Wednesday, March 14, 2007

Goalie Wins - A New Perspective

As I have demonstrated repeatedly in this space, goaltenders are often evaluated by their win record. Many fans think this is logical, since wins are the main objective in hockey, and therefore it seems that the goalie with the most wins would be the best. If the goalie's contribution determined 100% of the win or loss then this would be correct, but the problem is not the inapplicability of using wins as a measure (for example, one of baseball sabermetricians' key metrics is WARP, or wins above replacement), it is that wins are impacted very heavily by other factors. Goaltending is probably worth 20-25% of the game of hockey, and at the pro level the difference between the best and the worst goalies is marginal, something on the level of a save every two games. The skaters, therefore, have much more of an impact on the final result than the goaltenders.

Another argument made is that some goalies play to the score, and they know how to make the "big saves" when their team needs them, which shows up in their number of wins. This is a common cliche used when attempting to give credit to goalies for their high win totals. This may be a little bit true - goalies likely bear down a little more in close games than in blowouts, but I doubt the effect is large. And what about a goalie whose team never scored a single goal for him? He would never have the chance to make a "big save" to keep that one goal lead - is it therefore his fault that his team never won a game? The studies I have seen of goalie performance with one goal leads or other similar situations did not reveal certain goalies to be persistently clutch. This sentiment appears to be largely a result of selective memory bias for fans of goalies on good teams.

A problem with alternative goalie stats, like GAA or save percentage, is that one or two bad performances can have a large impact on the final season totals. An average starting goalie could let in 10 goals on a particularly bad day, which could end up dropping his seasonal save percentage .005 or more and raising his season GAA by 0.15-0.20 goals. In terms of winning or losing, however, there is little difference between letting in 5 goals and 10 goals, since both result in a very high chance of losing the game. This is a benefit of wins - not overly penalizing a goalie for the results from any particular game.

Another advantage of wins over counting stats is that wins do not (for the most part) need to corrected for era. Throughout the NHL's history, season lengths have varied, but for the last 25 years this variance has been small (between 80 and 84 games). In contrast, the level of goalscoring and the average goalie save percentage have changed drastically over the same period, making it difficult to compare most goalie statistics from the 1980s with those from the late 1990s and early 2000s. The only major change has been the introduction of a shootout in 2005, which affects only regular season results.

It would be nice to have a stat that retains the intuitive logic of wins and losses, yet focuses more closely on actual goaltending performance, without taking all of the other team factors into account. Shot totals are the most significant hidden variables in wins, since teams cannot score without shooting, and teams that take more shots have a greater chance of scoring more goals. A team outshot 50-20 is very unlikely to win, even if their goaltender plays an outstanding game.

I have developed a stat that takes out the impact of era and shot totals, and is presented in terms of wins and losses. I call it "Head-to-Head Wins". I go through game by game, and measure the save percentage of each goalie against his counterpart at the other end. The goalie with the higher save percentage gets a win, and the one with the lower save percentage gets a loss. That is the stat in its basic form. I also have an adjusted version of the statistic, based on the number of times shorthanded in the game, opponent quality, and team shot quality allowed. After these adjusting factors are applied to each goalie's save percentage, the results are again compared to determine the winner or loser. For the adjusted stat, I also count any game where the save percentages differ by .005 or less as a tie, since there really is no real difference between the two goalies' performances.

This stat is more focused on the goalie's actual performance than raw wins, yet it maintains some of the benefits of the traditional statistic. Why should a goalie be rewarded for a win, even though the guy at the other end outplayed him? The goalie who plays best gets the W, regardless of the final score. This allows us to compare goalies more directly, find out who is dominant against their peers, and find out who is winning only because of their teammates. I plan to use this statistic to look at recent regular season performance, as well as use it in a comprehensive study of playoff performance.

No comments: