Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

[Bhikukhu]

I've admittedly only been to one Frozen Four out east, at QU, but I don't think that was sold out, and I think UNH is a bigger rink.

There are definitely others that can give you a more informed answer than me though!

EDIT: QU holds 3,200 and UNH holds 7,500... So there you go haha -- so there's pretty much no way this sells out

Awesome! Thank you much!!!

[TonyTheTiger20]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

BU is 11th in RPI but 15th!! in PWR. That's nuts.

That's what happens when you're 0-6 against TUCs.

[ARM]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

BU is 11th in RPI but 15th!! in PWR. That's nuts.

That's what happens when you're 0-6 against TUCs.

That's good for BC, though, as playing an opponent with a strong RPI helps, while an opponent's TUC, COp, and H2H are meaningless.

[ne7minder]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

Doesn't it sell out every year? I'd like to be able to pay like 35 per ticket for the whole weekend or whatever it usually is instead of the 50 per ticket I had to pay at Quinnipiac 2 years ago just to watch the championship game... Though it was well worth it to see my school beat the monster that was Minnesota

I have only ever seen the weekend sell out when the games are in Minnesota. I have no idea how much seating is at this years rink but don't imagine it will be a problem. OTOH, if you are planning on it anyway why not by the weekend pass when they go on sale & then not worry about it?

[TonyTheTiger20]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

That's good for BC, though, as playing an opponent with a strong RPI helps, while an opponent's TUC, COp, and H2H are meaningless.

True, although until we lose a game our RPI is going to be entirely made up of whoever the one best team we played is -- with "best" being a bizarre weighting of the best OppOppWin% and best OppWin%

Right now that team is Cornell for those of you who care, followed closely by Duluth, based mostly on those teams' exceptional OppOppWin%.

RPI is weird.

[ARM]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

True, although until we lose or tie a game our RPI is going to be entirely made up of whoever the one best team we played is -- with "best" being a bizarre weighting of the best OppOppWin% and best OppWin%

FYP.

RPI isn't so much weird as just plain unfit for anything meaningful. Just about any other measurement system would work better, so of course, RPI is what the NCAA uses.

[pokechecker]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

Percentage of opponent's SOG that are goals scored:

and the difference, ranked from high to low:
Wisconsin 11.03% 2.45% 8.58%
Minnesota 16.50% 7.95% 8.55%
Boston College 13.43% 5.21% 8.22%
Harvard 9.91% 4.15% 5.76%
Quinnipiac 10.15% 6.27% 3.88%
Northeastern 13.42% 10.00% 3.42%
Robert Morris 10.70% 7.77% 2.93%
Princeton 9.48% 6.65% 2.83%

Connecticut 8.22% 5.69% 2.53%
Clarkson 8.47% 6.58% 1.89%
Bemidji State 8.17% 6.69% 1.48%
Colgate 9.16% 7.87% 1.29%
St. Cloud State 11.52% 10.77% 0.75%
Lindenwood 8.10% 7.86% 0.24%

Maine 8.03% 8.11% -0.08%
Penn State 7.18% 7.38% -0.20%
Cornell 10.17% 10.51% -0.34%
Dartmouth 7.67% 8.01% -0.34%
Rensselaer 7.18% 7.73% -0.55%
Boston University 9.73% 10.48% -0.75%
St. Lawrence 9.35% 11.30% -1.95%
Union 5.08% 7.17% -2.09%
North Dakota 8.17% 10.66% -2.49%
Merrimack 9.06% 11.61% -2.55%
Providence 9.75% 12.33% -2.58%
Minnesota Duluth 7.26% 9.87% -2.61%
Syracuse 9.27% 12.15% -2.88%
Mercyhurst 8.73% 11.92% -3.19%
Yale 10.45% 13.70% -3.25%
Ohio State 9.07% 13.03% -3.96%
Minnesota State 6.78% 11.46% -4.68%
RIT 6.73% 11.93% -5.20%
Brown 5.85% 11.18% -5.33%
Vermont 6.65% 12.15% -5.50%
New Hampshire 6.81% 13.38% -6.57%

[TonyTheTiger20]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

FYP.

Ha, well, a tie will feel like a loss, so there you go

[TonyTheTiger20]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

Side note!

According to the transitive property, BC would beat Minnesota by 6 goals

[mnpokecheck]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

Side note!

According to the transitive property, BC would beat Minnesota by 6 goals

At least 6...BC should also have at least a couple of transitive National Championships.....raise the banners.

[bc6696]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

I saw the comparison between Desbien and Burt in the "Wednesday" column and began to wonder how much stronger the schedule was for goalies at the top of the womens hockey statistically.
What I used to come up with my comparison was the following: Minutes Played x OPP RPI = Game Rating I then took the sum of all Game Ratings/TMP to come up with each goalie's avg OPP RPI.
I did it this way to take into account games where the goalie did not play a full game, in order to get a more accurate picture of their real difficulty of play.

I did one last step in the far right column (I sure this will stir some complaints about my process), I took each goalie's (SV% x AVG OPP RPI)/AVG RPI OF ALL TEAMS FOR ALL GAMES. I don't think that reflects exactly what I was trying to accomplish - see how goalies would fare if they all played the same schedule - but I think it points out how some goalies may be over/under rated based on their SOS.

I am fully aware there are many flaws in this: Home/Away, Injuries, Strength of the team in front of each goalie, ect. This was just for FUN!

With all that said I did found some of the results surprising

I have the goalies (Top 14 in GAA + a few others) ranked by SOS

Let me know your thoughts - go easy on the criticism, my skin isn't as thick as TTT


Goalie OPP RPI/GP GAA SV % Rel SV %
Shelby*Amsley-Benzie 0.5352 1.6 0.929 0.992
Emerance*Maschmeyer 0.5251 1.71 0.944 0.989
Amanda*Leveille 0.5232 1.33 0.934 0.975
Elaine*Chuli 0.5219 2.32 0.940 0.979
Brittni*Mowat 0.5174 1.62 0.941 0.972
Kimberly*Newell 0.5137 1.52 0.946 0.970
Shea*Tiley 0.5098 1.37 0.931 0.947
Katie*Burt 0.5023 1.14 0.946 0.948
Brittany*Bugalski 0.5003 2.04 0.923 0.922
Sydney*Rossman 0.4975 0.95 0.947 0.940
Ann-Renée*Desbiens 0.4958 0.61 0.966 0.956
Lovisa*Selander 0.4930 2.04 0.936 0.921
Melissa*Black 0.4826 2.61 0.934 0.900
Meghann*Treacy 0.4816 2.50 0.921 0.885
Ashlynne*Rando 0.4810 2.11 0.923 0.886
Celine*Whitlinger 0.4777 1.55 0.945 0.901
Sarah*McDonnell 0.4775 1.55 0.926 0.883
Jessica*Dodds 0.4722 2.24 0.925 0.872
Nicole*Hensley 0.4714 2.36 0.927 0.872
Jetta*Rackleff 0.4635 2.26 0.935 0.865

*Sorry about the chart - I need to learn how to carry the formatting over to this forum
**AVG RPI FOR ALL 35 DI Teams = .50098 (Through 2/15/2016)

[ARM]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

Interesting attempt. My quick thoughts:

a) If you want to rank strength of teams, don't start with RPI to do so. It is one of the poorer tools available. For this exercise, something like Robin Lock's WCHODR, especially the offensive component, might makes sense:
http://it.stlawu.edu/~chodr/wchodr/current.html

b) For judging goaltenders, is the opponents' overall strength that important? IMO, the strength of the opponents' offense is a big factor. Rather than starting with RPI, would it be better to factor in the opponents' scoring average. If a goalie plays against a team that normally averages 1.5 goals and shuts them out, that may not be as big a deal as holding a team that averages 5.0 goals to 1 goal. For example, BU is a slightly above average team overall, but its offense ranks fifth in the country, and that says more about the Terriers' ability to produce goals against a random goaltender.

c) It's possible that any method used may not really tell the full story about Desbiens. Glance at her season overall; she doesn't give up goals. She's allowed more than one goal three times: 3 goals once, 2 goals twice. The 3-goal game versus UND was the only time all season where an opponent exceeded its scoring average against her; everyone else was below its scoring average. For the season as a whole, UND scored 5 goals in four games, so for the season, UND was also well below its usual average. That may be a better measure of goaltenders, their ability to hold an opponent below its norm.

[jb20]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

Interesting attempt. My quick thoughts:

a) If you want to rank strength of teams, don't start with RPI to do so. It is one of the poorer tools available. For this exercise, something like Robin Lock's WCHODR, especially the offensive component, might makes sense:
http://it.stlawu.edu/~chodr/wchodr/current.html

b) For judging goaltenders, is the opponents' overall strength that important? IMO, the strength of the opponents' offense is a big factor. Rather than starting with RPI, would it be better to factor in the opponents' scoring average. If a goalie plays against a team that normally averages 1.5 goals and shuts them out, that may not be as big a deal as holding a team that averages 5.0 goals to 1 goal. For example, BU is a slightly above average team overall, but its offense ranks fifth in the country, and that says more about the Terriers' ability to produce goals against a random goaltender.

c) It's possible that any method used may not really tell the full story about Desbiens. Glance at her season overall; she doesn't give up goals. She's allowed more than one goal three times: 3 goals once, 2 goals twice. The 3-goal game versus UND was the only time all season where an opponent exceeded its scoring average against her; everyone else was below its scoring average. For the season as a whole, UND scored 5 goals in four games, so for the season, UND was also well below its usual average. That may be a better measure of goaltenders, their ability to hold an opponent below its norm.

Very interesting numbers. It is almost impossible to compare goalies due to the disparity of the league. Some goalies are averaging over 35 shots per game and posting 940 save percentages. Lots of goalies are facing 10 -15 good scoring chances per game. That is more than some of the goalies shots per game. Keep the numbers coming very interesting.
.

[TonyTheTiger20]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

a) If you want to rank strength of teams, don't start with RPI to do so. It is one of the poorer tools available. For this exercise, something like Robin Lock's WCHODR, especially the offensive component, might makes sense:
http://it.stlawu.edu/~chodr/wchodr/current.html

That's a good point, and yeah I think WCHODR's offense component is probably a really good place to go.

I like 6696's attempt to remove games where the goalie didn't play in figuring out goalie's SOS. Interested particularly with how that affect's Burt because Switaj's games were all against the bottom of the barrel.

[bc6696]

Re: Fun With Numbers: 2016 Pairwise Predictor, What-Ifs, and Other Goodies

Interesting attempt. My quick thoughts:

a) If you want to rank strength of teams, don't start with RPI to do so. It is one of the poorer tools available. For this exercise, something like Robin Lock's WCHODR, especially the offensive component, might makes sense:
http://it.stlawu.edu/~chodr/wchodr/current.html

b) For judging goaltenders, is the opponents' overall strength that important? IMO, the strength of the opponents' offense is a big factor. Rather than starting with RPI, would it be better to factor in the opponents' scoring average. If a goalie plays against a team that normally averages 1.5 goals and shuts them out, that may not be as big a deal as holding a team that averages 5.0 goals to 1 goal. For example, BU is a slightly above average team overall, but its offense ranks fifth in the country, and that says more about the Terriers' ability to produce goals against a random goaltender.

c) It's possible that any method used may not really tell the full story about Desbiens. Glance at her season overall; she doesn't give up goals. She's allowed more than one goal three times: 3 goals once, 2 goals twice. The 3-goal game versus UND was the only time all season where an opponent exceeded its scoring average against her; everyone else was below its scoring average. For the season as a whole, UND scored 5 goals in four games, so for the season, UND was also well below its usual average. That may be a better measure of goaltenders, their ability to hold an opponent below its norm.

I like your suggestions, my only reservation to applying the opponents offensive prowess is - What if they play a weak schedule? Won't that inflate their offensive stats and rank?
Ex: Robert Morris - Ranked ninth nationally in offense, but if you look at their schedule they play one of the weakest. 3 weekends against top ten teams-1.6 GFA, rest of season 3.15 GFA. Are they a good offensive team or do they play an easy schedule?

I think I need a much bigger algorithm to figure this out!!