Quote:
Originally Posted by Bingo
Totally agree, it's like your golf game it can always get better.
But given
a) the fact that it's not just every shot from within say home plate and
b) it's applied objectively across every game and every team
I wouldn't say it can be discounted as a pretty damn good summary either.
|
See this is what I wonder. Say an average goalie throws up a .915. People generally say .905 is bad and .925 is great. You are looking at what causes that 1 percent deviation from average.
In the example I gave you, the Corsica model can’t, as far as I know, predict whether 8 percent of goals are 5 percent likely to happen or 33 percent.
Many people have thrown up their hands and said there is no such thing as shot quality, but we know that’s not true. Still their best models had enough noise to drive the R2 into meaningless territory.
I acknowledge that they are trying to put some context so it is not simply shot location, note that I didn’t dispute that.
I just don’t know, from that list, that what they are capturing is adequate to capture context, and generate the most meaningful results.