Don't Rate Defenseman By Their GA/60 Metric

Wanted to get a very Grama (from Rounders) gorilla math look at a potential expected goals against per 60 metric for d-men at even strength.

The thought process was simple. Using Corsi events, since we have the most data on that from a single season in terms of shot attempts; first finding how many Corsi events lead to a goal on the NHL average goalie - 24.354.

Then, calculating each players CA60, and dividing it by 24.354. And finally, drawing a comparison to their actual GA60. 

For the analysis portion, I removed any defensemen who played less than 20 games this season, to allow a solid enough amount of time for goalies to make a consistent difference in the number. With an R^2 of just .2476, our actual GA60, and our Goalie Average Replacement Expected GA60 (GAReGA60) proved that a goalie can make a huge difference on a defensemen's GA numbers (no surprise) since the variance is so low. 

The question then becomes, what does this equation actually tell us?

Are the d-men who have supremely negative performance values bad? Or are they just the victim of poor goaltending? On the other hand, are the d-men who have supremely positive performance values good? Or are they just the beneficiary or great goaltending?

Let's apply (some) subjectivity to this, and take a look at the names that we see. Our top-5 most overrated players (players who's GAReGA60 - GA60  are the lowest) are Eric Brewer, Oscar Klefbom, Chris Summers, Jared Cowen, and Lubo Visnovsky. Our most underrated players are Sergei Gonchar, Grant Clitsome, Matt Tennyson, Kevan Miller, and Dan Boyle. 

I like this metric because it is quick and easy. But is it enough? Where is the true signal on performance. It seems too easy.

Yes, we are taking individual performance into account by using individual CA60, individual GA60, but the basis we are applying to it is a league-wide standard, an average replacement. Can we find a true signal in the data?

Of all the metrics on War-On-Ice, the one that correlates the highest to this new Performance metric is OFenSv% (on-ice unblocked save percentage) at a nearly perfect correlation of 0.97.

And this makes sense, but it confirms our belief that these players with a high performance metric are most-likely benefiting from timely goaltending as we know that players on ice have very little impact on goalie save percentage. 

Our performance metric has little to do with the defenseman, and a lot to do with the goalie. We can therefore surmise that players with a high performance metric can be overrated if we are weighing their actual GA60 metric in player analysis. 

One more check will be to look at defensemen who played one season on one team and the next on another, and comparing their GA60. If the R^2 variance is low, we can really chalk it up to being a goalie-based (or other outside factor, but we know it's goalies) skill rather than a player-based skill.

Suspicions confirmed. Goals against per 60 minutes of ice time is in no way, shape, or form an individual metric.