Advanced Stats

[TatteredTornNFrayed]

They assume every team that's winning while outshot is lucky and will eventually start to lose, which is absurd.

I haven't read the whole thread yet, but wanted to say I strongly agree with the absurdity of basing so much on just shot counts.

And I think a review of all teams winning percentages when Outshooting and when Outshot By, there generally has seemed to be pretty convincing evidence for quite a few recent years that, if anything, more teams win when being outshot than the other way around.

So that right there tells you that stats based only on shots in a game already have some major flaws, especially if they aggregate them over the whole game, and don't take situation into account.

I've read several arguments how they've statistically determined that shot quality is not repeatable, so that is sort of their argument why shot counts are so important, but I don't think those statistical arguments ever quite convinced me. I'll admit I'm no statistical expert, but the analysis always seemed a bit circular to me, and relies heavily on some questionable assumptions.

 

[dwkdnvr]

I haven't read the whole thread yet, [...]

So that right there tells you that stats based only on shots in a game already have some major flaws, especially if they aggregate them over the whole game, and don't take situation into account.

You need to read the thread. The stats that are most frequently quoted are far more specific than simply 'aggregate shots' - they are shots 5-on-5 when the game is 'close' (even or 1-goal differential). This is the metric that has been found to most closely correlate to success/winning.
There are several effect that this eliminates, specifically the lopsided nature of shot totals during PP/PK situations, plus the 'score effect' of teams that are behind launching far more shots than teams 'protecting the lead'.
It's still a highly imperfect stat, but it's not bad.

The main things about advanced stats that I have problem with aren't really the stats but how people use them.
- PDO still seems forced to me, yet it gets a lot of attention, maybe because getting to say 'regression to the mean' makes you sound smart. The whole PDO argument seems to basically boil down to 'everyone is about average in the long run'.
- some people seem to neglect that all the advanced stats analysis still only 'explains' about ~40% of the correlation to winning. Useful, sure, but it leaves a lot on the table and leaves lots of room for outliers, but it gets thrown around as 'proof' more than it should.