Just out of curiosity, I play a lot of EA’s NHL and it has a stat for time on attack. A lot of the time when matchmaking fails and I play a weak opponent I’ll crush them in time on attack, but I’m a selective shooter so shots can be fairly equal. It’s just that my shots are one-timers and other high danger shots, while the opponent’s shots are clearing attempts hitting the goalie etc.
So in that context the shots tell absolutely nothing, while time on attack shows who’s dominating. Wouldn’t that hold true for real games as well? Selective shooters will have worse Corsi so wouldn’t time on attack (literally possession) be a better measuring tool for possession?
@garret9
Sorry, I'm going to give a long answer first and then a short answer after... sorry
Long Version: Corsi History Lesson
Long ago, hockey bloggers, mostly fans of Albertan NHL teams, were looking for a way to help evaluate a player's effectiveness beyond their scoring. They knew that +/- sucked so they needed something better but along the same idea. They had been tracking scoring chances as a community project to study that kind of stuff.
Tim Barnes (a blogger that went by Vic Ferrari, now with the Washington Capitals) overheard an interview with (I think) the AGM of the Buffalo Sabres. The gentleman mentioned that the Sabres use shot attempts to get a general idea of goaltender workload.
Skipping ahead a bit, Barnes wrote an article titled "Possession is Everything." He showed the shot differentials for 3 games, noted that it matched his perception of the tilt in the ice for those games. He named it after Jim Corsi, saying that the person being interviewed was Jim, but in reality it was just because he didn't think the AGM's name matched and liked Jim's moustache. Ironically, while the person interviewed was not Corsi, it was his idea to look at shot attempts.
Skipping some more steps, we found out stuff about Corsi. Corsi predicting success better than goals or other similar stats. Corsi correlated highly with possession and time on attack. Corsi correlated highly with scoring chances, and didn't need to be tracked manually (and that differences in scoring chances and Corsi tends to regress towards Corsi).
When it comes to measuring itself, shot volume, Corsi is the best.
Shot volume matters, so it predicts success, and it seems to do so better than some other things (like possession*, goals, etc.) and especially so when adjusted for things like score effects and such.
Other things matter too, and so Corsi is both imperfect and can be improved upon. Interestingly, sometimes adding more doesn't always make it better.
Short version
Corsi correlates with possession, but that's more anecdotal (unless you want to specifically talk about possession).
True puck possession is nice, but the real reason we care about Corsi is that it both measures something that matters (shot volume) and helps you know which players/teams are likely to fall down or bounce back.
We have one *public* year of attack TOI, but my company has also tracked the same thing.
Turns out attack ice time and possession both are inferior to Corsi in evaluating players... but is still more information so it is useful.
He has published the proofs of all of those statements he made. They are beyond convincing. They are compelling.
