Bracketology 2023

[BearLover]

Yes, a better model would be a ton of work.  What if Wayne Gretzky, Jr was on a team, but injured for the first 20 games of the season and the team went .500.  Since he got back, he's averaging 2.5 points per game and the team is undefeated.  Where would your "better" model rank them headed into the playoffs?  Maybe the team hired a better strength and conditioning coach halfway through the season?   What if your star players also care about academics and the team always tanks during exam weeks....and there are midterms the week of the quarterfinals?

Models will never capture the richness and complexity of life.  Once you accept that fact, you can paradoxically enjoy looking at models again, because you are content in your certainty that it only tells a part of the story and just agree with yourself to leave it at that.

My problem with the models at issue is that they don't even accurately tell the part of the story they claim to tell. Yes, they don't account for injuries, or changes in personnel, or exams. But they don't claim to do those things. Rather, these models claim to predict future likelihoods based entirely on a team's past success. But they don't even do that properly, because the probabilities the models spit out do not accurately reflect the actual probability that one should get from looking purely at past success.  

As to adamw's point—I want to preface this by saying I love CHN and go there 30 times a day. There are many awesome features of the site. But when it comes to the model, my view is that if it isn't doing what it claims to do (provide realistic probabilities of future results), and we know it does not do this, then it would be better to not publish the model at all. Otherwise, publishing it just misleads people.

I don't have the training necessary to help improve the model. I hope someone does who is willing to assist. But until that happens, better not to publish the model at all, IMO.

[CU77]

KRACH could be adjusted to do better, as I outlined above. KRACH is a very simple model.

[upprdeck]

The first question would be on what context are the models not working?  As a whole or for a team?

Look at the probablility matrix thing

Cornell is like 60% to get in..   the model could have us at 90% of the time winning in a sweep and losing to Harvard.  That would get us in. Maybe

But it also has to account that there could be 3 upsets someplace else and we wouldnt get in.  

The first thing you have to do is eyeball what you thing will happen and then compare it to the model.  Then you can decide if its wrong.  You also have to do that across 100s of games.

And if a model is successful more than X% that makes it a good model. What do we consider that X to be

Look at horse racing.   Favs win like less than 30% of the time..  So that means across millions of peoples models they can only get the fav right 30% and that includes tons of races where there really is no toss up choice..   What is the stat where races have favs that are not clear favs like less than 10% or lower? And thats in races with clear things to measure and known past results and 1000s of histories to compare to and training notes to look at.

[ugarte]

what if i told you that all of what has come before and all that comes after has been foretold. *turns chair around* *flips hat around* and that one man gave his life...

[Trotsky]

I'm sure we'll solve this problem here.

Shorter Hume: We make shit up and try our best.

Shorter Shorter Hume: Drop the puck.

[adamw]

KRACH could be adjusted to do better, as I outlined above. KRACH is a very simple model.

The door is open ...

[adamw]

As to adamw's point—I want to preface this by saying I love CHN and go there 30 times a day. There are many awesome features of the site. But when it comes to the model, my view is that if it isn't doing what it claims to do (provide realistic probabilities of future results), and we know it does not do this, then it would be better to not publish the model at all. Otherwise, publishing it just misleads people.

I don't have the training necessary to help improve the model. I hope someone does who is willing to assist. But until that happens, better not to publish the model at all, IMO.

I generally agree with this philosophy. Which is why I cannot stand publishing "Bracketology" articles that tell you what the Brackets would be like if the season ended today - because they don't (especially when said articles don't acknowledge that point). It's also why I can't stand polls.

However, I disagree that the CHN Probability Krach-based Matrix thingy is doing that.

The Matrix is intended to play out the schedule and see who finishes where. Automatically. It's like "You Are the Committee" but done for you, because you can't eyeball and figure it out, and because it's a pain to manually input results - and also then proceed forward with brackets based on the first set of results you put in.

I don't think anyone in their right mind really believes we're predicting the future.  It's meant to extrapolate probable Pairwise scenarios - not "predict winners" per se.

[CU77]

KRACH could be adjusted to do better, as I outlined above. KRACH is a very simple model.

The door is open ...

I don't care enough to do it myself. In any case Massey probabilities are likely better, as they use more input (scores, locations, times).

[David Harding]

KRACH could be adjusted to do better, as I outlined above. KRACH is a very simple model.

The door is open ...

I don't care enough to do it myself. In any case Massey probabilities are likely better, as they use more input (scores, locations, times).

This comic says it well:  https://www.gocomics.com/thebuckets/2023/03/08

[marty]

KRACH could be adjusted to do better, as I outlined above. KRACH is a very simple model.

The door is open ...

I don't care enough to do it myself. In any case Massey probabilities are likely better, as they use more input (scores, locations, times).

This comic says it well:  https://www.gocomics.com/thebuckets/2023/03/08

Cloudy with a high wind warning.

[adamw]

KRACH could be adjusted to do better, as I outlined above. KRACH is a very simple model.

The door is open ...

I don't care enough to do it myself.

Congrats - you're in a growing club.

In any case Massey probabilities are likely better, as they use more input (scores, locations, times).

Times? We do have a home/away version of KRACH. It's not much different. I should use it though.

Scores have always been an extremely dubious and debatable thing to use in hockey. I go back and forth on it.

[upprdeck]

few sports have things like goals scored in the end of hockey games.

What other sport has a mechanism where teams can totally change the way they play to try and score goals at the end of games, that can lead to the other team scoring like ENGs in hockey..  even Soccer where teams throw everything forward seldom does it mean the other team scores.. But in hockey a 3-1 game can turn into a 4-3 game that had little to do with how the 95% of the game was played.

[Dafatone]

There are metrics such as Corsi and Fenwick. But those annoy me because "what if we measured all shots and not just shots on goal" isn't a big enough breakthrough to deserve slapping someone's name on it.

[CU77]

In any case Massey probabilities are likely better, as they use more input (scores, locations, times).

Times? We do have a home/away version of KRACH. It's not much different. I should use it though.

Scores have always been an extremely dubious and debatable thing to use in hockey. I go back and forth on it.

Times = early season vs late season results.

Scores are data, completely ignoring that data is almost surely subobtimal.

The fancy-stats people in pro hockey likely have good ideas about rankings.

As it stands now, Massey's match-up tool can be used to predict win probabilities for any possible game. I doubt that it's possible to significantly improve on those.

[Al DeFlorio]

few sports have things like goals scored in the end of hockey games.

What other sport has a mechanism where teams can totally change the way they play to try and score goals at the end of games, that can lead to the other team scoring like ENGs in hockey..  even Soccer where teams throw everything forward seldom does it mean the other team scores.. But in hockey a 3-1 game can turn into a 4-3 game that had little to do with how the 95% of the game was played.

Basketball is like that when the two-minute foul line parade starts for the team that played best for the first 38 minutes.  Unwatchable now for me.