View Full Version : NCAA Selection Process

Numbers

11-03-2013, 10:28 PM

Hey gang,

Just checking on something else that I want to make sure I have right. Is it correct that the TUC part of the comparisons for the tournament selection has been totally axed?

Also, I am not quite understanding the bonus part of the RPI calculation. The press release from September seems to say there a 5.00 point and decreasing bonus for wins/ties against high ranking teams, but I am missing how the decimal points fit together.

Thanks.

Numbers

MavHockey14

11-03-2013, 11:48 PM

Hey gang,

Just checking on something else that I want to make sure I have right. Is it correct that the TUC part of the comparisons for the tournament selection has been totally axed?

Also, I am not quite understanding the bonus part of the RPI calculation. The press release from September seems to say there a 5.00 point and decreasing bonus for wins/ties against high ranking teams, but I am missing how the decimal points fit together.

Thanks.

Numbers

I was told there'd be no math...:p

FlagDUDE08

11-04-2013, 09:04 AM

Hey gang,

Just checking on something else that I want to make sure I have right. Is it correct that the TUC part of the comparisons for the tournament selection has been totally axed?

Also, I am not quite understanding the bonus part of the RPI calculation. The press release from September seems to say there a 5.00 point and decreasing bonus for wins/ties against high ranking teams, but I am missing how the decimal points fit together.

Thanks.

Numbers

Let's see how I can explain this, at least based upon what my calculator does (shameless plug: http://www.wa2aea.com for a free download ); because I e-mailed Board about the calculations on this site and it was mentioned to me that they will be looking into their calculations come January:

A team's PairWise Ranking is calculated using three factors: RatingsPI, Common Opponents, and Head-to-Head play, with RatingsPI serving as the tiebreaker (hence why I have it worth 1.1 on my app). All 59 teams are given a PairWise Ranking.

As for the bonus, take the RatingsPI of all 59 teams BEFORE any quality bonus is calculated, on a scale of 0 to 100 (as opposed to 0 to 1; simply take the RatingsPI you'd normally see from 0 to 1 and multiply it by 100). Now, order these teams by decreasing RatingsPI. This is the reason why I leave this as an option for sorting in my application. The #1 team is worth 5.00 points to any team that beats it, #2 worth 4.75, #3 worth 4.50, and so on. This order of which teams are worth how many points will remain the same regardless of if a team advances position due to quality wins. Currently as an example, if you include Sunday's game, BU jumps from 21st to 12th solely due to quality wins, but BU is not retroactively worth quality wins bonus points, so a team like RPI is out of luck there from beating BU, because they are considered 21st, and therefore not worth a quality win bonus. Also, remember that a tie (which includes shootout win/loss) is officially defined as one-half win and one-half loss, so if you tied the #1 team, it would be worth 2.50 points, #2 worth 2.375, #3 worth 2.25, and so on.

Numbers

11-04-2013, 09:31 AM

Let's see how I can explain this, at least based upon what my calculator does (shameless plug: http://www.wa2aea.com for a free download ); because I e-mailed Board about the calculations on this site and it was mentioned to me that they will be looking into their calculations come January:

A team's PairWise Ranking is calculated using three factors: RatingsPI, Common Opponents, and Head-to-Head play, with RatingsPI serving as the tiebreaker (hence why I have it worth 1.1 on my app). All 59 teams are given a PairWise Ranking.

As for the bonus, take the RatingsPI of all 59 teams BEFORE any quality bonus is calculated, on a scale of 0 to 100 (as opposed to 0 to 1; simply take the RatingsPI you'd normally see from 0 to 1 and multiply it by 100). Now, order these teams by decreasing RatingsPI. This is the reason why I leave this as an option for sorting in my application. The #1 team is worth 5.00 points to any team that beats it, #2 worth 4.75, #3 worth 4.50, and so on. This order of which teams are worth how many points will remain the same regardless of if a team advances position due to quality wins. Currently as an example, if you include Sunday's game, BU jumps from 21st to 12th solely due to quality wins, but BU is not retroactively worth quality wins bonus points, so a team like RPI is out of luck there from beating BU, because they are considered 21st, and therefore not worth a quality win bonus. Also, remember that a tie (which includes shootout win/loss) is officially defined as one-half win and one-half loss, so if you tied the #1 team, it would be worth 2.50 points, #2 worth 2.375, #3 worth 2.25, and so on.

So, FlagDude,

That seems a huge bonus to me. I remember before, the bonus was (.005, .003, .001) which in this calculation that you are doing would equate to .5 because it is 5 points out of 1000. But, you are doing 5 points out of 100 for a win against Minnesota? Wow!

I like the idea that it grades out as you go down the list, but it still seems a work in progress.

Also, can you explain the math on the 1.2 and 0.8 business?

Here is what I don't understand:

Is it "Road Games count 0.8 and Home Games count 1.2"? or, is it "Road Wins and Home Losses count 1.2, and Home Wins and Road Losses count 0.8"?

Because the latter seems contrived to me.

I understand that there needs to be some adjustment for playing most OOC games at home, but that is a totally out of left field way to do it.

Thanks for the help.

Numbers

Numbers

11-04-2013, 09:32 AM

Let's see how I can explain this, at least based upon what my calculator does (shameless plug: http://www.wa2aea.com for a free download ); because I e-mailed Board about the calculations on this site and it was mentioned to me that they will be looking into their calculations come January:

A team's PairWise Ranking is calculated using three factors: RatingsPI, Common Opponents, and Head-to-Head play, with RatingsPI serving as the tiebreaker (hence why I have it worth 1.1 on my app). All 59 teams are given a PairWise Ranking.

As for the bonus, take the RatingsPI of all 59 teams BEFORE any quality bonus is calculated, on a scale of 0 to 100 (as opposed to 0 to 1; simply take the RatingsPI you'd normally see from 0 to 1 and multiply it by 100). Now, order these teams by decreasing RatingsPI. This is the reason why I leave this as an option for sorting in my application. The #1 team is worth 5.00 points to any team that beats it, #2 worth 4.75, #3 worth 4.50, and so on. This order of which teams are worth how many points will remain the same regardless of if a team advances position due to quality wins. Currently as an example, if you include Sunday's game, BU jumps from 21st to 12th solely due to quality wins, but BU is not retroactively worth quality wins bonus points, so a team like RPI is out of luck there from beating BU, because they are considered 21st, and therefore not worth a quality win bonus. Also, remember that a tie (which includes shootout win/loss) is officially defined as one-half win and one-half loss, so if you tied the #1 team, it would be worth 2.50 points, #2 worth 2.375, #3 worth 2.25, and so on.

So, FlagDude,

That seems a huge bonus to me. I remember before, the bonus was (.005, .003, .001) which in this calculation that you are doing would equate to .5 because it is 5 points out of 1000. But, you are doing 5 points out of 100 for a win against Minnesota? Wow!

I like the idea that it grades out as you go down the list, but it still seems a work in progress.

Also, can you explain the math on the 1.2 and 0.8 business?

Here is what I don't understand:

Is it "Road Games count 0.8 and Home Games count 1.2"? or, is it "Road Wins and Home Losses count 1.2, and Home Wins and Road Losses count 0.8"?

Because the latter seems contrived to me.

I understand that there needs to be some adjustment for playing most OOC games at home, but that is a totally out of left field way to do it.

Thanks for the help.

Numbers

EDIT: Sorry for the duplicate post.

FlagDUDE08

11-04-2013, 09:54 AM

So, FlagDude,

That seems a huge bonus to me. I remember before, the bonus was (.005, .003, .001) which in this calculation that you are doing would equate to .5 because it is 5 points out of 1000. But, you are doing 5 points out of 100 for a win against Minnesota? Wow!

I like the idea that it grades out as you go down the list, but it still seems a work in progress.

Also, can you explain the math on the 1.2 and 0.8 business?

Here is what I don't understand:

Is it "Road Games count 0.8 and Home Games count 1.2"? or, is it "Road Wins and Home Losses count 1.2, and Home Wins and Road Losses count 0.8"?

Because the latter seems contrived to me.

I understand that there needs to be some adjustment for playing most OOC games at home, but that is a totally out of left field way to do it.

Thanks for the help.

Numbers

You're welcome, and I can go through an explanation on that.

The latter is true, and is also true in other cases where it exists, such as the RatingsPI calculation for basketball.

The weights were put in to encourage schools to have a reason to travel instead of just playing at home. The premise is that a team is expected to win at home and lose on the road. Why should you then be punished more for losing on the road, something you were expected to do (unknown what the probability spread in the long term actually is), than you would be for losing at home? That doesn't make sense. This is why, when the outcome of the game is determined, the weight is set for both teams and points apportioned accordingly. If the home team wins, this was expected, so the game is worth 0.8 points to both teams, with the home team receiving credit for all the points because they still won. If the away team wins, this was NOT expected, so there is a bonus for "beating the odds", in that the game is worth 1.2 points to both teams, and naturally the away team gets credit for all of the points. A tie is still defined as one-half win and one-half loss, so technically BOTH outcomes happen. The potential of points becomes 1.0, as both the 0.8 and 1.2 weights are applied with credit given according to whether you were the home or away team, and use the distributive property at the end to take the one-half.

BTW, I'm not sure if you're on Version 0.3 of my calculator or not, but that is out.

Numbers

11-04-2013, 10:10 AM

You're welcome, and I can go through an explanation on that.

The latter is true, and is also true in other cases where it exists, such as the RatingsPI calculation for basketball.

The weights were put in to encourage schools to have a reason to travel instead of just playing at home. The premise is that a team is expected to win at home and lose on the road. Why should you then be punished more for losing on the road, something you were expected to do (unknown what the probability spread in the long term actually is), than you would be for losing at home? That doesn't make sense. This is why, when the outcome of the game is determined, the weight is set for both teams and points apportioned accordingly. If the home team wins, this was expected, so the game is worth 0.8 points to both teams, with the home team receiving credit for all the points because they still won. If the away team wins, this was NOT expected, so there is a bonus for "beating the odds", in that the game is worth 1.2 points to both teams, and naturally the away team gets credit for all of the points. A tie is still defined as one-half win and one-half loss, so technically BOTH outcomes happen. The potential of points becomes 1.0, as both the 0.8 and 1.2 weights are applied with credit given according to whether you were the home or away team, and use the distributive property at the end to take the one-half.

BTW, I'm not sure if you're on Version 0.3 of my calculator or not, but that is out.

Thanks FD. I like the explanation. Thanks. I just hate the whole idea though. NCAA is trying to tweak a formula that really as arbitrary (RatingsPI). The 25%, 21% and 54% is really just a guess. There is no inherent reason for those number. They are just guesses that seem to give good answers. The 1.2/0.8 is the same way. It's trying to force a bad tool to do a good job.

(FWIW - If they go to that great of lengths, and all the extra calculating, they should just to a home/away Bradley-Terry. It's no more confusing that the RPI with all those adjustments, and it would give a more accurate result. And, I hate the name KRACH, so I try to say "Bradley-Terry".)

FlagDUDE08

11-04-2013, 10:22 AM

Thanks FD. I like the explanation. Thanks. I just hate the whole idea though. NCAA is trying to tweak a formula that really as arbitrary (RatingsPI). The 25%, 21% and 54% is really just a guess. There is no inherent reason for those number. They are just guesses that seem to give good answers. The 1.2/0.8 is the same way. It's trying to force a bad tool to do a good job.

(FWIW - If they go to that great of lengths, and all the extra calculating, they should just to a home/away Bradley-Terry. It's no more confusing that the RPI with all those adjustments, and it would give a more accurate result. And, I hate the name KRACH, so I try to say "Bradley-Terry".)

Would you prefer what basketball has, 25/50/25 with a weighting of 1.4/0.6? Of course, they also don't use PairWise, but rather subjectively vote on who makes the tournament.

Do you have a link with the Western Bia- er KRACH formula?

Numbers

11-04-2013, 10:30 AM

Would you prefer what basketball has, 25/50/25 with a weighting of 1.4/0.6? Of course, they also don't use PairWise, but rather subjectively vote on who makes the tournament.

Do you have a link with the Western Bia- er KRACH formula?

I would not prefer what basketball has. 25/50/25 is also arbitrary. They are just number pulled from nowhere.

I have no problem with the home/road thing. That is a good idea. But, again, the 1.2/0.8 thing is arbitrary.

I can find a link with the math for a home/road dependent Bradley-Terry, but I don't know of anyone who is actually using it.

And, I assume you really do know there is no east/west bias in the method. The only 'bias' comes from the lack of a huge number of OOC games, which should be reduced this year given that the B10 schedule is only 20 games long.

FlagDUDE08

11-04-2013, 10:48 AM

I would not prefer what basketball has. 25/50/25 is also arbitrary. They are just number pulled from nowhere.

I have no problem with the home/road thing. That is a good idea. But, again, the 1.2/0.8 thing is arbitrary.

I can find a link with the math for a home/road dependent Bradley-Terry, but I don't know of anyone who is actually using it.

And, I assume you really do know there is no east/west bias in the method. The only 'bias' comes from the lack of a huge number of OOC games, which should be reduced this year given that the B10 schedule is only 20 games long.

We only call it Western Bias because it's only the westerners who want to use it because the PWR is supposedly filled with Eastern Bias. :p:D

Numbers

11-04-2013, 10:56 AM

We only call it Western Bias because it's only the westerners who want to use it because the PWR is supposedly filled with Eastern Bias. :p:D

Oh. Thanks. I wouldn't say PWR has Eastern bias at all. I think it's contrived math.

And I am starting to think that NCAA keeps it because they are stubborn or they fear they can't explain Bradley. But what the final RPI comes out to be now is hardly explainable either.

Patman

11-04-2013, 11:46 AM

Oh. Thanks. I wouldn't say PWR has Eastern bias at all. I think it's contrived math.

And I am starting to think that NCAA keeps it because they are stubborn or they fear they can't explain Bradley. But what the final RPI comes out to be now is hardly explainable either.

Right, but they understand the inputs and rationale.

With B-T its harder. It's a maximum likelihood solution but then you'd have to explain likelihood.

Butler (KRACH) spends the time to sort of turn it into a heuristic by mentioning the preservation of wins such that the count number is equal to the number of wins predicted over the same schedule.

Bradley-Terry don't go over this. Either because they didn't know, didn't care to know, or didn't see it was relevant. B-T was providing their method to assess farm animal scoring in a pairwise manner... Further B-T got their own model wrong in the sense that they should have had a random effects model... But GLMMs don't appear until later (1970s).

IMO, my current view is that B-T was western biased in the sense of the SEC bias, likely nonexistent but appearance is a nature of scheduling and statistical leverage.

purpleinnebraska

11-04-2013, 01:18 PM

B-T was providing their method to assess farm animal scoring in a pairwise manner...

IMO, my current view is that B-T was western biased in the sense of the SEC bias, likely nonexistent but appearance is a nature of scheduling and statistical leverage.

Well yes, if it has to do with farm animals, it clearly has a western/SEC bias.

Numbers

11-04-2013, 02:03 PM

Flag,

More RPI related math questions. Since this calculation uses a home and away weighting, do the opponents records get the same treatment before their percentages are added to make the base line RPI?

FlagDUDE08

11-04-2013, 02:08 PM

Flag,

More RPI related math questions. Since this calculation uses a home and away weighting, do the opponents records get the same treatment before their percentages are added to make the base line RPI?

Only the team's win percentage receives the weighting. The calculation of the OWP and OOWP does not include any sort of weighting whatsoever. Just your simple 1.0 win, 0.5 tie, 0.0 loss.

Drew S.

11-04-2013, 03:09 PM

What would people think about a champions league type setup with say two wildcards? I think with a setup like this teams would have a better idea where they stand and wouldn't rely so much on what other teams do. This would be a massive improvement over what they are doing now.

Numbers

11-04-2013, 03:37 PM

MODS - I am thankful that FlagDude08 has answered my initial question. There have been many, many threads through the years about PWR vs KRACH vs other options. I fear that I should not have mentioned such, and I am not sure we need another such thread here.

As OP (and I am not sure that even makes a difference, sorry) it seems fine to me if this thread is closed for the present.

FlagDUDE08

11-04-2013, 05:17 PM

MODS - I am thankful that FlagDude08 has answered my initial question. There have been many, many threads through the years about PWR vs KRACH vs other options. I fear that I should not have mentioned such, and I am not sure we need another such thread here.

As OP (and I am not sure that even makes a difference, sorry) it seems fine to me if this thread is closed for the present.

You are able to actually close the thread yourself, if you wish. However, it may be smart to keep it open in case others have questions about how all of this is calculated.

Ralph Baer

11-04-2013, 05:34 PM

You are able to actually close the thread yourself, if you wish. However, it may be smart to keep it open in case others have questions about how all of this is calculated.

I thought that you can only close threads yourself in the Scores Forum and I think in the Alumni Forum.

FlagDUDE08

11-09-2013, 10:38 AM

As of the games finish November 8th, the top 20 teams from which the Quality Wins Bonus is derived are:

5.00 Quinnipiac

4.75 Minnesota

4.50 St. Cloud State

4.25 Miami

4.00 Providence

3.75 Michigan

3.50 Boston College

3.25 Northeastern

3.00 Ferris State

2.75 LSSU

2.50 Bowling Green

2.25 North Dakota

2.00 Wisconsin

1.75 UMASS Lowell

1.50 Notre Dame

1.25 Western Michigan

1.00 Brown

0.75 St. Lawrence

0.50 Vermont

0.25 Minnesota Duluth

RPI, Bemidji State, and Boston University are within 0.50 RatingsPI points of this cliff.

The current Tournament Field is as follows:

Boston College

Miami

Quinnipiac

Minnesota

Providence

St. Cloud State

Notre Dame

Michigan

North Dakota

Wisconsin

UMASS Lowell

St. Lawrence

Northeastern

LSSU

Ferris State

AHA Champ (Air Force)

Powered by vBulletin® Version 4.2.2 Copyright © 2019 vBulletin Solutions, Inc. All rights reserved.