The View From Wisconsin
Just a random set of rants from a Sports Fan from Wisconsin.
Saturday, October 01, 2005
Hockey Win Shares
Now that hockey's back, I've been playing around a bit with the Goals Created formula. I sat down to analyze the 25 best players in NHL history, using goals created, on a season-by-season basis.
Mind you, I've always known that the GC formula has issues in seasons prior to 1967. Two key components of the formula – power play goals and plus/minus rating – were not kept (or were kept on a spotty basis) before the NHL expanded. I knew going in that I'd have to adjust things a bit, but it wasn't until I started tackling the problem that I realized I needed to make an adjustment to the formula itself.
For those of you who haven't bothered to look back through this blog for my first post about GC (or haven't checked out my webpage explaining it), Goals Created measures three areas: goals that a player directly participates in scoring (goals and assists), goals that a player causes to be scored against his team by his actions on the ice (penalty minutes times a league constant), and goals that the player indirectly participates in scoring (plus/minus rating). The existing league constant is considered to be league power play goals, divided by league-wide penalties in minutes; however, I made the decision to change this to league goals scored per penalty minute. It was easier to figure, and it could be determined for all seasons going back to 1917 in the NHL (and in most other leagues going back to 1922).
Plus/minus rating, however, wasn't as easy. I somewhat hoped that I could go back to a certain point where teams began to use more than six to eight players on their rosters, but I realized it was impossible to do that. So, I replaced plus/minus in pre-1967 seasons with "figured goal differential". I took the goals scored by the player's team, multiplied that by the number of games the player played for that team, and divided that total by the total number of games played by the team. It did result in some large goal differential numbers, but it also adjusts for the fewer games played overall in past seasons. For example, Joe Malone's stellar 1917-18 season was 56.2 goals created – a pretty good number, even by today's standards.
However, I still wanted to try and adjust the numbers so you could attempt to judge players across eras – like whether or not Rocket Richard was as good as Gretzky or Lemieux. So, I crunched numbers and decided to try to change the GC totals by multiplying them by the league goals per game total for each season, and then divide that total by a constant factor – 3.21 goals per game, the average over the entirety of organized major professional hockey from 1909 to 1924. Let's just say that this adjustment didn't leave me satisfied, though. My thinking was this: I'm essentially penalizing a guy for scoring a lot of goals in a season where goals were fast and frequent, while rewarding someone in an era where goaltenders were stopping everything who managed to eke out a few points.
I then had a revelation: how about using the GC total as the numerator in the Pythagorean Wins formula? Or, better, yet, Goals Created per game? Then, you could use league goals per game as the denominator, and figure the player's winning percentage, based on his goals created. Better still, you could multiply the winning percentage by the number of games he played, and get a win total. At first, I thought to call this resulting formula "offensive wins", but when I plugged in the numbers, I though elsewise. I mean, is it really appropriate to call something "offensive" when the person who holds the best single-season mark in major league history is a defenseman?
That was what sold me on the name "Win Shares." It means, to a certain extent, that the given player's performance contributed to a given number of shares of wins, among his teammates. Maurice Richard had 28.0 win shares in 1955-56, while his brother Henri had 21.2 – both for the Canadiens. It's not much wonder why that team won so many games in the NHL that season. Phil Esposito had 32.7 win shares for the Bruins in 1970-71 – a total that would be considered astronomical during any time period – but was overshadowed by Bobby Orr's record-setting 38.8 win shares. Either way, Boston's dispatching of the Blues in the finals that year was far from surprising.
When I totalled up Win Shares from 2003-04, and compared them with actual points in the standings, they correlated very well – somewhere at the .93 level, which is a nearly direct relationship. It also proved out that you could almost gauge a player's performance, over an 80-game season, based on his Win Shares total. 10 to 14 win shares would be a solid season over 80 games. 15-19 would be all-star worthy; and a definite Norris candidacy for a defenseman. 20-24 win shares would start to get you some nods for the Hart Trophy; 25-29 would make you a serious Hart candidate. 30-plus seasons are rare; those are the years other teams just throw up their hands in surrender after they play against this guy. As far as I know, no one has ever achieved a 40 win share season.
By the bye: Win Shares says that the best player in hockey history is (surprise surprise) Gordie Howe, who amassed 385.7 win shares in his career. Gretzky is the only other player with more than 300 win shares, totalling out at 356.6 in his career. The next highest total surprised me, in a way; but after consideration it made sense: Jean Beliveau, with 280.4 win shares. Other 200-plus win share totals are Bobby Hull (268.9), Phil Esposito (232.7), Henri Richard (220.7) and Mario Lemieux (214.8). Guy Lafleur (198.7) and Rocket Richard (195.1) just barely miss the 200 mark.
Another thing with Win Shares: it appears that a player becomes a qualified Hall of Fame candidate when he hits the 100-win share mark. Jarome Iginla, through 2004, has a career 48.1 win share total, while Paul Kariya has 92.7 win shares. Mark Messier's total is 159.0, though he only topped 10 or more win shares six times during his career (four times with Edmonton, twice with the Rangers).
Mind you, I've always known that the GC formula has issues in seasons prior to 1967. Two key components of the formula – power play goals and plus/minus rating – were not kept (or were kept on a spotty basis) before the NHL expanded. I knew going in that I'd have to adjust things a bit, but it wasn't until I started tackling the problem that I realized I needed to make an adjustment to the formula itself.
For those of you who haven't bothered to look back through this blog for my first post about GC (or haven't checked out my webpage explaining it), Goals Created measures three areas: goals that a player directly participates in scoring (goals and assists), goals that a player causes to be scored against his team by his actions on the ice (penalty minutes times a league constant), and goals that the player indirectly participates in scoring (plus/minus rating). The existing league constant is considered to be league power play goals, divided by league-wide penalties in minutes; however, I made the decision to change this to league goals scored per penalty minute. It was easier to figure, and it could be determined for all seasons going back to 1917 in the NHL (and in most other leagues going back to 1922).
Plus/minus rating, however, wasn't as easy. I somewhat hoped that I could go back to a certain point where teams began to use more than six to eight players on their rosters, but I realized it was impossible to do that. So, I replaced plus/minus in pre-1967 seasons with "figured goal differential". I took the goals scored by the player's team, multiplied that by the number of games the player played for that team, and divided that total by the total number of games played by the team. It did result in some large goal differential numbers, but it also adjusts for the fewer games played overall in past seasons. For example, Joe Malone's stellar 1917-18 season was 56.2 goals created – a pretty good number, even by today's standards.
However, I still wanted to try and adjust the numbers so you could attempt to judge players across eras – like whether or not Rocket Richard was as good as Gretzky or Lemieux. So, I crunched numbers and decided to try to change the GC totals by multiplying them by the league goals per game total for each season, and then divide that total by a constant factor – 3.21 goals per game, the average over the entirety of organized major professional hockey from 1909 to 1924. Let's just say that this adjustment didn't leave me satisfied, though. My thinking was this: I'm essentially penalizing a guy for scoring a lot of goals in a season where goals were fast and frequent, while rewarding someone in an era where goaltenders were stopping everything who managed to eke out a few points.
I then had a revelation: how about using the GC total as the numerator in the Pythagorean Wins formula? Or, better, yet, Goals Created per game? Then, you could use league goals per game as the denominator, and figure the player's winning percentage, based on his goals created. Better still, you could multiply the winning percentage by the number of games he played, and get a win total. At first, I thought to call this resulting formula "offensive wins", but when I plugged in the numbers, I though elsewise. I mean, is it really appropriate to call something "offensive" when the person who holds the best single-season mark in major league history is a defenseman?
That was what sold me on the name "Win Shares." It means, to a certain extent, that the given player's performance contributed to a given number of shares of wins, among his teammates. Maurice Richard had 28.0 win shares in 1955-56, while his brother Henri had 21.2 – both for the Canadiens. It's not much wonder why that team won so many games in the NHL that season. Phil Esposito had 32.7 win shares for the Bruins in 1970-71 – a total that would be considered astronomical during any time period – but was overshadowed by Bobby Orr's record-setting 38.8 win shares. Either way, Boston's dispatching of the Blues in the finals that year was far from surprising.
When I totalled up Win Shares from 2003-04, and compared them with actual points in the standings, they correlated very well – somewhere at the .93 level, which is a nearly direct relationship. It also proved out that you could almost gauge a player's performance, over an 80-game season, based on his Win Shares total. 10 to 14 win shares would be a solid season over 80 games. 15-19 would be all-star worthy; and a definite Norris candidacy for a defenseman. 20-24 win shares would start to get you some nods for the Hart Trophy; 25-29 would make you a serious Hart candidate. 30-plus seasons are rare; those are the years other teams just throw up their hands in surrender after they play against this guy. As far as I know, no one has ever achieved a 40 win share season.
By the bye: Win Shares says that the best player in hockey history is (surprise surprise) Gordie Howe, who amassed 385.7 win shares in his career. Gretzky is the only other player with more than 300 win shares, totalling out at 356.6 in his career. The next highest total surprised me, in a way; but after consideration it made sense: Jean Beliveau, with 280.4 win shares. Other 200-plus win share totals are Bobby Hull (268.9), Phil Esposito (232.7), Henri Richard (220.7) and Mario Lemieux (214.8). Guy Lafleur (198.7) and Rocket Richard (195.1) just barely miss the 200 mark.
Another thing with Win Shares: it appears that a player becomes a qualified Hall of Fame candidate when he hits the 100-win share mark. Jarome Iginla, through 2004, has a career 48.1 win share total, while Paul Kariya has 92.7 win shares. Mark Messier's total is 159.0, though he only topped 10 or more win shares six times during his career (four times with Edmonton, twice with the Rangers).
Copyright © 2004, 2005, 2006, 2007, 2008, Joseph Houk. All rights reserved.
