Quote:
Originally Posted by Enoch Root
Thanks for this - I have been meaning to get around to posting something similar. But it isn't just small sample sizes that need to be considered here.
You listed the top and bottom teams for CF stats, but they don't play each other very often, so let's look at an 'average good team' and an 'average bad team' (and assume their numbers are 52% vs 48%). The argument for CF is that, over time, the 52% team will perform significantly better than the 48% team.
But going back to your example, the expected shot attempts for those two teams would be 57 vs 53, on average. Should we expect the team making 57 attempts per game to have more success than the team making 53 shot attempts?
All else equal, yes. But there's the rub. All else is not equal. Quality of offense, quality of shot attempts, quality of defensive coverage, and quality of goaltending are not equal.
Are they equal enough to still expect the team with more attempts to be more successful? That is the 64,000 dollar question. And the answer is 'sometimes'.
IF (and that is a big if) teams are relatively equal in those qualitative metrics, then the shot attempts will rule the day, over time. However, are they equal enough, often enough for us to make any valid assessments as to whether the shot attempts are predictive?
IMO, the answer to that is no. They are valid enough to give us data that shows a positive correlation with winning. However, that data is not strong enough, and consistent enough, to filter out quality of competition. A better goalie, for instance, would largely offset the expected shot attempt differential (and that is just one variable).
So we are left with correlative data that gets misinterpreted as predictive data.
And that's the problem. There is nothing wrong with the data. The problem is that we cannot isolate it enough to draw consistently valid conclusions. And yet, people do, because they assume that stats are in fact predictive.
|
Agree with all of that for sure.
Would add though that shot attempts really isn't the stat to hang your hat on with the advances in quality added to xGF and xGA etc.
And for sure it's not predictive, but that doesn't make the history with a decent sample size important, because "all other things held equal" is a decent way to start looking at things going forward.
If a hockey team consistently gets out played you should be concerned as a team or a fan. Won't mean you can predict the next loss, but you certainly can't be happy either relying on a goaltender night in and night out or a shooting percentage from your skaters that isn't sustainable versus historical averages.
So agree ... all stats have to be taken with a grain of salt as the past doesn't equal the future in anything. But if a team like Vancouver over states their playoff success based on the game results and game scores and ignores the fact that they were almost run out of the building it's at their own peril, especially since said playoff metrics were pretty close to the regular season.