by PuckStopsHere on 08/25/11 at 03:37 PM ET
One of the big achievements in baseball sabermetrics is Bill James win shares system. This is a system that assigns wins to the individual players on baseball teams. It works very well to rate individual players. One goal in hockey sabermetrics is to come up with a similar system. Of course there are differences between hockey and baseball which make building a hockey system more difficult, but that hasn`t stopped Justin Kubatko at hockey-reference.com from trying. I am going to look at his system and its results. I think it is an important goal, but we are not able to do it properly yet. As a result, he has built an approximate system which can give unreasonable results in some cases.
Kubatko makes three conceptual changes from the Bill James system. First in the baseball system, a win is worth 3 win shares. In the hockey system one point in the standings is worth one point share. The hockey version of this is problematic because of point scheme that the NHL uses.
In the NHL, some games are worth two total points and others are worth three. The extra point occurs only if the game is tied at the end of regulation time. This is not something that is caused by any individual player, yet there is an extra point available that will be credited to individual players becomes available. This point is often given out for pretty arbitrary reasons - winning a shootout. I think that it is a failure to try to link up a point share system with the NHL`s point system because it is a poorly defined system. It is probably most accurate to link it to a more logical point system for teams. Perhaps two points for a win, one point for a game that goes to a shootout (win or loss) and no points for loss (including an overtime loss) is more meaningful.
Kubatko also breaks the perfect link between wins and win shares. In baseball, a win is worth exactly three win shares. A team that wins 100 games has exactly 300 win shares among their players. In the hockey system this is approximate. A team with 100 points will have about 100 point shares, but it may be a slightly higher or lower total. This is a necessity in part because of the NHL point system. It isn`t possible to meaningfully have some games worth two point shares and others worth three for a reason so arbitrary as if it is tied at the end of regulation.
The third difference is that Kubatko allows for players to have negative point shares. If a player plays poorly enough to cost his team wins (for example Brian Elliott), he should be given a negative point share total because he cost his team points. In the baseball system, the worst total a player can have is zero win shares regardless of how poorly he plays. This is something problematic with the baseball system that is potentially improved here.
The calculation of point shares is approximate and it becomes more and more approximate as we go back into history where sufficient stats (such as ice time) were not properly recorded. In the future I will discuss the details of these calculations and look at some of its results and the systematic biases that were involved.
Add a Comment
Please limit embedded image or media size to 575 pixels wide.
Most Recent Blog Posts
About The Puck Stops Here
Who am I? A diehard hockey fan.
Why am I blogging? I want to.
Why are you reading it? ???