Quote:
Originally Posted by Gaskal
Case in point - corsi is quite awful while used as a predictive model. It is much better seen as a performance measure.
|
Precisely. Goal differential is not a predictive measure of expected outcomes; it is a post-hoc measure of actual outcomes.
Shot-differential stats are, at most, useful as a diagnostic tool. What they tell us about this year's Flames is that the team was good at getting opportunities to put the puck on the opponent's net, and piss-poor at converting them into goals. What they tell us about several individual players on the team is that their shooting percentage declined dramatically from their career norms. Some of those players are quite young: if Sam Bennett can't put the puck in the net, it isn't because old age is catching up with him. The question is what went wrong with him and several other guys.
So we have to look at second-order diagnostics. This, in theory, is where the eye test comes in. But unless you're a professional video coach, the chances are extremely remote that you have the time to watch all 1,271 games in an NHL regular season and break them down. And that, approximately, is what you would have to do not only to assess your own players, but to have a full understanding of the competition. There are two teams on the ice in every game, and if you're content to analyse just one of them, you are at most getting half the picture.
So we're back at metrics – trying to apply uniform standards to many different people's eye tests so that the results can be aggregated in a reliable way. It's a hell of a problem, and no general solution has been found.