Quote:
Originally Posted by Bingo
Show my your analysis then.
I didn't discredit anyone. I'm saying I think it should be looked at. It would be prudent.
Arriving at a conclusion without an analysis isn't an analysis
|
You don't need to make yet another analysis.
Try a thought experiment.
Take corsi. Now, it's a correlation, so right on that basis it should not be trusted.
But hey, why the hell not.
So let's say that corsi has a direct effect on game points. That is, the higher a team's corsi, the more points they should have. Points are the majority measure by which teams make the playoffs (ties in points notwithstanding, but for this exercise it's not important).
I'm not going to bother to do the math because it's so obvious, but basically this means that the Flames should finish second this year.
So everybody knows that this is ridiculous. What to do?
Well, you can refine your data so that you're getting more data on different things. The one thing I have noticed is that the fancy stats guys in general are rather poor lateral thinkers. So then we have spurious things like PDO.
In general, a great data validity exercise is to take your dataset and generate a model with it. Does the model match with reality? It's really easy with sports because the dataset are so tiny.