What's new
By:
USCHO Fan Forum

This is a sample guest message. Register a free account today to become a member! Once signed in, you'll be able to participate on this site by adding your own topics and posts, as well as connect with other members through your own private inbox!

  • The USCHO Fan Forum has migrated to a new plaform, xenForo. Most of the function of the forum should work in familiar ways. Please note that you can switch between light and dark modes by clicking on the gear icon in the upper right of the main menu bar. We are hoping that this new platform will prove to be faster and more reliable. Please feel free to explore its features.

The 2014 Pairwise, Bracketology and History Thread`

  • Thread starter Thread starter Priceless
  • Start date Start date
Re: The 2014 Pairwise, Bracketology and History Thread`

Numbers said:
Thanks, Jim. I can't see how it would work to remove a home win and not a road win. Each is worth the same in the oppo and oppo,oppo win percentage. Each counts as 1.000 for a team's own win percentage. One counts 1.2 times, the other 0.8. So I can't reverse engineer that either.
Click to expand...

As you noted, we've all implemented it to only apply the weighting to the win%. As an extreme example, if the win% is beneficial (pulls the RPI up) but the opp win% and opp opp win% are harmful (push the RPI down), it's possible for a win with a .8 weighted win% to have the overall effect of pulling the RPI down while a game against the same opponent with a 1.2 weighted win% could pull it up. There are actually some more subtle quirks that come from one factor being weighted and the others not, but that's the easiest one to understand.

But, as I said in that previous post, I check each win individually, too. So, there's some more subtle difference between how Reilly calculates the drops and how USCHO and I do.
 
Re: The 2014 Pairwise, Bracketology and History Thread`

JimDahl said:
As you noted, we've all implemented it to only apply the weighting to the win%. As an extreme example, if the win% is beneficial (pulls the RPI up) but the opp win% and opp opp win% are harmful (push the RPI down), it's possible for a win with a .8 weighted win% to have the overall effect of pulling the RPI down while a game against the same opponent with a 1.2 weighted win% could pull it up. There are actually some more subtle quirks that come from one factor being weighted and the others not, but that's the easiest one to understand.

But, as I said in that previous post, I check each win individually, too. So, there's some more subtle difference between how Reilly calculates the drops and how USCHO and I do.
Click to expand...

Oh, thanks again Jim. I see now the difference. In previous years, with no weighting, you could do this:

Game 1: .25*w1+.21*o1+.54*oo1 +
Game 2: .25*w2+.21*o2+.54*.oo2 (where w1 is either 1.0, 0.5 or 0.0 for W, T, or L) and so on and add them all up and then divide by the number of games your team in question played. Thus, it was easy to see the effect of removing a game: Just do the addition again without that game involved.

Now, however, it’s way more complicated. I am not sure how you do this one. Can you help me get it?

The definition is still: win%(weighted)*.25 + oppwin%*.21 + oppoppwin%*.54, right? But now, since each part of each game doesn’t have the same weighting, you can’t just remove one game as above and redo the addition.

I wondered at one time whether it went like this:

Game 1: (weighting factor of 1.2, 1.0 or 0.8)*(.25*w1 + .21*o1 +.54*oo1). (ie – the weighting factor applies to whole game) And do the sum like that.
In that case, the oppwin% would not be weighted itself, but rather only through the weighting of the game. It would be like: OK, you have a maximum available to you for this game. The entire contribution of this game is home/road weighted….. This would make it easy to do the calculation, and make it easy to see which games should be removed. You would have to remove all games against an opponent in that scenario. But apparently it is not like that.
 
Re: The 2014 Pairwise, Bracketology and History Thread`

Right, instead of being able to calculate a per game RPI and take the average of those, if you have different weights for the components you instead need to calculate each component separately and then combine them into an RPI. To remove a game, you either need to recalculate the whole thing without that game or remove the effects of that game from each of the components then recombine them.

As to your last paragraph, the memo wasn't clear on how to weight the games. I would kind of prefer fully weighting all components, because I liked the idea of each game having an RPI. It's not clear which is correct, so for now I'm just doing what USCHO does (which is just weight the win%)
 
Re: The 2014 Pairwise, Bracketology and History Thread`

ExileOnDaytonStreet said:
Update:

[table="width: 600"]
[tr]
[td][/td]
[td]Total Apps[/td]
[td]Current Programs[/td]
[td]Apps / Program[/td]
[td]Apps/Program / Year[/td]
[td]Conf. Apps / Year[/td]
[td]2014 Proj. Apps[/td]
[td]2014 Proj. Apps/Program[/td]
[/tr]
[tr]
[td]B1G[/td]
[td]30[/td]
[td]6[/td]
[td]5.00[/td]
[td]0.50[/td]
[td]3[/td]
[td]3[/td]
[td]0.50[/td]
[/tr]
[tr]
[td]NCHC[/td]
[td]43[/td]
[td]8[/td]
[td]5.38[/td]
[td]0.54[/td]
[td]4.3[/td]
[td]2[/td]
[td]0.25[/td]
[/tr]
[tr]
[td]WCHA[/td]
[td]10[/td]
[td]10[/td]
[td]1.00[/td]
[td]0.10[/td]
[td]1[/td]
[td]1[/td]
[td]0.10[/td]
[/tr]
[tr]
[td]HE[/td]
[td]41[/td]
[td]11[/td]
[td]3.73[/td]
[td]0.37[/td]
[td]4.1[/td]
[td]5[/td]
[td]0.45[/td]
[/tr]
[tr]
[td]ECAC[/td]
[td]23[/td]
[td]12[/td]
[td]1.92[/td]
[td]0.19[/td]
[td]2.3[/td]
[td]4[/td]
[td]0.33[/td]
[/tr]
[tr]
[td]AHA[/td]
[td]13[/td]
[td]12[/td]
[td]1.08[/td]
[td]0.11[/td]
[td]1.3[/td]
[td]1[/td]
[td]0.08[/td]
[/tr]
[/table]
Click to expand...

Another update for this week. Still have the Beanpot games to get through, but I don't expect that to change much:


[table="width: 600"]
[tr]
[td][/td]
[td]Total Apps[/td]
[td]Current Programs[/td]
[td]Apps / Program[/td]
[td]Apps/Program / Year[/td]
[td]Conf. Apps / Year[/td]
[td]2014 Proj. Apps[/td]
[td]2014 Proj. Apps/Program[/td]
[/tr]
[tr]
[td]B1G[/td]
[td]30[/td]
[td]6[/td]
[td]5.00[/td]
[td]0.50[/td]
[td]3[/td]
[td]3[/td]
[td]0.50[/td]
[/tr]
[tr]
[td]NCHC[/td]
[td]43[/td]
[td]8[/td]
[td]5.38[/td]
[td]0.54[/td]
[td]4.3[/td]
[td]3[/td]
[td].38[/td]
[/tr]
[tr]
[td]WCHA[/td]
[td]10[/td]
[td]10[/td]
[td]1.00[/td]
[td]0.10[/td]
[td]1[/td]
[td]1[/td]
[td]0.10[/td]
[/tr]
[tr]
[td]HE[/td]
[td]41[/td]
[td]11[/td]
[td]3.73[/td]
[td]0.37[/td]
[td]4.1[/td]
[td]5[/td]
[td]0.45[/td]
[/tr]
[tr]
[td]ECAC[/td]
[td]23[/td]
[td]12[/td]
[td]1.92[/td]
[td]0.19[/td]
[td]2.3[/td]
[td]3[/td]
[td].25[/td]
[/tr]
[tr]
[td]AHA[/td]
[td]13[/td]
[td]12[/td]
[td]1.08[/td]
[td]0.11[/td]
[td]1.3[/td]
[td]1[/td]
[td]0.08[/td]
[/tr]
[/table]

The great normalization continues, although naturally everyone in double digits is still in a precarious situation.
 
Re: The 2014 Pairwise, Bracketology and History Thread`

There will be a more thorough post after the Crockpot tomorrow night.

For now, the brackets are:
Code: 
[B]Bridgeport (Yale)	Worcester (Holy Cross)		Cincinnati (Miami)	St Paul (Minnesota)[/B]
Union			Boston College			Ferris State		Minnesota
Lowell			St Cloud			Quinnipiac		Wisconsin
Cornell			Providence			Michigan		Northeastern
Vermont			Air Force			Duluth			Colgate

Bracket integrity among the 3 and 4 seeds is tossed out the window in favor of attendance since Wisconsin-Michigan and Lowell-Providence would have fouled things up anyway.
 
Re: The 2014 Pairwise, Bracketology and History Thread`

Dutchman said:
If I am readings things correctly, the RPI for CollegeHockeyNews ( http://www.collegehockeynews.com/ratings/ncaapwcr.php ) is slightly different than for USCHO/SiouxSports.com ( http://siouxsports.com/hockey/rankings/pwr/ ) Its not by much. Both claim that they are calcualting the way the NCAA committee wants things calculated. Any thoughts on this?
Click to expand...

yeah... hard to tell and we won't know for sure until we get to the end of the year and there's enough of a difference to matter.
 
Re: The 2014 Pairwise, Bracketology and History Thread`

Dutchman said:
If I am readings things correctly, the RPI for CollegeHockeyNews ( http://www.collegehockeynews.com/ratings/ncaapwcr.php ) is slightly different than for USCHO/SiouxSports.com ( http://siouxsports.com/hockey/rankings/pwr/ ) Its not by much. Both claim that they are calcualting the way the NCAA committee wants things calculated. Any thoughts on this?
Click to expand...

I meant to post about this in this thread; I blogged a little about it earlier last week. In short, the NCAA memo was ambiguous about how to apply the RPI weightings and there are two pretty legitimate ways to do it. USCHO has chosen one and CHN has chosen the other. I've been running both and they're close enough that long range forecasts are pretty comparable, but differences of a couple ranking positions are common enough that YATC or tournament possibilities forecasts will differ depending which you use. Hopefully we won't be waiting until selection day to find out which is right.
 
Re: The 2014 Pairwise, Bracketology and History Thread`

JimDahl said:
I meant to post about this in this thread; I blogged a little about it earlier last week. In short, the NCAA memo was ambiguous about how to apply the RPI weightings and there are two pretty legitimate ways to do it. USCHO has chosen one and CHN has chosen the other. I've been running both and they're close enough that long range forecasts are pretty comparable, but differences of a couple ranking positions are common enough that YATC or tournament possibilities forecasts will differ depending which you use. Hopefully we won't be waiting until selection day to find out which is right.
Click to expand...

Thanks !!!
 
Re: The 2014 Pairwise, Bracketology and History Thread`

JimDahl said:
I meant to post about this in this thread; I blogged a little about it earlier last week. In short, the NCAA memo was ambiguous about how to apply the RPI weightings and there are two pretty legitimate ways to do it. USCHO has chosen one and CHN has chosen the other. I've been running both and they're close enough that long range forecasts are pretty comparable, but differences of a couple ranking positions are common enough that YATC or tournament possibilities forecasts will differ depending which you use. Hopefully we won't be waiting until selection day to find out which is right.
Click to expand...

We chose the approach of applying it only to the winning percentage and not OWP and OOWP for a few reasons, one of which is that the NCAA does that in its basketball RPI, although they use .6 and 1.4 instead of .8 and 1.2 for the multipliers for squeakball.
 
Last edited:
Re: The 2014 Pairwise, Bracketology and History Thread`

Ed Trefzger said:
We chose the approach of applying it only to the winning percentage and not OWP and OOWP for a few reasons, one of which is that the NCAA does that in its basketball RPI, although they use .6 and 1.4 instead of .8 and 1.2 for the multipliers for squeakball.
Click to expand...

to me that's the most logically consistent way. Even if there's a more accurate reading I don't necessarily trust the NCAA to closely read what they wrote :p
 
Re: The 2014 Pairwise, Bracketology and History Thread`

Patman said:
to me that's the most logically consistent way. Even if there's a more accurate reading I don't necessarily trust the NCAA to closely read what they wrote :p
Click to expand...

Here's a problem: It is written on such a way that it can not only be interpreted by US in different ways, but it can be interpreted by this year's committee one way and next year's committee another way. In that case, we can't even be sure we'll know the correct answer after Selection Sunday this year. Isn't that fun?
 
Re: The 2014 Pairwise, Bracketology and History Thread`

JimDahl said:
I meant to post about this in this thread; I blogged a little about it earlier last week. In short, the NCAA memo was ambiguous about how to apply the RPI weightings and there are two pretty legitimate ways to do it. USCHO has chosen one and CHN has chosen the other. I've been running both and they're close enough that long range forecasts are pretty comparable, but differences of a couple ranking positions are common enough that YATC or tournament possibilities forecasts will differ depending which you use. Hopefully we won't be waiting until selection day to find out which is right.
Click to expand...

I wrote to The Minnesota AD, who is on the committee earlier about the QWB divisor, and he answered me, so I wrote him again last week to ask him about this.

The essence of my question was:
Is the calculation: 25%*SUM(winning %age for own games with home/road weighting) + 21%*SUM(opp win% with no weighting)/#ofgames + 54%*SUM(oppoppwin% with no weighting)/#ofgames

Or, is it: SUMoverallgames of {home/road weighting}*(25%*win/loss/tie + 21%*oppfor that game win%age + 54%*Oppforthatgame'soppwin%). and then divide that SUM by the #of games played, counted according to the weighting.

Hopefully he will respond to my question, but it will take awhile, because he will want to talk to the people who actually do the calculation for them.

Both seem possible to me.

And, also, I do not know what it means on CHN where it says something about the RPI listing not reflecting the home/road weighting...
 
Re: The 2014 Pairwise, Bracketology and History Thread`

Numbers said:
I wrote to The Minnesota AD, who is on the committee earlier about the QWB divisor, and he answered me, so I wrote him again last week to ask him about this.

The essence of my question was:
Is the calculation: 25%*SUM(winning %age for own games with home/road weighting) + 21%*SUM(opp win% with no weighting)/#ofgames + 54%*SUM(oppoppwin% with no weighting)/#ofgames

Or, is it: SUMoverallgames of {home/road weighting}*(25%*win/loss/tie + 21%*oppfor that game win%age + 54%*Oppforthatgame'soppwin%). and then divide that SUM by the #of games played, counted according to the weighting.

Hopefully he will respond to my question, but it will take awhile, because he will want to talk to the people who actually do the calculation for them.

Both seem possible to me.

And, also, I do not know what it means on CHN where it says something about the RPI listing not reflecting the home/road weighting...
Click to expand...

That has been corrected. I read somewhere that Adam Wodon said it was an oversight - that when CHN added in the bonus calculations that they forgot to remove that disclaimer.

I found the quote from Wodon's site: http://www.collegehockeynews.com/news/2014/02/03_bracket_abcs_early_edition.php
You have to read the comments section at the end of the article.
 
Last edited:
Re: The 2014 Pairwise, Bracketology and History Thread`

Thanks to the Build Your Own Rankings Calculator we now have 11 years of Pairwise data to study. In that time there have been 20 teams that qualified thanks to the autobid and 156 who qualified by being ranked high enough (autobid or not) to make the NCAA tournament.

Of the 156 teams, 127 (81.4%) that qualified as of the post-Beanpot Pairwise would have made the eventual field. No teams that were ranked in the top 6 after the Beanpot dropped out of the eventual field. Three teams that were ranked #7 and two more that were #8 fell out of the rankings. "Only" 34 of the 44 (77.3%) teams ranked 9-12 qualified.

The 2010 Northern Michigan squad was ranked #23 and still earned at an-large bid. The 2007 UMass Minutemen were #21. The 2013 Wisconsin Badgers joined the group as a #22 seed. Every other team has been in the top 20 post-Beanpot.

Some teams that are on the doorstep have a track record of success in the last month of the season. #18 North Dakota has come from outside the field three times to qualify. #16 Maine and #24 New Hampshire have done it twice. Those three, along with #9 Cornell, are the only teams to do it more than once.

On the other side, Denver has fallen out three times. Ohio State, Northern Michigan and Colorado College have done it twice. Of those currently in the field, #14 Minnesota-Duluth has done it on two occasions.

Historically, the field should look like this.
Teams in bold are locks (historically)
Teams from 7 and 24 are still in the running.
Teams below 25 need to autobid.

Boston College
Minnesota
Union
Ferris State
Quinnipiac
Mass.-Lowell
----
Wisconsin
St. Cloud State
Cornell
Michigan
Providence
Northeastern
Vermont
Minnesota-Duluth
Colgate
-
Maine
Yale
North Dakota
Clarkson
Notre Dame
Minnesota State
Western Michigan
Denver
New Hampshire
----
Ohio State
Alaska-Anchorage
St. Lawrence
Brown
Rensselaer
Bowling Green
Nebraska-Omaha
Air Force
Lake Superior
Michigan Tech
Miami
Alaska-Fairbanks
Bentley
Northern Michigan
Mercyhurst
Connecticut
Massachusetts
Michigan State
Harvard
Bemidji State
Merrimack
Boston University
Robert Morris
Canisius
Colorado College
Niagara
Dartmouth
Penn State
Holy Cross
RIT
American Int'l
Princeton
Sacred Heart
Alabama-Huntsville
Army

The results of the Crockpot did not alter this week's bracketology.
 
Re: The 2014 Pairwise, Bracketology and History Thread`

Priceless said:
Here's a problem: It is written on such a way that it can not only be interpreted by US in different ways, but it can be interpreted by this year's committee one way and next year's committee another way. In that case, we can't even be sure we'll know the correct answer after Selection Sunday this year. Isn't that fun?
Click to expand...

you missed my point Priceless. What is most likely is they'll follow established formulation. Just because it could be taken in altenative ways does not mean it will unless there is a specific rub in this that is different for hockey than basketball. If the language is identical then they'll apply the identical method, correct or not.

I think the NCAA would find it far easier to change two numbers than an entire algorithm.
 
UnionHockeyManiac said:
Jayson Moy's weekly Bracketology Blog: http://www.uscho.com/bracketology/2...and-a-lot-of-moves-make-for-good-story-lines/

I'm ready for the NCAA Tourney to begin and would be very happy if the brackets end up looking like that.
Click to expand...

With that current bracket 3 out of the 4 four seeds could definitely upset the number one seeds with UMD going 1-1-1 this year against Minnesota. Colgate going 2-1 against Ferris St. and a solid Providence team that just fell to a 4th seed with last nights loss (no disrespect to Union). This would make for one difficult bracket to predict as always but would also be fine with this bracket as a 'gate fan. Lots of hockey left however.
 
Re: The 2014 Pairwise, Bracketology and History Thread`

Priceless said:
Historically, the field should look like this.
Teams in bold are locks (historically)
Teams from 7 and 24 are still in the running.
Teams below 25 need to autobid.

Boston College
Minnesota
Union
Ferris State
Quinnipiac
Mass.-Lowell
----
Wisconsin
St. Cloud State
Cornell
Michigan
Providence
Northeastern
Vermont
Minnesota-Duluth
Colgate
-
Maine
Yale
North Dakota
Clarkson
Notre Dame
Minnesota State
Western Michigan
Denver
New Hampshire
----
Ohio State
Alaska-Anchorage
St. Lawrence
Brown
Rensselaer
Bowling Green
Nebraska-Omaha
Air Force
Lake Superior
Michigan Tech
Miami
Alaska-Fairbanks
Bentley
Northern Michigan
Mercyhurst
Connecticut
Massachusetts
Michigan State
Harvard
Bemidji State
Merrimack
Boston University
Robert Morris
Canisius
Colorado College
Niagara
Dartmouth
Penn State
Holy Cross
RIT
American Int'l
Princeton
Sacred Heart
Alabama-Huntsville
Army
Click to expand...

I would argue that Ohio State, if they win their final 8 regular season games, would probably be a #2 seed and would not fall out as an At Large. A few of the teams directly below them, should they run the table and go deep into their tournament, could also qualify as an At Large. Just because it may not have happened in the past 11 years doesn't mean it can't. Remember that PWR metrics have changed (no TUC and addition of bonus) so you can't necessarily compare this season's volatility to what was experienced in the past. So your statement that teams below #25 need to autobid, while probabilistically true, is not 100% accurate.
 
You must log in or register to reply here.
Back
Top