Quote:
Originally Posted by FanIn80
I think what people might be missing about "goals saved above expected" is that it's not a measurement of goals vs saves or shots etc... it's a measurement of "expected goals" vs all goals allowed.
What's an "Expected Goal" (xG)?
What is "Goals Saved Above Expected" (GSAx)?
So you take the total of all expected goals a goalie has face, and you subtract all goals they've allowed. If you end up with a positive result, it means the goalie has allowed less goals than they were expected to. A negative means they've allowed more goals than they were expected to.
It doesn't mean that all goals against are considered equal. It means if you allow 60 goals, but only 40 of them were actually expected to go in (based on the above definition of an Expected Goal), then you've allowed 20 more goals than you should have. If you allow 40, but you were expected to allow 60, then you've saved 20 more goals than you should have.
Shot selection, location, etc is already factored into whether an expected goal has occured or not, so trying to argue that "all goals are not created equal" is moot in this case. The stat is already aware of that, and has already factored it all in.
Edit: I don't mean to sound "matter of fact" about the validity of the stat itself, I'm only confident in my understanding of what the stat is and how to apply it (in addition to other factors) in my thinking about a goalie's abilities. I'm more than willing to be convinced that the stat is flawed, but I need logical evidence to support the idea... and the idea has to originate from a demonstrated understanding of what the stat actually represents.
Edit 2: The quoted definitions above are from Monepuck (they also provide a "learn more" link to Natural StatTrick for xGs).
|
How the stat is flawed, I have explained many times. So will do the short version.
An expected goal is just a refinement on shot percentage, based on shot context.
The ice is broken in to various locations. Shots from the slot, for example, have higher probability of going in than shots from the point. Other things are considered. What type of shot, time since previous events. That’s the basis of the model - using the context of the shot, based on things that are measured, only up until the time the puck leaves the stick
Absolutely nothing is measured or therefore matters in those models with respect to where the shot is aimed. Consider two shots of identical context - one shot into the goalie’s logo and the other a snipe over his shoulder under the bar.
That’s the goalie’s challenge. To stop the puck where it’s going after it’s shot. If it isn’t measured, it’s a shortcoming of the model
Defensive positioning, pressure the shooter is under - not included in the model.
The model effectively takes the average shot which is say 90.5 likely to be stopped, and puts it into buckets where there are, say, 78 percent likely, or 99 percent likely.
So, basically the time and space and opportunity to aim, and the defensive breakdowns can pretty much torpedo the stat
**
It’s a good stat, GSAx is an improvement over GSAA because it fine tunes based on measurable things (shot context)
Where the shot is taken from is factored in
Where the shot is aimed isn’t factored in
I have argued that team defense has a lot more to do with a goalie having good stats or not
**
Since you reference Moneypuck, I will share some data from Moneypuck
What we can do is look at this GSAx stat. I will propose look at the top 20 goalies. To avoid low volume statistical outliers, let’s set the threshold at 20 games played. And to even the basis for comparison, let’s look at GSAx per 60. Sound good?
Ok. So what do we see
FLA - Stolarz - .801
LA - Rittich - .607
WPG - Brossoit - .582
WPG - Hellebuyck - .557
BUF - Levi - .496
VAN - Demko - .438
BOS - Swayman - .430
STL - Hofer - .379
NYR - Quick - .372
BOS - Ullmark - .370
NYI - Varlamov - .342
TOR - Jones - .323
STL - Binnington - .301
TOR - Woll - .294
CGY - Markstrom - .290
FLA - Bobrovsky - .274
NYR - Shesterkin - .222
WAS - Lindgren - .222
NAS - Lankinen - .206
LAK - Talbot - .206
So what do we see?
Out of the top 20 goalies in the league, 14 came from 7 teams with 2
LAK, FLA, WPG, BOS, NYR, STL, TOR
The remaining came from 6 teams with 1
CGY, BUF, VAN, NYI, WAS, NAS
Let that list sink in.
I think it pretty darn clearly illustrates that the stat has shortcomings. Not in what it measures - calculations are calculations - but more so in how people interpret it and the weight they give it
People want badly for it to be a really good gauge of an individual goalie’s play. Sorry. It really is not. It reflects team play.
It just doesn’t make sense that, if it was really a reflection of the individual goalie’s skill, that in a league of 32 teams, 7 teams would have both their starter and backup within that top 20