Pairwise tweak idea

I will preface this post by saying I fully endorse KRACH as a better system.

Right now the PWR system is so heavy weighted on adjusted-RPI. Therefore the goal of considering CoP and H2H comparisons is to “overcome” that bias. Here’s the concern. For example what if Dartmouth wins the second game against us in a few weeks? The PWR will be tied 2-2 and the stronger RPI team still wins. In fact a “much weaker” team has to win H2H by a margin of 3+ to flip the comparison... which is almost insurmountable.

(Yes my example favors our favorite team this year. Think past this season.)

My proposed solution is simple: count each H2H as 1.5 comparisons. Could be just enough to make the PWR game more interesting at end of the year. Thoughts?

KenP
I will preface this post by saying I fully endorse KRACH as a better system.

Right now the PWR system is so heavy weighted on adjusted-RPI. Therefore the goal of considering CoP and H2H comparisons is to “overcome” that bias. Here’s the concern. For example what if Dartmouth wins the second game against us in a few weeks? The PWR will be tied 2-2 and the stronger RPI team still wins. In fact a “much weaker” team has to win H2H by a margin of 3+ to flip the comparison... which is almost insurmountable.

(Yes my example favors our favorite team this year. Think past this season.)

My proposed solution is simple: count each H2H as 1.5 comparisons. Could be just enough to make the PWR game more interesting at end of the year. Thoughts?

My thoughts are that your analysis is flawed. You forget the records against common opponents. Using the CU/Dar PWR data as of today, We get 1 point for being ahead on RPI and 1 for having the better common opponent record. Dartmouth has 1 point for a H2H win and would get a second for the second win.

The weaker RPI team in your example could flip the comparison if they had a better record against common opponents. It doesnt require more H2H wins than are possible.

I’m thinking of this specific type of situation with Cornell-Dartmouth. David beats Goliath twice. RPI & CoP will both go to Goliath. Goliath wins comparison.

I'd ditch the pairwise entirely. It's pretty much just RPI right now. Why not just use RPI?

Dafatone
I'd ditch the pairwise entirely. It's pretty much just RPI right now. Why not just use RPI?

Because it’s inferior to KRACH. If the idea on the table is “just use a calculated ratings system,” then just use KRACH. We’re already living with a RPI-dominated ranking.

RichH

Dafatone
I'd ditch the pairwise entirely. It's pretty much just RPI right now. Why not just use RPI?

Because it’s inferior to KRACH. If the idea on the table is “just use a calculated ratings system,” then just use KRACH. We’re already living with a RPI-dominated ranking.

I can get behind just using KRACH. I think KRACH is better than RPI, but as the ECAC tends to be worse than other major conferences, RPI usually makes us look better than KRACH.

But in either case, why mess with all the other stuff in the pairwise when we can just use the computer ranking?

Again, let's preface the discussion that PWR is going to stay. We have a system that is very similar to RPI. By definition the other comparisons exist to address "intangibles" which may not show up in RPI alone.

Common Opponents - the net result of this comparison is to adjust for non-conference play. We beat Michigan State therefore we have a let up against the rest of the Big 10. Within a conference the CoP will almost always go to the stronger team (it aligns with RPI)

Head-to-Head - can help adjust pairwise, especially for non-conference play.

The point of my thread is that a top 10 team who loses multiple games to a bottom dweller team in their conference is going to will still win the pairwise comparison:
Comp -- Strong Team -- Weak Team
RPI -- 1 -- 0 (Reflects the stronger team)
CoP -- 1 -- 0 (Aligns with RPI)
H2H -- 0 -- 2
Strong team wins 2-2 on basis of higher RPI.

If the H2H is 2+ in favor of the weaker team IMO that should be enough to flip a comparison. In the current system it is not. Hence my proposed tweak.

The problem with giving more weight to head-to-head or even common opponents is that they focus on just a few games rather than a team's performance over the entire season. It's good that it's difficult for the anecdotes to overwhelm the rest of the data.

jtwcornell91
The problem with giving more weight to head-to-head or even common opponents is that they focus on just a few games rather than a team's performance over the entire season. It's good that it's difficult for the anecdotes to overwhelm the rest of the data.

But what about THE EYE TEST???!!!

jtwcornell91
The problem with giving more weight to head-to-head or even common opponents is that they focus on just a few games rather than a team's performance over the entire season. It's good that it's difficult for the anecdotes to overwhelm the rest of the data.

I agree with this, but I think the reason this comes up every now and again is because there are two different ways of thinking about the Pairwise. Since it's based on pair-wise comparisons featuring a handful of fairly arbitrary comparison criteria, it means you can think of each individual comparison as a mechanism for answering the following two questions:

1) Which team is the better overall team?
2) Which team is more likely to win if they play each other in the tournament?

If you focus on question 1, you would rather not have head to head or common opponents be over-weighted. In fact, a purist would just get rid of them, because they're effectively over-weighting the value of a specific subset of games in ways that can subvert the global ranking.

If you focus on question 2, you'd want to boost the weighting of head to head (and maybe common opponents), because this theoretically tells you more about the matchup between those two specific teams.

I honestly don't know what the ranking committee is trying to do with the pairwise, or even if they follow the pairwise process as it's laid out on CHN/USCHO, but given that it all seems to resolve to RPI and that other factors (L10 or L16 or whatever it was) have been eliminated I don't think we're going to see movement towards weighting H2H more strongly. For the specific application of NCAA tournament selection I think this is a good thing.

Tom Lento

jtwcornell91
The problem with giving more weight to head-to-head or even common opponents is that they focus on just a few games rather than a team's performance over the entire season. It's good that it's difficult for the anecdotes to overwhelm the rest of the data.

I agree with this, but I think the reason this comes up every now and again is because there are two different ways of thinking about the Pairwise. Since it's based on pair-wise comparisons featuring a handful of fairly arbitrary comparison criteria, it means you can think of each individual comparison as a mechanism for answering the following two questions:

1) Which team is the better overall team?
2) Which team is more likely to win if they play each other in the tournament?

If you focus on question 1, you would rather not have head to head or common opponents be over-weighted. In fact, a purist would just get rid of them, because they're effectively over-weighting the value of a specific subset of games in ways that can subvert the global ranking.

If you focus on question 2, you'd want to boost the weighting of head to head (and maybe common opponents), because this theoretically tells you more about the matchup between those two specific teams.

I honestly don't know what the ranking committee is trying to do with the pairwise, or even if they follow the pairwise process as it's laid out on CHN/USCHO, but given that it all seems to resolve to RPI and that other factors (L10 or L16 or whatever it was) have been eliminated I don't think we're going to see movement towards weighting H2H more strongly. For the specific application of NCAA tournament selection I think this is a good thing.

My philosophy is that a pairwise comparison is supposed to compare teams against one another. And if Team A outplays Team B 2-0 (or wins 2 of 3 or 3 of 4 etc.) IMO Team A deserves to win that individual comparison. (Regardless of RPI or CoP)

Tom Lento
I honestly don't know what the ranking committee is trying to do with the pairwise, or even if they follow the pairwise process as it's laid out on CHN/USCHO,

they do