Quote:
Originally Posted by Street Pharmacist
Yes, but I'm not sure what your getting at here.
Corsi = all picks directed at net full stop
Fenwick = all picks directed at net minus the ones blocked
I don't understand the point of Fenwick.
As in, Ok you got dominated but it's fine because you blocked a lot of shots? You still got dominated
|
They're two different concepts. You have to understand that when people are making these stats they're trying to model the game, they're not gods gift to hockey.
Corsi is an attempt to easily make a reliable possession stat out of readily available data, since shot attempts are readily available and have been shown to strongly correlate with possession time. As its critics will be quick to tell you, it carries no information about shot quality whatsoever. Corsi is a stat that tells you which team is carrying play, not necessarily which team is more likely to score.
Unfortunately, figuring out which team is more likely to score is what's important when you want to make a good educated guess about which team is more likely to win the hockey game. No one has figured out how to do that yet, but one way people are trying to get there is by trying to establish a model for how hockey works. In this model shot attempts generally follow a normal distribution, so, with enough samples, for every bad shot on goal there will generally be a good shot on goal centered around a mean-quality shot. (I haven't explained that too well, but the statistical foundation
is solid. If you want to learn more check out some normal distribution videos on youtube.) Over a large enough sample size a teams shooting percentage will approach the true percent-chance that any given shot will be a goal. It makes sense to discount blocked shots in this model because any shot that is blocked will not become a chance to score. So if a team has 20 unblocked shots and is shooting at 10%, Fenwick would predict that the bell curve for goals-for for such a game would be centered around 2 and have some standard deviation.
Basically statisticians are trying to simplify hockey down to a game similar to this: Imagine if for every unblocked shot in a game you and an opponent rolled a die, and every time you rolled a one you scored a point. If you were rolling 12 times and got 5 points, while your opponent rolled 18 times and got 4 points, he would lose, but he would be right in saying that your point production is unsustainable.
My issue with analytics right now is that I think Fenwick is just a really noisy stat and hockey is practically just a little too chaotic to be modeled the way it is being modeled. The general concept isn't completely meritless though.
As for PDO, unfortunately the stat does make some sense, although it is hardly to be taken as gospel. In the above "Dice roll" analogy, a high pdo is like a number that shows how often you have been rolling ones. If you've been rolling ones half the time, with enough repetitions you
will regress back to rolling ones 1/6th of the time.
(Damn you Law of Large Numbers!) In reality I don't think PDO is the best stat because A) some goaltenders
are better than others and can sustain higher save percentages for long periods of time,B) some teams are able to convert their chances and make chances better than others and C) some teams are able to play defense in a way that doesn't allow quality shots on goal as often as other teams. Teams that are good in those three areas
will have high PDO's that are
completely sustainable.
On the flipside, some teams really are just getting lucky. There's definitely some grey area there.
So basically this advanced stats stuff is not complete garbage, but it is definitely not gospel either.