ELynah Forum

Quote from: jtwcornell91 on November 08, 2025, 12:22:35 AMI'm not sure which message to reply to, and I don't have time to wade into the discussion at the moment, except to remind Adam that we actually implemented a version of KRACH that automatically fit the home-ice advantage, which we called KASA.  ("KRACH Adjusted for Site Advantage", a backronym which Ken Butler helped us construct.)

Also, a previous debate on the eLF did lead to a paper entitled "Prediction and Evaluation in College Hockey Using the Bradley-Terry-Zermelo Model": https://arxiv.org/abs/2001.04226

well aware on both counts

Quote from: BearLover on November 13, 2025, 11:53:50 PM

Quote from: The Rancor on November 13, 2025, 11:41:26 PMWHO CARES

It's illegal to post about Cornell hockey on a Cornell hockey forum now

I always love the "hey, I'm just asking" defense.

Quote from: abmarks on November 06, 2025, 07:56:41 PMI was trying to ask Adam if he could recalculate the historic home advantage with a krach adjustment.   He had given us the actual numbers from past season(s) as a measure of the "true" home advantage in terms of wins.

I thought that the historic evaluation would be more meaningful if krach adjusted to show us how much home advantage there really is historically.

It takes me 30 seconds to run a query on the database to get winners of home and road games. It takes many times longer than that to do what you're asking

Quote from: BearLover on November 06, 2025, 10:48:12 AMAnyway, related to the above and also something I was wondering during the podcast—the guest mentioned that coaches and fans have put their faith in these models because its outputs seem acceptable. That is, we see the 16 teams the Pairwise spits out, they pass the smell test, we move on with our lives and don't question the model. But I'm wondering if anyone has ever made a more rigorous attempt to quantify if the model is picking teams properly. I'm not sure how this would look or if it's even possible, but it did strike me as a little spooky that all this time we've been entrusting a computer model whose outputs we don't even have a means of judging past "the smell test."

Decided to split this up into two answers.

I mean, how it would like, can go a million different ways, and I think anyone close to it, does all sorts of tests to tinker around with what it would mean if this was tweaked or that was tweaked. Certainly the Committee/Tim Danehy do. But there is no god-like answer, so I don't even know how you'd answer that question. On what basis would you judge?

Quote from: BearLover on November 06, 2025, 10:48:12 AMJust to clarify what I mean above—

KRACH (and the Pairwise, and the NPI) are built to pick the 16 teams most deserving of the NCAA tournament (and their order). That is a very different thing than a predictive model. KRACH may well be the best model we have at choosing NCAA teams, but it's not built to be predictive and shouldn't be used that way.

KRACH et al are mostly just a function of winning percentage and strength of schedule. As they should be, because that's how we should choose who has earned an NCAA bid. But you get absurd results when you extrapolate that to future performance. With a large enough, and representative enough, sample, this would start to even out, but in college hockey this isn't possible (because seasons are very short and teams only play a small subset of other teams). 

You can't just plug in two teams' win% and SOS and try to predict which one will win. (Try doing that with college hockey games, you'd lose a ton of money.) There's wayyyyy too much luck and other factors that affect the tiny sample of games on which win% and SOS are based. Again, this is not a knock on KRACH-it's doing its job just fine. Its job is not to predict future performance.

You continue to make a really odd assertion that we (I) don't seem to know this. We're not curing cancer dude. Our Probability Matrix, which I assume you refer to, is just for poops and giggles.  That said, it does provide some utility, because it plays out the schedule based on who wins and loses, including conference tournament brackets. That could get pretty complicating to do by hand. Ergo - SOMETHING has to be used to pick the winner of each game. Thus KRACH. And with 10,000 simulations, it gives people an idea.

No one - ever - claimed that KRACH was some magic potion that predicted the future. You're arguing straw men.

This is also why myself - and others - have been working on using better models, with KRACH as the starting point. But we all have other lives, and it's not easy. There's also no sense of urgency, because the state of the world won't change based upon whether our Matrix is improved.

Other sites have simply taken all the possible win/loss combinations and played things out. Our Matrix decided to do Monte Carlo simulations instead, which allowed us to play out the season much further in advance. That's pretty much it. And I never claimed otherwise.

Quote from: Jim Hyla on November 06, 2025, 02:01:04 PMMisconception means you had the wrong idea.

Being misled implies that someone was actually leading you in that direction. That your misconception was due to others.

Misinterpreting means you made an assumption that was incorrect. So your misconception was due to your own misinterpretation.

Unless you can show where you were led astray, I think that you misinterpreted and weren't misled.

for some people to admit this, would actually require an iota of humility. So - don't hold your breath.

Quote from: Chris '03 on November 06, 2025, 08:25:08 AMThe gnashing of teeth around selection criteria and the relative closeness of teams ranked 5 and 12 or whatever, is part of why I remain in the neutral site NCAA camp.  A team could host a tournament game because it had an extra road game because of the way its conference scheduled vs its opponent who had a home game instead or because its AD insisted on an extra home date to cover expenses at a small school?  Neutral sites (in general) work to mitigate that issue.

Agreed. I've made this point about 5 bazillion times in recent years, including this podcast, and every "debate" with David Carle and others. People who are not very math savvy are making these decisions, and also making comments like "if the math is good enough to pick the teams, why not good enough to seed the teams for home ice," not realizing the inherent flaw in that logic. I would have a much easier and enjoyable time debating the topic, if people just admitted the math was far from perfect, and so there's no way to know how to seed teams 5-12 really, so giving 5-8 home ice advantage is a double whammy. And then just say you still want home-ice Regionals for other reasons. Don't try to defend the math of it. ... But alas that will apparently never sink in for many.

Quote from: BearLover on November 05, 2025, 08:55:34 PMWell, I came to this forum to tell people about the fascinating podcast I just listened to on CHN, and what do I see?

Quote- leave it to BL to twist things around into oblivion, and if they don't fit into the narrowly defined specifications of his liking, then everyone must be terrible or doing something wrong or misleading people. No one is being misled. No one thinks they are being misled. As I said, they endlessly debate, with each other, what the right thing is. If 1.2/0.8 isn't working out, they'll change it. So far it's close enough. And if it was so easy to "game the system" by scheduling nothing but road games - why aren't teams doing it? Like I said, listen to the podcast.

Huh? I was, quite literally, misled. I thought home/away weighting was intended to be accurate. Why wouldn't I? The rankings are meant to do their best to capture the most deserving teams for the NCAAs, so obviously I would expect each of the components to serve that same purpose??? I would guess many others had the same misconception.

Quote- As for KRACH - saying it's not "remotely" a perfect measure, is silly (go figure). Where's John Whelan when I need him? It's actually - as far as I'm concerned (with some needed tweaks to what we publish) - the best method there is. Listen to Tim discuss this on the podcast, where he gives a counter-point to that, while also saying that, with enough data, it would be the best system.

But that's the key flaw, isn't it? There's nowhere near enough data. KRACH may be mathematically elegant and it may do the best job at ranking teams at the end of a short season based on past performance, but that is VERY different from accurately predicting future outcomes. There simply are nowhere near enough games in a season and most of them are intra-conference. If one were to build a betting model to predict college hockey games, it would look wildly different from KRACH. Obviously! Because KRACH is trying to rank teams the best it can based on a tiny sample of past performance, it isn't trying to tell you who  will come out on top in the future.

Anyway, the podcast is great. I encourage everybody listen.

hmm - thanks ... heh.  Sorry - didn't realize you were referring to KRACH in a future sense, because that was not the context of this conversation. It's OK to just admit you misspoke - because your doubling down to say you were referring to the future, is also silly, given that there is nothing that can predict the future, and nobody says it can. But it's something - and it's better than anything else. And that's all we have.

Glad I could clear up the misconceptions about the home/road weighting.

Quote from: abmarks on November 04, 2025, 11:46:05 PMAdam - I'm assuming you didn't catch one of my questions earlier.

Any chance of looking at home advantage and adjusting for quality of opponent by applying krach?  The ultimate question isn't home advantage. It's whether being at home makes you more likely than you should be to win vs a specific opponent.

I think this is the math for this, someone correct me if I am wrong:

for each game result, a team is likely to win at a percentage of  Ke =Kh/(Kh+Kr)

where Ke is the expected win percent, Kh is the home teams krach and Kr is the road teams krach.

You can calculate a season total for Krach expected home wins by adding up the Ke for all home games by played by a team.

Over or under expectation is the difference between the Ke and the actual win percentage from those home games. 

Admittedly I don't know how to handle ties  when krach adjusting in this model. Krach ratios tell you how often to expect a win, but I don't think I've ever seen a mention of how to use them to predict the tie percentage.  We know that if you are 80% win probability a tie is underperforming, but not as much as a loss, but using any point system for WLT is arbitrary. Is there a theoretical way to predict the tie ratio as well as the win ratio?

Side notes first ...

- not sure who injected the remark about innumeracy into my quote to make it look like I said it, but I didn't. That said, most people are, in general.

- listen to our latest podcast, with NPI architect Tim Danehy if you want more insight

- leave it to BL to twist things around into oblivion, and if they don't fit into the narrowly defined specifications of his liking, then everyone must be terrible or doing something wrong or misleading people. No one is being misled. No one thinks they are being misled. As I said, they endlessly debate, with each other, what the right thing is. If 1.2/0.8 isn't working out, they'll change it. So far it's close enough. And if it was so easy to "game the system" by scheduling nothing but road games - why aren't teams doing it? Like I said, listen to the podcast.

- As for KRACH - saying it's not "remotely" a perfect measure, is silly (go figure). Where's John Whelan when I need him? It's actually - as far as I'm concerned (with some needed tweaks to what we publish) - the best method there is. Listen to Tim discuss this on the podcast, where he gives a counter-point to that, while also saying that, with enough data, it would be the best system.

- To finally answer the question ... certainly you can do a lot of complex formulations to make things work, and that's what I mean about tweaking KRACH for home/road. But when it comes to something like NPI, they want the criteria to be also somewhat understandable. And since it's all "close enough" - albeit arbitrary in many ways - it works out. Just as the Pairwise did.

- I have 2 math PhDs helping me tweak our KRACH calculation to do exactly things like that, and more.  At the end of the day, however, almost anything you do will have subjective components to them. For example, having a "recency bias" is doable, but a philosophical discussion. Having a "Quality Win Bonus" is done now, but that's a philosophical decision, not based on real math, per se. Same for H/R weights, and OT weights. They're not always gunning for perfect math. "Better" math, but not perfect. Because there are other considerations.  I mean, you either accept that or you don't.  Sorry if anyone believed otherwise - but I tell you the truth, and you can get all agitated about it, or not. Up to you.

repeating ... listen to the Podcast.

Quote from: BearLover on November 04, 2025, 11:03:15 AM

Quote from: adamw on November 04, 2025, 09:53:24 AM

Quote from: BearLover on November 03, 2025, 04:12:35 PMSo my question here was whether the current home/away weighting in the NPL is accurate. It assumes home teams should win 60% of the time, all else being equal. Sounds like in effect it's closer to 53.5%. If I have that right, the 1.2/0.8 weighting should be lessened. Obviously, you'd need to cross-check this against more years to confirm last year wasn't a fluke.

This is all a wash if every team plays the same ratio of games home:away.

Nobody ever believed the home/road weighting was accurate. 100% accuracy was literally never the goal. The fact that it's not accurate is not even a question.

Umm, ok? Obviously no one would expect it to be perfectly accurate. But I would have expected it to be somewhat based in reality, given they had to choose a number. Why 1.2/0.8 instead of 1.1/0.9 or 1.3/0.7? It's pretty clearly implied that the question of "whether the weighting is accurate" also includes the question of "if it's not accurate, then by how much?"

They literally debate 6 ways to Sunday every different possibility all the time. Conference games vs. NC - OT weights - etc... You want the minutes of all the meetings?  I'm telling you the general gist is that it wasn't intended to be accurate. It was intended to encourage bigger programs to play road non-conference games.  So no one has cared about tweaking 1.2/0.8 to fit whatever the exact home ice advantage is every year - which fluctuates.

Quote from: BearLover on November 03, 2025, 04:12:35 PMSo my question here was whether the current home/away weighting in the NPL is accurate. It assumes home teams should win 60% of the time, all else being equal. Sounds like in effect it's closer to 53.5%. If I have that right, the 1.2/0.8 weighting should be lessened. Obviously, you'd need to cross-check this against more years to confirm last year wasn't a fluke.

This is all a wash if every team plays the same ratio of games home:away.

Nobody ever believed the home/road weighting was accurate. 100% accuracy was literally never the goal. The fact that it's not accurate is not even a question.

Quote from: BearLover on November 02, 2025, 11:43:42 PM

Quote from: adamw on November 02, 2025, 11:37:06 PM

Quote from: BearLover on November 02, 2025, 12:18:41 PMChoosing teams for the national tournament based on a formula that is designed to maximize things other than picking the most qualified teams is absolutely nuts! I'm wondering how off the 1.2/0.8 split is though. I could imagine it's close to the "true" advantage...

National home ice advantage last year was .5377 ... which is the lowest in at least 10 years. I think that's also skewed by all the "bigger" teams that still host "smaller" teams in most instances where two such teams meet. But that's somewhat speculative.

Also I assume it is counting teams that host home playoff games by virtue of having a better record.

If I have time I'll eventually calculate what home ice advantage was within the ECAC regular season last year. Small sample but it would account for most of these issues. If someone else wants to take a crack at it, feel free.

my bad not including ties in a WL% formula. Doesn't change it much nationally ... .5345

the home ice advantage in just ECAC regular-season games was ... negative - by a lot ... .4394

I checked this for all conferences just to make sure I wasn't doing something wrong ... the national WL% for regular-season games in any conference was .... .4960

This lends credence to the idea that non-conference games are largely by "bigger" teams playing "smaller" teams.

The home WL% in the Big Ten - BTW - was ... .5563

take that for what it's worth

(FYI - I did not do goofy points % math with 3-2-1 points, etc... - Just 2 for a W, 1 for T through OT - so Ws including OTWs)

Quote from: BearLover on November 02, 2025, 12:18:41 PMChoosing teams for the national tournament based on a formula that is designed to maximize things other than picking the most qualified teams is absolutely nuts! I'm wondering how off the 1.2/0.8 split is though. I could imagine it's close to the "true" advantage...

National home ice advantage last year was .5377 ... which is the lowest in at least 10 years. I think that's also skewed by all the "bigger" teams that still host "smaller" teams in most instances where two such teams meet. But that's somewhat speculative.

Quote from: BearLover on November 02, 2025, 11:26:24 AMWhat is the accurate home/away split? You'd think someone would have figured it out by now. One crude way to do it would be to take all ECAC teams and compare their in-conference home record to their in-conference away record. 1.2/0.8 sounds large (that seems to indicate a team at home is 50% more likely to win than when they're away?). I.e., Cornell winning 4/10 games on the road is as easy as them winning 6/10 games at home.

As for the change to the playoff weighting—-it seems incorrect to say the old formula "punished" or "dinged" teams for getting home games. Rather, the old formula included ann equalizing factor of home games with respect to the pairwise/NPI rankings, and the new formula removes this equalizing factor. Which is to say, if the home/away split was accurate, there was no "punishment." And if it was inaccurate, then the fix should be changing the split to make it more accurate.

Your conclusion about the tooth-and-nail 4/5 playoff series seems backwards. If a team gets the 4-seed and get home ice, isn't that a big advantage, and now, under the new rule, the 5-seed gets severely punished for barely missing out on home ice because they are no longer are protected by the home/away split?

(Plus, the 4-seed still gets the bonus of home games with respect to actually winning your conference tournament. Now, they get this benefit AND they get the benefit that they are more likely to win for their NPI ranking.)

My issue with the new rule is it disproportionately benefits teams who play their conference semis/finals at home sites. Now these teams get the benefit of home ice for up to two extra rounds, without their advantage being accounted for in the NPI. Whereas a team like Cornell is actually disadvantaged, relatively speaking.

*we do benefit in a vacuum from this new rule in the sense that we usually get a round of ECAC home games. But other teams who are competitive for an at-large spot benefit more.

Well - it wasn't my decision.  But if the "real" home/road is really like 1.02/0.98 - then it most certainly is a punishment for earning home ice advantage. The 1.2/0.8 was put in to encourage top teams to schedule road non-league games.  It wasn't meant to be mathematically accurate.

Your argument about 4/5 logic being backwards would be true if the 1.2/0.8 was real.

The rest - we'll see how it plays out.  Cornell is going to get other advantages from the new system.

Quote from: CU2007 on November 02, 2025, 12:09:08 AM

Quote from: adamw on October 27, 2025, 01:07:22 PM

Quote from: upprdeck on October 25, 2025, 10:51:14 AMClarkson has beaten PSU/NDAK and also got dominated by a bad RIT team and lost to Canisius who also beat Colgate, but Colgate has tied BU and beaten Maine?  Canisius lost to LIU which is that teams only win.

Also since NPI has replaced the pairwise. What tweaks were made to make that different?

It's complicated

https://www.collegehockeynews.com/info/?d=npi

Adam - I found the summary useful but noted the following line regarding home and away game weightings: For postseason conference tournament games, there is no weighting.

Is this true if the game is played at the home rink of the higher seed rather than a neutral site? And if so, what is the rationale there?

Is

yes it's true for all conference tournament games. The rationale stems from conversations with CC coach Kris Mayotte on our podcast a couple seasons ago, after his team barely lost out on an NCAA bid because it lost a best-of-3 home playoff series. He thought the team was getting punished for earning home ice, and then losing a tooth-and-nail 4/5 series. I thought he had a great point and it wasn't just sour grapes - and was glad to see it get taken to the Committee and go from there. Other coaches had said similar things over the years, but none had suggestion that kind of specific solution.  As you know 1.2/0.8 isn't really an accurate home/away split as it is, so to get dinged for it in a 4/5 series in the postseason seems pretty unfair.