Page 5 of 6
Posted: Wed Jan 30, 2013 11:34 am
by ghshockeyfan
MNHockeyFan wrote:I was surprised to see the huge impact that Mounds View's loss last night had on their KRACH ranking. They dropped 443 points, from 819 to 376 (-54%)! I guess this is because White Bear Lake was ranked so far below them, at #58.
At the same time I was also surprised that the impact on WBL for their win was relatively minor. They only gained 2 points, going from 10.5 to 12. 5 (+19%). Their ranking did climb 15 places, from #58 to #43.
Would appreciate hearing your explanation, ghshockeyfan, as to why a huge upset like this, so late in the season, appears to impact the losing team (the favorite) so much more than the winning team (the underdog).
The losing team had far more to lose than the winning team to gain in an upset from the KRACH perspective.
Why? Because, as I understand it, KRACH takes WBLs best prior loss and turns that into a win. So, let's just say they lost to a team ranked 45 before last night. Beating the #1 team essentially makes that #45 loss a W. Maybe I'm wrong on this, but it should make sense as to why it didn't help WBL that much if KRACH handles this in such a way.***
MV on the other hand now has their best W become a L. I don't know the best team on their schedule that they beat before, but essentially that becomes a L in KRACH's eyes so-to-speak.
This is how it was explained to me by some experts some time ago. Maybe I misinterpreted what they were saying, but in the extreme case like MV (undefeated #1), it kind of makes sense as to why this happened as it did in KRACH.
***There's a bit more detail to this but I'm going to leave it out, for now, as it could complicate the discussion and really doesn't add much value to the high-level idea.
Posted: Thu Jan 31, 2013 1:58 pm
by sinbin
I'll throw in my $0.02. I think that KRACH looks at the ratio of wins to losses (or winning % to 1- winning %). So, if a team is 22-0-1, the W/L ratio is 45.0 (22.5/0.5). If they lose, their ratio becomes (22.5/1.5) = 15.0, so they get cut down by 67%. The actual number cited below is 54%. Then, adjust by the change in MV's SOS. Then, consider the interaction of all teams and how every team's numbers change after every game, whether they play or not (all that iterative logistic regression stuff) and I think that 54% makes sense. In fact it's only because of MV's almost perfect record and that it is so late in the season that their rating changed so dramatically (the model is highly leveraged based winning %/fewest losses). It may not make perfect sense intuitively, but that's what the mathematical model gives us.
Posted: Fri Feb 01, 2013 7:03 am
by ghshockeyfan
sinbin wrote:I'll throw in my $0.02. I think that KRACH looks at the ratio of wins to losses (or winning % to 1- winning %). So, if a team is 22-0-1, the W/L ratio is 45.0 (22.5/0.5). If they lose, their ratio becomes (22.5/1.5) = 15.0, so they get cut down by 67%.
Slight adjustment to the math inputs based on KRACH having a phantom tie for all teams as part of its algorithm.
With that in mind - the before WBL record for MV in KRACH is 22-0-2 => W/L Ratio is 23/1 = 23. And, similarly, after WBL record is 22-1-2 => W/L Ratio is 23/2 = 11.5 => cut down by 50% (11.5/23). from there...
sinbin wrote:The actual number cited (earlier) is 54%. Then, adjust by the change in MV's SOS. Then, consider the interaction of all teams and how every team's numbers change after every game, whether they play or not (all that iterative logistic regression stuff) and I think that 54% makes sense. In fact it's only because of MV's almost perfect record and that it is so late in the season that their rating changed so dramatically (the model is highly leveraged based winning %/fewest losses). It may not make perfect sense intuitively, but that's what the mathematical model gives us.
The "adjustment" is what gets at the final ranking value and it's not very straightforward. I tried to get at this directly with my earlier post but it may be that the math route is best understood. Some find it easier to simply think of upsets as offsetting earlier weaker losses for the winner and earler better wins for the loser. This is all part of the adjustment phase as what's above is only considering one team without comparing that to everyone else for an actual ranking value.
Updated 01/31 AM 26672
Posted: Fri Feb 01, 2013 7:20 am
by ghshockeyfan
Updated 01/31 AM
Posted: Fri Feb 01, 2013 8:30 am
by D6 Girls Fan
I was told there would be no math.
Posted: Fri Feb 01, 2013 11:29 am
by sinbin
GHS, thanks. That explains one question I always had and that is how does KRACH account for teams with 0 losses and 0 ties, since you can't divide by 0? So, it makes the calculations somewhat inaccurate based on actual records, but since this is a model anyway, once you have enough data, it doesn't impact the results as much (except for the outliers).
P.S. D6, there will be a short KRACH quiz to award bonus points for the State Tournament You Pick 'Em challenge, so please pay attention.
Posted: Fri Feb 01, 2013 6:12 pm
by ghshockeyfan
sinbin wrote:GHS, thanks. That explains one question I always had and that is how does KRACH account for teams with 0 losses and 0 ties, since you can't divide by 0? So, it makes the calculations somewhat inaccurate based on actual records, but since this is a model anyway, once you have enough data, it doesn't impact the results as much (except for the outliers).
Yep - you got it... I could probably remove the phantom tie now as I believe everyone has at least 0.5 wins and 0.5 losses - but I'll double check and - yes - slight impact but usually not enough to change anything too much once removed.
Posted: Fri Feb 01, 2013 6:13 pm
by ghshockeyfan
D6 Girls Fan wrote:I was told there would be no math.
LOL - I actually did when I read this!
Updated 02/01 AM 27112
Posted: Fri Feb 01, 2013 6:18 pm
by ghshockeyfan
Updated 02/01 AM
Updated 02/02 AM 27292
Posted: Sat Feb 02, 2013 8:55 am
by ghshockeyfan
Updated 02/02 AM
Updated 02/03 AM 27646
Posted: Sun Feb 03, 2013 10:28 am
by ghshockeyfan
Updated 02/03 AM
Posted: Sun Feb 03, 2013 1:03 pm
by goaliepad
I have noticed the Forest Lake girls record you have is 11-10-2, and it really is 13-10-2.
Posted: Sun Feb 03, 2013 1:16 pm
by allhoc11
[quote="goaliepad"]I have noticed the Forest Lake girls record you have is 11-10-2, and it really is 13-10-2.[/quote
If I remember correctly this was covered before, and their two wins over non-mn teams don't factor in to the equation, because they are not in the data base (i.e. no ranking for SOS, or other comparative scores). The same was true for some other teams who have played non-mn teams, I believe.
Posted: Sun Feb 03, 2013 1:38 pm
by ghshockeyfan
allhoc11 wrote:goaliepad wrote:I have noticed the Forest Lake girls record you have is 11-10-2, and it really is 13-10-2.
If I remember correctly this was covered before, and their two wins over non-mn teams don't factor in to the equation, because they are not in the data base (i.e. no ranking for SOS, or other comparative scores). The same was true for some other teams who have played non-mn teams, I believe.
This is correct.
Updated 02/06 PM - End of Reg Season
Posted: Wed Feb 13, 2013 8:28 am
by ghshockeyfan
Updated 02/06 PM - End of Reg Season
Posted: Wed Feb 13, 2013 9:21 am
by sinbin
It would be interesting to see year-end KRACH with teams still alive highlighted (or those eliminated crossed-out) going forward to see how the KRACH and actual games match up.
Posted: Thu Feb 14, 2013 3:15 pm
by luckyEPDad
With the huge rating swings near the top and goofy results near the bottom it looks like SOS needs a little tweaking. A 2 win team should not be ranked higher than a 14 win team just because they lost games to Buffalo and NWC.
Posted: Thu Feb 14, 2013 3:41 pm
by rwb1351
luckyEPDad wrote:With the huge rating swings near the top and goofy results near the bottom it looks like SOS needs a little tweaking. A 2 win team should not be ranked higher than a 14 win team just because they lost games to Buffalo and NWC.
SOS is an average of the strength of teams they play, regardless of outcome. Whether they win or lose that game shouldn't impact their SOS rating.
SOS has nothing to do with how good a team is; rather, it is a reflection of how good the teams they play are.
In theory, a team could win 0 games but still have the most difficult schedule (often not seen in reality because the best teams don't like to play the worst teams).
Posted: Thu Feb 14, 2013 7:17 pm
by luckyEPDad
rwb1351 wrote:luckyEPDad wrote:With the huge rating swings near the top and goofy results near the bottom it looks like SOS needs a little tweaking. A 2 win team should not be ranked higher than a 14 win team just because they lost games to Buffalo and NWC.
SOS is an average of the strength of teams they play, regardless of outcome. Whether they win or lose that game shouldn't impact their SOS rating.
SOS has nothing to do with how good a team is; rather, it is a reflection of how good the teams they play are.
In theory, a team could win 0 games but still have the most difficult schedule (often not seen in reality because the best teams don't like to play the worst teams).
SOS is one of the things that goes into the KRACH equation. How much it affects the KRACH rating is what I'm talking about. Right now I think SOS is being given disproportionate weight.
Posted: Sat Feb 16, 2013 5:01 pm
by ghshockeyfan
luckyEPDad wrote:With the huge rating swings near the top and goofy results near the bottom it looks like SOS needs a little tweaking. A 2 win team should not be ranked higher than a 14 win team just because they lost games to Buffalo and NWC.
I expect huge differences for the KRACH rating when comparing teams at the top and those teams at the bottom. In fact, those at bottom could be lower but I only go to a certain number of decimal places.
If a 2 win team plays opponents all ranked above x and a 14 win team plays teams all ranked below x, then I can see ranking the 2 win team above the 14 win team.
Not sure if this is the case in the scenario cited - but just a point worth considering.
Updated 02/16 - End of Sections
Posted: Sat Feb 16, 2013 5:02 pm
by ghshockeyfan
Updated 02/16 - End of Sections
Rankings???
Posted: Sat Feb 16, 2013 7:35 pm
by fishfreak11
How can Hibbing have a lower ranking than proctor after beating them in sections especially when they were 9 and #12 before this section final?? Does not seem correct
Posted: Sat Feb 16, 2013 7:46 pm
by luckyEPDad
ghshockeyfan wrote:luckyEPDad wrote:With the huge rating swings near the top and goofy results near the bottom it looks like SOS needs a little tweaking. A 2 win team should not be ranked higher than a 14 win team just because they lost games to Buffalo and NWC.
I expect huge differences for the KRACH rating when comparing teams at the top and those teams at the bottom. In fact, those at bottom could be lower but I only go to a certain number of decimal places.
If a 2 win team plays opponents all ranked above x and a 14 win team plays teams all ranked below x, then I can see ranking the 2 win team above the 14 win team.
Not sure if this is the case in the scenario cited - but just a point worth considering.
The example I looked at had a 2 win team loosing against Buffalo, NWC and a bunch of other teams that were similar in strength to the team that won 14 games. If Worthington were to lose to BSM it wouldn't make them a much better team, but I bet it would jump their KRACH ranking at least 20 spots.
Re: Rankings???
Posted: Sat Feb 16, 2013 9:37 pm
by ghshockeyfan
fishfreak11 wrote:How can Hibbing have a lower ranking than proctor after beating them in sections especially when they were 9 and #12 before this section final?? Does not seem correct
When other "upsets" happen, it can impact every team. My guess is that some upsets happened, those teams that lost were knocked down the rankings, and so too were their earlier opponents impacted (to some degree as those teams are not longer as highly ranked).
Also - one game does not a season make. KRACH looks at the entire season, not just the section final. This could be considered a flaw - but it is what it is.
Posted: Sat Feb 16, 2013 9:45 pm
by ghshockeyfan
luckyEPDad wrote:ghshockeyfan wrote:luckyEPDad wrote:With the huge rating swings near the top and goofy results near the bottom it looks like SOS needs a little tweaking. A 2 win team should not be ranked higher than a 14 win team just because they lost games to Buffalo and NWC.
I expect huge differences for the KRACH rating when comparing teams at the top and those teams at the bottom. In fact, those at bottom could be lower but I only go to a certain number of decimal places.
If a 2 win team plays opponents all ranked above x and a 14 win team plays teams all ranked below x, then I can see ranking the 2 win team above the 14 win team.
Not sure if this is the case in the scenario cited - but just a point worth considering.
The example I looked at had a 2 win team loosing against Buffalo, NWC and a bunch of other teams that were similar in strength to the team that won 14 games. If Worthington were to lose to BSM it wouldn't make them a much better team, but I bet it would jump their KRACH ranking at least 20 spots.
An easy way to investigate this would be to look at the before and after of exactly this sort of scenario. Pick a weaker team that played a very strong one, look at the weak team's rank/rating before playing the strong team, and then look at the after. Assuming they lost (and not confusing all the other games going on at the same time that also indirectly impact every other team) - my guess is not much movement at all for the losing team. We could hunt for an example or we could run a hypothetical phantom game with exactly this scenario and see what happens. My guess is that there are plenty of examples already existing and here is every ranking for this season to use for before and after numbers if you can find a game & its date that is similar to what you described: