Pairwise and Bracketology 2013 Edition

[Patman]

Not sure I understand this comment in context Patman. Are you saying the current system is or is not good? are you commenting on the site at playoffstatus.com? Do you know those people? Sorry. I am slow. I don't understand....

It is my assumption from looking at that site, and from what FS wrote, that they have their own system, and they are not aware that it doesn't match the PWR. And, that they don't know how the PWR works. Do you know more about that?

No, I mean a model based on score prediction in tandem with proper PWR execution would end a lot of this discussion.

A few years ago I was giving probs on such a model at the qf stage but I made myself look bad with errors in the program somewhere. Either in game merging or pair wise algo

Despite all this it's doable but its not the most fluid piece of programming as far as I've been able to either discern or muster.

 

[Numbers]

Re: Pairwise and Bracketology 2013 Edition

No, I mean a model based on score prediction in tandem with proper PWR execution would end a lot of this discussion.

A few years ago I was giving probs on such a model at the qf stage but I made myself look bad with errors in the program somewhere. Either in game merging or pair wise algo

Despite all this it's doable but its not the most fluid piece of programming as far as I've been able to either discern or muster.

Oh, so I get it. You are not talking about playoffstatus.com at all, but in general, what might happen at this point in the season, and you propose:

1) A predictive model with some depth, so that it uses not only won/lost/tied game results, but goals for/against, etc, to predict the outcomes of the conference tourneys.

2) Once the outcomes are predicted, then do the PWR calculation on them.

And, I assume since there are still so many options, right now you would do a Monte Carlo. But, perhaps next week, it would be a full up weighted average of all possible results?

Wow. That would be some programming.

 

[Patman]

Oh, so I get it. You are not talking about playoffstatus.com at all, but in general, what might happen at this point in the season, and you propose:

1) A predictive model with some depth, so that it uses not only won/lost/tied game results, but goals for/against, etc, to predict the outcomes of the conference tourneys.

2) Once the outcomes are predicted, then do the PWR calculation on them.

And, I assume since there are still so many options, right now you would do a Monte Carlo. But, perhaps next week, it would be a full up weighted average of all possible results?

Wow. That would be some programming.

I could do it, I won't. There are different challenges posed at each stage... Mostly it's keeping all the team ID values straight. In the playoffs the hard part is handling the flexibility.

---

So, hierarchy of evaluation... In terms of utility. The model part is the easiest as I know I have software that works (and free at that!)

Pairwise
Conf Tournaments
Regular season w/o effective tiebreakers
Actual tie-breakers
In season tournaments...

In that order in the sense that it becomes available for use progressively earlier in the season. As it is now, were I to waste a weekend I could crank something out. I've done that before.

Oddly enough the modeling goes similarly.

Results and teams
Home ice by game
Score by game
Duration of game.

In that case it's more a matter of model refinement.

 

[Patman]

Re: Pairwise and Bracketology 2013 Edition

As of right now, i don't have an input file but I'm close to being able to calculate pairwise again. Of course I'm going to assume there are more bugs here than a Lousiana swamp.

 

[LTsatch]

Re: Pairwise and Bracketology 2013 Edition

Nice analysis by Wodon http://www.collegehockeynews.com/news/2013/03/13_bracket_abcs_playoff_edition.php

 

[Wisko McBadgerton]

Re: Pairwise and Bracketology 2013 Edition

Nice analysis by Wodon http://www.collegehockeynews.com/news/2013/03/13_bracket_abcs_playoff_edition.php

I can't speak for all of that article but in the case of WI he says "Wisconsin has a vested interest in Colorado College remaining a TUC, something that will be difficult to do if CC loses its playoff series to Denver.

As WI is 0-2 against CC and CC falling off would move WI from 8-9-3 to 8-7-3 (.5277) flipping that component against 6 teams ahead of them, I believe that would be incorrect.


Interesting side note: It was pointed out in another thread that if UNO were to be swept by MN St. this weekend their RPI could potentially fall to .4999. Crazy. Affecting WI, UND, and ND I believe.

 

[Patman]

Re: Pairwise and Bracketology 2013 Edition

I can't speak for all of that article but in the case of WI he says "Wisconsin has a vested interest in Colorado College remaining a TUC, something that will be difficult to do if CC loses its playoff series to Denver.

As WI is 0-2 against CC and CC falling off would move WI from 8-9-3 to 8-7-3 (.5277) flipping that component against 6 teams ahead of them, I believe that would be incorrect.


Interesting side note: It was pointed out in another thread that if UNO were to be swept by MN St. this weekend their RPI could potentially fall to .4999. Crazy. Affecting WI, UND, and ND I believe.

.4999 is a value that'd make everybody nervous... it'll dither back and forth as the results come in.

 

[Wisko McBadgerton]

Re: Pairwise and Bracketology 2013 Edition

.4999 is a value that'd make everybody nervous... it'll dither back and forth as the results come in.

Yeah, I assume that's the number if those were the only two games played. Just struck me as it would be a heck of a twisting of the knife for a team like WI that could end up out (provided they still do some winning) on a turn of events like that. Or for anybody, of course-- .4999!!!

 

[Numbers]

Re: Pairwise and Bracketology 2013 Edition

Yeah, I assume that's the number if those were the only two games played. Just struck me as it would be a heck of a twisting of the knife for a team like WI that could end up out (provided they still do some winning) on a turn of events like that. Or for anybody, of course-- .4999!!!

I actually hope this happens, because someone at the NCAA might decide that the TUC Cliff is an issue that needs to be addressed.




On another note, in addition to what Wodon wrote, one interesting thing here is that if Michigan beats Western Michigan, they will become a TUC, and that will even out Notre Dame's TUC record, and bring their PWR more in accord with their RPI rank.
I am not sure I want that to happen (Gophers and Wolverines don't get along too well), but that is the reality.

 

[FlagDUDE08]

Re: Pairwise and Bracketology 2013 Edition

I actually hope this happens, because someone at the NCAA might decide that the TUC Cliff is an issue that needs to be addressed.

The TUC cliff will always be an "issue" for a team that is on the fence. Just pick a way to do it and stick with it.

 

[bottomdweller]

Re: Pairwise and Bracketology 2013 Edition

Curious, just because I want Lowell is to remain in this lovely Eastern time zone for the NCAA; if UNH gets swept by PC in the HE QFs, what are the chances they are on the outside looking in? I need them to not be a 4 if Lowell is a 1, and I'd actually just prefer them to watch from their couches.

 

[hockey east fan]

Re: Pairwise and Bracketology 2013 Edition

Curious, just because I want Lowell is to remain in this lovely Eastern time zone for the NCAA; if UNH gets swept by PC in the HE QFs, what are the chances they are on the outside looking in? I need them to not be a 4 if Lowell is a 1, and I'd actually just prefer them to watch from their couches.

Looks like UNH is pretty safe...but...they could drop to a 4-seed if swept by PC.

 

[JimDahl]

Re: Pairwise and Bracketology 2013 Edition

Curious, just because I want Lowell is to remain in this lovely Eastern time zone for the NCAA; if UNH gets swept by PC in the HE QFs, what are the chances they are on the outside looking in? I need them to not be a 4 if Lowell is a 1, and I'd actually just prefer them to watch from their couches.

I actually wrote a little about UNH this week. In short, you want them to win at least one.

New Hampshire PWR outlook for this weekend

 

[Numbers]

Re: Pairwise and Bracketology 2013 Edition

For any that wondering about the site mentioned on the prior page of this thread (post #533-536), called playoffstatus.com:

I sent them an email asking about their methods. They replied that they use their own ranking system to choose the field when they do the %ages, not the PWR. They also added a very brief, non-detailed description of their ranking system. It could be similar to KRACH, but there is insufficient detail to know that. They compute the remaining games in the same way that Jim Dahl does, with Monte Carlos, but they add a home/away factor. I can't comment on the correctness of their factor, because their explanation doesn't give enough detail to know how heavy the factor is. They also said that next year they may change to the PWR for Hockey.

In short, playoffstatus.com is of no value to us reading this site in these last couple of weeks.

 

[Steve_MN]

Re: Pairwise and Bracketology 2013 Edition

For any that wondering about the site mentioned on the prior page of this thread (post #533-536), called playoffstatus.com:

I sent them an email asking about their methods. They replied that they use their own ranking system to choose the field when they do the %ages, not the PWR. They also added a very brief, non-detailed description of their ranking system. It could be similar to KRACH, but there is insufficient detail to know that. They compute the remaining games in the same way that Jim Dahl does, with Monte Carlos, but they add a home/away factor. I can't comment on the correctness of their factor, because their explanation doesn't give enough detail to know how heavy the factor is. They also said that next year they may change to the PWR for Hockey.

In short, playoffstatus.com is of no value to us reading this site in these last couple of weeks.

Thanks for the update. I can completely understand them using their own system to figure the winning percentages to "simulate" the games, but NOT running through PWR at the end is just just idiotic. It's KNOWN that this is how the tournament is chosen, and not using it to come up with the NCAA tournament probabilities boggles the mind.

 

[kingdobbs]

Re: Pairwise and Bracketology 2013 Edition

I actually hope this happens, because someone at the NCAA might decide that the TUC Cliff is an issue that needs to be addressed.

I still think they'll come back with some combination of "it's a manufactured 'message board' issue" and "there's always going to be a cliff of some kind."

 

[Numbers]

Re: Pairwise and Bracketology 2013 Edition

Thanks for the update. I can completely understand them using their own system to figure the winning percentages to "simulate" the games, but NOT running through PWR at the end is just just idiotic. It's KNOWN that this is how the tournament is chosen, and not using it to come up with the NCAA tournament probabilities boggles the mind.

Very true Steve. In their defense, and I am sure you looked their site, their site is not hockey specific. So, I think what is happening is that they developed their ranking system, and they apply the same system to all sports. I could give them that, but in general, their response did not suggest to me that they are very well acquainted with hockey's situation.

 

[Steve_MN]

Re: Pairwise and Bracketology 2013 Edition

Very true Steve. In their defense, and I am sure you looked their site, their site is not hockey specific. So, I think what is happening is that they developed their ranking system, and they apply the same system to all sports. I could give them that, but in general, their response did not suggest to me that they are very well acquainted with hockey's situation.

Exactly my thought. Their methodology may work quite well for the conference standings and such, since there's no real "trick" to figuring out what happens at the end. For arguments sake, I'm going to assume they use some form of home ice modified KRACH (obviously, I don't know that this is the case, but it's probably something at least vaguely like that), which is as good a method as I can think of to do any useful simulations on the scale they seem to be attempting. How they can justify putting up tourney participation probabilities without taking their final results (however those are calculated) and running them through PWR (which is really pretty simple to do, from a programming standpoint, just look at how many simulators are out there already)... that I just don't get.

Again, without more information on how they calculate their "Meaningful Win Percentage" (which I seriously doubt we'll ever see) it's difficult to say whether it's a decent methodology or not. It may well be a useful tool to predict multiple game outcomes over the course of a couple weeks, which could be incredibly useful around here. But not putting it through PWR at the end?

 

[Rtn2GoldCountry]

Re: Pairwise and Bracketology 2013 Edition

Is Q a lock to be the # 1 seed overall, or could the Gophers jump them with a Broadmoor Trophy and some help from Cornell? Similar qeustion for Miami, are they locked into a # 1 seed, or could the fall out of the top 4 of the pairwise?

 

[Numbers]

Re: Pairwise and Bracketology 2013 Edition

Is Q a lock to be the # 1 seed overall, or could the Gophers jump them with a Broadmoor Trophy and some help from Cornell? Similar qeustion for Miami, are they locked into a # 1 seed, or could the fall out of the top 4 of the pairwise?

I don't believe the Gophers can jump QU. The reason is that QU will always be ahead in TUC and ComOpp. Priceless has written that Miami is almost guaranteed a #1. I have not researched that.