John t whelan ranking simulator

[FlagDUDE08]

Re: John t whelan ranking simulator

After games ending December 6th:

Quality Wins rate:

5.00 St. Cloud State
4.75 Minnesota
4.50 Providence
4.25 Michigan
4.00 Ferris State
3.75 Quinnipiac
3.50 Boston College
3.25 Clarkson
3.00 LSSU
2.75 New Hampshire
2.50 Cornell
2.25 UMASS Lowell
2.00 Notre Dame
1.75 Minnesota State Mankato
1.50 Union
1.25 Northeastern
1.00 Minnesota Duluth
0.75 Miami
0.50 Wisconsin
0.25 Bowling Green

And the tournament field:

Minnesota
St. Cloud State
Providence
Michigan

Ferris State
Quinnipiac
Boston College
LSSU

UMASS Lowell
New Hampshire
Clarkson
Cornell

Notre Dame
Minnesota Duluth (assuming .500 is OK, if not move Northeastern up and throw in Union)
Northeastern
AHA Champ (37 - Mercyhurst)

[FlagDUDE08]

Re: John t whelan ranking simulator

Every team now has a win under their belt. Here's the standings after December 7th:

Quality Wins rate:

5.00 St. Cloud State
4.75 Minnesota
4.50 Providence
4.25 Ferris State
4.00 Michigan
3.75 Boston College
3.50 Clarkson
3.25 Quinnipiac
3.00 UMASS Lowell
2.75 Cornell
2.50 Union
2.25 New Hampshire
2.00 Wisconsin
1.75 Northeastern
1.50 LSSU
1.25 Minnesota State Mankato
1.00 Miami
0.75 North Dakota
0.50 Vermont
0.25 Notre Dame

And the tournament field:

St. Cloud State
Minnesota
Providence
Ferris State

Michigan
Boston College
Quinnipiac
Clarkson

UMASS Lowell
New Hampshire (assuming 500 is allowed)
Union
Cornell

Northeastern
Wisconsin
LSSU
AHA Champ (36 - Mercyhurst)

Bubble teams:
Miami
Minnesota State Mankato
Colgate

[Numbers]

Re: John t whelan ranking simulator

I have some information for Reilly Hamilton, Jim Dahl, and FlagDude.

I heard back from the Minnesota Asst AD, who is on the NCAA Hockey Committee. As far as the QWB points go, this is how you divide:

(Total of QWB points)/(Number of All Games). If you have games removed from your RatingsPI calculation because they are wins, but still reduce your RatingsPI, those games still count in the QWB divisor.

I believe the reasoning is like this:
QWB is supposed to encourage scheduling of strong OOC competition. If you exclude the games from the divisor, that goes against what you are trying to encourage.

All for now.

[Priceless]

Re: John t whelan ranking simulator

This program, SS.com and College Hockey News all have different results when looking at RPI and pairwise. Hopefully there can be a meeting of the minds to determine which is correct before the season concludes. It will be chaos if there are still questions about this new formula in March.

[Ed Trefzger]

Re: John t whelan ranking simulator

This program, SS.com and College Hockey News all have different results when looking at RPI and pairwise. Hopefully there can be a meeting of the minds to determine which is correct before the season concludes. It will be chaos if there are still questions about this new formula in March.

This is why we haven't published USCHO's PWR yet, because we want it to be correct. We expect to have it up around Jan. 1 if not before. At least one of the sources you mentioned is not using the new formula so you can safely assume it is incorrect.

[RHamilton]

Re: John t whelan ranking simulator

I have some information for Reilly Hamilton, Jim Dahl, and FlagDude.

I heard back from the Minnesota Asst AD, who is on the NCAA Hockey Committee. As far as the QWB points go, this is how you divide:

(Total of QWB points)/(Number of All Games). If you have games removed from your RatingsPI calculation because they are wins, but still reduce your RatingsPI, those games still count in the QWB divisor.

I believe the reasoning is like this:
QWB is supposed to encourage scheduling of strong OOC competition. If you exclude the games from the divisor, that goes against what you are trying to encourage.

All for now.

Thanks for the info; I missed this before.

This is close to the way I have it implemented. Instead of "number of all games" I'm using "total possible weight of all games" after accounting for home/away weighting.

My non-cached, auto-updating PWR standings are here: http://pwr.reillyhamilton.com/pwr.php . Still matches up exactly with Jim's version, for which I am thankful.

[Numbers]

Thanks for the info; I missed this before.

This is close to the way I have it implemented. Instead of "number of all games" I'm using "total possible weight of all games" after accounting for home/away weighting.

My non-cached, auto-updating PWR standings are here: http://pwr.reillyhamilton.com/pwr.php . Still matches up exactly with Jim's version, for which I am thankful.

Reilly, I should apologize. I am still not sure. I looked again again at my correspondence. All I can ascertain is that removed games count. My specific correspondence was that the divisor IS home/away weighted.

My correspondence was that the process is:
1) Compute RPI using weighting for home/road results
2) Delete any game with wins that lower the team's RPI
3) Add QWB, using (Total of QWB pts)/(weighted total of games including games deleted in step 2)

What was not made clear was:

1) When computing the Opp% and OppOpp%, do you use the weighting system?
2) When computing QWB, is it definite that the weighted number of games is used?

For myself, I note that RHamilton and JimDahl must be using exactly the same algorithm, but FlagDude is doing something different.

Given Ed's comment above, CHN must be doing something different as well.

[Priceless]

Re: John t whelan ranking simulator

Only teams that have a .500 record or better qualify for an at-large berth. If it comes down to it, will the NCAA consider a team's raw W/L record or the modified W/L record which accounts for home/road games?
Also, is the NCAA only using the modifications for non-conference games? It seems that the idea is to force teams to do better scheduling. A team has no control over its conference schedule, only out of conference games.

[FlagDUDE08]

Re: John t whelan ranking simulator

Only teams that have a .500 record or better qualify for an at-large berth. If it comes down to it, will the NCAA consider a team's raw W/L record or the modified W/L record which accounts for home/road games?
Also, is the NCAA only using the modifications for non-conference games? It seems that the idea is to force teams to do better scheduling. A team has no control over its conference schedule, only out of conference games.

The NCAA only considers conferences in terms of autobids.

[Priceless]

Re: John t whelan ranking simulator

The NCAA only considers conferences in terms of autobids.

This is a major mistake, and a deviation from the past when they accounted for conferences when adjusting RPI (bonus points) and a "bad win" in the conference playoffs.

[JimDahl]

Re: John t whelan ranking simulator

Reilly, I should apologize. I am still not sure. I looked again again at my correspondence. All I can ascertain is that removed games count. My specific correspondence was that the divisor IS home/away weighted.

My correspondence was that the process is:
1) Compute RPI using weighting for home/road results
2) Delete any game with wins that lower the team's RPI
3) Add QWB, using (Total of QWB pts)/(weighted total of games including games deleted in step 2)

What was not made clear was:

1) When computing the Opp% and OppOpp%, do you use the weighting system?
2) When computing QWB, is it definite that the weighted number of games is used?

For myself, I note that RHamilton and JimDahl must be using exactly the same algorithm, but FlagDude is doing something different.

I didn't use the weights for Opp% and OppOpp%. Basketball doesn't and I think it makes more sense not to (given that they're strength of schedule indicators).

I think the QWB divisor is weighted. From the memo: "Once the total bonus points have been calculated they are divided by the total weighting of all games played per the RPI calculation (with road wins/home losses weighted with a factor of 1.2 and home wins/road losses weighted with a factor of 0.8)."

If someone can give me a dump from FlagDude's output (e.g. table of PWR/RPI/record/win%/opp%/oppopp%, etc...) I'm happy to take a look and see if anything jumps out at me. I've implemented it in a few languages now, so that plus matching RHamilton is increasing my confidence that there aren't significant calculation bugs, but there certainly could be some differing assumptions in that the specification is quite ambiguous.

And thanks Numbers -- I thought it a pretty good guess that you don't drop harmful wins from the QWB, but nice to get official confirmation.

[FlagDUDE08]

Re: John t whelan ranking simulator

I didn't use the weights for Opp% and OppOpp%. Basketball doesn't and I think it makes more sense not to (given that they're strength of schedule indicators).

I think the QWB divisor is weighted. From the memo: "Once the total bonus points have been calculated they are divided by the total weighting of all games played per the RPI calculation (with road wins/home losses weighted with a factor of 1.2 and home wins/road losses weighted with a factor of 0.8)."

If someone can give me a dump from FlagDude's output (e.g. table of PWR/RPI/record/win%/opp%/oppopp%, etc...) I'm happy to take a look and see if anything jumps out at me. I've implemented it in a few languages now, so that plus matching RHamilton is increasing my confidence that there aren't significant calculation bugs, but there certainly could be some differing assumptions in that the specification is quite ambiguous.

And thanks Numbers -- I thought it a pretty good guess that you don't drop harmful wins from the QWB, but nice to get official confirmation.

On the application, click on a team on the left side, and you'll get all of that data.

[Numbers]

Re: John t whelan ranking simulator

JimDahl and RHamilton - Just a note, and you probably are aware of this, but the last I checked, your current Final RPI numbers for all teams are identical, except for Bentley and Holy Cross, and I am not sure why that is.

[Numbers]

Re: John t whelan ranking simulator

Also, FlagDude, do you have a recent "Results" file available? I can't find anything past Dec 8 on your website.

Thanks.

[FlagDUDE08]

Re: John t whelan ranking simulator

Also, FlagDude, do you have a recent "Results" file available? I can't find anything past Dec 8 on your website.

Thanks.

I do not have a recent one.