My problem with statistical models in hockey is that they are often times black boxes. A whole host of statistical information is inserted into the black box, and it spits out a single easily expressed number. But the whole process of how that number was achieved is shrouded, either through being too complex or being proprietary.
I prefer simple advanced stats. Corsi is just the number of shot attempts either for or against. It doesn't express as much as a full statistical model, but I know what it means. I know its strengths and weaknesses. It doesn't lie, it just doesn't attempt to tell the full story. Since I inherently understand what it means, I can combine it with other factors (eye test, deployment stats, QoC stats, etc) to better understand the full picture myself.
With a full model, I don't know which of those factors the model weighted and how heavily, because its all in a black box. I can get a sense that the model heavily favors X while ignoring Y by looking at how it rates certain players. But that's a bit of work to just understand the model. I don't think understanding the model is the point. I think the point of a model is for people to substitute its opinion for their own. That doesn't work in a place where discussion can get as granular as it can here (ideally). We end up discussing the model more than the player, and that is inherently less interesting to me.
