Friday, January 28, 2011

NHL Goalie Pay & Performance at the All-Star Break

In the Hockey News, there is a chart of each team's top two goalies pay and below the caption asks the following question: "If goaltending is the most important position in hockey, why do so many of the NHL's bottom-feeders slot up top with their blue ice budgets?".

That is a great question, so I thought that I would use the data on the total salaries each team's top two goalies and compare that with each team's performance, first to see if the idea is on target and to then try to explain these counter-intuitive results.

In order to figure this out, I collected each NHL team's standings data (Games Played (GP), Wins (W), Losses (L), OT, Goals For (GF), Goals Against (GA), Standings Points (PTS)) from nhl.com as of today (January 28th, 2011). Then I included the salary data from the Hockey News (January 31, 2011 edition p. 10) and ran a correlation between standings points and goalie salary. The correlation is -0.2777. Hence the writer is on target saying the teams with higher salaried top two goalies are negatively correlated with standings points at the 2010-2011 NHL All-Star break. As I tell my undergraduate students repeated, correlation does not imply causation. We have to work harder to draw conclusions (i.e. statistical inference) than just calculating the correlation between each NHL team's standings points and the combined salary of the top two NHL goalies on each team. To that I now turn.

In order to make a conclusion using statistical techniques, I decided to employ a linear regression on the data that I collected. The regression that I ran was standings points are a function or are determined by Goals For, Goals Against and Games Played. The result was that each variable is statistically significant (i.e. statistically is different from zero at least at a 95% confidence level) and each variable was of the expected sign. The estimated coefficient on Goals For and Games Played were positive (i.e. more goals for and more games played result in more standings points) and the estimated coefficient on Goals Against was negative (i. e. more goals against results in fewer standings points). Yet, when I then included goalie salary, the estimated coefficient was statistically insignificant (i. e. statistically likely to be zero or not relevant). So we can conclude that salary is not a factor in determining standings points up to the All-Star Break this year. Even if we adjust standings points for the number of games played, there still is no statistically significant relationship between team top two goalie salary and team performance.

OK, so there does not seem to be a statistical link between team top two goalie pay and team performance. What about goals against and salary? Shouldn't teams with higher goalie salaries have lower Goals Against? I would think so, but that is not what I find. I find that the estimated coefficient is negative and statistically insignificant - again meaning that there is no statistical relationship between Goals Against and NHL goalie payroll. Nor do I find a statistical relationship between Wins, Losses and NHL goalie payroll.

All of these findings are consistent with the paper Dave Berri and I published in the Journal of Sports Economics last year. We find that goalie salary and goalie performance are not statistically related. Why? Mainly because goalie performance is highly variable from period to period, game to game and season to season.

This is not a dig at NHL goalies. I think that it is really difficult to find an individual who can game after game consistently stop the puck from entering the net at the NHL level, because doing so against the level of NHL talent is so good, not because NHL goalies are not good. What makes people conclude the opposite is that they focus on those goalies (such as Martin Brodeur or Dominic Hasek) who for many years perform at an above average level, they they believe that all goalies can do so. If we look at Brodeur's stats this year, we see that he is (using our NHL goalie measure of Wins Above Average) is only better than Brian Elliott, Dan Ellis and Nikolai Khabibulin.

That leaves us with the conclusion, that if goalies performance is highly variable, NHL GM's should spend more of their capped payroll on other positions and hope for the best at the goalie position.

Thursday, January 27, 2011

NHL Goalie Evaluation at the All-Star Break

OK, my interests are not exclusively about NCAA FBS football. I have done a number of research projects on other sports, and here I turn away from NCAA FBS football, and turn to NHL goalie evaluation.

My good friend and co-author David Berri and I published a paper in the Journal of Sports Economics in 2010 on how GM's evaluate NHL goalies. Basically what we found was that GM's are good at evaluating current player performance by using Vezina voting for the best NHL goalie and using past player performance in setting NHL salaries. Unfortunately, GM's on the whole are not so good at setting salaries in relation to the goalies future performance.

I have used this model to evaluate NHL goalies at the end of the 2009-2010 regular season, and have decided to pick up with this evaluation again. So to be clear, here is how we measured NHL goalie performance. WAA is our measure of NHL goalie productivity and is based on the absolute value (since a goal against has a negative effect on team wins) of the marginal value of a goal against divided by two (since each win is worth two standings points) times the number of shots on goal that goalie faces times the difference in the save percentage of the goalie and the average save percentage of all goalies for that season (or in this case up to the trading deadline).

So here are the results for each goalie. The goalie data comes from nhl.com. Hopefully, there is no surprise as to the best goalie in the league (Tim Thomas), who is having a sensational season up the All-Star break. Some of the other results (both positive and negative) were surprising to me.

Player
Team
WAA
Tim Thomas
BOS
4.754
Jonas Hiller
ANA
2.736
Ondrej Pavelec
ATL
2.358
Pekka Rinne
NSH
2.099
Henrik Lundqvist
NYR
1.985
Roberto Luongo
VAN
1.875
Marc-Andre Fleury
PIT
1.796
Tomas Vokoun
FLA
1.714
Carey Price
MTL
1.591
Cam Ward
CAR
1.570
Niklas Backstrom
MIN
1.365
Jonathan Quick
LAK
1.142
Sergei Bobrovsky
PHI
1.103
Semyon Varlamov
WSH
1.079
Kari Lehtonen
DAL
0.969
Ilya Bryzgalov
PHX
0.968
Dwayne Roloson
NYI, TBL
0.832
Corey Crawford
CHI
0.824
Tuukka Rask
BOS
0.788
Brent Johnson
PIT
0.689
James Reimer
TOR
0.631
Cory Schneider
VAN
0.618
Scott Clemmensen
FLA
0.567
Ryan Miller
BUF
0.485
Andrew Raycroft
DAL
0.457
Devan Dubnyk
EDM
0.424
Cedrick Desjardins
TBL
0.412
Brian Boucher
PHI
0.384
Martin Biron
NYR
0.377
Kevin Poulin
NYI
0.359
Anders Lindback
NSH
0.334
Alex Auld
MTL
0.330
Anton Khudobin
MIN
0.311
Martin Gerber
EDM
0.271
Jose Theodore
MIN
0.243
Antti Niemi
SJS
0.238
Jason LaBarbera
PHX
0.225
Michal Neuvirth
WSH
0.169
Matt Climie
PHX
0.050
Richard Bachman
DAL
0.040
Braden Holtby
WSH
-0.023
Pascal Leclaire
OTT
-0.056
Curtis McElhinney
ANA
-0.065
Joey MacDonald
DET
-0.072
Jacob Markstrom
FLA
-0.081
Mark Dekanich
NSH
-0.081
Michael Leighton
PHI
-0.081
Robin Lehner
OTT
-0.101
Mike McKenna
NJD
-0.102
Jhonas Enroth
BUF
-0.173
Peter Mannino
ATL
-0.192
Chris Osgood
DET
-0.216
Jaroslav Halak
STL
-0.218
Nathan Lawson
NYI
-0.244
Henrik Karlsson
CGY
-0.306
Mike Brodeur
OTT
-0.353
Patrick Lalime
BUF
-0.355
Jimmy Howard
DET
-0.409
Mathieu Garon
CBJ
-0.442
Antero Niittymaki
SJS
-0.522
Craig Anderson
COL
-0.628
Johan Hedberg
NJD
-0.661
Jonathan Bernier
LAK
-0.680
Marty Turco
CHI
-0.856
Jean-Sebastien Giguere
TOR
-0.865
Miikka Kiprusoff
CGY
-0.985
Justin Peters
CAR
-1.052
Chris Mason
ATL
-1.086
Steve Mason
CBJ
-1.111
Peter Budaj
COL
-1.149
Ty Conklin
STL
-1.173
Mike Smith
TBL
-1.184
Jonas Gustavsson
TOR
-1.299
Rick DiPietro
NYI
-1.318
Martin Brodeur
NJD
-1.372
Brian Elliott
OTT
-1.496
Dan Ellis
TBL
-1.662
Nikolai Khabibulin
EDM
-1.928

Wednesday, January 26, 2011

Final NCAA FBS Top 25

OK, I am rather late with the top 25. I was on vacation and then the semester started, and this got dropped to the bottom of the to do list. Anyway, here is the production model's top 25 for the NCAA FBS for the 2010 year.

Rank
School
1
Boise State
2
TCU
3
Oregon
4
Alabama
5
Ohio State
6
Northern Illinois
7
Stanford
8
Oklahoma State
9
Wisconsin
10
Nevada
11
Hawai'i
12
Oklahoma
13
Auburn
14
UCF
15
Iowa
16
West Virginia
17
Arkansas
18
Air Force
19
Virginia Tech
20
Florida State
21
Tulsa
22
Missouri
23
Pittsburgh
24
Nebraska
25
Miami (Florida)

Later on this year I plan on looking at competitive balance in the NCAA FBS and also post the entire 120 NCAA FBS schools ranking. Who knows, maybe I will also find some time to do some other things as well.

Thursday, January 20, 2011

2010-2011 Bowl Results

Well the model went 24-11 this college bowl season. Not bad, given there were some significant upsets (in my opinion - notably North Carolina State defeating West Virginia and Washington defeating Nebraska). Hopefully, tomorrow I will have the final production numbers done and put the production model's final Top 25 on the blog.

Monday, January 10, 2011

BCS Title Game

Finally the BCS title game as chosen by the BCS polls and computer models pitting the #3 most productive team in the nation - Oregon - against the #12 most productive team in the nation - Auburn.

BCS Title Game: Auburn (13-0) vs. Oregon (12-0)
Auburn (#12) overall in production has an offense that is ranked #7 and a defense that is ranked #57. Auburn also has a SOS of 49.85, with their best win over #4 Alabama in terms of overall production.

Oregon (#3) overall in production has an offense that is ranked #1 and a defense that is ranked #8. Oregon also has a SOS of 75.00, with their best win over Stanford (ranked #11 in overall production).

What will be interesting to see is how well Oregon's offense and Auburn's defense play. To me this is the most interesting match up on the field. I think whichever side of this match-up has the best game will determine the outcome, which may not be the conventional wisdom. Thus based only on the production model Oregon is favored over Auburn. So let's stay with the model and choose Oregon.

Sunday, January 9, 2011

Kraft Fight Hunger Bowl

Let's look at the tale of the tape in tonight's Kraft Fight Hunger Bowl.

Kraft Fight Hunger Bowl: Boston College (7-5) vs. Nevada (12-1)
Boston College (#61) overall in production has an offense that is ranked #91 and a defense that is ranked #13. Boston College also has a 60.75, with their best win against #35 Clemson and BC's worst loss against #50 Notre Dame.

Nevada (#7) overall in production has an offense that is ranked #6 and a defense that is ranked #52. Nevada also has a SOS of 80.23, with their best win against #1 Boise State and their only loss coming to #6 Hawai'i.

Thus based only on the production model Nevada is favored over Boston College, so I expect that Nevada will be the winner of the Kraft Fight Hunger Bowl this year.

Saturday, January 8, 2011

BBVA Compass Bowl

Here is the numbers for tonight's BBVA Compass Bowl.

BBVA Compass Bowl: Kentucky (6-6) vs. Pittsburgh (7-5)
Kentucky (#38) overall in production has an offense that is ranked #29 and a defense that is ranked #58. Kentucky also has a SOS of 64.67, with their best win against #30 Louisville and their worst loss against #87 Mississippi.

Pittsburgh (#26) overall in production has an offense that is ranked #51 and a defense that is ranked #9. Pittsburgh also has a SOS of 55.58, with Pitt's best win over #30 Louisville and Pitt's worst loss against #50 Notre Dame.

Thus based only on the production model Pittsburgh is favored over Kentucky, so the model predicts that Pittsburgh will win this year's BBVA Compass Bowl.

Friday, January 7, 2011

Cotton Bowl

Here is the tale of the tape for tonight's Cotton Bowl.

Cotton Bowl: Texas A&M (9-3) vs. LSU (10-2)
Texas A&M (#31) overall in production has an offense that is ranked #33 and a defense that is ranked #47. Texas A&M also has a SOS of 55.03, with their best win over #15 Oklahoma and thier worst loss against #25 Missouri.

LSU (#36) overall in production has an offense that is ranked #62 and a defense that is ranked #17. LSU also has a SOS of 53.92 with LSU's best win over #4 Alabama and LSU's worst loss against #18 Arkansas.

Thus based only on the production model Texas A&M is favored over LSU, so the model predicts that Texas A&M will win this year's Cotton Bowl.

Thursday, January 6, 2011

Go Daddy Bowl

Update: As of now, the NCAA FBS production model is 21-9. Here are the numbers for the Go Daddy Bowl.

Go Daddy Bowl: Miami (OH) (9-4) vs. Middle Tennessee (6-6)
Miami (OH) (#55) overall in production has an offense that is ranked #60 and a defense that is ranked #48. Miami (OH) also has a SOS of 70.85, with their best win against #9 (in terms of production) Northern Illinois and their worst loss against #73 (in terms of production) Cincinnati.

Middle Tennessee (#85) overall in production has an offense that is ranked #87 and a defense that is ranked #74. Middle Tennessee also has a SOS of 89.83 with their best win against #71 (in terms of production) Florida International and their worst loss against #119 Memphis - ouch! Middle Tennessee may be the worst team playing in a bowl game this year.

Thus based only on the production model Miami (OH) is favored over Middle Tennessee. Not a game that is high on my list of priorities tonight.

Tuesday, January 4, 2011

Sugar Bowl

Here is the tale of the tape for tonights Sugar Bowl.

Sugar Bowl: Ohio State (11-1) vs. Arkansas (10-2)
Ohio State (#5) overall in production has an offense that is ranked #16 and a defense that is ranked #5. Ohio State also has a SOS of 63.42 with OSU's best win over #17 Iowa and worse (and only) loss over #8 Wisconsin.

Arkansas (#18) overall in production has an offense that is ranked #12 and a defense that is ranked #42. Arkansas also has a SOS of 56.25, with Arkansas' best win over #27 Georgia and worst loss over #12 Auburn.

Thus based only on the production model Ohio State is favored over Arkansas, so the model predicts that Ohio State will defeat Arkansas in this year's Sugar Bowl.

Monday, January 3, 2011

Orange Bowl

UPDATE: The model went 6-0 on Saturday, giving the overall tally 19 correct and 9 incorrect. Not bad given the amount of what I would consider upset's during this years bowl season. But enought of that, let's look at this year's Orange Bowl which puts two team's equally matched against each other, with the best of the ACC against the second best PAC 10 team. Here's how they stack up.

Orange Bowl: Virginia Tech (11-2) vs. Stanford (11-1)
Virginia Tech (#16) overall in production has an offense that is ranked #14 and a defense that is ranked #33. Virginia Tech also has a SOS of 64.69, with VT's best win against Florida State (#22 in terms of production) and their worst loss was against FCS James Madison - which for all FCS schools I rank 121.

Stanford (#11) overall in production has an offense that is ranked #8 and a defense that is ranked #26. Stanford also has a SOS of 71.75 with Stanford's best win coming against Arizona (#21 in terms of overall production) and their worst (and only loss) against (#3 Oregon in terms of overall production).

Thus based only on the production model Stanford is favored over Virginia Tech. So the model predicts that Stanford will be the winner of this year's Orange Bowl.

Saturday, January 1, 2011

January 1st Bowls

Happy New Year!

The model went 2-2 yesterday, bring the running total to 13-9 for the bowl season. So let's take a look at today's games.

Ticket City Bowl: Northwestern (7-5) vs. Texas Tech (7-5)
Northwestern (#76) overall in production has an offense that is ranked #80 and a defense that is ranked #58. Northwestern also has a SOS of 71.00, with their best win over #17 (based on the production model) Iowa and their worst loss against #97 (based on the production model) Purdue.

Texas Tech (#75) overall in production has an offense that is ranked #40 and a defense that is ranked #97. Texas Tech also has a SOS of 60.08, with their best win over #25 (based on the production model) Missouri and their worst loss against #82 (based on the production model) Iowa State.

Thus based only on the production model Texas Tech is slightly more productive than Northwestern, thus the model predicts that Texas Tech will win the Ticket City Bowl.


Outback Bowl: Penn State (7-5) vs. Florida (7-5)
Penn State (#65) overall in production has an offense that is ranked #67 and a defense that is ranked #53. Penn State also has a SOS of 51.92, with their best win over #46 (in overall productivity) Temple and their worst loss against #48 (in overall productivity) Illinois.

Florida (#33) overall in production has an offense that is ranked #66 and a defense that is ranked #11. Florida also has a SOS of 51.75, with their best win over #27 (in overall productivity) Georgia and their worst loss against #41 (in overall productivity) Mississippi State.

Thus based only on the production model Florida is more productive than Penn State, thus the production model predicts that Florida will win the Outback Bowl.


Capital One Bowl: Michigan State (11-1) vs. Alabama (9-3)
Michigan State (#24) overall in production has an offense that is ranked #25 and a defense that is ranked #36. Michigan State also has a SOS of 65.08, with their best win over #8 (based on the production model) Wisconsin and their worst (and only) loss against #17 (based on the production model) Iowa.

Alabama (#4) overall in production has an offense that is ranked #15? and a defense that is ranked #2. Alabama also has a SOS of 62.33, with their best win over #18 (based on the production model) Arkansas and their worst loss against #39 (based on the production model) South Carolina.

Thus based only on the production model Alabama is favored over Michigan State, thus the model predicts that Alabama will win the Capital One Bowl.


Gator Bowl: Michigan (7-5) vs. Mississippi State (8-4)
Michigan (#51) overall in production has an offense that is ranked #22 and a defense that is ranked #81. Michigan also has a SOS of 56.75, with their best win over #44 (from the production model) Connecticut and their worst loss against #65 (from the production model) Penn State.

Mississippi State (#41) overall in production has an offense that is ranked #43 and a defense that is ranked #43. Mississippi State also has a SOS of 52.00, with their best win over #27 (from the production model) Georgia and their worst loss against #36 (from the production model) LSU.

Thus based only on the production model Mississippi State is favored over Michigan, thus the model predicts Mississippi State as the winner of the Gator Bowl.


Rose Bowl: TCU (12-0) vs. Wisconsin (11-1)
TCU (#2) overall in production has an offense that is ranked #5 and a defense that is ranked #4. TCU also has a SOS of 76.00, with their best win over #23 (in terms of overall production) Air Force.

Wisconsin (#8) overall in production has an offense that is ranked #10 and a defense that is ranked #15. Wisconsin also has a SOS of 72.83, with their best win over #5 (in terms of overall production) Ohio State and their worst (and only) loss against #24 (in terms of overall production) Michigan State.

Thus based only on the production model TCU is favored over Wisconsin, thus the model predicts that TCU will win the Rose Bowl.


Fiesta Bowl: Oklahoma (11-2) vs. Connecticut (8-4)
Oklahoma (#15) overall in production has an offense that is ranked #9 and a defense that is ranked #39. Oklahoma also has a SOS of 50.38, with their best win over #10 (from the production model) Oklahoma State and their worst loss against #31 (from the production model) Texas A&M.

Connecticut (#44) overall in production has an offense that is ranked #65? and a defense that is ranked #27. Connecticut also has a SOS of 64.92, with their best win over #13 (in terms of overall productivity) West Virginia and their worst loss against #96 (in terms of overall productivity) Rutgers.

Thus based only on the production model Oklahoma is favored over Connecticut, so the model predicts that Oklahoma will win the Fiesta Bowl.