Division I Rutter Computer Ranking Top Ten for Monday, January 14th 2008

[LakersFan]

Rutter Division I Women's Ice Hockey Computer Ranking Top Ten

Through Games of Sunday, January 13th.

Rating is based on wins, loses, and ties only. Margin of victory and game location are not included in the rating. A team's rating from the previous year is used as an initial weight using a Bayesian approach.

Rank Team Rating Previous Record
1 Harvard 2.1913 1 16-1-0
2 New Hampshire 1.7458 2 19-3-0
3 Minnesota Duluth 1.7085 3 18-3-1
4 St. Lawrence 1.1926 6 15-7-0
5 Minnesota 1.1817 5 15-4-3
6 Mercyhurst 1.1614 4 15-5-2
7 Wisconsin 0.8581 8 14-6-2
8 Dartmouth 0.8214 7 11-5-2
9 Connecticut 0.7761 9 15-4-3
10 St. Cloud State 0.6088 NR 13-5-4

Dropped from top ten: Clarkson

For more information, visit http://math.bd.psu.edu/faculty/rutter/WomensRankings.html

Copyright 2007 Michael A. Rutter

This work is licensed under the Creative Commons Attribution-Share Alike 3.0 License. To view a copy of this license, visit http://creativecommons.org/licenses/by-sa/3.0/us/; or, (b) send a letter to Creative Commons, 171 2nd Street, Suite 300, San Francisco, California, 94105, USA.

[RStarr]

ahhhh, suck.

I was viewing the thread and I wondered how Clarky didnt drop out when SCSU won two over OSU....In the end I decided I didnt really care and all that mattered is that they were still there.

...it was last weeks thread.....so i stand corrected. :(

[BCWH Superfan joe]

The numbers aren't too pretty for BC. They know what they have to do.

[puckster]

Here are the Women's College Hockey Offensive/Defensive Ratings (WCHODR) for 1/13

-------------------------------------------------------------------------------
Division I Offense Defense Overall
Rank Team Record Rating Rank Rating Rank Rating
-------------------------------------------------------------------------------
1 Harvard 16 1 0 3.83 5 0.88 1 0.872
2 New Hampshire 19 3 0 4.08 2 1.21 3 0.841
3 Minnesota Duluth 18 3 1 4.21 1 1.27 4 0.841
4 Wisconsin 14 6 2 3.35 9 1.12 2 0.805
5 St. Lawrence 15 7 0 3.95 4 1.77 6 0.759
6 Mercyhurst 15 5 2 4.05 3 1.93 9 0.746
7 Minnesota 15 4 3 3.46 7 1.81 7 0.712
8 Dartmouth 11 5 2 3.68 6 2.44 15 0.648
9 Clarkson 14 5 3 2.36 19 1.53 5 0.635
10 Princeton 8 8 4 2.55 16 1.88 8 0.601

WCHODR is a computer generated rating system applied to women's collegiate hockey. It produces an Offensive, a Defensive, and an Overall rating for Division I Women's Ice Hockey teams based on past game results.

WCHODR places equal emphasis on offense and defense and provides an overall rating for each team based on its offense/defense ratio. A team's Offensive rating is based on the goals it has scored, with an adjustment for the quality of the defenses it has faced to score those goals. Similarily, the Defensive rating uses goals allowed and adjusts for the offensive ratings of the opponents. An adjustment is also made for home-ice advantage. An Expected Winning Pct is computed to reflect a team's outlook when playing a balanced schedule against all teams.

A key feature of WCHODR is that it may be used to forecast future scores.

WCHODR also provides forecasts for upcoming games.

Further info can be found at:
http://it.stlawu.edu/~chodr/wchodr/current.html

[LakersFan]

The WCHODR rankings is the work of Robin Lock, a St. Lawrence statistics professor (and hockey player). Big difference between his work and mine is that margin of victory/score is a key component of Robin's work. Robin has been doing his work a lot longer than I have, and his method is very sound.

You can see the effect of adding score by looking at Wisconsin. They are much higher in Robin's rankings due partly to the fact they don't give up a lot of goals.

The one thing that goes against my gut feeling is that home ice is worth 1.057 goals a game (per his website). This number seems high to me. If we look back to last years NCAA quarterfinals, three road teams won and Wisconsin beat Harvard in 4OT (and therefore should have been up 2 goals since they played two "games"). Perhaps this number is inflated when a highly ranked team plays a poorly ranked team, when the effect is more pronounced. When two team are evenly matched, I think home ice has less of an effect. Maybe this opinion is biased by watching Mercyhurst split many of its series with similar strength teams this season.

[puckster]

Yes, the Wisc has a stronger Defensive rating, bringing them up higher. Same can be said for Princeton which places them 10th in this rating and 14th in yours

I like the forecasting ability in this work. Last week he was at 84% and 79% on the year. Not bad.
http://it.stlawu.edu/~chodr/wchodr/results.html#Summary

[dave1381]

The one thing that goes against my gut feeling is that home ice is worth 1.057 goals a game (per his website)... Maybe this opinion is biased by watching Mercyhurst split many of its series with similar strength teams this season.
Maybe your gut is right about Mercyhurst's home ice, and there is heterogeneity. For an estimate for a model of homogenous home-ice advantage, I think it's ok.

Harvard in 06-07, for instance, was 14-1-1 at home and 9-7-1 road/netural (1-1 neutral), so home ice might matter for some teams more than others.

As for NCAA results, last year is a small sample size. Home teams were 7-1 before last year.

Also, look at the breakdown of the home,road = total losses of the top 5 teams this season.

UNH 1,2 = 3
Harvard 0,1 = 1
UMD 1,2 = 3
Minnesota 0,4 = 4
Mercyhurst 1,4 = 5

So that looks like some strong home-ice advantage to me.

[ARM]

Big difference between his work and mine is that margin of victory/score is a key component of Robin's work.The big problems with trying to make too much of margin of victory are

a) some teams start to work on other things when they get a big lead, others attempt to increase that lead. I don't know how he uses margin, but if the first goal of margin doesn't receive a lot more weight than the 5th goal, or even the 9th goal, then the model is IMO flawed.

b) every once in a while, a team will have "one of those days" -- think Harvard in the 2003 ECAC championship game. Nothing goes right for the team, and it looks like they were badly outplayed. So his model will conclude that UMD was 5 goals better than Minnesota for their first game in Duluth, while people who witnessed the game will tell you that wasn't the case.

The one thing that goes against my gut feeling is that home ice is worth 1.057 goals a game (per his website).Is that a number he derived for women's hockey, or has men's hockey been included. Mens hockey means more fans, more fans should mean a larger home-ice advantage.

As for NCAA results, last year is a small sample size. Home teams were 7-1 before last year.So in three years (still a very small sample size), home teams are 8-4. Because the home team is supposedly also the better team, I don't see that as a strong indication that home ice is that valuable.

[LakersFan]

Comments on comments....

A quick closer look at Robin's website reveals that margin of victory per se is not used in Robin's rankings, just number of goals scores and goals allowed. He explicitly states that who wins or losses the game is not included, just how many goals are scored (and against who). This is a technical distinction, the winner of the game is not included in the model, just the number of goals scored. So if the game ends in a tie, for example, the model does not change, since the outcome is not reverent.

However, the website does not mention a "cap" on the number of goals. To me, this implies an 11-0 win is better than a 5-0 win. It could be a detail that Robin has left out on his webpage, so I will look into this further.

Women's games are modeled separately from men. The home ice advantage for men is 1.025 goals, not really much of a difference.

A couple of years ago, I included home ice in my rankings, and did not find any effect. Haven't tried it again on recent data, however. It is possible that home ice may increase your offensive output, but has little bearing the probability of winning the game (that sounds like something a math professor would say). For the 323 games played so far this year, the average MOV was 2.47 goals (median = 2 goals). 37% of games were ties or decided by goal. Since close games only account for about a third of all games played, this could be one reason why a home ice-advantage does not appear in my ranking system. Sounds like I need to look at this again.

My ranking system can be used to predict the probability that one team beats another team (or a tie). You need a normal table to calculate this, however. Robin's system speaks in terms of goals, so you can quickly compare teams.

[LakersFan]

Just finished a quick email conversation with Robin. Two points to make:

1. I misinterpreted the home ice advantage value. It is the number you multiply by, not add. So really home ice increases you goal output by 5.7%. This pretty much renders all discussion about home ice advantage moot. Sorry about that.

2. There is no correction for blowouts, so an 11-0 game will increase your offensive rating while increasing your opponent's defensive rating. Robin also points out that the opposite is true, that if a blowout is predicted and you only score 4 goals instead of 7, your rating will decreases. To quote Robin: "No surprise that the NCAA would be reluctant to use such a system!".

Regardless of how you feel about including the number of goals scored, it still is an interesting metric. Considering Robin and I approach the same problem for two different ideas, it is interesting to compare the results.

[ARM]

Apparently teams Rutter rankings can either a) drop due to victories over lesser ranked opponents; or b) results of past opponents can still impact teams to a significant extent at this point of the season.

Minnesota dropped from 1.1817 on Monday to 1.1762 today after 1 game versus UND. Perhaps the BC loss to BU also hurt, although the Gophers played both teams so that result should be somewhat buffered. SCSU also drops from 0.6088 to 0.5940 despite a win over Vermont. The Rutter rankings may either need a correction factor or some form of "don't penalize wins" logic.

[dave1381]

Apparently teams Rutter rankings can either a) drop due to victories over lesser ranked opponents; or b) results of past opponents can still impact teams to a significant extent at this point of the season.

Minnesota dropped from 1.1817 on Monday to 1.1762 today after 1 game versus UND. Perhaps the BC loss to BU also hurt, although the Gophers played both teams so that result should be somewhat buffered. SCSU also drops from 0.6088 to 0.5940 despite a win over Vermont. The Rutter rankings may either need a correction factor or some form of "don't penalize wins" logic.

ARM -- A team's Rutter ranking can't go down after a win, all else equal. It's just a coincidence that both happened to fall. KRACH also has that property.

I think you're right in guessing that the Northeastern-BC result is what caused the declines you witnessed. This was a huge upset. The team's difference in Rutter Rankings was a full point, which means BC had an 80% chance of a win and an 8% chance of a tie. Since this estimation is nonlinear, one upset like this can have a big impact on everyone.

So why was Minnesota affected? A couple reasons. One reason is the BC-Northeastern result hurt BC more than it helped Northeastern, I'm guessing. This tends to be a common feature of such systems. Another factor is that BC played Minnesota twice and tied them once, where Minnesota beat Northeastern, so Minnesota is hurt more by this result. SCSU then suffered because it's ranking is closely tied to Minnesota's.

[ARM]

Dave, Good explanation. Minnesota dropped from 1.1891 on Sunday morning to 1.1852 on Monday with BU and Northeastern losing and BC with another tie. The BC tie shouldn't hurt much, because it came vs a team ranked higher than the Eagles and their "Rutter" went up, but the NU loss to Maine dropped the Huskies from -.8228 to -.9369.

That may be a question regarding the Rutter rankings as a potential tool for selecting an NCAA field -- how much effect should a result between teams like Maine and Northeastern have on who makes the field and who does not?

[LakersFan]

That may be a question regarding the Rutter rankings as a potential tool for selecting an NCAA field -- how much effect should a result between teams like Maine and Northeastern have on who makes the field and who does not?

If it decides the relative strength of your opponents, it should have an effect. You want every team you play to win the rest of your games. It makes those wins "better."

RPI has the same effect. When ever BU, BC, and Northwestern lose, Minnesota's RPI will go down. When the win, it will go up. My rankings work the same way, but it takes into account who they beat/lose to a little more than RPI. If these results didn't mater, than the teams with the eight highest win percentages would make the tournament.

[ARM]

If it decides the relative strength of your opponents, it should have an effect. You want every team you play to win the rest of your games. It makes those wins "better."I understand the math behind the logic. In practice on the women's side, I don't think that a win or loss for Maine vs Northeastern tells us anything about the relative strength of NCAA-quality teams that played and defeated either. Ideally, I'd like to see the system place more weight on games between teams at or near the top 8 than those at or near the bottom 8. It might matter less in men's hockey where the ideal of "on any given Saturday ..." is closer to reality. The problem in women's hockey is that the "huge" upsets of teams > 20 places apart in the rankings are so rare that most can remember the Vermont over Connecticut type results from year to year.

RPI has the same effect. When ever BU, BC, and Northwestern lose, Minnesota's RPI will go down. When the win, it will go up. My rankings work the same way, but it takes into account who they beat/lose to a little more than RPI. If these results didn't mater, than the teams with the eight highest win percentages would make the tournament.First, unless we are going to discuss basketball, let's leave the Wildcats from Evanston out of this. ;) I think RPI is buffered to a greater extent. In any case, the real mistake for a team like Minnesota is scheduling a team like Northeastern in the first place. They will have no impact in the TUC. They may hurt in COP even if Minnesota wins, because it adds 1-0 to their comparison while another team might be able to count 2-0 or 3-0. The slight gain in RPI (or Rutter, if that were used) due to a win pales in comparison to the huge hit that will occur in the case of a loss. If the goal of the formula is "to provide motivation for NCAA-hopeful teams to never schedule a nonconference opponent in the bottom third of the rankings", then I'd say it meets the objective. The bottom line for the Gophers is that they put together a rather weak nonconference schedule, didn't win out, and can only blame themselves.