2013-2014 Women's D-I PairWise Contentions and Affirmations

[dave1381]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

To be clear, my prediction is Mercyhurst & RMU for the last two slots. Mercyhurst gets #7 based largely on a strong performance against Syracuse for a strong TUC, seriously. Whatever, but that's the criteria. And then I've discussed the RMU-Quinnipiac at length before, but the criteria are 2-2 and the absolute clearest one is head-to-head. There's almost no way in my mind the committee uses a less than .0050 RPI edge for Quinnipiac as the tiebreaker for the comparison (that gap amounts to about a 1-point difference in the standings after adjusting for SOS -- while in comparison there's 2-point gap in the head-to-head results).

Talking about UND before, I'm just saying the criteria is loose enough that one could rationalize just about anything, and UND clearly is the best of these four teams by a mile, and the only reason UND fails to get in is the completely nonsensical way the performance vs. TUC is measured, when UND has the best performance by far against Minnesota and Wisconsin this year. Absent that recognition against the WCHA top 2, UND has only been comparable to other bubble teams in their performance against the next tier of teams. In reality, UND's common opponent advantage over Mercyhurst is far too tiny to justify UND's selection.

You can't justify UND's selection without either a better measure of overall strength than the RPI or a better measure of performance vs. TUC than simple record.

Anyway, enjoy this tournament where the teams get decided by who is exceptional against #15 and not the #1 and #2 teams in every statistical sound ranking available (Rutter/KRACH/WCHODR). Thanks but not thanks committee for not revising the criteria and the travel situation.

(p.s. note "finishing strong" is not a criterion so that won't hurt RMU.

I expect the bracket will be RMU@Minnesota, Mercyhurst@ECAC champ. For the #3-#6 pairings involving Wisconsin/ECAC loser/BC/Harvard, I trust whatever was posted further down in this thread).

 

[brookyone]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

Bump up:

http://www.ncaa.com/liveschedule/2014/03/09

6:00 PM Eastern...5:00 PM Central.

Don't forget...spring ahead!

 

[Eeyore]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

To be clear, my prediction is Mercyhurst & RMU for the last two slots.

You may well be right but I can't overstate just how pathetic that is. The only reason that Quinnipiac loses the comparison with Mercyhurst and is tied with Robert Morris to begin with is that the TUC criterion is so bad. And the single RMU win over Quinnipiac is so heavily overweighted just to give the extra point to RMU in the first place that to then also use it as the tiebreaker is ridiculous. Saying that it's the only criterion that RMU wins handily betrays a lack of understanding of what the numbers are saying. Robert Morris does not win that criterion handily; in fact, they win it by the smallest margin possible. They only look like they win it handily because record vs. TUC and COP are denominated in thousandths of a point and RPI is denominated in ten-thousandths of a point, while HtH record is denominated in integers. That's an artifact of the calculation method, not something that has actual meaning.

 

[Numbers]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

.......
I expect the bracket will be RMU@Minnesota, Mercyhurst@ECAC champ. For the #3-#6 pairings involving Wisconsin/ECAC loser/BC/Harvard, I trust whatever was posted further down in this thread).

Thanks for all the input through the whole process. We all gain from those who are knowledgeable.

I agree about the bracket, but I will write like this:
RMU/Quinn @ Minn (8 v 1) // {Harvard (if BC loses), BC (if they win and Cornell wins), or Cornell (if BC and Clarkson win)} @ Wisc(if Cornell wins), BC(if Cornell loses and BC wins), or Corn(if BC & Cornell both lose) (5 @ 4)
Mercyhurst @ ECAC Champ (7 @ 2) // {BC(if they lose) or Harvard} @ {Clarkson (if they lose) or Wisconsin} 6 @ 3

And this bracket pleases me because it is not West v West, West v West, East v East, East v East. The absence of NoDak and its obligatory bus ride to Minneapolis, and the presence of Mercyhurst and RMU(potentially) as crossovers requires wither Harvard or BC to fly to Madison. I like that for variety.

Also, note that the bracket goes by 'integrity' (whatever that is, with the mess at 7 & 8).

 

[Numbers]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

You may well be right but I can't overstate just how pathetic that is. The only reason that Quinnipiac loses the comparison with Mercyhurst and is tied with Robert Morris to begin with is that the TUC criterion is so bad. And the single RMU win over Quinnipiac is so heavily overweighted just to give the extra point to RMU in the first place that to then also use it as the tiebreaker is ridiculous. Saying that it's the only criterion that RMU wins handily betrays a lack of understanding of what the numbers are saying. Robert Morris does not win that criterion handily; in fact, they win it by the smallest margin possible. They only look like they win it handily because record vs. TUC and COP are denominated in thousandths of a point and RPI is denominated in ten-thousandths of a point, while HtH record is denominated in integers. That's an artifact of the calculation method, not something that has actual meaning.

Eeyore, I assume you would go with Quinn 7th and Mercyhurst 8th? And, if so, could I offer a wish that the committee chooses to fly Quinn to Minneapolis, and bus Mercyhurst to Potsdam or Ithaca, rather than flying Mercyhurst to Minneapolis and busing Quinn to Potsdam or Ithaca. Or even worse, swapping 6 and 7 in order to fly Mercyhurst and Quinn to Minneapolis and Madison with Cornell, Clarkson, Harvard and BC left to play each other. Just for the sake of something different - get Quinn out west.

 

[Eeyore]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

Eeyore, I assume you would go with Quinn 7th and Mercyhurst 8th? And, if so, could I offer a wish that the committee chooses to fly Quinn to Minneapolis, and bus Mercyhurst to Potsdam or Ithaca, rather than flying Mercyhurst to Minneapolis and busing Quinn to Potsdam or Ithaca. Just for the sake of something different - get Quinn out west.

I'm fine either way. The three teams are pretty close but if the committee actually understood the criteria that they use they'd see that Quinnipiac actually performs the best, albeit by very narrow margins, by those metrics. If you are going to use objective criteria to make important decisions, you really ought to go to the trouble to understand them.

To be honest, this process mildly offends me as a hockey fan. It seriously offends me as someone who took the trouble to get a four year degree in statistics.

 

[TonyTheTiger20]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

Hey can one of you mathy types help me out here?

I'm 99% of the way done with an RPI predictor that will spit out EXACT results (rather than the guesstimates on my old sheet) but for the life of me I can't figure out how the RPI calculates the OppWin% and OppOppWin%...

This is what I'm doing -- Let's say a team -- I'll call them State -- played 6 games, 2 each against:

Tech -- Win% of .700
College -- Win% of .600
A&M -- Win% of .500

To calculate opponent's winning percentage, I'm doing:

(.7+.7+.6+.6+.5+.5)/6 -- is this not correct??

Then for OppOppWin% I'm taking each of *those* results and calculated for each team, and then dividing THAT by the number of games played. I'm just not matching up to USCHO's RPI amounts (and I am removing bad wins as well, so that's not it).

What the frigging hell am I missing here?

 

[Numbers]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

Hey can one of you mathy types help me out here?

I'm 99% of the way done with an RPI predictor that will spit out EXACT results (rather than the guesstimates on my old sheet) but for the life of me I can't figure out how the RPI calculates the OppWin% and OppOppWin%...

This is what I'm doing -- Let's say a team -- I'll call them State -- played 6 games, 2 each against:

Tech -- Win% of .700
College -- Win% of .600
A&M -- Win% of .500

To calculate opponent's winning percentage, I'm doing:

(.7+.7+.6+.6+.5+.5)/6 -- is this not correct??

Then for OppOppWin% I'm taking each of *those* results and calculated for each team, and then dividing THAT by the number of games played. I'm just not matching up to USCHO's RPI amounts (and I am removing bad wins as well, so that's not it).

What the frigging hell am I missing here?

TTT -
I think it goes for OppWin%, for each game, you take the Win%age of that opponent (Delete the games played versus the team you are evaluating, so in your example, if you are Calculating BC's RPI, and need Tech's win%age, don't figure in TechvBC games to get Tech's win%age). You add all those #s, one for each game, and then divide by the number of games (So, it's Avg of OppWin%age).
For OppOppWin%age, same thing. You average them, but here you delete no games (repeat, don't delete any games), because that gets too complicated with how everything melds together.

So, I think you have it right for the OppOpp, but for the Opp part, delete the games played between the 2 teams (Like calculating TUC record - you leave out games between the teams in question).

And, the remove bad wins thing is very complicated, because you have to calculate the RPI first, then go back and calculate it without each game (one more calculation for each game), and then remove the bad wins.

Clear?

 

[Numbers]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

I'm fine either way. The three teams are pretty close but if the committee actually understood the criteria that they use they'd see that Quinnipiac actually performs the best, albeit by very narrow margins, by those metrics. If you are going to use objective criteria to make important decisions, you really ought to go to the trouble to understand them.

To be honest, this process mildly offends me as a hockey fan. It seriously offends me as someone who took the trouble to get a four year degree in statistics.

A 4-year degree in statistics is enough to make me sit up and take notice, so I ran a full scale comparison of all criteria. Granted, some of this gets subjective, because the TUC games are against different strength teams in differing amounts of games, and the common opponents are, too.

But, my conclusion is that:
Quinnipiac should win the compare with Robert Morris
Mercyhurst should win the compare with Robert Morris
Quinnipiac should win (very slightly) with Mercyhurst

So, I get Quinn 7 and Mercyhurst 8. Which, for the bracket, would require Mercyhurst @ Minny, Quinn @ ECAC Champ, Harv/BC/Corn @ Wisc, Plus one all eastern game. Both QU and MU have played Clarkson and Cornell a few times already, but I would prefer this stay, and someone else fly to Madison.

 

[dave1381]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

You may well be right but I can't overstate just how pathetic that is. The only reason that Quinnipiac loses the comparison with Mercyhurst and is tied with Robert Morris to begin with is that the TUC criterion is so bad. And the single RMU win over Quinnipiac is so heavily overweighted just to give the extra point to RMU in the first place that to then also use it as the tiebreaker is ridiculous. Saying that it's the only criterion that RMU wins handily betrays a lack of understanding of what the numbers are saying. Robert Morris does not win that criterion handily; in fact, they win it by the smallest margin possible. They only look like they win it handily because record vs. TUC and COP are denominated in thousandths of a point and RPI is denominated in ten-thousandths of a point, while HtH record is denominated in integers. That's an artifact of the calculation method, not something that has actual meaning.

I don't think you're understanding. The committee's logic is like so: suppose your math tells you that if team A and team B played the exact same schedule, team A would be 22-5-3, and team B would be 22-6-2, so they are 1 point apart in the standings (this would be an RPI difference of .0050). Team B went 1-0-1 head-to-head against team A, so they are 2 points apart in the head-to-head standings. You then select team B over team A. I don't think that's so objectionable.

 

[dave1381]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

A 4-year degree in statistics is enough to make me sit up and take notice, so I ran a full scale comparison of all criteria. Granted, some of this gets subjective, because the TUC games are against different strength teams in differing amounts of games, and the common opponents are, too.

But, my conclusion is that:
Quinnipiac should win the compare with Robert Morris
Mercyhurst should win the compare with Robert Morris
Quinnipiac should win (very slightly) with Mercyhurst

So, I get Quinn 7 and Mercyhurst 8. Which, for the bracket, would require Mercyhurst @ Minny, Quinn @ ECAC Champ, Harv/BC/Corn @ Wisc, Plus one all eastern game. Both QU and MU have played Clarkson and Cornell a few times already, but I would prefer this stay, and someone else fly to Madison.

I'm sure you can rationalize anything, but if you're going to start to do anything more sophisticated with the TUC criterion other than assigning a weight of 0 to 1 and looking at the overall TUC record (which I believe is the best model of what the committee does), then I would expect your more sophisticated TUC criterion would also have to put North Dakota into the tournament.

 

[Numbers]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

I don't think you're understanding. The committee's logic is like so: suppose your math tells you that if team A and team B played the exact same schedule, team A would be 22-5-3, and team B would be 22-6-2, so they are 1 point apart in the standings (this would be an RPI difference of .0050). Team B went 1-0-1 head-to-head against team A, so they are 2 points apart in the head-to-head standings. You then select team B over team A. I don't think that's so objectionable.

Hi Dave. I understand what you are saying. It makes sense in a way.

But, Eeyore is saying that if you have such a thing as head-to-head standings, it is impossible for any team to be less than 2 points ahead. So, those 2 points really mean the same thing as (comparing to RPI), "Anywhere from .0005 to somewhere in the neighborhood of .0100 apart." It's a matter of scale in the meaning of the numbers.

If you want to apply the " 1 pt = .005 RPI rule" I will show you the trouble you get. If that is your equivalence, then the difference between one game played between 2 teams spans all the way from NoDame at 9th to Northeastern at 14th. And, remember, this is the slimmest separation possible in pts head-to-head. I don't think anyone really want to make that equivalence. Could you really say that 1 h2h win passes 6 teams? That doesn't seem right.

So, Eeyore is saying that that one H2H point is the cheapest point in all the PRI calculation. If you weighed everything perfectly, it would not be worth one point. And, then, to use it as tiebreaker besides compounds the problem.

That's what he means by, "If the committee understood what their objective numbers really measured..."

Which is not really an argument on my part. You and he are seeing 2 different things, and I am trying to help you out.

 

[Numbers]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

I'm sure you can rationalize anything, but if you're going to start to do anything more sophisticated with the TUC criterion other than assigning a weight of 0 to 1 and looking at the overall TUC record (which I believe is the best model of what the committee does), then I would expect your more sophisticated TUC criterion would also have to put North Dakota into the tournament.

Thanks for the comments. I have a question about this, and it's a vague question, relating to a past year. I seem to remember a time a few years ago, when someone described that the committee would look at the records (they can't use anything but the criteria mentioned), and then for their own evaluation, decide that the TUC part didn't have much real meaning, because someone had lost 4 times to UMD and 4 times to UW, or something like that. And, it seemed the committee did exactly that. Am I remembering that wrong?

 

[dave1381]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

Here's how I would model how the committee actually weights the criteria:

H2H: Award a full point for each head-to-head win.

RPI: Award a full point for a difference of .0100 or more. Award X/.0100 points for a difference of less than .0100 (having the factor be more like .0080 would be better for a 36-game schedule but it won't matter for the comparisons below)

COP: When summing win pct. vs. each COP, award a full point for a difference of 1 or more, otherwise reward the difference.

TUC: This is more guesswork, but reward 1 point if one team is a full game ahead of the other in TUC record. Reward 0 if the SOS of TUC games cannot be compared.

So I gather that the committee has discretion to give Mercyhurst 0 weight for it's TUC record advantage over Quinnipiac. But the committee I don't think has the discretion to say that Quinnipiac did better vs. TUC by some other TUC metric that more seriously adjusts for SOS.

So here's my evaluation of the comparisons using this model:

RMU vs. Quinnipiac
H2H: 1 point for RMU
COP: .025 points for Qpac
RPI: .45 points for Qpac
TUC: 0 points for RMU
RMU wins comparison by .525 points + whatever points it gets for TUC

Mercyhurst vs. Quinnipiac
COP: .308 points for Mercyhurst
RPI: .190 points for Qpac
TUC: 0 points for Mercyhurst
Mercyhurst wins comparison by .108 points + whatever points it gets for TUC

Mercyhurst vs. RMU
RPI: .260 points for Mercyhurst
COP: .475 points for RMU
TUC: at least .215 points for Mercyhurst
Mercyhurst wins comparison barely

Let's also look at UND. No one can compare TUC with UND, so give it zero weight.

UND vs. Mercyhurst
RPI: .650 points for Mercyhurst
COP: .125 points for UND
TUC: 0 points for Mercyhurst
Mercyhurst wins by .525 plus whatever it gets for TUC

UND vs. RMU
RMU wins every category so no reason to bother with the calculation

UND vs. Quinnipiac
RPI: .840 points for Quinnipiac
COP: .583 points for UND
TUC: 0 points for Quinnipiac
Quinnipiac wins by .257 plus whatever it gets for TUC

You can change the weights slightly, but it looks crystal clear to me that RMU and Mercyhurst are in and Quinnipiac are out. And note with RMU vs. Quinnipiac, it's not just about the individual comparison, but RMU also compares favorably relative to Quinnipiac in the comparsion vs. each of Mercyhurst and UND.

 

[dave1381]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

Numbers - yes, I agree the committee gives and has given a lot of weight to head to head, and you can make a fair argument that head-to-head gets too much weight in the process. And yes, teams are unusually bunched together in RPI on the bubble right now, so a head-to-head win can swing a lot of comparisons. But I think the committee wants to give a lot of weight to head-to-head becomes it seems more transparent, and avoids inevitable press comments like "But we beat these guys 5-1 on the ice and we were close otherwise so how can we not be selected? How can you use all this nonsense math instead of going by what happened on the ice?" (hypothetical, and I'm not endorsing that point of view)

 

[Numbers]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

Here's how I would model how the committee actually weights the criteria:

H2H: Award a full point for each head-to-head win.

RPI: Award a full point for a difference of .0100 or more. Award X/.0100 points for a difference of less than .0100 (having the factor be more like .0080 would be better for a 36-game schedule but it won't matter for the comparisons below)

COP: When summing win pct. vs. each COP, award a full point for a difference of 1 or more, otherwise reward the difference.

TUC: This is more guesswork, but reward 1 point if one team is a full game ahead of the other in TUC record. Reward 0 if the SOS of TUC games cannot be compared.

So I gather that the committee has discretion to give Mercyhurst 0 weight for it's TUC record advantage over Quinnipiac. But the committee I don't think has the discretion to say that Quinnipiac did better vs. TUC by some other TUC metric that more seriously adjusts for SOS.

So here's my evaluation of the comparisons using this model:

RMU vs. Quinnipiac
H2H: 1 point for RMU
COP: .025 points for Qpac
RPI: .45 points for Qpac
TUC: 0 points for RMU
RMU wins comparison by .525 points + whatever points it gets for TUC

Mercyhurst vs. Quinnipiac
COP: .308 points for Mercyhurst
RPI: .190 points for Qpac
TUC: 0 points for Mercyhurst
Mercyhurst wins comparison by .108 points + whatever points it gets for TUC

Mercyhurst vs. RMU
RPI: .260 points for Mercyhurst
COP: .475 points for RMU
TUC: at least .215 points for Mercyhurst
Mercyhurst wins comparison barely

Let's also look at UND. No one can compare TUC with UND, so give it zero weight.

UND vs. Mercyhurst
RPI: .650 points for Mercyhurst
COP: .125 points for UND
TUC: 0 points for Mercyhurst
Mercyhurst wins by .525 plus whatever it gets for TUC

UND vs. RMU
RMU wins every category so no reason to bother with the calculation

UND vs. Quinnipiac
RPI: .840 points for Quinnipiac
COP: .583 points for UND
TUC: 0 points for Quinnipiac
Quinnipiac wins by .257 plus whatever it gets for TUC

You can change the weights slightly, but it looks crystal clear to me that RMU and Mercyhurst are in and Quinnipiac are out. And note with RMU vs. Quinnipiac, it's not just about the individual comparison, but RMU also compares favorably relative to Quinnipiac in the comparsion vs. each of Mercyhurst and UND.

This is valuable information, Dave. Are these figures somehow based on the past few years, say 2 or 3, results? If so, it's a great predictive tool.

And, to me, it also points out again the problems that we all know the PWR system has. As an example, in the above, each win h2h is 1 point. It would seem clear that, to compare, say Robert Morris and OSU, on the basis of one game, and give RoMo 1 point is all you can do if you count that way. But, if they played 5 times, and the score was 3-2, it would still be 1 point. Those 2 one-point factors really don't have the same meaning. And, it's hard to quantify the difference, in my mind. One cannot go by win%age, because then you give 1.000 for the first win, and you give 0.20 for the 3-2 record. So, it seems hard to know how to use the h2h in the right way.

Common opponents are difficult to know what to do with also. In the case of QU and RoMo, each played RIT. QU was 1-0, RMU was 3-2. In the ComOpp category, QU gets 1.000, RoMo gets 0.600. Even though, as above, there is probably not that much difference between the teams. How does one use the ComOpp correctly? I don't know. If you go back to total games, then it can skew the other way, where someone is 6-0 against a pure bottom feeder, and someone is 0-4 against Wisconsin, and the record ends up not showing much.

So, really, what we are left with is that RPI should be the favorite tool of the committee. It is supposed to be a "whole body of work" evaluator. But, of course, it has flaws, too. That is why I look at the men's RPI, with its 25/21/54 and the QWB, and the 'remove games' feature, and think "This is contrived. I think the math guys chose all this approximate something else. Like, maybe they looked at KRACH, and thought, 'how can we choose the %ages to make this come out close'. " I think that because the NCAA will never use KRACH. But, they might try to fake it.

So, really the whole PWR thing is kind of fake anyway.

And, you are right, you can massage your own interpretation of these numbers to get whatever conclusion you want. You can sort of subjectively say (I am going to look at Mercy/RoMo now), "TUCs. Well, throw out the RIT results because they are identical. Both teams played 5/6 games against OSU and Syracuse combined. Those 2 teams have similar strength. Mercy's results are slightly better. RoMo went 2-1-1 against BU/QU, and Mercy went 1-1-2 against Clark/Corn. That's a 1/2 game difference, but given that the Clark/Corn is stronger as opponents, I think that is about equal. And, thus you say that Mercy gets a very slight edge for TUCs" You can always massage that kind of thing.

Unfortunately, doing some sort of that kind of analysis is the only way to really do it - because the pure numbers won't tell you much.

Maybe, subjective analysis on the Mercy/RoMo compare goes like this: TUCs as above. We will give a very very slight edge to Mercyhurst (slightly better against OSU/Syr and the rest a wash when adjusting for SoS in some way). ComOpp: Throwing out all identical results, we are left with these differences: RoMo has a win v OSU, where Mercy has a loss. Mercy has a tie against Lindenwood where RoMo has a loss. Mercy has an extra win against Maine. Mercy is 4-0-1 against Syracuse, RoMo is 1-1-2. Now qualitatively, the difference in results against Syracuse, Lindenwood, and Maine should at least balance the results against OSU. So, this should be a draw, or again a slight edge to Mercyhurst. Head-to-head is a draw. Mercyhurst is ahead on RPI. So, a qualitative/subjective analysis gives this compare to Mercyhurst. I believe that Eeyore would consider this way.

As the committee does, however, is your numbers, which yield a much tighter result. The difference between the 2 methods really comes down to one thing: The amount of weight the games against OSU receive.

Again, that's not a good system.

 

[dave1381]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

This is valuable information, Dave. Are these figures somehow based on the past few years, say 2 or 3, results? If so, it's a great predictive tool.

I think the model I described matches up pretty well with what the committee has done since 2008. Note that previously they didn't officially use the new common opponents criterion. I couldn't tell the difference between when the committee was downweighting the old criterion and using the new one, but ARM confirmed that up until this year the committee had been downweighting the old criterion (or so they say). Also note that in past years I think record vs. TUC has played a larger role, like it definitely made the difference in 2011 when BC was selected as host over Minnesota and someone else when they were really close. Here the TUC record tips in the direction of the balance of the other criteria in most cases (the Mercyhurst-RMU comparison is an exception) so overall it doesn't make much of a difference.

But yes, I agree the PWR system has always put a big weight on H2H. I mean one way you could maintain a PWR system using KRACH is to get a KRACH rating for each criterion, then calculate the expected win pct H2H for each criterion and add them up, with the actual H2H win pct. So if KRACH predicted Mercyhurst would win 70% of the time vs. RMU, but RMU beat Mercyhurst 3 of 4 times (75%), then RMU would win the comparison. But the problem there is you'd put way too much on one game for teams who played each other once. So yes, it's more about the whole problem with PWR system, which is why a lot of KRACH advocates just want a straight KRACH selection system.

 

[Numbers]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

I think the model I described matches up pretty well with what the committee has done since 2008. Note that previously they didn't officially use the new common opponents criterion. I couldn't tell the difference between when the committee was downweighting the old criterion and using the new one, but ARM confirmed that up until this year the committee had been downweighting the old criterion (or so they say). Also note that in past years I think record vs. TUC has played a larger role, like it definitely made the difference in 2011 when BC was selected as host over Minnesota and someone else when they were really close. Here the TUC record tips in the direction of the balance of the other criteria in most cases (the Mercyhurst-RMU comparison is an exception) so overall it doesn't make much of a difference.

But yes, I agree the PWR system has always put a big weight on H2H. I mean one way you could maintain a PWR system using KRACH is to get a KRACH rating for each criterion, then calculate the expected win pct H2H for each criterion and add them up, with the actual H2H win pct. So if KRACH predicted Mercyhurst would win 70% of the time vs. RMU, but RMU beat Mercyhurst 3 of 4 times (75%), then RMU would win the comparison. But the problem there is you'd put way too much on one game for teams who played each other once. So yes, it's more about the whole problem with PWR system, which is why a lot of KRACH advocates just want a straight KRACH selection system.

Right. And, it's the same with the men, although they have more parity, so a year like this or last, where UMinn is far ahead and that renders some metrics useless, is more rare with the men.

And, then, the next problem is the way the committee locks itself in to the bracket. For example, this year I am happy for no North Dakota in the field (at least we think not). I understand the reason for being frugal, but it does not give good results in the west, because there are few possible matchups which are bus-able. So, you are stuck with UND/Minn, or UMD/Minn or UMD/Wisc or UMN/Wisc if those teams are on the top half/bottom half of your field, simply because the rules require it. Never mind that UND/Minn have played 11 times in the last 2 years already.

But at least this year it looks like we get something new. If you are right, and RoMo makes it and Cornell wins tomorrow, I would like to see Mercyhurst @ Minnesota and Robert Morris @ Cornell and BC @ Clarkson and Harvard @ Wisconsin. I think that would be all new matchups.

 

[Eeyore]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

I don't think you're understanding. The committee's logic is like so: suppose your math tells you that if team A and team B played the exact same schedule, team A would be 22-5-3, and team B would be 22-6-2, so they are 1 point apart in the standings (this would be an RPI difference of .0050). Team B went 1-0-1 head-to-head against team A, so they are 2 points apart in the head-to-head standings. You then select team B over team A. I don't think that's so objectionable.

I understand fine; my guess is that you are correct as to what the committee is going to do. I just think the committee's logic is defective.

 

[Cali]

Re: 2013-2014 Women's D-I PairWise Contentions and Affirmations

How can the NCAA get this one so wrong?

http://www.ncaa.com/news/icehockey-...8/rochester-inst-heads-ncaa-tourney-after-two