PDA

View Full Version : Annual change to PWR announced.



Pages : [1] 2

Alton
08-18-2011, 01:45 PM
The annual tweak that the Men's Ice Hockey Committee makes to the PWR has been announced, but it is pretty much unintelligible.

The NCAA Men's Ice Hockey Committee announced...


That, effective with the 2012 championship, the common opponent comparison in the Rating Percentage Index (RPI) be adjusted so that the record versus common opponents be looked at individually rather than collectively when providing the advantage for the awarding of a comparison point in the RPI.
http://fs.ncaa.org/Docs/DI_Champs_Sports_Mgmt_Cab/2011/September_2011/Supp_17_MIH.pdf

Two issues--

(1) There is no "common opponent comparison" in the Rating Percentage Index (RPI). I think we can assume that they meant the selection criteria, which uses the method of pairwise comparisons, and is simulated on this website by the PWR.

(2) There are a few ways of looking at records versus common opponents individually rather than collectively. The one that makes the most sense is the one that is based on the same principle as calculating the opponents' winning percentage in RPI.

Let's look at the Common Opponents comparison between Michigan and Minnesota-Duluth from last season (before the tournament).
against Colorado College--Michigan was 1-0, Minnesota-Duluth was 0-1-1
against Lake Superior--Michigan was 2-0, Minnesota-Duluth was 0-0-1
against Michigan Tech--Michigan was 1-0, Minnesota-Duluth was 4-0
against Minnesota--Michigan was 0-1, Minnesota-Duluth was 1-1-2
against Nebraska-Omaha--Michigan was 1-1, Minnesota-Duluth was 1-1
against Northern Michigan--Michigan was 2-0, Minnesota-Duluth was 1-0
against Wisconsin--Michigan was 0-0-1, Minnesota-Duluth was 3-1.

Last year, the point in the pairwise comparison was won by Michigan because Michigan was 7-2-1 (.750) against these 7 teams and Minnesota-Duluth was 10-4-4 (.667). This year, I think we will be figuring Michigan's record against common opponents as the average of their winning percentages against each team (1.000, 1.000, 1.000, .000, .500, 1.000, .500) for an average of .714; Minnesota-Duluth would average (.250, .500, 1.000, .500, .500, 1.000, .750), for a .643 percentage. Michigan would still win the comparison, but both would have different percentages. There are probably some comparisons out there that will switch because of this. It will take away the advantage of some teams playing multiple games against easier opponents.

Unfortunately, the informational item does not confirm that this is exactly how the common opponents point will be calculated--the calculation method is an educated guess on my part.

JF_Gophers
08-18-2011, 01:49 PM
Annual post that the Selection Committee does not use PWR. PWR is a tool others use to mimick what the Selection Committee uses. ;) :D

Alton
08-18-2011, 01:53 PM
As I stated in my post..."I think we can assume that they meant the selection criteria, which uses the method of pairwise comparisons, and is simulated on this website by the PWR." Yes, I know. I just thought a concise headline with a technical inaccuracy would be clearer to the average reader of this board.

:D

Driftryder
08-18-2011, 01:58 PM
Eastern Bias

goldy_331
08-18-2011, 02:59 PM
Eastern Bias

Oo, oo, my turn - "Simple Math!"

MUhawks628
08-18-2011, 03:07 PM
Wuts a PWR :confused:








(Thanks for the explanation)

Patman
08-18-2011, 03:44 PM
See... they say individually... how do we know they're not doing something like

sum_{opponent_i \in common opponents}{%Win_{team A,opponent_i}>%Win_{team B,opponent_i}}

and then see which is larger... that is compare each common opponent head to head and score each of those those micro-PWR as wins and losses... the one with the more micro-PWR wins gets the comparison point.

Alton's probably right... but when you start lawyering it you can get into trouble.

edit: I can't imagine the NCAA committee calculates this all themselves... I don't suppose a member institution can compel a code release?

Ralph Baer
08-19-2011, 04:01 AM
The original link appears to be broken (although there is cached copy). Perhaps they realize that it could have been written better.

FlagDUDE08
08-19-2011, 07:24 AM
It almost looks like what they're talking about is the number of common opponents in which there is a higher record. Let's take Alton's example:

against Colorado College--Michigan was 1-0, Minnesota-Duluth was 0-1-1
against Lake Superior--Michigan was 2-0, Minnesota-Duluth was 0-0-1
against Michigan Tech--Michigan was 1-0, Minnesota-Duluth was 4-0
against Minnesota--Michigan was 0-1, Minnesota-Duluth was 1-1-2
against Nebraska-Omaha--Michigan was 1-1, Minnesota-Duluth was 1-1
against Northern Michigan--Michigan was 2-0, Minnesota-Duluth was 1-0
against Wisconsin--Michigan was 0-0-1, Minnesota-Duluth was 3-1.

Michig: (1.000, 1.000, 1.000, 0.000, 0.500, 1.000, 0.500)
Duluth: (0.250, 0.500, 1.000, 0.500, 0.500, 1.000, 0.750)

In this comparison, Michigan has two common opponents in which they have a higher record (Colorado College, LSSU), while UMD has two opponents higher (Minnesota, Wisconsin), while the other three are the same record percentage, and thereby a wash. Michigan has 2, Duluth has 2, therefore no comparison point is awarded for common opponents. The comparison would change.

That's my theory, at least. Perhaps the idea is just having to play a certain opponent a number of times hurts a team in COP too much. It could be said that RPI losing to Colgate helped RPI much more than beating Colgate and playing Cornell in the playoffs, where we already had a bad record, and potentially losing to them twice would have destroyed our COP.

Alton
08-19-2011, 07:53 AM
The original link appears to be broken (although there is cached copy). Perhaps they realize that it could have been written better.

Hmm. The whole folder has been taken down. I am hoping that this file will reappear in a clearer form.

Patman and Flagdude have both come up with the other possibility: awarding the comparison point to the team with the better record against more common opponents. Unfortunately, the way the announcement was phrased was completely unclear. I think the possibility that I posted in the original post is more likely, since it is analogous to how they calculate "Opponent's record" in the RPI, but there is no way of knowing right now what they did mean.

Hopefully it will be more clear if and when the NCAA re-posts the file.

FlagDUDE08
08-19-2011, 12:29 PM
Hmm. The whole folder has been taken down. I am hoping that this file will reappear in a clearer form.

Patman and Flagdude have both come up with the other possibility: awarding the comparison point to the team with the better record against more common opponents. Unfortunately, the way the announcement was phrased was completely unclear. I think the possibility that I posted in the original post is more likely, since it is analogous to how they calculate "Opponent's record" in the RPI, but there is no way of knowing right now what they did mean.

Hopefully it will be more clear if and when the NCAA re-posts the file.

Perhaps they don't want mathematical geniuses to figure out who makes the tournament before they announce it.

slurpees
08-19-2011, 12:56 PM
I am totally confused.

Ralph Baer
08-19-2011, 01:38 PM
It almost looks like what they're talking about is the number of common opponents in which there is a higher record. Let's take Alton's example:

against Colorado College--Michigan was 1-0, Minnesota-Duluth was 0-1-1
against Lake Superior--Michigan was 2-0, Minnesota-Duluth was 0-0-1
against Michigan Tech--Michigan was 1-0, Minnesota-Duluth was 4-0
against Minnesota--Michigan was 0-1, Minnesota-Duluth was 1-1-2
against Nebraska-Omaha--Michigan was 1-1, Minnesota-Duluth was 1-1
against Northern Michigan--Michigan was 2-0, Minnesota-Duluth was 1-0
against Wisconsin--Michigan was 0-0-1, Minnesota-Duluth was 3-1.

Michig: (1.000, 1.000, 1.000, 0.000, 0.500, 1.000, 0.500)
Duluth: (0.250, 0.500, 1.000, 0.500, 0.500, 1.000, 0.750)

In this comparison, Michigan has two common opponents in which they have a higher record (Colorado College, LSSU), while UMD has two opponents higher (Minnesota, Wisconsin), while the other three are the same record percentage, and thereby a wash. Michigan has 2, Duluth has 2, therefore no comparison point is awarded for common opponents. The comparison would change.

That's my theory, at least. Perhaps the idea is just having to play a certain opponent a number of times hurts a team in COP too much. It could be said that RPI losing to Colgate helped RPI much more than beating Colgate and playing Cornell in the playoffs, where we already had a bad record, and potentially losing to them twice would have destroyed our COP.

That is what I think was meant -- and I am due to be right one of these days. ;)

FlagDUDE08
08-19-2011, 02:05 PM
I am totally confused.

For those who were told there'd be no math:

Magic is used to determine whether or not your team will be in the NCAA tournament if it does not win an automatic qualifier. ;)

MUhawks628
08-19-2011, 02:08 PM
For those who were told there'd be no math:

Magic is used to determine whether or not your team will be in the NCAA tournament if it does not win an automatic qualifier. ;)

Well shucks, why didn't you just say so in the first place :p

The Sicatoka
08-19-2011, 03:13 PM
The NCAA Men's Ice Hockey Committee announced...


That, effective with the 2012 championship, the common opponent comparison in the Rating Percentage Index (RPI) be adjusted so that the record versus common opponents be looked at individually rather than collectively when providing the advantage for the awarding of a comparison point in the RPI.

:confused:

I'd give a link but this site is hatin' on the wav suffix so here:
nifter.com/sound_effects/cartoons_animations_sound_effects/homer_simpson/homer_simpson_33_NifterDotCom
add the dot-wav at the end yourself.

unofan
08-19-2011, 03:57 PM
Hmm. The whole folder has been taken down. I am hoping that this file will reappear in a clearer form.

Patman and Flagdude have both come up with the other possibility: awarding the comparison point to the team with the better record against more common opponents. Unfortunately, the way the announcement was phrased was completely unclear. I think the possibility that I posted in the original post is more likely, since it is analogous to how they calculate "Opponent's record" in the RPI, but there is no way of knowing right now what they did mean.

Hopefully it will be more clear if and when the NCAA re-posts the file.

I'm with Ralph and flagdude on this one. Sounds like they'll do mini comparisons and tally up who has the better record against each individual common opponent, so a team can't simply rack up 5 wins against a weak sister of the poor to offset their 0-1 record against 4 other opponents.

Numbers
08-19-2011, 04:04 PM
I believe that Flag Dude has it right. And, if I am thinking clearly, the reason would be something like this:

Suppose Underachieving U (UU) is in the same conference as Nearly In the Tourney Tech (NITT). Suppose NITT played UU 4 times in the reg season, and swept them, and then sweeps again in their conference playoffs. NITT gets 6-0 added to their common opponents criterion. Now, suppose that, in another conference, Always Unlucky U (AUU) played one conference game against the school that all agree had the least talent in the nation (which is UU), and they won. AUU gets 1-0 added to the comparison criterion. The result is that NITT gets a boost due only to scheduling. And the result is that NITT (this is such a nitpicky deal), gets in the tourney, and AUU (Awwwww!!) gets left out again.

I think it's a good idea, but really what the NCAA criterion needs is a broader based solution to this problem, applying some sort of KRACH scale to both common opponents and to TUCs.

Ah, but that has been discussed so many times before.....

Alton
08-19-2011, 04:05 PM
I'm with Ralph and flagdude on this one. Sounds like they'll do mini comparisons and tally up who has the better record against each individual common opponent, so a team can't simply rack up 5 wins against a weak sister of the poor to offset their 0-1 record against 4 other opponents.

OK, but I think the RPI-method works for that as well. If a team's record against those 5 teams is (1.000, .000, .000, .000, .000), they will be considered a .200 team for the common opponents comparison, even if their record is 5-4 (.556). I think the mini-comparison method is less than ideal--it doesn't tell you how much better team A did than team B against teams X, Y, and Z. For example, let's say team A is (1.000, .667, .250) and team B is (.000, .750, .333) against the same teams. I think it would be more fair to award the comparison to team A than to team B, even though team B wins 2 of the 3 mini-comparisons.

Either way, this change is an improvement for teams in the stronger conferences.

Numbers
08-19-2011, 04:07 PM
Or, what UNOFAN said