Deady, I mean this in the nicest way possible, you’re not the smartest guy here so cut the bullshit. You know exactly what you’re posting and why you’re posting it.
You can’t even admit that you cherry pick stats let alone admit that you’ve ever been wrong. The attempted “you’re not a dentist” remark is a bad look and it’s insulting.
Well, I'm smarter than Beef, so there's that.
I don't "cherry pick," I give the rationale behind every inflection point I try to identify.
Data mining (picking the model that provides the desired result) is a no-no, but there is a legitimate form of data mining, which requires accounting for the order in which you test models, but that's far more sophisticated than the modeling used by hockey sites.
The problem is that how a team performs does change over a season as personnel change and players adjust to schemes, so you can't assume an average over the whole season is a valid predictor of future performance. So you want to identify any sustained changes in performance and the reasons for those changes. This is SOP in time series modeling, testing for inflection points (where the underlying model changes), but would probably be difficult to implement in the hockey stat context.
To me, there are three inflection points this season, the first 10 games, where everyone agrees the team looked like shit but was carried by Hart, the next 16 games, when they got comfortable with Torts' scheme but the lines and D-pairs were in constant flux, and starting in Dec 7, when JVR returned, York came up and the lines and D-pairs became set (with Allison replacing MacEwen a week or so later).
To me, if I want to judge the players and the team going forward, Dec 7 is the starting point, b/c the players are in a "comfort zone" and can't blame adjustments or getting used to their linemates for any struggles. And we see a number of players settle down at that point, Frost, Laughton, especially.
Practically all those "metrics" are flawed since they rely on naive linear correlations.
There are some serious issues with biased measures, left out variable error, etc. And chemistry matters but is hard to measure.
The "more sophisticated" metrics have similar weaknesses, they go into deeper detail but then have smaller samples.
But the real problem is that the available raw data is fundamentally flawed, so at best all you can get from them is some suggestive numbers that need to be validated by eyeballs. Proprietary team statistics should be better, b/c they have the money to do more detailed analysis (who was actually responsible for good or bad) and thus work with better raw data. But since it's proprietary, who knows?