Quote:
Originally Posted by Enoch Root
There seems to be a theme in the last number of posts suggesting that those on the -anti advanced stats' side of the argument don't understand the stats or think they are flawed.
While there are some valid arguments that there are some marginal flaws, that is not the problem at all. The stats are fine. What they are quantifying is fine.
The primary problem is that the people wielding them often don't understand how to utilize stats for analysis. Too often it is a case of: here's a stat showing an imbalance, therefore this is going to happen.
Doesn't work that way.
|
I agree with this for the most part but I have to disagree that what's being quantified is accurate with regards to Corsi or Fenwick quantifying "possession". Those two stats quantify shot attempts and not possession, full stop. Sure, they can be used as a model to help understand possession, but they are not able to measure anything other than shot attempts. The only way to measure possession right now is to get a stop watch and time the number of minutes a team has possession of the puck during the game (and I'm sure some teams have employed that strategy).
For example, say you are trying to figure out how rich a group of 1000 people is, but for some reason, they don't want to tell you how much money they make. You find a correlation between wealth and the price of vehicle they drive. So you create a model in which you figure out what type of vehicle each one of those 1000 people drive and figure out where each one of them stands in terms of wealth. Sure it might make sense and it might be an accurate model for 95‰ of the population, but if one rich guy is happy driving a 1995 Accord or one not so rich guy puts all his money into a brand new Benz, then that throws everything out of whack.
Similarly, if a team employs a strategy in which they don't try to limit shot attempts, but try to limit scoring chances instead and instead of randomly firing pucks on net they hold on to the puck a bit longer in order to get into good scoring lanes, then it throws the whole corsi/Fenwick possession model out of whack as well.
In the vehicle wealth model, you can't say you're measuring wealth. By measuring the price of the vehicle you're measuring vehicle price and interpreting the data to figure figure out wealth.
Similarly, corsi/Fenwick aren't measuring possession no matter which way someone tries to spin it.