Announcement

Collapse
No announcement yet.

Pairwise = RPI

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    Re: Pairwise = RPI

    Not quite, Shirtless. Just to take an example, If you have, say, .002 QWB points and an opponent has none, this only switches the comparison when you're within 0.002 of each other, which is to say almost never. But if 0.002-0.000 QWB points gave you a full point, then that combined with, say Common Opponents will flip a comparison. So small differences in QWB between two teams mean almost nothing now, but would be huge in the TUC-like system. Of course, big QWB differences count just as you say. Yale is getting a whopping 0.0071 QWB points for four different wins and that is helping them a lot. But the higher your QWB, the harder it is for them to drop through a few odd results. As teams drop, the effects drop which is what eliminates the cliff.

    Comment


    • #17
      Re: Pairwise = RPI

      Originally posted by 5mn_Major View Post
      The PWR and KRACH are notoriously different for women's D1. The WCHA is stacked and deep with talent (with every NC winner except that of 2014). So the WCHA SOS are off the charts, typically includes all of the top 5 or 6...and top rated teams in the east have SOS in the range of >10. So Wisconsin has only lost to Minnesota...and outside of that beating everyone else. This gives it the same record as BC (who has the best record). UW has a much tougher schedule than BC even if you take Minnesota out of the numbers. Yet, the PWR doesn't have them 1st (as having the best combination of quality wins), it doesn't have them 2nd (after Minnesota), it doesn't even have them 3rd...it has them 4th. That is 4th even though UW has the 2nd toughest schedule in the country and the team at 3rd has the 20th toughest schedule.

      Many feel PWR penalizes SOS vs. KRACH.
      KRACH has the ability to differentiate between groups... to some extent. Any time when you have a game there is a winner and a loser... duh. But how do you reconcile that when you compare teams? Anything involving a win% or an average of win% then a game between two GOOD teams has to be a net loss. Likewise a game between two bad teams is a net win. Whereas, KRACH, if the collective performance is good enough the net effect of a game between two good teams leaves them relatively the same within realm of good teams... and same the other way.

      This effect is real.

      ---

      I think the decreased "schedule effect" induced by the broadening of out of conference opponents has helped reign this in... but if you want to still see in action... look at basketball RPI vs. computer systems.
      BS UML '04, PhD UConn '09

      Jerseys I would like to have:
      Skating Friar Jersey
      AIC Yellowjacket Jersey w/ Yellowjacket logo on front
      UAF Jersey w/ Polar Bear on Front
      Army Black Knight logo jersey


      NCAA Men's Division 1 Simulation Primer

      Comment


      • #18
        Re: Pairwise = RPI

        Originally posted by goblue78 View Post
        Not quite, Shirtless. Just to take an example, If you have, say, .002 QWB points and an opponent has none, this only switches the comparison when you're within 0.002 of each other, which is to say almost never. But if 0.002-0.000 QWB points gave you a full point, then that combined with, say Common Opponents will flip a comparison. So small differences in QWB between two teams mean almost nothing now, but would be huge in the TUC-like system. Of course, big QWB differences count just as you say. Yale is getting a whopping 0.0071 QWB points for four different wins and that is helping them a lot. But the higher your QWB, the harder it is for them to drop through a few odd results. As teams drop, the effects drop which is what eliminates the cliff.
        It's all about softening the TUC cliff but still making wins over quality opponents mean something. I do wish there was a 4 component to allow jumping a team with a better adjusted RPI without playing H2H since the schedules are so insular and game locations matter. How do we compare MTU to UML? MTU wins the common opp, UML wins the RPI, UML wins the comparison on RPI tiebreaker...that seems very boring and overly simplistic. Without finding a way to jump UML in RPI it's impossible for MTU to pass UML. But at the same time, should MTU being 2-1 vs Michigan while UML is 0-1 be a huge determining factor? Probably not.
        Michigan Tech Legend, Founder of Mitch's Misfits, Co-Founder of Tech Hockey Guide, and Creator/Host of the Chasing MacNaughton Podcast covering MTU Hockey and the WCHA.

        Sports Allegiance: NFL: GB MLB: MIL NHL: MIN CB: UW CF: UW CH: MTU FIFA: USA MLS: MIN EPL: Everton

        Comment


        • #19
          Re: Pairwise = RPI

          Originally posted by Patman View Post
          KRACH has the ability to differentiate between groups... to some extent. Any time when you have a game there is a winner and a loser... duh. But how do you reconcile that when you compare teams? Anything involving a win% or an average of win% then a game between two GOOD teams has to be a net loss. Likewise a game between two bad teams is a net win. Whereas, KRACH, if the collective performance is good enough the net effect of a game between two good teams leaves them relatively the same within realm of good teams... and same the other way.

          This effect is real.

          ---

          I think the decreased "schedule effect" induced by the broadening of out of conference opponents has helped reign this in... but if you want to still see in action... look at basketball RPI vs. computer systems.
          To be fair, there is still a rather large separation of east and west.

          Michigan Tech has no common opponents with Harvard, Yale, Vermont, Merrimack, Providence, Colgate and Quinnipiac (I stopped after the top 20).
          Mankato: BU, UML, Vermont, Providence, BC.
          BGSU: BU, UML, Denver, BC
          Michigan Tech Legend, Founder of Mitch's Misfits, Co-Founder of Tech Hockey Guide, and Creator/Host of the Chasing MacNaughton Podcast covering MTU Hockey and the WCHA.

          Sports Allegiance: NFL: GB MLB: MIL NHL: MIN CB: UW CF: UW CH: MTU FIFA: USA MLS: MIN EPL: Everton

          Comment


          • #20
            Re: Pairwise = RPI

            Originally posted by Patman View Post
            Please don't lecture to me
            Mea culpa.

            Was speaking, generally, and not to you specifically, although, that's what I actually did.

            Comment


            • #21
              Re: Pairwise = RPI

              Originally posted by Red Cows View Post
              Mea culpa.

              Was speaking, generally, and not to you specifically, although, that's what I actually did.
              and I tend to over-react
              BS UML '04, PhD UConn '09

              Jerseys I would like to have:
              Skating Friar Jersey
              AIC Yellowjacket Jersey w/ Yellowjacket logo on front
              UAF Jersey w/ Polar Bear on Front
              Army Black Knight logo jersey


              NCAA Men's Division 1 Simulation Primer

              Comment


              • #22
                Re: Pairwise = RPI

                Dude, not to go totally off topic here, but:
                Originally posted by 5mn_Major View Post
                So Wisconsin has only lost to Minnesota...and outside of that beating everyone else.
                False.
                Yet, the PWR doesn't have them 1st (as having the best combination of quality wins), it doesn't have them 2nd (after Minnesota), it doesn't even have them 3rd...it has them 4th.
                It had them barely 4th, before QU lost tonight. Now it has them third. KRACH had and still has Wisconsin 3rd. It also has BC soundly in 1st.

                Also, Wisconsin went 0-3-1 against Minnesota; you are high on some insane cocktail of drugs if you think Wisconsin should be ranked #1 in *any* women's hockey ranking right now.

                smgdh.
                Last edited by TonyTheTiger20; 01-21-2015, 11:03 PM.
                Grant Salzano, Boston College '10
                Writer Emeritus, BC Interruption
                Twitter: @Salzano14


                Click here for the BC Interruption Pairwise, KRACH, and GRaNT Calculators

                Comment


                • #23
                  Re: Pairwise = RPI

                  Originally posted by Shirtless Guy View Post
                  I think you are missing the importance of the QWB. The only reason UNO is ahead of BGSU is because of the QWB and that can be adjusted by results independent of those two teams. If a team like Denver falls back a bit and UNO no longer gets a QWB for sweeping Denver on Jan 9/10 they lose a 0.028 bump in RPI. Michigan Tech is only ahead of Vermont now because of Michigan continuing to improve and MTU getting a QWB for sweeping Michigan in October. The QWB is basically replacing the TUC cliff and softening it by giving more benefit to defeating difficult opponents, instead of treating every team above the cliff the same, there is wide variety.

                  There is a bonus of 0.06 to RPI for winning at Mankato (#1). There is a bonus of 0.002 to RPI for winning at home vs Quinnipiac (#20). There is no bonus for beating #21 and lower.




                  I don't think it is a whole lot different when you consider the adjusted RPI is the tiebreaker so its like getting a whole point for having the most QWB.
                  Vermont has one win against a team with a winning record. Im glad we are that high but how are we that high?

                  Comment


                  • #24
                    Re: Pairwise = RPI

                    Originally posted by catmandoo View Post
                    Vermont has one win against a team with a winning record. Im glad we are that high but how are we that high?
                    Maybe you didn't play the best competition, just the right ones.

                    Comment


                    • #25
                      Re: Pairwise = RPI

                      Originally posted by catmandoo View Post
                      Vermont has one win against a team with a winning record. Im glad we are that high but how are we that high?
                      Vermont is probably benefiting from the RPI being based largely on Opponent's Opponent's Win % instead of Opponent's. The traditional RPI is 25% Winning%, 50% Opponent's W%, 25% Opp Opp's W%.
                      The current hockey formula is 25/21/54. Vermont drops to 18th if you change the calc to the older 25/50/25

                      Comparing the top 16 of each RPI - 25/21/54 to 25/50/25:
                      UMD climbs 3 spots from 5 to 2
                      Harvard climbs 3 spots from 7 to 4
                      UNO drops 3 spots from 2 to 5
                      BU climbs 1 spot from 8 to 7
                      UND drops 4 spots from 4 to 8
                      Merrimack climbs 7 spots from 17 to 10
                      Michigan climbs 3 spots from 15 to 12
                      Yale drops 3 spots from 10 to 13
                      Denver drops 1 spot from 13 to 14
                      Providence drops 1 spot from 14 to 15
                      Vermont drops 6 spots from 12 to 18
                      Michigan Tech Legend, Founder of Mitch's Misfits, Co-Founder of Tech Hockey Guide, and Creator/Host of the Chasing MacNaughton Podcast covering MTU Hockey and the WCHA.

                      Sports Allegiance: NFL: GB MLB: MIL NHL: MIN CB: UW CF: UW CH: MTU FIFA: USA MLS: MIN EPL: Everton

                      Comment


                      • #26
                        Re: Pairwise = RPI

                        Originally posted by FlagDUDE08 View Post
                        Maybe you didn't play the best competition, just the right ones.
                        I think it's all because RPI emphasizes Opps Opps W% and since Vermont hasn't played BC, BU or UML yet, that means their conference mates have played them ALOT.
                        Michigan Tech Legend, Founder of Mitch's Misfits, Co-Founder of Tech Hockey Guide, and Creator/Host of the Chasing MacNaughton Podcast covering MTU Hockey and the WCHA.

                        Sports Allegiance: NFL: GB MLB: MIL NHL: MIN CB: UW CF: UW CH: MTU FIFA: USA MLS: MIN EPL: Everton

                        Comment


                        • #27
                          Re: Pairwise = RPI

                          Fighting Sioux 23, the calc changed from 25/50/25 to 25/21/54 to drastically reduce the number of games that negatively impact a team's RPI so that less games would get excluded when they changed that rule. The new RPI calc is most definitely 25/21/54:
                          http://www.collegehockeynews.com/info/?d=krach
                          http://www.uscho.com/rankings/rpi/d-i-men/
                          Michigan Tech Legend, Founder of Mitch's Misfits, Co-Founder of Tech Hockey Guide, and Creator/Host of the Chasing MacNaughton Podcast covering MTU Hockey and the WCHA.

                          Sports Allegiance: NFL: GB MLB: MIL NHL: MIN CB: UW CF: UW CH: MTU FIFA: USA MLS: MIN EPL: Everton

                          Comment


                          • #28
                            Re: Pairwise = RPI

                            Originally posted by TonyTheTiger20 View Post
                            Dude, not to go totally off topic here, but:

                            False.
                            It had them barely 4th, before QU lost tonight. Now it has them third. KRACH had and still has Wisconsin 3rd. It also has BC soundly in 1st.

                            Also, Wisconsin went 0-3-1 against Minnesota; you are high on some insane cocktail of drugs if you think Wisconsin should be ranked #1 in *any* women's hockey ranking right now.

                            smgdh.
                            So here's solid evidence that the PWR does not take into account SOS.

                            Say the women's Q and BC played each other every game for the rest of the season. Even though they are PWR 1 and 4 respectively, both teams would see a drop in PWR scores with losses. An even 5-5 split would be devastating for both team's PWR (also reducing their SOS for each other and reducing their PWR even more than it would have been). Outcome could be one team goes from an easy top PWR overall to having a tough road outcome for the NCAAs and the other could go from having a home NCAA series to being knocked out of the tourney altogether. If you're either team, why would you want this outcome?
                            Go Gophers!

                            Comment


                            • #29
                              Re: Pairwise = RPI

                              Originally posted by 5mn_Major View Post
                              So here's solid evidence that the PWR does not take into account SOS.

                              Say the women's Q and BC played each other every game for the rest of the season. Even though they are PWR 1 and 4 respectively, both teams would see a drop in PWR scores with losses. An even 5-5 split would be devastating for both team's PWR (also reducing their SOS for each other and reducing their PWR even more than it would have been). Outcome could be one team goes from an easy top PWR overall to having a tough road outcome for the NCAAs and the other could go from having a home NCAA series to being knocked out of the tourney altogether. If you're either team, why would you want this outcome?
                              I am not saying PWR is good. It's pretty crap. But my point is that saying "PWR is crap because Wisconsin is only 3rd" is pretty asinine given that KRACH also has them -- by a considerable margin in both directions -- in third.
                              Grant Salzano, Boston College '10
                              Writer Emeritus, BC Interruption
                              Twitter: @Salzano14


                              Click here for the BC Interruption Pairwise, KRACH, and GRaNT Calculators

                              Comment


                              • #30
                                Re: Pairwise = RPI

                                Originally posted by Red Cows View Post
                                The tweak exists. KRACH:

                                http://www.collegehockeynews.com/new...18_thelast.php

                                http://www.collegehockeynews.com/new.../27_nomore.php

                                http://www.collegehockeynews.com/info/?d=krach

                                Although Pairwise and KRACH do not look terribly different from one another at the moment.
                                Those are some interesting articles, Red Cows. For years, I've posted that there are improved statistical systems available (e.g., KRACH) for evaluating team rankings and ultimately, tournament team selection. Most respondents to my posts "didn't get it" (IMO) and would cite "simple math" or other things for keeping the current system. In a way, with all the patches and fixes, it is a bit like keeping Windows 95 to the present day, with patches, updates, etc. It is really simple to simply say the currently used system and all that comes with it served its purpose, its time to move on.

                                There is one problem which likely is slowing progress. The level of sophistication or comprehension of certain quantitative concepts, or lack thereof, must be addressed. How many will "vote" for a system them do not understand? The use of high level ratio data (which KRACH has) requires a certain level of statistical comprehension most of us might lack. Educate everyone? No. Decrease the sense of a threat somehow? Yes. One way is to show it is not a radical change, simply an improvement.
                                No more slow starts to the season; we got over that hurdle. Now, Get it on!!! GO UND!!!

                                New nick is coming up. I recommend Spirithawks.

                                Character is what a person is; reputation is what he is supposed to be. (Abbott)

                                Comment

                                Working...
                                X