Friday 2/25 other scores

[DeltaOne81]

In the last game of the night, UAA ties MTU 2-2, and UAA drops as a TUC, vaulting Minn back up to tied for 3rd with us. They take the head to head comparison so they're 3rd, we're 4th.

[A-19]

that these irrelevant games with obscure teams are impacting our quest for a #1 seed is absolutely ridiculous.

at this point though, we'll keep doing our part with the W's and prove to the committee we deserve it. if we win through the ECAC champs, that'd be a ridiculous unbeaten streak, and i would have to assume the committee would recognize the quality of play in the second half of the season to justify a #1 seed. then again, alot of distance from now til then, and some real quality opponents (probably clarkson in the playoffs @ lynah, dartmouth and 'gate/harvard in albany).

-mike

[Will]

The committee will believe a #1 seed is justified if the math in the Pairwise Rankings supports us being there, plain and simple.  And really, that's the way it should be.

[Trotsky]

[Q]A-19 Wrote:
that these irrelevant games with obscure teams are impacting our quest for a #1 seed is absolutely ridiculous.[/q]

Irritating, but not ridiculous.  PWR is particularly clunky, with the artificial distinctions of TUC and record against common opponents, but even a more sophisticated system would have this effect.  There are no irrelevant games.  Every decision between every pair of teams bends and shapes the positioning of every team relative to every other.  That's as it should be.

[KenP]

Even though I understand the math, it is ridiculous that Cornell's aspirations for a #1 seed hinge upon Western Michigan and Lake Superior State slugging it out to determine if WMU is just above average or just below average.  The "math" would be better served by changing TUC qualification to RPI (KRACH) over 0.5000 (100)  AND a winning record.

[ninian '72]

Up to a point.  The .500 cutoff is arbitrary and a bit of a blunt instrument.  A better algorithm would not allow subtle shifts across an arbitrary criterion to have this kind of an impact.  However, an alternative - such as weighting wins/losses by the RPI of an opponent -  may be too mathematically impenetrable for lay users of this stuff.

[Josh '99]

[Q]KenP Wrote:
Even though I understand the math, it is ridiculous that Cornell's aspirations for a #1 seed hinge upon Western Michigan and Lake Superior State slugging it out to determine if WMU is just above average or just below average.  The "math" would be better served by changing TUC qualification to RPI (KRACH) over 0.5000 (100)  AND a winning record.[/q](RPI > .5000 and Win% > .5000) is just as sharp a cutoff, and would produce the exact same type of situations, as just using (RPI > .5000).  (For instance, Brown has Win% > .500, and will likely finish that way, but their TUC status depends on just as many seemingly extraneous factors as that of WMU and UAA.)

[jtwcornell91]

[Q]Trotsky Wrote:

 [Q2]A-19 Wrote:
that these irrelevant games with obscure teams are impacting our quest for a #1 seed is absolutely ridiculous.[/Q]
Irritating, but not ridiculous.  PWR is particularly clunky, with the artificial distinctions of TUC and record against common opponents, but even a more sophisticated system would have this effect.  There are no irrelevant games.  Every decision between every pair of teams bends and shapes the positioning of every team relative to every other.  That's as it should be.[/q]

Yeah, but the derivative d(significance)/d(RPI) shouldn't blow up at .500.

[KenP]

Okay, John, you're starting to sway me towards a KRACH-only method.  That would combine the RPI issue (additive not multiplicative), TUC issue (meaningless breakpoints), and COP issue (teams play COP different numbers of times).  But one thing I like about PWR is it's strong weighting of head-to-head results. If the #4 ranked team was 0-2-1 against the #5 team, wouldn't you feel the #5 team deserves the higher seed?  (Note, I used "feel" instead of "think".)

[jkahn]

[Q]KenP Wrote:

 If the #4 ranked team was 0-2-1 against the #5 team, wouldn't you feel the #5 team deserves the higher seed?  (Note, I used "feel" instead of "think".)
[/q]

No.  The H-to-H games do have a fairly significant effect on the teams' relative KRACH which needed to be made up elsewhere for Team #4 to be ahead of Team #5.  By way of example, if baseball team A was 13-6 vs. Team B but finished 2 games behind Team B in the division, would you feel that Team A deserved first place?

[jtwcornell91]

[Q]KenP Wrote:

 Okay, John, you're starting to sway me towards a KRACH-only method.  That would combine the RPI issue (additive not multiplicative), TUC issue (meaningless breakpoints), and COP issue (teams play COP different numbers of times).  But one thing I like about PWR is it's strong weighting of head-to-head results. If the #4 ranked team was 0-2-1 against the #5 team, wouldn't you feel the #5 team deserves the higher seed?  (Note, I used "feel" instead of "think".)
[/q]

Well, if you really wanted to capture the spirit of a league playoff seeding, you could use KRACH with some sort of a PWC "tie-breaker" if two teams are close in RRWP and/or HHWP.  The trick would be how to handle a chain of "almost ties".  Suppose the #4 team is close to the #3 team, and the #5 team is close to the #4 team, but the distance between #5 and #3 is greater than your threshhold.  You don't want a three-team tiebreaker since #5 should not pass #3, but which tie do you break first?  If #4 wins the tiebreak with #3 you never evaluate the tiebreak between #4 and #5.  But then the same thing comes up in any tie-breaker system.

Okay, how's this.  Rank teams by KASA. If a team's KASA RRWP is less than .0125 above the team below them (one tie in a 40-game schedule), throw the teams into a tie-breaker using
1) head-to-head record
2) head-to-head home ice disadvantage (e.g., if you tied them in their barn you win the tiebreak)
3) KASA

So here we go http://www.uscho.com/rankings/?data=kasa
1. CC .8585
2. Denver .8320
3T. Wisconsin .7962
3T. Minnesota .7935

Tiebreaker:
Minnesota is 3-1 vs Wisconsin

3. Minnesota .7935
4T. Wisconsin .7962
4T. Michigan .7872

Tiebreaker:
Wisconsin is 1-0 vs Michigan

4. Wisconsin .7962
5. Michigan .7872
6T. Cornell .7733
6T. BC .7716

Tiebreaker:
BC is 1-0 vs Cornell

6. BC .7716
7T. Cornell .7733
7T. UNH .7617

Tiebreaker:
Cornell and UNH have not played; keep the KASA ordering

7. Cornell .7733
8T. UNH .7617
8T. BU .7501

Tiebreaker:
UNH is 1-0 vs BU

8. UNH .7617
9. BU .7501
10. OSU .7332
11T. UML .7173
11T. North Dakota .7102

Tiebreaker:
UML and NoDak have not played; keep the KASA ordering

11. UML .7173
12. UND .7102
13T. Harvard .6852
13T. NMU .6845

Tiebreaker:
NMU and Harvard tied on neutral ice; keep the KASA ordering

13. Harvard .6852
14. NMU .6845

[DeltaOne81]

How about we just KRACHify the PWR and get it done with. That gets rid of RPI problems, lessens the TUC problem by weighting the lower teams less anyway, and eliminates the lack of SOS in the CoP. Keep the head-to-head, although I feel it should probably only be worth 1 anyway, and tada!

List everything in RRWP, and if there's a tie in that (about as likely as a tie in RPI), then deal with it.

Your math above still picks an "arbitrary" cutoff for what is a tie. Not too much better (well, okay, better, but still arbitrary).

[jtwcornell91]

[Q]DeltaOne81 Wrote:

 How about we just KRACHify the PWR and get it done with. That gets rid of RPI problems, lessens the TUC problem by weighting the lower teams less anyway, and eliminates the lack of SOS in the CoP. Keep the head-to-head, although I feel it should probably only be worth 1 anyway, and tada![/q]

You mean like this?

http://slack.net/~whelan/tbrw/2005/cgi-bin/rankings.cgi?dispPWR=true;PWCdetails=true;PCTweight=25;OPPweight=50;OOPweight=25;topqual=15;homebon=.0010;neutbon=.0020;roadbon=.0030;rpifudge=playoff;PWCtb=RRWP;PWCtbwt=1;PWCh2hwt=1;PWCh2h=total;PWCtucwt=1;TUCdefcrit=rpi;TUCdefrel=ge;TUCdefcut=.500;PWCtuccrit=rrwp;PWCtucomit=true;PWClastwt=0;PWClastnum=16;PWClastcrit=pct;PWCcomwt=1;PWCcommingm=1;PWCcommintm=1;PWCcomcrit=rrwp;scoresel=current;scores=

BTW, I just fixed a bug that was keeping the "count H2H as one criterion" option not work.  Now it does.

[DeltaOne81]

I am aware of your script and yup, like that.

I realize it doesn't produce vastly different results, but it does at least produce them in a way that is less prone to swings and turns and random stupidities.