Around D-I tonight

UND blows a 3-0 lead after 1 to loose 5-3 (w/ ENG) to CC .

Gotta also root for the long shot UAA against Minn, which is probably at about the end of the first, but I can't find details.

1-0 Minnesota in the first - not sure of the time; just saw the post on USCHO

You are doing a lot of wishing if you want UAA to win a game...they are really bad.

Seeing as Minn just went up 2-0, yeah... but maybe playing them will pull Minn's RPI down anyway .

Results since Tuesday, by ranking on Tuesday:

1 Maine
2 NoDak L-CC
3 CC W-NoDak
4 Cornell L-Colgate
5 UNH
6 BC W-UMass
7 Michigan W-Ferris
8 Minny W-UAA
9 Ferris L-Michigan
10 OSU T-NoDame

If Maine beats UNH tomorrow, we win the TUC category versus UNH and therefore pick up another comparison win. Also, should UND lose again, we'd have a chance of picking up the comparison against them, if their RPI drops enough. If not, we'd tie the comparison (though apparently they're going with the Whelan method of comparison ties now).

By playing UAA tonight, Minn's RPI dropped 0.0044, shifting things around enough to put us in 6th in the PWR, instead of the bottom of a 3-way tie for 5th . Minn still wins the comparison though, with an RPI 0.0014 higher than ours. Their RPI should drop again by playing UAA again, but even a win over Colgate would probably drop ours (though Colgate has a slightly higher RPI than UAA), so it'll be close.

I need to sleep .

DeltaOne81 '03 wrote:

Also, should UND lose again, we'd have a chance of picking up the comparison against them, if their RPI drops enough. If not, we'd tie the comparison (though apparently they're going with the Whelan method of comparison ties now).

I'm not sure what you mean by that. Individual comparisons are never tied, because RPI is the tie-breaker. If you mean "breaking ties in the PWR" that's a concept that's only meaningful to those of us who are trying to rank the teams according to total comparison wins. Right now, the old-style PWR table still breaks PWR ties by RPI, while the comparison grid breaks them by individual comparisons. But in both cases, the tables indicate that teams are "tied in the PWR" because they win more comparisons. Remember when you say "they" that USCHO is not the NCAA.

Their RPI should drop again by playing UAA again, but even a win over Colgate would probably drop ours (though Colgate has a slightly higher RPI than UAA), so it'll be close.

Remember, RPI (1 part pct, 2 parts opct, 1 part oopct) is not RPI's measure of strength of schedule. It's better to use RPIStr (2 parts pct, 1 part opct), although leaving out head-to-head games actually hurts us here, since Colgate's win over us won't count in determining their winning percentage for these purposes. At any rate, UAA's winning percentage is so bad that their RPIStr is only .3012 compared to Colgate's .4322. So playing UAA hurts your strength of schedule worse than playing Colgate (or anyone else but Fairfield and Princeton).

The most encouraging item in John's posting above is that UAA will likely not join Wisconsin and Michigan Tech as TUCs under the new RPI scheme, which, IMHO, is much too skewed toward whom you've played as opposed to how you've done.

[Q]I'm not sure what you mean by that.[/Q]
The way I remember it on USCHO in the past (in fact, I distinctly remembet them stating it in one of their 'how the PWR works' pieces, is that if an individual comparison is tied, then it's tied. Nobody gets a point for it and RPI is only used to break overall ties.

While your method breaks individual ties with RPI (as well as overall ones). But now that USCHO has adopted your comparison grid structure, you can't have no one winning a comparison, or there will be no where to put the link to view it. So, taking a look at the BC-UNH comparison shows it tied at 2, yet given to UNH for the higher RPI (which does indeed count in their 24 comparison wins - if I counted their row correctly).

What I was saying about UND, is that they currently beat us 1-0 (on RPI only - other categories tied). If they lose tomorrow, we'll win the TUC category, making it 1-1. Under the "old" USCHO system, we may not have picked up a comparison win, but they would have lost one. Currently, unless they're RPI drops significantly, they'll still get the point.

But definitely route for Maine tonight, it'll give us the TUC category versus UNH, and give us that comparison 2-1. Perhaps our RPI will even advance enough (with respect to UNH's - currently only 0.0014 apart) to give us a 3-0 sweep. Plus a Maine win helps our RPI, albeit barely.

[Q] the new RPI scheme, which, IMHO, is much too skewed toward whom you've played as opposed to how you've done.[/Q]
Agreed, I know it's a MAAC team, but when Quinnipiac has 15 wins, and it's much higher than a team that's 1-17, I have issues with that. I don't care how good the teams you play are, if you only have 1 win in 20 games, you don't deserve anything. Perhaps the old 35-50-15 was too little weight... anyone up for 30-50-20?

Also, one would think the multiplicative KRACH would take care of this, but alas it doesn't. Quinnipiac (due to SOS I suppose) is listed lower than UAA. In fact it says UAA should beat QU about 7 out of 10 times. As I said once near the end of last year, my feeling is KRACH gives SOS too much weight, and this seems to back it up, IMHO.

DeltaOne81 '03 wrote:

[Q]I'm not sure what you mean by that.[/Q]
The way I remember it on USCHO in the past (in fact, I distinctly remembet them stating it in one of their 'how the PWR works' pieces, is that if an individual comparison is tied, then it's tied. Nobody gets a point for it and RPI is only used to break overall ties.

That is simply not true, and has never been. No PWR ever published by USCHO has ever contained a tied comparison. The pairwise comparison process, as described by the NCAA's Championship Handbooks, explicitly says that RPI is used in a tie-breaker when each team wins the same number of criteria:
[Q]
During the selection process, each of the above criteria will carry one point except head-to-head competition, which will carry the number of points equal to the net difference in the results of these games (e.g., if team A defeats team B three out of four games, team A would receive two points in the selection process). When comparing two teams, the team earning the most points will be selected.

If the point process provides a tie, the Rating Percentage Index will serve as the determining factor, regardless of the difference.
[/Q]

As evidence that USCHO has always done this as well, here is a link to a post on HOCKEY-L by Lee Urton, who wrote USCHO's original PWR program, discussing just this issue:
[lists.maine.edu]
(In fact, this post predates USCHO itself.)

I don't actually see this explicitly stated in the PWR FAQ, so I suppose it should be clarified.

[Q] the new RPI scheme, which, IMHO, is much too skewed toward whom you've played as opposed to how you've done.[/Q]
Agreed, I know it's a MAAC team, but when Quinnipiac has 15 wins, and it's much higher than a team that's 1-17, I have issues with that. I don't care how good the teams you play are, if you only have 1 win in 20 games, you don't deserve anything. Perhaps the old 35-50-15 was too little weight... anyone up for 30-50-20?

The problem is not that RPI weights strength of schedule too much or too little, but that it does it wrong. Even with the new weighting, our RPI would be higher if we had swept Quinnipiac instead of BU.

Also, one would think the multiplicative KRACH would take care of this, but alas it doesn't. Quinnipiac (due to SOS I suppose) is listed lower than UAA. In fact it says UAA should beat QU about 7 out of 10 times. As I said once near the end of last year, my feeling is KRACH gives SOS too much weight, and this seems to back it up, IMHO.

Only if you believe Quinnipiac really is better than Anchorage. Their schedules have been so different, it makes little sense just to compare their winning percentages. QU is 15-3-1 against the MAAC and 0-3 against everyone else; UAA is 0-0-1 against the MAAC and 1-16-6 against everyone else. Look at the breakdown of their opponents at
[slack.net]
and
[slack.net]
and consider that QU has 31 points and their 16th weakest game was against Canisius, while UAA has 9 points and their 5th weakest game was against Michigan Tech.

I'll yield you the UAA/Qu issue for now, but on the tiebreaking thing:

[Q]Every team with a .500 or better RPI (ratings percentage index) is called a "team under consideration," or TUC. The PWR method compares every TUC with every other such team, with the winner of each "comparison" earning one PWR point. After all possible comparisons are made, the points are totaled up and rankings listed accordingly.

For instance, if there are 24 TUCs, the greatest number of PWR points any one team could earn would be 23, by winning the comparison with each of the other 23 teams. Meanwhile, a team which lost all of its comparisons would, of course, have no PWR points.

Teams are then ranked by PWR point total, with ties broken by looking at the Ratings Percentage Index (RPI).[/Q]
I guess it doesn't say specifically, but the line I read as important is "with the winner of each 'comparison' earning one PWR point." It says nothing about breaking a tie within a comparison. On the other hand it *does* specifically mention breaking overall ties, so you'd think if they were to break individual comparison ties, they'd mention it as they did with the overall situation. The exception that proves the rule (if they have to mention the tiebreaking procedure in one situation, it would imply that there is none when they don't).

Also, I'm pretty damn sure I remember seeing tied comparisons go to nobody in the past, but obviously I can't prove that now.

I think UAA and RPI make for a better KRACH comparison.

RPI has beaten St. Cloud, Duluth, and Wisconsin--and won five other games, besides. UAA has beaten only Fairbanks in 25 attempts. And yet the wonderful KRACH algorithm ranks UAA higher than RPI. This, frankly, makes no sense whatsoever.

Nothing can account for this except gross exaggeration of strength-of-schedule in the algorithm.

AFAIC, whichever one ranks us better is the better ranking. Can we just get the NCAA to adopt that as the criteria? It's much simpler for everyone.

Criterion ties within an individual comparison are not broken. If Cornell and NoDak have the same TUC record, then neither team receives a comparison point.

If the total number of comparison points are tied, RPI (the rating system) is used as the tiebreaker, which makes RPI worth effectively 1.5 comparison points. As far as I know it's always been this way.

For example, if Cornell and NoDak have identical TUC records, Cornell has a better COp record, and they have never played each other, the winner of the RPI category will win the comparison. 1-1 in comparison points, with the RPI tiebreaker giving the overall comparison to NoDak if NoDak has the better RPI. 2-0 for Cornell if Cornell has the better RPI.

PWR points are not used by the NCAA. Each team receives a PWR point for every individual comparison won, and are ranked accordingly. I think you may be confusing comparison points with the PWR points on USCHO.

Al DeFlorio wrote:

I think UAA and RPI make for a better KRACH comparison.

RPI has beaten St. Cloud, Duluth, and Wisconsin--and won five other games, besides. UAA has beaten only Fairbanks in 25 attempts. And yet the wonderful KRACH algorithm ranks UAA higher than RPI. This, frankly, makes no sense whatsoever.

Nothing can account for this except gross exaggeration of strength-of-schedule in the algorithm.

Or the fact that you're conveniently ignoring UAA's seven ties. KRACH, RPI, just about any other method of evaluating a hockey team's record count a tie as half a win and half a loss. UAA has tied Denver, Mankato, Duluth twice and Wisconsin twice, and they've got three other points besides that. You're implying that RPI's record is 8 times as good as UAA's because they've won eight games to UAA's one, but if you look at points (i.e., twice wins plus ties), UAA has 9 points in 24 games, so they've gotten about one point for every four they've given up, while RPI has 18 points in 26 games, so they've got one point for every two they've given up. RPI's record is a little more than twice as good as UAA's, but their typical opponents are a little less than half as good, which is why the two have comparable KRACH ratings.

RPI beat SCSU and UMD, but they lost to Princeton twice. They beat Wisconsin but lost to Colgate. They tied Brown, but also tied Union. Keep in mind that this isn't HEAL where you get more credit for beating a good team than you lose for losing to a bad one. Both RPI and KRACH would rate them the same if they'd beaten all those bad teams and lost to the good ones, which means the break-even mark on their schedule is somewhere between Brown and Union. UAA's break-even point is Michigan Tech. (Our break-even point is Ohio State, BTW.)

No, John, nowhere did I imply RPI is "eight times better." You, in fact, incorrectly inferred that.

But they clearly have a much better record, have beaten several "real" teams, and should be rated higher than UAA by any rational ranking system--and certainly not beneath them.

RPI has a better record against a weaker schedule, which is why we have to use some sort of model to extrapolate how they would each do if they played the same schedule. You gave as support for "much better" the fact that UAA only has one win, which is a specious argument given all their ties. UAA has taken a total of six points from teams of the caliber of the "real" teams RPI has beaten. RPI has lost to the likes of Princeton (twice) and Colgate. If you take all the results of both teams into account, it doesn't seem at all unreasonable to me that UAA's raking is slightly better than RPI's.

I'm of the opinion that all a team can do is win the games on the schedule. Yes, I believe a 25-9 MAAC team is more tournament worthy than a 19-15 WCHA team, especially if that 19-15 WCHA team is only .500 and about 5th place in conference play. I also believe non-conference SoS is more important than overall. You can control your non-conference SoS to some extent, you have no control over your conference schedule. If (hypothetically) Holy Cross went out and scheduled BC, Cornell, Michigan, and Minnesota among others and fared reasonably well, they shouldn't be penalized excessively for being in the MAAC.

And yes, this is a carried over from Hoops, where the mid-majors get the screw job every year due to the RPI and its inherent bias towards the power conferences.

I agree with Ben on this, for the most part, and actually meant to say something like that myself before it got late and I forgot ...

Well I don't know about his exactly numbers, I certainly feel a 16-7-1 MAAC team has done more than a 1-18-7 WCHA team, unless that WCHA team has played CC 26 times. Maybe the MAAC team hasn't played as many teams, and based on their schedule you can say they are about the same, but it's not entirely the fault of the MAAC team that they played the lesser team. As Ben said, you can only beat who you play.

On a related sidenote, the WCHA plays more in conference games (28) than any other conference, so their schedule is the most insulated, possibly leading to exagerrations in S.O.S's. I'd love the see the selection committee deny a WCHA team a seeding for the reason, the rule that was supposed to be anti-MAAC and anti-CHA.

So I guess I agree with a Dartmouth fan ... speaking of which, whaddaya say your boys come down here tomorrow, Ben, instead of waiting another whole damn week .

Ben and Fred, you can only beat who you play, but you can only play who you schedule. If the MAAC remains a ghetto of mediocrity, they should not be taken seriously as national players by objective ranking systems or subjective observers.

Stephen F. Austin University is running away with the Southland Conference and could conceivably finish the season 29-3. When March Madness comes around should I care?

> but you can only play who you schedule.

But you can only play who schedules you.

John, we'll just have to disagree. Your arguments above about who's strong and who isn't are based on how they're ranked in these flawed ranking systems, so you're making in effect a circular argument.

DeltaOne81 '03 wrote:

As Ben said, you can only beat who you play.

Yes, but you have to actually beat them. Quinnipiac is 0-3 against a spectrum of non-conference opponents (from Maine to Lake State) and 16-4-1. Not even considering the three non-conference losses, they're still losing to MAAC teams at a clip of 3 for every 11 wins. Considering the MAAC's performance in non-conference games, that's not an impressive record.

On a related sidenote, the WCHA plays more in conference games (28) than any other conference, so their schedule is the most insulated, possibly leading to exagerrations in S.O.S's. I'd love the see the selection committee deny a WCHA team a seeding for the reason, the rule that was supposed to be anti-MAAC and anti-CHA.

Except that the guideline is not "insulated schedule", it's "competitive equity" based on "overall conference RPI". Insulated schedules are the reason why strength of schedule based on opponents' winning percentage can be dragged to .500. Considering that the WCHA has a winning record against every other conference and an overall interconference record better than all but Hockey East, insularity of schedule is more likely to hurt WCHA teams' strength of schedule.

The whole idea of proposing KRACH as an alternative to RPI is that it doesn't have this isolated-schedule-drags-your-SOS-to-.500 problem, so the committee doesn't need to have a special rule and make a subjective judgement call to overcome it. (When it's obvious, like Quinnipiac in 1998 and 1999, this is not a big problem, but in the case of Niagara in 2000, where we knew they were not as good as their PWR implied, but still good enough to be in the running for the tournament, there was no way of quantifying the correction, and it was obvious that they would get the last at-large bid regardless of how their final few games played out.)

Al DeFlorio wrote:

John, we'll just have to disagree.

Nice rhetorical trick of ostensibly ending the argument peacefully and then firing a parting shot that makes me look wrong if I don't respond to it.

Your arguments above about who's strong and who isn't are based on how they're ranked in these flawed ranking systems, so you're making in effect a circular argument.

Okay, first of all, you still haven't said what the supposed flaw in KRACH is, other than that a team with a good record can be ranked below a team with a bad one. Any system which accounts for strength of schedule has to to allow for this possiblity. (In contrast, I can list the flaws in RPI: beating a bad team can hurt your rating, an extended round-robin tournament among the top three teams will lower all of their ratings, teams with good records and bad RPIs are still considered strong opponents by the strength of schedule measure.)

As far as circularity, it's true that a justification of a particular team's rating that will fit into a forum posting will only show that everything is self-consistent, but the rankings don't exist in a vacuum. They're all determined together from all the results, and there's only one set of ratings which is consistent with everyone's results. And I think the recursion of defining the strength of an opponent with the rankings themselves is essential if you're going to judge the strengths of opponents who themselves play vastly different schedule. In broad strokes, if you don't believe KRACH to tell you the MAAC teams Quinnipiac plays are on average much worse than the WCHA teams UAA plays, consider the MAAC's overall non-conference records of 5-18-3 against the CHA and 2-34-4 against the other four conferences. Those are the results that contribute strongly to the low KRACH ratings of all those MAAC teams, and thus also to a Quinnipiac team that wasn't capable of winning 4/5 of their games against them, and lost all three times they played anyone else to boot.

I will occasionally try to follow the technical aspects of the ranking conversation until something glittery in the corner of my peripheral vision distracts me for a moment and then -- what was I saying?

But: is it true that KRACH will never penalize you for winning and never reward you for losing? It's a value judgment (what non-trivial statement isn't?), but that should be one of the 3 basics of a ranking system:

1. Never penalize you for winning and never reward you for losing.
2. Reward you more for beating a better* team. Penalize you more for losing to a worse* team.%
3. Eliminate as much as possible of the circularity inherent in the judgments of better and worse.

* To my untrained# logical eye there always has to be some circularity. If there wasn't and you had an objective standard for better or worse than just chuck the whole ranking and use that.

# No, you may not mention Godel numbers, John.

% Maybe. Or maybe a loss is a loss is a loss, and the rankings should emphasize wins and ties, period.

John, you always drag in the MAAC comparison, which is not the point. The problem is, in ensuring that MAAC teams fall to the bottom, both the "new" RPI and the KRACH algorithms overrate bad teams in good conferences, which is my complaint. The "insularity" issue has now been turned on its head in RPI.


Okay, first of all, you still haven't said what the supposed flaw in KRACH is, other than that a team with a good record can be ranked below a team with a bad one. Any system which accounts for strength of schedule has to to allow for this possiblity.


You, unfortunately, exaggerate my opinion--a typical JTW "rhetorical trick," by the way. It's a question of degree. The current implementations of both RPI and KRACH simply go too far in giving credit for strength of schedule. Low-end teams in the WCHA, in particular, are overrated as a result.


In contrast, I can list the flaws in RPI: beating a bad team can hurt your rating, an extended round-robin tournament among the top three teams will lower all of their ratings, teams with good records and bad RPIs are still considered strong opponents by the strength of schedule measure.)

Your logic flow implies--inaccurately--that I'm somehow arguing for RPI over KRACH while you're defending the honor of the latter. Where have I ever said RPI was good? Or better than KRACH? This year RPI is worse than ever, IMHO. Any algorithm that makes Michigan Tech a "team-under-consideration"--based on what they've done so far this year--has got to be flawed. Like you, I believe KRACH is better than RPI, but, as the heading of my posting above says, I believe KRACH is flawed, too, in sweeping wretched records under a strength-of-schedule carpet.

Greg wrote:

Or maybe a loss is a loss is a loss, and the rankings should emphasize wins and ties, period.

This is a point I go back and forth on, and am starting to lean toward the side expressed in your statement above, Greg.

Barnstable High School could lose just as easily to Colorado--the College or the Avs--as to Fairfield, and it isn't clear to me that they should get more credit--or less penalty--for the former. Either way, they're still Barnstable High School--and not competitive at any of those levels.

One would have to take into consideration the fact that team A might play 40 games and team B 30, so team A would have more chance to pile up wins and ties.

My sense is that the NCAA has recently done a variation on this for the lax tournament. If you weren't an automatic qualifier and you didn't have a win against any of the top four or so teams, you didn't make it. Cornell made it last year--despite losses to Princeton, Brown, and Georgetown, because it beat Syracuse. Hofstra had no such win and was left out, to much gnashing of teeth, despite an identical record and a one-goal loss to top-seed JHU. This approach is easier to implement, of course, in a sport with typically 12-13 game schedules rather than 30-40.

Greg wrote:

is it true that KRACH will never penalize you for winning and never reward you for losing?

Yes. I decided to actually sit down and prove that yesterday. I'm writing it up, but I need to tighten the proof that the inverse of a particular matrix has only positive elements. (Technically speaking, my proof only works when everyone's KRACH is already finite, but that's always true by the time we start talking about KRACH anyway. Plus I can see in rough terms why it must be true even in pathological situations, although in some of those cases a victory will have no impact on the ratings.)

It's a value judgment (what non-trivial statement isn't?), but that should be one of the 3 basics of a ranking system:

1. Never penalize you for winning and never reward you for losing.
2. Reward you more for beating a better* team. Penalize you more for losing to a worse* team.%
3. Eliminate as much as possible of the circularity inherent in the judgments of better and worse.

* To my untrained# logical eye there always has to be some circularity. If there wasn't and you had an objective standard for better or worse than just chuck the whole ranking and use that.

# No, you may not mention Godel numbers, John.

% Maybe. Or maybe a loss is a loss is a loss, and the rankings should emphasize wins and ties, period.

Conditions 1 and 2 are objective, and KRACH satisfies them both. (RPI satisfies condition 2 but not condition 1.)

The sentiment expressed in footnote % is the basis of HEAL and RHEAL, which obey 2a but not 2b.

The meaning of condition 3 is ambiguous and at odds with footnote *.

Another condition which could be added is: if you apply the rating system to teams playing a balanced round-robin schedule, you will always get the same ordering as if you used straight winning percentage. KRACH and RPI obey this condition, but HEAL and RHEAL (and PWR) do not.

Actually, Gary Hatfield wrote a paper a few years ago where he defined a set of desirable properties for a ranking system and proved that no system could satisfy them all. IIRC, one of those properties was that the order in which two teams are ranked should not change if neither of them plays. Any system that uses strength of schedule is doomed to fail there.

See
[lists.maine.edu]
[lists.maine.edu]
for more.

FWIW, I've never heard of Gödel numbers. I'm a physicist, not a mathematician. (Dammit, Jim!)

I think effectively what Al is trying to say (before his anger gets the better of him ) is that UAA is a poor team in a good conference, who would easily be poor enough to compile just as bad a record in a poorer conference, but the fact that they happen to loose to good teams, gives them a major benefit.

To make a concrete (exagerrated) example, let's say there's some heaven-given rating system that tells us that team X is ranked 50 (of 60).

Now let's put them in a 10 team conference with teams 1 -> 9 in the nation. And they compile a 1-15-2 record playing each twice (because, heck, everyone gets lucky sometimes in hockey). Now let's put them in a conference with 52->59. And in conference play they go 14-3-1 . Neither of these are unreasonable records for a #50 team in the supposed conference.

Now, when we compare situation 1 to reality, we see that the team would be ranked even higher than UAA, having played better teams. And in scenario 2, they'd be ranked even worse than QU, having played even worse teams. But these are reasonable situations for the same exact team to be in, and reasonable results for them to have.

Who you play is too important. SOS is too important, I know there's no direct weight for SOS, but obviously imbedded in KRACH. Yes, KRACH is better than RPI overall, but it still has problems.

And let me add one new aspect to this discussion... I think the only truly fair rating system would have to take winning/losing margin into account. It's all good to say that QU has an 0-3 OOC record, but they lost 2-1 to LLSU, 2-1 to NU, and 2-1 TO MAINE! It's not like they had no chance and were blown out of the water. They lost each of those games by one! Had they been totally outclassed OOC, you *might* have a point. But against a long-running #1 in the country, they were one goal shy! That deserves credit, IMNSHO.

I'm at a loss here, Al. JTW is consistently the most humane and patient poster on this forum. With all due respect to everyone else here, I don't think that there is a close second. That he hasn't thrown up his hands and refused to continue to engage you in a subject on which his expertise appears to vastly outpace yours (and probably even more vastly outpaces mine - my eyes glaze over at a lot of this) is a credit to him.

I know you will accuse me of infantilizing your opinion, (and you may be right), but I have no other way to describe your opinion. Your objection to the ranking systems appears to be no more sophisticated than "None of the statistical algorithms jibe with my preconceived notions of the meaning of a team's actual record and relative conference strength, and so they are all hopelessly flawed." You never seriously address the possibility that the worst team in the WCHA is actually better than a middle of the pack ECAC team.

While I was typing this out I decided to read the rest of the posts, and came across your point about Barnstable High School, and it has the patina of legitimacy to it, but, as JTW has pointed out, UAA is not 0-infinity, they have 8 ties in a very good conference. The ties count enough that it is not fair to compare them to a team that would lose to everyone, but chooses to lose to only the very best (as Fred hypothesizes on a different post).

I don't understand the nuts and bolts statistics to tell you if RPI and KRACH are internally consistent or accurate, but I can certainly tell you that JTW has much the better of this argument.

I haven't even hit "post" and I think I regret it already.

Well said, DeltaOne81.

No anger here. But it isn't clear to me why it's viewed as OK for someone to accuse me of a cheap rhetorical trick but it's viewed as "anger" when I point out the use of the same by him. Makes me feel like Hornby.

And I invite anyone to check the season records of UAA and Rensselaer, compare Rensselaer's eight wins and one tie against UAA's one win and seven ties--many against the same teams, incidentally--and conclude that UAA has had a better season.

I don't think JTW committed a cheap trick, nor that it is his wont, as you suggested. But that is neither here nor there.

I perused the schedules (and, more specifically, the common opponents) of UAA and RPI as you suggested. What I found is that I would instinctively think that RPI should be rated higher than UAA.

The common opponents were Wisconsin (RPI 1-0, UAA 0-0-2); Duluth (RPI 1-1, UAA 0-2-2); and Iona (RPI 1-0, UAA 0-0-1). In all of the common opponent comparisons, RPI performed better.

The remainder of RPI's wins and ties, however, leave much to be desired: RPI beat/tied schwag (Army, SLU, Mercyhurst) and mediocrity (Union, Brown, SCSU).

UAA's are a little more mixed. Ties at Mankato and, especially, against Denver are more impressive than anything on RPI's tab. (The UAA tie against Iona is just sad. UAA outshot Iona 33-15; Iona potted the tying goal with about a minute and a half left. Like I said, sad. And, to be fair, a win at SCSU is better than a tie at Mankato.)

All things considered, I think RPI has a more impressive profile than UAA, and would expect them to be ranked higher. But if we think that ratings systems aren't worth a damn every time they fail to echo our preconceived impressions, what is the point of statistical analysis? I'd trust KRACH's number crunching over my own impressions in Vegas. If I were crazy enough to bet on hockey.

I appreciate your checking, apple. Really.

But I do think that exaggerating what someone's said in order to trivialize it is an uncalled-for "trick" (word first used by John describing my posting) or, if you'd prefer, tactic. And I agree that John's usually a pillar of civility here.

I'd say Rensselaer's win over SCSU is the equivalent of UAA's ties with Denver and Mankato. Rensselaer did that in just two games with--at--SCSU. It took UAA four games with Denver and Mankato to get the two ties. KRACH, by the way, ranks SCSU between Denver and Mankato and RPI ranks them above both. But I won't accuse you of having a "preconceived notion" that a win against SCSU is "mediocre" while ties against Denver and Mankato are "more impressive." And I assure you, apple, that I analyzed carefully both team's results before raising the comparison as an issue above--nothing "preconceived" about it.

To recap, I do think KRACH is the best we have, but I believe it does give too much weight to whom you've played, just as the new RPI does, resulting in anomalies in the rankings. And I think it is legit to point out examples where that happens. How else can you check the validity of the rankings? How else can you fine-tune the algorithms? And, by the way, I don't think I, or anyone here, has ever said KRACH is "not worth a damn."

The real concern I have is with RPI/PWR, because important decisions are made on those rankings. What I see as an overreaction in resetting the RPI parameters to "fix" the MAAC/Niagara/CHA problem is quite likely going to hurt the ECAC over time unless it is moderated, and that does trouble me. When games against UNO and Michigan Tech and Notre Dame and Northeastern and Lowell might be counted in the TUC calculation, it could result in a significant advantage to bubble teams from those other leagues at selection time.

I think I confessed myself that the SCSU win was on par with Mankato. I am a bit surprised that any system would rank SCSU ahead of Denver (but I don't put any stock in my surprise). And I don't think it matters that it took more games for UAA to get the ties against Denver and Mankato (but I don't think it is a matter worth beating into the ground with an explanation).

Ultimately, I can live with a system that breaks down a little (if at all) when trying to figure out which team stinks less at the bottom of the rankings. I know that you were only using RPI-UAA as a symbolic data point, but if I ever indicate that I care that much about either team again, shoot me.

And the crowd thought that this might get out of hand . . .

U Maine is up 2-1 over UHN through the second period.

... and Thunderstix still suck!

[Q]Ultimately, I can live with a system that breaks down a little (if at all) when trying to figure out which team stinks less at the bottom of the rankings.[/Q]
In theory I agree, only it does matter when calculated how good the good teams are that have played them. A breakdown anywhere effects things elsewhere.

If a system underweights RPI and overweights UAA, then it will underweight us and overweight UND.

UHN ties it up on the PP, late in the 3rd.

Going to overtime

UMaine wins as Lawson scores with 6.8 on the clock

Maine scores with 6 seconds left in overtime!

The replays showed that the GWG was totally kicked in, but it counts and the result helps us.

Oh, and John:
[Q]Except that the guideline is not "insulated schedule", it's "competitive equity" based on "overall conference RPI". Insulated schedules are the reason why strength of schedule based on opponents' winning percentage can be dragged to .500. Considering that the WCHA has a winning record against every other conference and an overall interconference record better than all but Hockey East, insularity of schedule is more likely to hurt WCHA teams' strength of schedule.[/Q]

From the USCHO PWR FAQ (not enough acronyms):

Q: What about the MAAC and the CHA?

A: Starting with the 1998-1999 season, the committee voted in the right to override the results of comparisons in instances where a team's schedule was too insular (i.e. too many games solely against teams from its own conference) to draw a valid conclusion from the results of their games.

For example, teams in weaker conferences who only -- or mostly -- play teams from that same conference will win more comparisons than would otherwise happen, due to the fact that both winning percentage and strength of schedule are enhanced. By never playing "stronger" teams -- and, more importantly, because their opponents never play the "stronger" teams -- there is no way of telling just how accurate their strength of schedule indicator is.


If USCHO is simplying too much (to the point where they're basically lying), then I apologize, but it's definitely what is says. The WCHA plays (at least!) 28 of 34 games in conference (82.3%). The ECAC only plays 64.7%, even the Ivies only hit 75.8%. The MAAC only does 76.4% - and the random team I looked at, UConn, actually doesn't have any "NC" games against conference opponents.

Your mathematical point above does mean it would have to be a big down swing for them before the rule would apply, but I'd still like to see it. Might prompt them to cut down on the conference schedule a little bit.

I'm curious what you number crunchers would think about this.

Use Win% instead of the RPI as a tiebreaker within the PWR so rather than having the RPI count for 1.5 comparisons, it would only count the same as the others. If all else balances out, take the team that has won more games. Seems fair to me.

I think Quinnipiac would love that idea.

Okay, so that's two corrections we need to send to USCHO for their FAQ. The original statement from the NCAA, which USCHO is paraphrasing, was
[Q]
the committee noted that it reserves the right to evaluate each team based on the relative strength of their respective conference using the overall conference ratings percentage index (RPI) in determining competitive equity.
[/Q]
This comes from the following NCAA News item
[www.ncaa.org]
and does appear in my explanation of the selection procedure at
[slack.net]
(which I have not updated for 2003 because we don't really know how it'll work with 16 teams, and the NCAA still hasn't released this year's Championships Handbook on the web)