2024 Pairwise Predictor

[TonyTheTiger20]

Predictor is live:

https://www.bcinterruption.com/bosto...wise-predictor

Let me know if you have any questions. Small caveat that my numbers for Clarkson and St. Michael's for some reason aren't quite matching what USCHO has. I have given up figuring it out. For all I know my numbers are right. But they aren't far off anyway. EDIT: This has been resolved and my numbers should match the NCAA's and USCHO's numbers exactly. See post number 8 below.

Any bug reports or questions, post here or email me at grant dot salzano at gmail. Thanks!

 

[BowWowWow]

Thanks for creating this! Fun to experiment with it.

As fans on the WCHA thread noted, the Cornell vs Quinnipiac and UMD vs St. Cloud State series are probably suspenseful for Quinnipiac and UMD. On the predictor I see that if UMD wins their series in 2 games, without upsets except 1 Quinnipiac win max, UMD passes Quinnipiac to make the national tournament at large. I wonder how much head-to-head affects the results (head-to-head was 1 Quinnipiac win and 1 tie, at Quinnipiac).

 

[robertearle]

Thanks for creating this! Fun to experiment with it.

As fans on the WCHA thread noted, the Cornell vs Quinnipiac and UMD vs St. Cloud State series are probably suspenseful for Quinnipiac and UMD. On the predictor I see that if UMD wins their series in 2 games, without upsets except 1 Quinnipiac win max, UMD passes Quinnipiac to make the national tournament at large. I wonder how much head-to-head affects the results (head-to-head was 1 Quinnipiac win and 1 tie, at Quinnipiac).

At the bottom of the predictor, there's a tab that reads "Grid". If you click on that, at the top of the page that loads, you can select any two teams and see their particular Pairwise comparison.

With the defaults loaded on the Predictor page, the Duluth vs Quinnipiac comparison has Duluth winning the NPI component, Quinnipiac +1 for head-to-head (1-0-1 vs 0-1-1), and no 'common opponents. Because there are no common opponents, the comparison is then tied, Duluth "+1" for NPI and Q "+1" for head-to-head. If Q had won instead of tying that game, they'd have been "+2" and would have won the comparison. As it is, ties are broken in favor of the team with the better NPI.

So Duluth wins the pair. Change the conference playoff results, and you can get it to where Q's NPI is better than Duluth's and that makes Q win the comparison, and move up ahead of Duluth into the tournament field.

(As it sits right now, Q wins the comparison, because they have both the +1 head-to-head and a better NPI. It's the 'presumed' Duluth wins and Q losses that move Duluth ahead on NPI and into the NCAA field.)

 

[QuasiCosmos]

With the defaults loaded on the Predictor page, the Duluth vs Quinnipiac comparison has Duluth winning the NPI component, Quinnipiac +1 for head-to-head (1-0-1 vs 0-1-1), and no 'common opponents. Because there are no common opponents, ...

Isn't Colgate a common opponent for Duluth and Quinnipiac?

 

[QuasiCosmos]

With the defaults loaded on the Predictor page, the Duluth vs Quinnipiac comparison has Duluth winning the NPI component, Quinnipiac +1 for head-to-head (1-0-1 vs 0-1-1), and no 'common opponents. Because there are no common opponents, ...

Aren't Colgate & LIU common opponents for Duluth and Quinnipiac?

 

[BowWowWow]

Aren't Colgate & LIU common opponents for Duluth and Quinnipiac?

Yeah against commom opponents, it looks like:
UMD 2-0-0 against LIU, Quinnipiac 1-0-0 vs LIU
UMD 1-0-1 vs Colgate, Quinnipiac 0-2-0 against Gate

Thanks for the explanations about the grid comparisons, also, robertearle!

 

[robertearle]

Yeah against commom opponents, it looks like:
UMD 2-0-0 against LIU, Quinnipiac 1-0-0 vs LIU
UMD 1-0-1 vs Colgate, Quinnipiac 0-2-0 against Gate

Thanks for the explanations about the grid comparisons, also, robertearle!

The grid page at Grant's app doesn't show those common opponents, but the grid at USCHO does. That looks like a problem for Grant to look into. (My guess is that it's a 'cosmetic' problem, and the underlying comparisons are all being done correctly.)

But it ends up not really impacting anything here. Both teams won all the games vs LIU, so the winning percent for each is 100%, or 1.0. The win and tie by Duluth over Colgate means 1.5 wins and 0.5 loss, or .75 winning percent. Q gets a zero. Add those up, and Duluth wins the common opponents 1.75 to 1.0. That turns into a "+1" for Duluth for that comparison win.

So, Duluth is +1 on common opponents, Q is +1 head-to-head, meaning NPI decides who wins the comparison.

At the moment, Q has the better NPI, but the default "expectation" in Grant's predictor that Duluth beats St Cloud while Q loses to Cornell would move Duluth's NPI ahead of Q, flip the comparison, and put Duluth into the tournament.

Clearly this is one to watch this coming weekend.

More generally, the #6, #7, and #8 slots are all 'up for grabs'; It doesn't take much work in the predictor to get three in, and one out from the group of Cornell, St Lawrence, Q, and Duluth.

 

[TonyTheTiger20]

Hi all! I updated a few bugs tonight and just pushed the update out.

First:

The grid page at Grant's app doesn't show those common opponents, but the grid at USCHO does. That looks like a problem for Grant to look into. (My guess is that it's a 'cosmetic' problem, and the underlying comparisons are all being done correctly.)

That was a good one, thank you! Yes, it was cosmetic, the calculations had it coming from the right column but the display in the comparisons tab was pulling the wrong one. This should be resolved.

Second:
I got an email from someone out there letting me know that the "prefill higher seeds" option was advancing Sacred Heart instead of Franklin Pierce (FPU got the sweep this weekend). This has been resolved and I offer my sincere apologies to the Fightin' Pierces.

Third+Fourth, and these are big ones:
The NCAA made two changes to the calculations without updating the pre-champs manual this year.

• We all know about "bad wins" getting removed from the calculation. Well, starting this year, the NCAA is also removing "good losses" -- that is, if you are a bad team and you lose to a great team and your NPI goes up as a result, that result is now getting thrown out. This only really makes a meaningful difference in the NPIs of the lower ranked teams, but it does have a trickle-down (trickle-up?) effect on the better teams due to affecting SOSs.
• An even bigger change this year is that "partial bad wins" and "partial good losses" are now being removed. As you may know, a tie is calculated as a half-win and a half-loss. In prior years, if you were a bad team and you had a tie against a bad team, it just went into the results as a tie against a bad team. Now, however, if a full win would have lowered your NPI, the result goes in as a "half-loss" and "zero wins," and only counts overall as one-half of a game. So you can have teams who have a partial number of games played, which is kind of weird.
  
  This also affects overtime wins (67% of a win) and overtime losses (33% of a win) -- i.e. if you lose in overtime to a bad team, 0.33 wins are removed from your win total and 0.33 games are removed from your games total. Yes, I know, it's goofy. The end result is that tying or going to overtime against a bad team doesn't have quite the negative effect that it did before. (The reverse also applies to "good losses" as well).

The bad news about that last bullet point is that it seems kind of dumb to do that all of a sudden after not doing that with ties for all these years. But the good news though is that it means I've figured out the differences I had in my numbers relative to what the NCAA and USCHO had, and now my calculations match theirs exactly.

I spoke to someone at the NCAA and they said it was an oversight that these changes were missing from the manual and they plan to update it ASAP.

Please absolutely continue to send over any bugs you all find -- thank you all for your help!

 

[robertearle]

Grant, the updated version isn't loading correctly. I click 'Y' for defaults, and everything semi final and higher loads, but the first rounds are blank. If I try to manually select a HE quarterfinal, one of the choices is "#NA". If I try to select a WCHA quarterfinal, the choices are in the pull down, but choosing one doesn't show after it has been chosen.

Windows 10 Pc, Google Chrome. Likewise on a Samsung Android tablet, also Google Chrome.

 

[TonyTheTiger20]

Grant, the updated version isn't loading correctly. I click 'Y' for defaults, and everything semi final and higher loads, but the first rounds are blank. If I try to manually select a HE quarterfinal, one of the choices is "#NA". If I try to select a WCHA quarterfinal, the choices are in the pull down, but choosing one doesn't show after it has been chosen.

Windows 10 Pc, Google Chrome. Likewise on a Samsung Android tablet, also Google Chrome.

Woops, sorry about that. Forgot to reset the WHEA first round games in the "prefill" section when I was done testing. Try it now.

 

[robertearle]

Woops, sorry about that. Forgot to reset the WHEA first round games in the "prefill" section when I was done testing. Try it now.

I worked with a guy for a half-dozen years who would often as not, instead of calling it a bug, he'd say "it was a little oops. just a little oops."

Looks good.

 

[robertearle]

So I was playing around, just sort of randomly changing results when I hit on a pretty fun outcome:

Load the defaults. Then have Northeastern beat UConn; Minnesota beat Wisconsin in the semi, but Ohio State still win the championship; Clarkson takes 3 to beat Princeton but then beats Colgate for the championship; Colgate and St Lawrence need 3 to win their first rounds.

That gets you Duluth at #8, beating out Quinnipiac 0.58857 to 0.58819.

Now have St Anselm beat LIU in their championship.

It changes such that now Q is #8, 0.58764 to 0.58761.

Duluth and Quinnipiac, sweating out an NCAA tourney berth on a razor thin margin, hanging on the outcome of LIU vs St Anselm.

--------------

Adding, change HE back to UConn winning, and now it's Duluth 0.58784 and Q 0.58783. That's close enough that it could conceivably yield a different result if you just let the iteration loop run a little longer, or shorter?

 

[TonyTheTiger20]

I worked with a guy for a half-dozen years who would often as not, instead of calling it a bug, he'd say "it was a little oops. just a little oops."

Looks good.

They're just "happy little accidents" hahaha

 

[TonyTheTiger20]

Adding, change HE back to UConn winning, and now it's Duluth 0.58784 and Q 0.58783. That's close enough that it could conceivably yield a different result if you just let the iteration loop run a little longer, or shorter?

Oh boy that's about as close as I've ever seen it lol... I'll have to plug that in offline and give it some extra time to run to see if they start to converge with one or the other ahead.

I think online it runs pretty far on the web but not certain if it uses the same settings as offline or not.

 

[robertearle]

I haven't played with it a lot this morning, but Merrimack beating Maine last night damaged Quinnipiac some (Q had beaten Maine twice early in the year, so their 'opponent win percent' took a hit). Ways that Q used to get past Duluth, they don't anymore.

 

[BowWowWow]

The Pairwise changes after Friday's playoff games include UMD and Quinnipiac flipping. Duluth is now 8, and Quinnipiac 9. They were ranked that way in the poll. Duluth and Quinnipiac could both make it to the national tournament if SLU were to drop, for which UMD and Quinnipiac would have to do well and SLU would probably have to lose its series to Yale, in 3 games, after SLU has won the first. Pairwise teams 1-6 - Ohio State, Wisconsin, Clarkson, Colgate, Minnesota, and Cornell - appear to be a lock now for the national tourney. After a bit of experimenting, at least, I haven't found any scenario in which Cornell or the higher teams are out.

 

[BowWowWow]

The Pairwise changes after Friday's playoff games include UMD and Quinnipiac flipping. Duluth is now 8, and Quinnipiac 9. They were ranked that way in the poll. Duluth and Quinnipiac could both make it to the national tournament if SLU were to drop, for which UMD and Quinnipiac would have to do well and SLU would probably have to lose its series to Yale, in 3 games, after SLU has won the first. Pairwise teams 1-6 - Ohio State, Wisconsin, Clarkson, Colgate, Minnesota, and Cornell - appear to be a lock now for the national tourney. After a bit of experimenting, at least, I haven't found any scenario in which Cornell or the higher teams are out.

On a second try with the experimenting, Cornell wouldn't make it to the national tournament if St. Lawrence wins their series (in 2 games - or in 3 with a win in the semifinals), Quinnipiac beats Cornell in 3, and the ECAC and WCHA tournaments end with upset auto-bids. So, only the top 5 teams are a lock for the national tourney.

 

[QuasiCosmos]

Looking at the three leagues with automatic bids to the NCAA Tournament without a team in the top 7/8 PWR:

With Franklin Pierce upsetting LIU (PWR #18T) in the NEWHA Tournament, am I correct to deduce that the NEWHA will be represented in the NCAA Tournament by the winner of next Saturday's game between Stonehill (PWR #30) and Franklin Pierce (PWR #32)?

CHA Tournament champ Penn State (PWR #14) has already punched their ticket.

And the top four regular season finishers in Hockey East (PWRs #10, 12, 17, 18T) will battle next weekend for their league's automatic spot in the NCAA Tournament.

 

[robertearle]

Looking at the three leagues with automatic bids to the NCAA Tournament without a team in the top 7/8 PWR:

With Franklin Pierce upsetting LIU (PWR #18T) in the NEWHA Tournament, am I correct to deduce that the NEWHA will be represented in the NCAA Tournament by the winner of next Saturday's game between Stonehill (PWR #30) and Franklin Pierce (PWR #32)?

CHA Tournament champ Penn State (PWR #14) has already punched their ticket.

And the top four regular season finishers in Hockey East (PWRs #10, 12, 17, 18T) will battle next weekend for their league's automatic spot in the NCAA Tournament.

Yes.

The Stonehill - Franklin Pierce winner will be the #11 seed, and travel to the location of the #3 seed to take on the #6 seed.

Penn State will either be the #10 seed or #9 seed, depend on HE winner, and travel to the location of the #2 seed or Ohio State, the #1 seed. #10 and #7 will go to #2, #8 and #9 will go to Ohio St. And the reverse for the HE winner, depending whether their Pairwise rank higher or lower than Penn State.

#4 and #5 play one another without a Thursday play-in game at their site.

 

[TonyTheTiger20]

I updated the calculator to include games already played for less input.

Looking at BC's tournament possibilities should they win the league, because I enjoy suffering, I am pleasantly surprised to see that I can get Clarkson to #2 if Minnesota beats Wisconsin in the semifinal and Clarkson wins the ECAC (sending #10 BC to #2 Clarkson's regional). Would be nice to be able to go watch BC in the tournament in person (though obviously, long way to go to get to that point).

Playing with the other results, I can't quite get Colgate to #2 -- though I can get them to within 0.00009 of it!