PWR Rating is a JOKE

[Ken \'70]

[Q]jkahn Wrote:

 Why PWR is a joke (note: this is not an anti-Minnesota article, as I am a KRACH believer):
1) RPI, a very important factor in PWR, is not even internally consistent.  RPI is 25% your winning percentage, 50% your opponents' winning %, and 25 % your opponents' opponents winning %.  That last 75% is in there to factor in your strength of schedule.  However, if RPI is really a better measure of a team's strength than winning %, why isn't that final 75% similarly divided into 3 parts,  i.e., if a team had Cornell on its schedule,  Cornell's strength should be factored in as 25% Cornell's record, 50% Cornell's opponent's record and 25% Cornells opp-opp record to be consistent (not 2/3 winning % and 1/3 opp. win %).  I don't believe this would be right either, but just want to point out that opponent's record is very overweighted in the current system, and if RPI is supposedly a better measure than winning percentage, than something that resembles opponent's RPI should be the strength of schedule factor.
2) Here's a hypothethical.  One team plays every team in the country and has a .750 record.  Another team plays only the ten teams that have .750 records and ends up with a .500 record.  Which team is better?  It seems that the first is a .750 team and the second is just as good as the average .750 team.  I believe they are equal.  Yet RPI would rank the second team much higher.

3) Is it really better to lose to Michigan Tech and beat Denver, than the opposite?  I think if you go 1-1 it shouldn't matter whether you've beaten the strong team and lost to the weak or vice versa.  The TUC concept however, makes losses to weak teams count less than losses to strong teams.
4) The whole concept of the TUC cutoff introduces strange variations as opponents go in and out of TUC-land.  In the interest of brevity, I won't give examples - but merely indicate that a step function (TUC, non-TUC) causes all sorts of aberrations.
5) Any system where there is a possibility that you can look back at the end of the season and say that a team would have been better off tying or losing a game than winning it is not a good system.

In addition to the above, since non-conference records of teams in your conference plays a very significant role in strength of schedule (75% of the RPI), the ECAC is at a decided disadvantage because many of its teams are playing these games much earlier in the season than the opponents.  When we opened at OSU in '02-'03, I think it was their 7th game.[/q]

Where to begin?  At the beggining I guess...

1) I don't really get your logic here except that it's a complaint against RPI, of which there are many.  Are you suggesting RPI include 4 categories now, adding an OpOpOp?
2) You don't know what team RPI would rank higher because you haven't considered the OpOp records of either team.  The second team's opponents may have been playing cupcakes to get their .750 record, and the RPI would rank them lower than the first team.
3) TUC is only part of the PWR, and fits very nicely in the overall picture.  Losses to non-TUC teams figure in all of the other PWR components.  This component seeks to give particular weight to how a team does against the good teams.  RPI and COP include records against non-TUC.
4) Is your complaint about how a team gets to be TUC?  It used to be on winning record at or above .500.  It was changed a few years ago to RPI.  In either case, if you want to have a consideration of how a team does against good teams you need a way defining what a good team is.  And yes, teams move in and out of that catagory.  How could it be otherwise?
5) Under what circumstance does PWR reward a team for losing a game?

Non-conference records of a particular team are not necessarily of more or less importance than conference records.  They're only more important if your conference mates do lousy out of conference themselves.  If they do well out of conference, then your conference record may be more important (because of its high SOS) than your out of conference.

[DeltaOne81]

Oh, one other thing. I dunno, lately, I wouldn't be too afraid of meeting Minn, even in Mariucci. They've just not been playing well lately. That can turn around, of course, but the way they've been playing, they shouldn't be too scary for anyone anywhere. Sure, in one game, you could deflinitely lose, but they could definitely definitely lose.

[Ken \'70]

[Q]DeltaOne81 Wrote:

  [Q2]KeithK Wrote:
For those of you thinking Cornell might finish ahead of MN, forget it.  Cornell will lose the BC comparison it's now winning, will lose the MI comparison tonight if MI beats NO, can't flip MN itself, and is at risk of losing the OS comparison once OS plays MS and SLU drops from TUC (under the bonus scenario, SLU is already out of TUC without bonuses).

A final PWR with MN 4 and Cornell 5 is a nightmare for Cornell fans.   Almost any scenario in which MN finishes 5 is a nightmare for the committee.[/Q]
I don't know if we can finish ahead of them, but we *can* win our comparison against MN. All we need to do it take TUC. We can't do it ourself, but we can with a little help from Minn. We can get one more TUC win against SLU, assuming they can keep themselves a TUC.

Minn, meanwhile, can help make UAA a TUC, especially if they lose to them again tonight, and that would add 2 loses (hypothetically) to their TUC column. They also play SCSU, who just  beat CC tonight, and with the way Minn has been playing, they could easily lose those. Are at least one. At 2 loses to UAA, and 2 to SCSU, or even 1-3 total, plus a win for us, and I believe we could flip the comparison with Minn.

Keeping SLU a TUC might be harder than we'd like. There's no way to know if the committee ads RPI bonuses to everyone, or if they might just do it for TUCs - either intentionally, or just by neglect, thinking it wouldn't matter when it really would.



Edited 1 times. Last edit at 02/12/05 12:36PM by DeltaOne81.[/q]

First of all, KeithK didn't write the original quote, I did ("I don't care what you say about me as long as you spell my name right").

Second, all we need to do is take TUC and RPI!  While the winning % of MN and Cornell's Ops are about the same from now to then end of the RS, MN's OpOP is much stronger and they're starting out ahead of us (.5867 to .5847 under a 3-2-1 bonus as of this morning).

While it's not impossible that: AA becomes TUC plus MN loses 3 of 4 to AA and SCSU plus SLU stays TUC; it is extraordinarily low, isn't it?  

Hence my belief that, reasonably, we can't flip the MN comparison.

[jtwcornell91]

[Q]Ken '70 Wrote:
1) I don't really get your logic here except that it's a complaint against RPI, of which there are many.  Are you suggesting RPI include 4 categories now, adding an OpOpOp?
[/Q]
I think what Jeff's getting at is a "recursive RPI" something like the RHEAL rating that Wayne Smith calculates http://maine.edu/HEAL/healiter.html

For example, you could say a team's RRPI is 25% that team's winning percentage and 75% the average RRPIs of their opponents.  (Of course, if you're going to use recursion, why not switch to KRACH and be done with it.)

[Q]
3) TUC is only part of the PWR, and fits very nicely in the overall picture.  Losses to non-TUC teams figure in all of the other PWR components.  This component seeks to give particular weight to how a team does against the good teams.  RPI and COP include records against non-TUC.
[/Q]

But they also include records against TUC, so those games matter more.  Lately I've come around to the point of view that any reasonable rating system should be set up so that if you applied it to a balanced schedule (like the ECAC regular season) it would give the same standings as winning percentage.  The ECAC rewards wins against stronger teams, but only as a tiebreaker.

[Q]
4) Is your complaint about how a team gets to be TUC?  It used to be on winning record at or above .500.  It was changed a few years ago to RPI.  In either case, if you want to have a consideration of how a team does against good teams you need a way defining what a good team is.  And yes, teams move in and out of that catagory.  How could it be otherwise?
[/Q]

Well, you could use a gradual weighting, like calculating a weighted average winning percentage by opponents' RPI.  This is sort of what HEAL and RHEAL do, except they don't consider the strength of teams you lose to at all.

[Q]
5) Under what circumstance does PWR reward a team for losing a game?
[/Q]

Suppose Minnesota goes 4-0 against St. Cloud in the WCHA regular season and then draws them in the WCHA playoffs.  And suppose St. Cloud is right on the cusp of being a TUC, and Minnesota's record vs TUCs determines the outcome of a critical comparison.  It may be that if Minnesota sweeps SCSU, SCSU ends up not being a TUC and Minnesota loses the comparison.  But if they win in three games, the extra win helps SCSU stay a TUC, adding six wins and only one loss to Minnesota's record vs TUCs, flipping the comparison, while the hit in RPI due to the extra loss doesn't end up changing any comparisons.  Then Minnesota does better in the PWR because of losing the extra game than if they'd won the series 2-0 and not played it.

[Josh '99]

You can say what you want about the way Minnesota has been playing lately (and hell, I'll do the same), but come tournament time I still don't want to play them on their (200x100, incidentally) home ice.

[DeltaOne81]

[Q]Ken '70 Wrote:
While it's not impossible that: AA becomes TUC plus MN loses 3 of 4 to AA and SCSU plus SLU stays TUC; it is extraordinarily low, isn't it?  

Hence my belief that, reasonably, we can't flip the MN comparison.
[/q]
Sorry about the misquote. Orignally I had quoted my own post and became confused while editing it

Well, right now its not 3 out of 4, its 2 out of 3. Tonight is particularly important to maybe make UAA a TUC. So its tonight, and 1 out of 2, for a team that's not doing so great. SLU staying a TUC there's a distinct chance. Now I'm not saying that all of those are likely, but they're all very possible, and the possibility of all three is not "extraordinarily low". I wouldn't put even money on it, but I think at the rate that Minnesota has been going, I would be far from surprised if it happened.

[jkahn]

[Q] Ken '70 wrote in response to my post:

1) I don't really get your logic here except that it's a complaint against RPI, of which there are many. Are you suggesting RPI include 4 categories now, adding an OpOpOp?
[/q]
My point is that if the committee truly believes that RPI as currently defined, using only 25% of a team's winning percentage and 75% SOS, is a good measure of a team's strength - then when each opponent is factored into to SOS, shouldn't their strength factors be weighted similarly.   My main point is that RPI is internally inconsistent - for instance right now Cornell and Minnesota have the same RPI, but if RPI is a true measure of the team's abilities, then when calculating a third team's RPI, shouldn't playing Cornell or Minnesota be equal in terms of how it factors into that team's SOS which makes up 75% of its RPI.
As JTW suggests, I was moving toward recursive RPI, except that I deliberately stopped somewhat short of that, so the committee would have a simpler formula.  I'm really not advocating my formula, but just using it to explain faults in the current one.  Personally, I love the mathematical elegance of KRACH, and believe by far that it is the fairest ranking of teams.

One other point though.  The reason we have a national tournament is to see how teams in different parts of the country who don't play each other very often will do against each other.  As such, perhaps, given that six team conferences get one bid, a twelve team conference should be guaranteed two bids - it would help make for a more national tournament.

[Steve M]

I would say it's likely that SCSU takes at least one of two games against the Gophers.  They typically play Minny tough and took them to OT earlier in the season when the Gophers were on fire.  Also SCSU KOd CC on the road, and Minnesota has been struggling.  So if UAA can pull off another upset tonight, the chances for CU to overtake them are decent.

[adamw]

[Q]jkahn Wrote:
One other point though.  The reason we have a national tournament is to see how teams in different parts of the country who don't play each other very often will do against each other.  As such, perhaps, given that six team conferences get one bid, a twelve team conference should be guaranteed two bids - it would help make for a more national tournament.[/q]

That would happen if the Ivies split off, creating 2 separate 6-team conferences.  Something a few people have endorsed, but that I don't like.

[Scersk '97]

[Q]jmh30 Wrote:

 You can say what you want about the way Minnesota has been playing lately (and hell, I'll do the same), but come tournament time I still don't want to play them on their (200x100, incidentally) home ice.
[/q]

Neither would I be all too excited about playing them at the Mullins Center, where the dimensions are (a very strange) 200x95.  Unfortunately, I think that if we end up being a #1 seed, we'll likely get stuck in Amherst.

Being in Chicago, I'm cheering, selfishly, for Grand Rapids.  That would also mean that we're a #2 seed and have avoided meeting Wisconsin--the most undervalued team by PWR when compared against KRACH--in the first round and likely the second.  Personally, my hope is to be the #2 seed in Worcester:  first round payback vs. UNH and confirm a streak vs. BC in the second.  Or, even better, BU ends up as a #4 seed, cranking BC out of Worcester and we end up with, say, Minnesota in the second round, sent out East and playing on a smaller surface than they're used to.

[Ken \'70]

[Q]jtwcornell91 Wrote:

 1) I think what Jeff's getting at is a "recursive RPI" something like the RHEAL rating that Wayne Smith calculates

For example, you could say a team's RRPI is 25% that team's winning percentage and 75% the average RRPIs of their opponents.  (Of course, if you're going to use recursion, why not switch to KRACH and be done with it.)

2) Lately I've come around to the point of view that any reasonable rating system should be set up so that if you applied it to a balanced schedule (like the ECAC regular season) it would give the same standings as winning percentage.  The ECAC rewards wins against stronger teams, but only as a tiebreaker.

3) Well, you could use a gradual weighting, like calculating a weighted average winning percentage by opponents' RPI.  This is sort of what HEAL and RHEAL do, except they don't consider the strength of teams you lose to at all.

[Q2]
4) Under what circumstance does PWR reward a team for losing a game?
[/Q]
Suppose Minnesota goes 4-0 against St. Cloud in the WCHA regular season and then draws them in the WCHA playoffs.  And suppose St. Cloud is right on the cusp of being a TUC, and Minnesota's record vs TUCs determines the outcome of a critical comparison.  It may be that if Minnesota sweeps SCSU, SCSU ends up not being a TUC and Minnesota loses the comparison.  But if they win in three games, the extra win helps SCSU stay a TUC, adding six wins and only one loss to Minnesota's record vs TUCs, flipping the comparison, while the hit in RPI due to the extra loss doesn't end up changing any comparisons.  Then Minnesota does better in the PWR because of losing the extra game than if they'd won the series 2-0 and not played it.[/q]


1) I read this thread as a plaint against PWR as a totality, how its components play against each other and somehow work against determining the best teams.  I'm not a particular proponent of RPI over KRACH or another approach to performance vs SOS.  But one is needed, and it will not suffice by itself instead of the more comprehensive PWR.

2) To pick the best, and sort them in some way among themselves, you need a tool that measures their performance against other very good teams instead of the universe of teams. TUC does that.

3) Even with a gradual weighting you'll have teams moving up and down on a nightly basis.  Less volatile for sure, might be worth a look. (But it would sure be a lot less fun.  I've got my eye on WMU and BGSU right now, dropping those from TUC will cripple MI's TUC record and take a lot of uncertainty out of their comparison with Cornell - let's go OS and NM - yeeha!).

4) Losing to a higher ranked team doesn't always lower your RPI.  Merrimack lost to BC last night and their RPI went up.  Even if SCSU dropped from TUC the damage to MN's RPI would be significant and their COP comparisons would suffer.  In theory, OK it's possible.  But I can't imagine the outcome being so unambiguous, or a college team so craven, that they'd tank a game upon that prospect.

[Ken \'70]

[Q]Scersk '97 Wrote:

 Being in Chicago, I'm cheering, selfishly, for Grand Rapids.  That would also mean that we're a #2 seed and have avoided meeting Wisconsin--the most undervalued team by PWR when compared against KRACH--in the first round and likely the second.  Personally, my hope is to be the #2 seed in Worcester:  first round payback vs. UNH and confirm a streak vs. BC in the second.  Or, even better, BU ends up as a #4 seed, cranking BC out of Worcester and we end up with, say, Minnesota in the second round, sent out East and playing on a smaller surface than they're used to.[/q]


If MN makes the tournament they have to play in Minneapolis.  

[Scersk '97]

Oh, yeah...  duh.  I guess I'd hope for CC, then.  For me, the great victory in seeding will be to somehow avoid Denver and Wisconsin.

Actually, I'll edit myself again.   I'd rather avoid CC and Wisconsin and play Denver.  We're very good at holding teams off 5x5, but I want to avoid the teams that have similarly good special teams to ours.

[Ken \'70]

It's not the AA becoming TUC or MN losing 2 out of the next 3 that's extraordinarily remote, it's SLU staying TUC.  If you think MN is on a crummy streak, take a look at the Saints.