It's getting close. Oh man, it's getting close. It's so close I went to watch a Capitals-Flyers rookie game where Peter Laviolette was having a good look at his guys. Saw Paul Holmgren in his suit. Gahdangit. Wish I hadn't missed those Penticton tournament streams.
Anyway, catching up on a few things analytics-wise, and one of the ideas out there is that pucks toward the opposing net leads to pucks not toward your own net. Intuitively, this makes sense as we think Corsi is a proxy for puck possession (e.g. see here
During this past season, every game at FTF had a table of Corsi and chances. I loved--*loved*--reading Danny Boyle's line. His on-ice Corsi+ was sick. Just racked them up. His Corsi+ and Corsi- rates at the end of the season supported the idea that shot generation/puck possession led to shot suppression. Among Sharks defensemen (sans Burns) he had the best CF20 (Corsi For per 20 TOI) and 2nd best CA20 (Corsi Against per 20 TOI, to Marc-Eduoard Vlasic). League-wide, Boyle was among only 10 defensemen (+300 TOI, 5v5) to have a CF20 > 21 and CA20
In looking at defensemen league-wide (+300 TOI, 5v5) to see how others fared, I realized however that there were serious issues with this idea of shot generation --> shot suppression. I did a simple correlation test to see if great CF20 rates meant low CA20 rates. It didn't. It was negative but weak (~ -0.18). So, I took a step back and had a look at team Corsi+, Corsi-, and dCorsi.
Suppose we think greater Corsi+ numbers suppress Corsi-. This means that the more Corsi+ numbers increase, the more Corsi- decreases. In terms of shot differential, we should see a steep line. The chart below is from 2013, 5v5 play, all teams. It depicts the possible relationship between Corsi+ and Corsi-.
Pretty blob-like. This means the association is, at best, very weak. I did 5v5 and 5v5 Close play, all teams for NHL seasons 2007-2008 through 2013. In doing the season-only correlations, the strongest association was 2009-2010 5v5 (-0.24; .xls file with summary here; includes team points and PDO with Corsi For and Against).
[Stat note: if you collapsed all the numbers together, Corsi For very strongly correlates with Corsi Against (0.91). If you zoom out enough, we can very safely say all hockey games consist of Corsi events For and Against. This is what happens when you just scan the entire forest from a distance. You see a ton of trees. No chipmunks, no bird nests, no beavers building dams and shit.]
So, let's think through this. We have evidence that shot generation does not lead to shot suppression. Teams get the puck, advance, generate a shot. Think about what happens after a Corsi+ event. Puck caroms off glass, defensemen gathers, and makes a zone exit pass. Or, rebound to corner, forward gathers, and passes to the point for another shot. So, there is some kind of probability of another Corsi+ event after a Corsi+ event. In addition, there is some kind of (somewhat reduced) probability for a Corsi- event after a Corsi+ event. This makes sense given all the things that need to go right to exit the zone, get through the neutral zone, and then setup a Corsi+ event.
So, that the association is weak NHL-wide is not that unsurprising. At the same time, looking at things this way allows for us to think about the relationship between Corsi+ and dCorsi (and Corsi- and dCorsi). Obviously, there is some endogeneity here. There is definitely more to be said, but I can't think of much that's coherent.
Bottom line, I think, is that Corsi is important for us fans. It really is. I need to be able to rationalize losses. If anybody knows of stuff on shot suppression, I'd appreciate the links.