season.
Oh ok so not really influenced by the polls but rather the recruiting figures are influence by rankings that are produced by humans.
Tho I guess at the same rate, I guess it makes sense to factor that in. Especially schools that grab a lot of the top recruits.
It still normalizes very quickly tho. Uconn opened up the season 6th probably on the merits of back to back national championships. Lose 3 in a row in Maui and drop all the way out of the top 25 (26th).
Wins back to back over Baylor and Texas put them back at 13th.
It corrects very quickly based on actual results.
I feel like there's a certain ceiling of accuracy with all of this anyways. Usually initial KP is still more accurate than say the coaches/AP polls.
The one thing I really do like about Kenpom is at the end of the season he goes back and actually analyzes the results. He sees what teams he nailed and what ones differed the most (starting rankings to ending).
I feel like all systems should do this.
One thing is certain. The system is definitely well calibrated. If he says a team has a 55% chance of winning, they very well do have a 55% chance of winning. 2024 figures:
I feel like all systems should do this.
One thing is certain. The system is definitely well calibrated. If he says a team has a 55% chance of winning, they very well do have a 55% chance of winning. 2024 figures:
| Predicted Range | W-L | Win% | Exp Win% |
|---|---|---|---|
| 50-60% | 765-623 | 55.1 | 54.9 |
| 60-70% | 867-505 | 63.2 | 64.9 |
| 70-80% | 901-295 | 75.3 | 74.8 |
| 80-90% | 816-136 | 85.7 | 84.8 |
| 90-95% | 367-30 | 92.4 | 92.3 |
| 95-98% | 239-12 | 95.2 | 96.5 |
| 98-99% | 84-2 | 97.7 | 98.5 |
| 99-100% | 84-0 | 100.0 | 99.5 |
For sure no model is 100%. KenPom is used by a lot of gamblers. It’s fine for that. By nature, they should get more accurate by game 40 rather than game 10
I feel like a data science guy who is good at machine learning could destroy this problem. They wouldn’t necessarily be able to explain the metrics used to conclude the rankings, but they have years upon years of sample data to create ML models.
I still feel like Vegas is tough to beat.
I'm sure there's some instance of some wiz kid developing a system that consistently is better. Then he just looks at the Vegas line, looks at his model and bets on the games where the two differ the most.
But most of the time it's just hard to beat Vegas lol
I'm sure there's some instance of some wiz kid developing a system that consistently is better. Then he just looks at the Vegas line, looks at his model and bets on the games where the two differ the most.
But most of the time it's just hard to beat Vegas lol
Once again you assume I lack your intelligence because I disagree with your "OPINION". I'll let you in on a secret, me having a different opinion doesn't mean I can't understand metrics. I just don't agree with criteria used and the outcomes they project.
When you’ve beaten two of that top five with a road loss like the teams we beat and we are 8th? No ream has two wins like us. I saw where lutardi has us as a two seed. Lmao right now we are a solid 1 if the season ended today. We gonna have to earn respect the old fashioned way and make a big tournament run.
Well maybe overvalued is the wrong term. But until you win championships like it or not it will apply.
TBH I'm not sure why this or the comment replied to above should matter.
Gonzaga not winning a title or going to a bunch of Sweet 16s in the past should really have absolutely no bearing on the current year and the ranking this particular season.
He just had his theory he could get them in big games and somehow get them ready for March even though he never changed a thing so idk why he kept thinking it would work other than he hoped each new class would be good enough to dominate like pre 2018.
Yeah and I guess with two of those three games he was kind of forced to be locked in. We always had to be in the champions classic and the Clemson game was just that SEC ACC showdown thing. Not sure Cal really had a choice lol
Wow, I had no idea Sagarin was no longer in use. I also had no idea how hard it was to type Sagarin with autocorrect on.
