Most compresensive ranking/seeding method

[JonathanW_rivals]

There are flaws with every way we can try to rank teams (media polls, coaches polls, computer rankings, Quad system, etc..., even Vegas). So I have put together a way that uses all of these, ranking them within each category (AP poll, Coaches poll, NET rank, Q1 Wins, W2+ Losses, Q1 Win %, Vegas Odds, and the Average Massey Composite which averages out about 40 different computer rankings), but I did not weight them all equally. I weighted Vegas at 30%, NET at 20%, the AP & Coaches polls at 5% each, and each of the 3 Quad systems at 10% and the Massey composite at 10%.

What I get is:
1 Seeds: Gonzaga 1.55, Baylor 4.3, Arizona 4.85, UK 5.0
2 Seeds: Auburn 5.95, KU 7.0, Villanova 9.7, Duke 9.75
3 Seeds: UT 9.95, Purdue 10.2, Tex Tech 12.45, UCLA 14.3
4 Seeds: Illinois 15.3, Wisc 15.4, Houston 15.7, Ark 16.5
5 Seeds: Texas, Prov, St. Marys, UConn
6 Seeds: LSU, Alabama, Iowa, Ohio St
7 Seeds: USC, Murray St, Col. St., UNC
8 Seeds: Memphis, Marq, San Fran, Mich St

This would likely make our 2-seed KU, our 3-seed either Purdue or UCLA, our 4 could be any except Ark.

What hurts us is our Q1 Win % (8-6), with only Nova having a worse among the other 1-2 seeds.

 

[St.PatterSoN-54-]

There are flaws with every way we can try to rank teams (media polls, coaches polls, computer rankings, Quad system, etc..., even Vegas). So I have put together a way that uses all of these, ranking them within each category (AP poll, Coaches poll, NET rank, Q1 Wins, W2+ Losses, Q1 Win %, Vegas Odds, and the Average Massey Composite which averages out about 40 different computer rankings), but I did not weight them all equally. I weighted Vegas at 30%, NET at 20%, the AP & Coaches polls at 5% each, and each of the 3 Quad systems at 10% and the Massey composite at 10%.

What I get is:
1 Seeds: Gonzaga 1.55, Baylor 4.3, Arizona 4.85, UK 5.0
2 Seeds: Auburn 5.95, KU 7.0, Villanova 9.7, Duke 9.75
3 Seeds: UT 9.95, Purdue 10.2, Tex Tech 12.45, UCLA 14.3
4 Seeds: Illinois 15.3, Wisc 15.4, Houston 15.7, Ark 16.5
5 Seeds: Texas, Prov, St. Marys, UConn
6 Seeds: LSU, Alabama, Iowa, Ohio St
7 Seeds: USC, Murray St, Col. St., UNC
8 Seeds: Memphis, Marq, San Fran, Mich St

This would likely make our 2-seed KU, our 3-seed either Purdue or UCLA, our 4 could be any except Ark.

What hurts us is our Q1 Win % (8-6), with only Nova having a worse among the other 1-2 seeds.

Very compresensive indeed.


J/K Jon. Interesting stuff here.

 

[yabbadabbadoo_rivals213851]

No
That system is no good
Gotta figure something else out
Still keeps Duke at a two seed

 

[weatherbird]

There are flaws with every way we can try to rank teams (media polls, coaches polls, computer rankings, Quad system, etc..., even Vegas). So I have put together a way that uses all of these, ranking them within each category (AP poll, Coaches poll, NET rank, Q1 Wins, W2+ Losses, Q1 Win %, Vegas Odds, and the Average Massey Composite which averages out about 40 different computer rankings), but I did not weight them all equally. I weighted Vegas at 30%, NET at 20%, the AP & Coaches polls at 5% each, and each of the 3 Quad systems at 10% and the Massey composite at 10%.

What I get is:
1 Seeds: Gonzaga 1.55, Baylor 4.3, Arizona 4.85, UK 5.0
2 Seeds: Auburn 5.95, KU 7.0, Villanova 9.7, Duke 9.75
3 Seeds: UT 9.95, Purdue 10.2, Tex Tech 12.45, UCLA 14.3
4 Seeds: Illinois 15.3, Wisc 15.4, Houston 15.7, Ark 16.5
5 Seeds: Texas, Prov, St. Marys, UConn
6 Seeds: LSU, Alabama, Iowa, Ohio St
7 Seeds: USC, Murray St, Col. St., UNC
8 Seeds: Memphis, Marq, San Fran, Mich St

This would likely make our 2-seed KU, our 3-seed either Purdue or UCLA, our 4 could be any except Ark.

What hurts us is our Q1 Win % (8-6), with only Nova having a worse among the other 1-2 seeds.

Seems like these are double dipping a bit. If we are going to use the NET ranking, then there is no reason to use any of the Quad stuff because it is basically already being used in the NET ranking. Additionally, I don't think using the Vegas odds works because it is all about getting people to bet on each side.

And I will say that using a bunch of systems that all have some flaws doesn't necessarily start to negate those flaws, it can, in fact, magnify them.

People will complain regardless of the system used, even if it is some type of composite system. I am actually in favor of the NET system, as in addition to KenPom, it probably does the best job of ranking teams. There are flaws in them, but they basically have no bias to them (even though there will be some that say there is bias built in to the programming). At least with computer systems everybody is being rated the same, on the same criteria. The only thing I wished they would do is just do away with the committee, use the computer system rankings to seed the teams, put the top seeds in the closest regional, and then S-Curve it. Don't want to be sent across the country?? Then be a 1 seed.

You can see my post in the Burns thread. Regardless of whether you use the ranking, or look at quality wins, bad losses, whatever, there really is no debate, UK should be a 1 seed. And if you were being honest, Auburn should be the other 1 seed if we are just looking at top 30 wins, and outside the top 30 losses. If the committee doesn't get lazy, and just look at Q1 records, UK has a 1 seed locked up. They have the best road win of the group, only Baylor has a better worst loss, and that is debateable as it was #47 at home, versus #50 on the road. Only Auburn has a better neutral site win at #25, UK was #33. UK has beaten #7 by 18, #8 by 28, and #33 by 29. Nobody has more impressive wins than that.

 

[JonathanW_rivals]

Seems like these are double dipping a bit. If we are going to use the NET ranking, then there is no reason to use any of the Quad stuff because it is basically already being used in the NET ranking. Additionally, I don't think using the Vegas odds works because it is all about getting people to bet on each side.

And I will say that using a bunch of systems that all have some flaws doesn't necessarily start to negate those flaws, it can, in fact, magnify them.

People will complain regardless of the system used, even if it is some type of composite system. I am actually in favor of the NET system, as in addition to KenPom, it probably does the best job of ranking teams. There are flaws in them, but they basically have no bias to them (even though there will be some that say there is bias built in to the programming). At least with computer systems everybody is being rated the same, on the same criteria. The only thing I wished they would do is just do away with the committee, use the computer system rankings to seed the teams, put the top seeds in the closest regional, and then S-Curve it. Don't want to be sent across the country?? Then be a 1 seed.

You can see my post in the Burns thread. Regardless of whether you use the ranking, or look at quality wins, bad losses, whatever, there really is no debate, UK should be a 1 seed. And if you were being honest, Auburn should be the other 1 seed if we are just looking at top 30 wins, and outside the top 30 losses. If the committee doesn't get lazy, and just look at Q1 records, UK has a 1 seed locked up. They have the best road win of the group, only Baylor has a better worst loss, and that is debateable as it was #47 at home, versus #50 on the road. Only Auburn has a better neutral site win at #25, UK was #33. UK has beaten #7 by 18, #8 by 28, and #33 by 29. Nobody has more impressive wins than that.

Valid points. But in Statistics, the bigger the sample size, the more accurate the estimate to the truth. And in any kind of statistical analysis, the more variables you can include in your model, the more accurate your model. However there is a breaking point because at some point if you keep adding variables you will make it so that your model cannot converge (won't work).

I wrote a computer ranking algorithm probably 15 years ago, and it was one of the ones Massey put on his website, but I was to busy to keep it going. Every ranking (computer too) has it's biases, in what data is uses or doesn't use, in how it weights each part of data.

KenPom is one of many in the Massey composite. In my algorithm I used the raw game data from Ken Pomeroy's website. But there is no reason to think his is any better or worst than any other algorithm.

As for Vegas, I think it does have some advantages. As it is their prediction of the likelihood of a given team to win the title. But as you note, there is some bias there too, because for example you may see popular teams getting more money bet on them than they should (so probably applies to UK, Duke, UNC, and honestly maybe Gonzaga).

As for the NET ranking, I don't trust it, so I would rather get rid of it than the Quad stuff. Partly because we don't know what drives the NET ranking. But when you look at every component outside of MoV and Off/Def efficiency, there is no way NET should have Houston anywhere close to as high as they do.

 

[JonathanW_rivals]

No
That system is no good
Gotta figure something else out
Still keeps Duke at a two seed

They could drop by Sunday, as many top teams will have more chances at Q1 wins, and less chances at Q2+ loss than Duke this week. But it would probably have to be UT or Purdue to leap them, possibly Tex Tech if they won the B12.

 

[weatherbird]

Valid points. But in Statistics, the bigger the sample size, the more accurate the estimate to the truth. And in any kind of statistical analysis, the more variables you can include in your model, the more accurate your model. However there is a breaking point because at some point if you keep adding variables you will make it so that your model cannot converge (won't work).

I wrote a computer ranking algorithm probably 15 years ago, and it was one of the ones Massey put on his website, but I was to busy to keep it going. Every ranking (computer too) has it's biases, in what data is uses or doesn't use, in how it weights each part of data.

KenPom is one of many in the Massey composite. In my algorithm I used the raw game data from Ken Pomeroy's website. But there is no reason to think his is any better or worst than any other algorithm.

As for Vegas, I think it does have some advantages. As it is their prediction of the likelihood of a given team to win the title. But as you note, there is some bias there too, because for example you may see popular teams getting more money bet on them than they should (so probably applies to UK, Duke, UNC, and honestly maybe Gonzaga).

As for the NET ranking, I don't trust it, so I would rather get rid of it than the Quad stuff. Partly because we don't know what drives the NET ranking. But when you look at every component outside of MoV and Off/Def efficiency, there is no way NET should have Houston anywhere close to as high as they do.

Kentucky Baylor and Houston are 3-5 in both the NET and KenPom. Not knowing what drives the NET is part of what makes it good in my opinion. Because nobody knows what drives the calculation, it is harder for teams to scheme the system.

What I mean by bias is that humans tend to value teams more or less based on their personal feelings. The computer systems can't do that. Everybody is evaluated against the same criteria. Maybe the calculation values offensive rebounding more, or maybe 3 or percentage more, or maybe it's just efficiency. Whether we agree or not, at least all teams are evaluated the same. As with any mathematical formula, the result is only as good as the input. Junk in = Junk out. But if you read the articles on the NET and it's updates to the formula, they basically used Google to look at previous teams that did well in the tournament, and essentially google developed the formula, basically crunching the statistical data of previous teams. They did it again a couple years ago, and modified it slightly. It's not perfect, but nothing is ever going to be.

 

[KY Allen Hunt]

There are flaws with every way we can try to rank teams (media polls, coaches polls, computer rankings, Quad system, etc..., even Vegas). So I have put together a way that uses all of these, ranking them within each category (AP poll, Coaches poll, NET rank, Q1 Wins, W2+ Losses, Q1 Win %, Vegas Odds, and the Average Massey Composite which averages out about 40 different computer rankings), but I did not weight them all equally. I weighted Vegas at 30%, NET at 20%, the AP & Coaches polls at 5% each, and each of the 3 Quad systems at 10% and the Massey composite at 10%.

What I get is:
1 Seeds: Gonzaga 1.55, Baylor 4.3, Arizona 4.85, UK 5.0
2 Seeds: Auburn 5.95, KU 7.0, Villanova 9.7, Duke 9.75
3 Seeds: UT 9.95, Purdue 10.2, Tex Tech 12.45, UCLA 14.3
4 Seeds: Illinois 15.3, Wisc 15.4, Houston 15.7, Ark 16.5
5 Seeds: Texas, Prov, St. Marys, UConn
6 Seeds: LSU, Alabama, Iowa, Ohio St
7 Seeds: USC, Murray St, Col. St., UNC
8 Seeds: Memphis, Marq, San Fran, Mich St

This would likely make our 2-seed KU, our 3-seed either Purdue or UCLA, our 4 could be any except Ark.

What hurts us is our Q1 Win % (8-6), with only Nova having a worse among the other 1-2 seeds.

Okay, how does your system work historically? The selection committee does change ranking methodology, and the humans on the committee, so is there any evidence that your method has more predictive power than any other scheme?

 

[JonathanW_rivals]

Okay, how does your system work historically? The selection committee does change ranking methodology, and the humans on the committee, so is there any evidence that your method has more predictive power than any other scheme?

Lol, I just came up with it today. And don't have the time or energy to go back to look at prior years.