# GSAA: a step in the right direction, but hardly essential

February 27, 2014, byA little over a week ago, one of my go-to publications, InGoal Magazine, released an interesting article, entitled GSAA: An Essential Statistic for Evaluating Goaltenders, touting a new advanced metric for analyzing goaltending, called GSAA (Goals Saved Above Average). The author, Greg Balloch, does a nice job of breaking down the specific methodology that goes into determining how many goals a goaltender saves above the league-average. Here is Greg’s explanation of the mechanics from the article:

You take the league’s average save percentage and apply it to the amount of shots a particular goalie has faced. You get a number of goals that the average goalie in that league would have surrendered if they faced the same number of shots as the goaltender in question. That number gets compared to the number of goals surrendered by that goaltender, and a plus/minus is created. If a goalie is in the positive, that is how many goals they have saved compared to a league-average goalie. If they are in the negative, then it is safe to assume that they are performing worse than how a league-average goaltender would perform in the same situation.

He discusses the application of this stat in the same context as WAR in baseball. For those unfamiliar with WAR, here is a link to Fangraphs’ explanation of the formula. While I agree with Greg that conceptually GSAA is similar to WAR in that it creates a quantifying statistic as it relates to league average production, it still lags far behind its baseball counterpart. WAR is an aggregate formula which incorporates several other advanced statistical calculations with their own independent formulas and creates a snapshot of production in relation to the rest of the league. There are statistically significant variables calculated into the individual formulas of the components of WAR, and thus is much more reliable than the fairly context-neutral GSAA. GSAA simply normalizes save percentage over a consistent amount of shots.

Greg does a similarly nice job pointing out the pro’s and con’s of the application of GSAA, but these pro’s are very anecdotal and not statistically significant due to the lack of context. Sure, you can make an assessment of how much a team relies on a goaltender based on this stat, or who is being pummeled on a nightly basis, but it doesn’t really tell you anything about how well the goaltender is playing in a higher shooting environment due to lack of shot quality, location analysis, or game context data.

Compared to a newer statistic like DIGR (Defense Independent Goalie Rating), which I reviewed here prior to joining the staff full time, GSAA offers a more tangible approach, but less useful information. DIGR actually utilizes heat maps for shot location, game context (PP, PK, etc.), and shot type (backhand, deflection, slap shot, etc.) to predict league average performance. Its biggest weakness is the predictive rather than observational nature of the statistic.

Thus far, no advanced statistic has been able to really get at the essence of goalie analysis, and although GSAA can be useful in limited contexts, it still isn’t accounting for nearly enough variables to be reliable in advanced scouting or predictive analyses. Yes, it’s better than GAA or raw Save Percentage. However, while it does have its place amongst valuable statistics, in this goaltender’s opinion, it is far from essential.

“but it doesn’t really tell you anything about how well the goaltender is playing in a higher shooting environment due to lack of shot quality, location analysis, or game context data.”

Isn’t the difference of location and shot quality between teams over a large sample size negligible? It’s the reason we largely ignore it with Corsi/Fenwick.

For players, yes, for goalies, no. Fenwick and Corsi are using shot attempts to quantify puck possession, so the conversion rate can be normalized without affecting the overall intent of the stat.

Over a large enough sample size (like career-size), GSAA can tell you a lot about long term success, but very little on a season to season basis. For example, this year Ben Bishop is far and away the best goalie in the NHL. Save percentage and GAA back that up. However, it tells us nothing about how much his defense helps his numbers. If I were valuing him for a possible long-term extension, these stats would be useless to me.

Take two similar poor puck possession teams: Team A plays a low zone collapse and takes relatively few penalties. They give up a lot of shots, but mostly from the point and from wide angles. Team B has a weak, aggressive blue line that gives up odd man rushes and takes a lot of penalties. Most of their shots come from the slot area and off lateral passes. Two very different environments for goalies, but similar shot totals. I would love a stat that could help me quantify that.

I assume you’ve read through this stuff, but in case you haven’t:

http://www.sportsnet.ca/hockey/nhl/mediocre-play-of-rangers-lundqvist-wont-last/

http://www.sportsnet.ca/hockey/maple-leafs-reimer-showing-unique-ability/

I’m surprised no one has taken these stats to the next level and added a shot location factor. It affects goalies (shot location effect on skaters is negligible).

Well, there’s this:

http://www.boysonthebus.com/2013/11/12/piecing-the-shot-puzzle-together/

He’s not using (x,y) though, just x.

That’s definitely a start, but there’s a next step involved which (at least in my opinion) combines these two stats to get an expected SV%.

What about tonight’s opponent? The Hawks are coming to town and we’re shorthanded & full of distractions involving rumors. I see this game becoming a set back in this teams confidence. Kane is on a mission & Sharp is sharp. Toews had a wonderful tournament. I hope Nash has his legs tonight.

It’s not great. And its based on averages not a replacement player which WAR

One problem with this – and I don’t know whether or not this is significant or not – is the effect the goalie himself has on the number of shots. Goalies can pursue the puck in the offensive zone, play the puck, control rebounds, etc. Another issue is low percentage shots. Against some goalies, any shot is a good idea. OTOH, you don’t want to give a Brodeur the puck.

Martin St. Louis asks out of Tampa, waives his NTC for a move to NY. Rangers offer Callahan, but Yzerman counters requesting an additional prospect or pick. No deal.

Now, Kesler asks out of Vancouver apparently. At face value, Vancouver, too, probably asks for more than Ryan Callahan. However, does the possible reunion of Cally and Torts push this into the realm of possibility without the Rangers having to give up additional pieces?

Callahan would fit well in Van reunited with Torts. Kesler obviously would be a big addition in NY, reunited with AV. Is a straight up swap doable?