Pair wise Issues

While I don't claim to be an expert on Pairwise calculations, I do question how conference strength carries so much weight in rankings. It may seem that
I am only looking at this from a Cornell perspective but it affects other teams too. I'm not saying conference strength shouldn't matter but when a team
specifically St Cloud who has 14 losses, overall is one game over .500 and is under .500 in their own conference with 10 losses is ranked 12th, a little too much emphasis is placed on the league itself. From tie to time the committee tweaks the criteria and might want to take a look at this since almost every year the selection committee is in 100% agreement with Pairwise.

wakester2468
when a team specifically St Cloud who has 14 losses, overall is one game over .500 and is under .500 in their own conference with 10 losses is ranked 12th, a little too much emphasis is placed on the league itself.

What is your objective criteria for saying that indicates too much emphasis? The math actually has a pretty good theoretical justification, and it is possible for SOS to punish a deserving team. As reductio ad absurdum let's imagine an independent who plays only teams in the top 10 all season. If they wound up a shade under .500 that would suggest they fully deserved to be in the top 10 themselves.

What makes elynah a site to enjoy following regularly is the ability to express differing points of view.
We obviously disagree and I'm fine with that and enjoy expressing my opinion without reservation. You only gain knowledge by what you learn from others along the way. Politicians would be wise to do the same.

wakester2468
What makes elynah a site to enjoy following regularly is the ability to express differing points of view.
We obviously disagree and I'm fine with that and enjoy expressing my opinion without reservation. You only gain knowledge by what you learn from others along the way. Politicians would be wise to do the same.

...and yet you offer no suggestions as to why your "differing point of view" is valid in light of Trotsky's quite apt example. Why wouldn't his reductio ad absurdum example make sense, and if it does, how does it differ substantially enough from St. Cloud's current circumstances that their mathematical ranking is objectively wrong?

Beeeej

wakester2468
What makes elynah a site to enjoy following regularly is the ability to express differing points of view.
We obviously disagree and I'm fine with that and enjoy expressing my opinion without reservation. You only gain knowledge by what you learn from others along the way. Politicians would be wise to do the same.

...and yet you offer no suggestions as to why your "differing point of view" is valid in light of Trotsky's quite apt example. Why wouldn't his reductio ad absurdum example make sense, and if it does, how does it differ substantially enough from St. Cloud's current circumstances that their mathematical ranking is objectively wrong?

I think he disagrees subjectively, so you're not going to get evidence. That's okay, he did nicely say that he didn't like the system. Nothing wrong with that.

Jim Hyla

Beeeej

wakester2468
What makes elynah a site to enjoy following regularly is the ability to express differing points of view.
We obviously disagree and I'm fine with that and enjoy expressing my opinion without reservation. You only gain knowledge by what you learn from others along the way. Politicians would be wise to do the same.

...and yet you offer no suggestions as to why your "differing point of view" is valid in light of Trotsky's quite apt example. Why wouldn't his reductio ad absurdum example make sense, and if it does, how does it differ substantially enough from St. Cloud's current circumstances that their mathematical ranking is objectively wrong?

I think he disagrees subjectively, so you're not going to get evidence. That's okay, he did nicely say that he didn't like the system. Nothing wrong with that.

*shrug*

In actuality, I agree with both of you. My opinion certainly was subjective but I do have an opinion on how the ranking system could be changed.
The NCAA football selection committee incorporates both statistical and a subjective human factor. Until November 1st, the only rankings which frankly
don't mean much are the polls. Thereafter, weekly the committee meets weekly to establish a top 20 list until the last games are over and the final 4
are selected. While conference strength is a factor, it isn't necessarily the end all. A team with two losses in the strongest conference while potentially being one of the top four doesn't get a birth. Thus, record does matter. While this might be too much for college hockey to undertake, something similar
could be undertaken. Another difference between the two sports is there is no factor for how teams are performing down the stretch or how injuries impact
the viability of a team in the tournament. I would prefer to see on February 1st, the beginning of ratings. While it's not the same with automatic bids granted to all conferences, the at large teams need personalized analysis along with the metrics.
So you asked and here is what I think is a reasonable alternative.

The selection committee doesn't actually have to follow the pairwise, right? So presumably they could boot St. Cloud out if they are a .500 team but they are something like 14th in the PWR?

wakester2468
In actuality, I agree with both of you. My opinion certainly was subjective but I do have an opinion on how the ranking system could be changed.
The NCAA football selection committee incorporates both statistical and a subjective human factor. Until November 1st, the only rankings which frankly
don't mean much are the polls. Thereafter, weekly the committee meets weekly to establish a top 20 list until the last games are over and the final 4
are selected. While conference strength is a factor, it isn't necessarily the end all. A team with two losses in the strongest conference while potentially being one of the top four doesn't get a birth. Thus, record does matter. While this might be too much for college hockey to undertake, something similar
could be undertaken. Another difference between the two sports is there is no factor for how teams are performing down the stretch or how injuries impact
the viability of a team in the tournament. I would prefer to see on February 1st, the beginning of ratings. While it's not the same with automatic bids granted to all conferences, the at large teams need personalized analysis along with the metrics.
So you asked and here is what I think is a reasonable alternative.

The problem that most of us will have with something like this, is that it's too subjective, whereas PWR is basically known to all and based on the numbers.

No system, will satisfy everyone.

Ok last word on this, at least for now, NCAA basketball might actually be a better example to use. It is similar in that there are automatic bids
to all D1 conferences and then the committee selects the remaining teams. They use many criteria but ultimately they are not locked into them. Thus,
like it or not on selection day which is a banner event for basketball,one doesn't know for certain who is in the field. Lets compare it the hockey
where it's a foregone conclusion except where teams will play. If football and basketball feel this is the best system,I'm not sure why hockey should differ.

wakester2468
Ok last word on this, at least for now, NCAA basketball might actually be a better example to use. It is similar in that there are automatic bids
to all D1 conferences and then the committee selects the remaining teams. They use many criteria but ultimately they are not locked into them. Thus,
like it or not on selection day which is a banner event for basketball,one doesn't know for certain who is in the field. Lets compare it the hockey
where it's a foregone conclusion except where teams will play. If football and basketball feel this is the best system,I'm not sure why hockey should differ.

Exactly because it's subjective and unknown. The committee in basketball not only gets to select the at-large teams based on their own criteria, there are no limits to those criteria and they don't have to be revealed. Particularly with only 60 teams in Div. I hockey, it seems unnecessarily opaque, and potentially cruel, to tell you that you have to perform at a certain level to make the NCAA tournament - only to be told at the end that you didn't make it, and nobody is interested in telling you why. This way, the teams know what they have to do - pending of course the math of other teams' results, but that's always the case. You either performed well enough relative to the other teams or you didn't. A foregone conclusion is a good thing.

NCAA Hockey feels this is the best system, and frankly I think basketball ought to follow - but they won't, because there are TV ratings and a lot more money involved.

I find this discussion very interesting and applies towards one's perspective even outside of sports.
Personally, i think statistics are a great tool to help make ones decisions. That being said I will never solely use them
eliminating factors statistics can't recognize. To take away the human element in life troubles me. I appreciate hearing
all sides of subject though.

I do think this year demonstrates a bit of an issue with pairwise that could use tweaking. Keep in mind that pairwise these days is just RPI (unless you beat a team head to head AND in common games, so the RPI chart and the pairwise usually have one or two differences in total) and RPI is pretty much just a function of your record, your opponents' record, and your opponents' opponents' record. With a little tweaking for quality wins, road/neutral/home wins, and a bump to make sure wins against scrubs don't actually hurt you.

Most of the ECAC has been okay this year. Sucks is excelling (ew), and union, slu, and us have been hovering at or above the cutoff for much of the season. Q isn't that far behind, Clarkson and Yale are competent, Dartmouth and Princeton are alright (and Princeton keeps beating top teams somehow). The only issue is that Colgate is very bad and RPI and brown are hilariously terrible. The entire conference is being dragged down because our last place teams, instead of being in the usual bottom 5-15 of college hockey, are in the very bottom few. While I like RPI generally, maybe we need more of a bump for wins over bad teams. Why should we suffer so much just because brown has a few fewer wins than usual?

KRACH often has less respect for us and for the ECAC than the pairwise does, but this year KRACH has us in 11th. They also have st cloud in 16th. And brown and RPI in the third to last and second to last spots in all of college hockey.

i thought that hockey was deciding the teams strictly by PWR now?

wakester2468
I find this discussion very interesting and applies towards one's perspective even outside of sports.
Personally, i think statistics are a great tool to help make ones decisions. That being said I will never solely use them
eliminating factors statistics can't recognize. To take away the human element in life troubles me. I appreciate hearing
all sides of subject though.

I guess I'm just not sure why you think Pairwise takes away the human element. The kids on the skates are still determining, through their performance, who goes to the tournament. Under your suggested framework, someone on the committee might be able to say, "Hey, Cornell never gets very far in the tournament, so let's not include them even though they finished in the 12 spot. Send #17 Boston College instead, we'll get better ratings that way." Does that really increase the quality of the selection process for you?

upprdeck
i thought that hockey was deciding the teams strictly by PWR now?

It's strictly by pairwise (I think), but PWR changed a little so that RPI is, by far, the strongest factor.

There's RPI, common opponents, and head to head record. But RPI is the tiebreaker, so the only scenario in which RPI isn't the decider is if you're behind a team in RPI, but beat them in both common opponents AND head to head.

Currently, there are 3 hiccups in the rankings where a lower RPI team is one spot ahead of where they should be, and the rest of the PWR resembles RPI.

wakester2468
i think statistics are a great tool to help make ones decisions. That being said I will never solely use them
eliminating factors statistics can't recognize. To take away the human element in life troubles me. I appreciate hearing
all sides of subject though.

Often the human element is simply various unconscious cognitive biases. When a statistical system which is also well-founded in theory conflicts with people's gut instinct it frequently indicates that the gut instinct is wrong and not that the model is missing something "intangible."

I would not want to fly to the moon based on how the pilot "felt" about trajectory.

Now the choice of factors and their weighting is always subject to review and improvement. But they aren't arbitrary as long as the model's creators have given sufficient thought to all the aspects of the phenomena they wish to model. The way to improve the model is to suggest factors that are not measured, or give reasons why the various elements should be weighted differently. Otherwise one is comparing the model to metaphysical perfection and concluding that's it's just an approximation. Well of course it is.

Trotsky

wakester2468
i think statistics are a great tool to help make ones decisions. That being said I will never solely use them
eliminating factors statistics can't recognize. To take away the human element in life troubles me. I appreciate hearing
all sides of subject though.

Often the human element is simply various unconscious cognitive biases. When a statistical system which is also well-founded in theory conflicts with people's gut instinct it frequently indicates that the gut instinct is wrong and not that the model is missing something "intangible."

I would not want to fly to the moon based on how the pilot "felt" about trajectory.

Now the choice of factors and their weighting is always subject to review and improvement. But they aren't arbitrary as long as the model's creators have given sufficient thought to all the aspects of the phenomena they wish to model. The way to improve the model is to suggest factors that are not measured, or give reasons why the various elements should be weighted differently. Otherwise one is comparing the model to metaphysical perfection and concluding that's it's just an approximation. Well of course it is.

To take the human element to the absurd, the national champion in lacrosse was, at one time, just picked by committee. And they always picked a Baltimore area team. After-all, everybody knew those teams were stronger than the northern teams. Well in 1971 they started the playoffs, and guess what happened.

We all have our biases, conscious or not. I'd rather have the biases put into designing the system, before the season starts, rather than picking after the season.

Trotsky

wakester2468
when a team specifically St Cloud who has 14 losses, overall is one game over .500 and is under .500 in their own conference with 10 losses is ranked 12th, a little too much emphasis is placed on the league itself.

What is your objective criteria for saying that indicates too much emphasis? The math actually has a pretty good theoretical justification, and it is possible for SOS to punish a deserving team. As reductio ad absurdum let's imagine an independent who plays only teams in the top 10 all season. If they wound up a shade under .500 that would suggest they fully deserved to be in the top 10 themselves.

And this is one of the biggest reasons why I believe having a strong ECAC is Good.

My problem with pairwise is it relies too heavily on RPI, and RPI relies too heavily on whom you've played and does not look at how you've actually done against the top teams you've played.

The best example is the 2007 lacrosse selection, where Hopkins got a #3 seed and Cornell was graciously moved up to #4 by the committee even though our statistical numbers (RPI, SOS, etc.) said we should have been #5 or #6. Why did Hopkins get #3 even though their won-loss was 9-4? Because they played five games against the other seven NCAA-seeded teams, giving them a high RPI. But when you peeled the onion what you saw was a 1-4 record against those other seeded teams, with the one win coming against Maryland, the #7-seeded team. So, how does a team that goes 1-4 against top eight teams with the sole win over #7 (and a loss to #8) get to be #3? Because of the way RPI is calculated.

Cornell, on the other hand, was unbeaten and also with one win against a seeded team, #1 Duke, at Duke, but no other games against seeded teams. So, if Cornell had played and lost to two or three other seeded teams, they might have been able to get a higher seed than Hopkins, solely because of whom they would have lost to. Nuts, in my opinion.

There seems to be mixed feelings on this subject. Interesting with the just released USCHO coaches poll, the biggest spread between Pairwise and the Poll is St Cloud.
Pairwise 12 poll 18. Second biggest difference is North Dakota. To be fair there is a noticeable difference for Union too.
Looks like that damn human thing strikes again.

rpi has bigger issues in bball than hockey just because there are so many more teams and the margin of error is so much larger. you can skew RPI in bball by playing 10 teams in the 100 range and not in the 200 range but not actually be any better

hockey has issues in the other direction in that many times the best team on any given night does not win when 1-3 goals is the result most nights.

I wonder why hockey doesnt try experiments to increase scoring. bigger nets would reward the better teams so that one player couldnt decide the outcome as much as team play did,

there are tons of factors in a game that watching you can see who the better team is, and almost all of them are not represented in any of the computer models that rank teams.

case in point you can go weeks without seeing a fluky basket go in, but you see a wayward shot go in in hockey almost every game and often its the only way to score is to hope for a weird bounce.

Dafatone
KRACH often has less respect for us and for the ECAC than the pairwise does, but this year KRACH has us in 11th. They also have st cloud in 16th. And brown and RPI in the third to last and second to last spots in all of college hockey.

I prefer KRACH because it prefers us

I think one cause of this pairwise issue is that we have more conferences on average, yielding more auto-bids. Right now, it looks like we're going to have at least two autobids that would not make it as at-larges. Before the Big 10, I felt like it was normally just the AHA that wouldn't make it in.

wakester2468
What makes elynah a site to enjoy following regularly is the ability to express differing points of view.
We obviously disagree and I'm fine with that and enjoy expressing my opinion without reservation. You only gain knowledge by what you learn from others along the way. Politicians would be wise to do the same.

Sorry but if you think there is even a remote chance for politicians today to have and/or demonstrate anything remotely resembling wisdom, then I feel bound to completely ignore your posts

Al DeFlorio
My problem with pairwise is it relies too heavily on RPI, and RPI relies too heavily on whom you've played and does not look at how you've actually done against the top teams you've played.

Actually, the RPI does, these days, factor in how you've done against top teams. It's part of the Quality Win Bonus formula. Which also factors in home/road.

There an argument to be made that this shouldn't be there. Why is it any better to beat a top team than lose to a bottom team. You are rewarding "good wins" but not penalizing "bad losses." I've never understood the difference really. But most people like good wins to be rewarded, so there you go.

Having a bunch of different criteria in the Pairwise besides RPI again (at one point, there were 5, including Record in Last 16 games) - would be more fun and give people like me more to write about. But I'm not sure if it's better or worse in terms of a method for objectively ranking team quality.