<< Previous Article Inside The VUKOTA Proj... (10/07) << Previous Column Behind The Net (09/25) Next Column >> Behind The Net (12/09) Next Article >> Illegal Curve (10/08)
 October 7, 2009 Behind The Net A Basic Goaltending Model by Gabriel Desjardins Printer- friendly
 There have been a lot of efforts in the past to quantify the impact of shot quality on a goaltender’s save percentage. For example, Alan Ryder used a distance-based model (pdf), while I used a system based on x-y coordinates. Both systems got a bit bogged down by observation errors made by the scorers at different rinks. While it’s clear that some coaches can affect shot quality with their defensive systems – Jacques Lemaire and Ken Hitchcock spring to mind – the impact is less than what we see in shot quality models. I’d like to set aside shot quality for the moment and just look directly at save percentage. In my mind, there are two main skills embodied in save percentage: even-strength (5v5) and short-handed (4v5). I don’t include two-man advantages in that category – the number of shots faced by goaltenders at 5-on-3 and the expected shooting percentage are so high that it doesn’t resemble hockey anymore. Additionally, goaltenders face so few shots when their teams have the man-advantage that there’s little difference between the best and worst goaltenders. This plot shows the distribution of goaltender save percentage at 5v5 and 4v5 over the last two seasons, 500 shots minimum: I’m going to arbitrarily set replacement level at one-half of a standard deviation below the mean in each category. This puts approximately 60 full- and part-time goaltenders above replacement level. The replacement level save percentage is .938 at 5v5 and .899 at 4v5. I can then determine the approximate contributions of every goaltender relative to replacement level over the past two seasons. I took the difference between the goaltender's save percentage at each strength and subtracted it from replacement level. Then, I multiplied the difference by the number of shots faced at each strength. Finally, I normalized the goaltender’s performance to 82 games, using his ice time at 5v5, 4v5 and 5v4, and divided that by 6 to estimate Wins Above Replacement. The list is pretty much what you’d expect – Tim Thomas, Nicklas Backstrom, Marc-Andre Fleury and Roberto Luongo at the top, along with the unheralded Tomas Vokoun: ```Goaltender Wins/82 Thomas 8.0 Vokoun 7.6 Backstrom 7.2 Fleury 5.9 Luongo 5.3 Lundqvist 5.0 Huet 4.9 Biron 4.3 Miller 4.2 Bryzgalov 4.1 Brodeur 4.1 Giguere 4.1 Conklin 4.0 Ellis 3.8 Ward 3.3 Price 3.3 Nabokov 3.1 Mason 2.7 Roloson 2.7 Auld 2.3 Lehtonen 2.3 Khabibulin 1.8 Smith 1.7 Kiprusoff 1.2 Theodore 1.1 Turco 0.8 Gerber 0.5 Toskala -0.9 Budaj -0.9 Osgood -1.4 Hedberg -6.3 ``` This list isn’t substantially different than the save percentage leaders over the last couple of seasons, but it does adjust for goaltenders who've had enough bad luck to face quite a few 5-on-3s or who've had power-play defensemen surrender a number of breakaways. It also gives us a reasonable estimate of how many goals a netminder could prevent, and what this is worth to his team. I’ll build on this model to incorporate the impact of defensive play on expected save percentage and hopefully get a more complete picture of goaltending. Gabriel Desjardins is a contributor to Puck Prospectus and runs the statistical hockey site behindthenet.ca. You can contact him at info at behindthenet.ca.