Quoting: Dan10900
In the end goals are what matter, and yes, player do consistently underperform (Hornqvist Tkachuk already) or overperform (Kane* Laine to a smaller extent) and you can adjust for that accordingly, but that doesn't mean in the majority of cases (especially smaller sample sizes... Kane's xGF all of a sudden falling off) xGF is completely useless (especially things like RAPM xGF where it adjusts for QoT QoC usage etc) and defensively xGA is 100% the best metric we currently have available, it shows to the best of our ability how many chances a player limits etc, and you don't have to worry about having a terrible goalie tanking a players defense, and the unexplained factor in most RAPM models is 80% offensive/TM's S% bc the variance in opponents shooting/offensive talent largely balances out
My critique of using expected results as being directly related to a player's ability is that the measurement cannot take into account enough variables to accurately indicate a player's skill or value. Often times analytic people on Twitter (Micah, EV twins, Dom, etc.) will use the rhetoric that "we understand the model has flaws, but it's the best available so we will rely on it until we can develop better models." This mostly comes into effect with WAR/GAR models and micro statistics (zone entries, exits, etc.)
I could pull up any player's xGA and the only thing I'd be able to definitively say is that they were on the ice for when those results happened. It cannot account for positioning, systems, situation, etc., and those limitations effectively make the stat useless with regards to player evaluation. I think it's nice to look at after watching games, but there needs to be context to the stat, which there currently isn't any. The concept of "something is better than nothing" doesn't matter if the something is used incorrectly.
Quoting: Dan10900
The issue with that logic is both Drai's GF-GA (bad stat ik) RAPM GF-xGA are negative or barely positive (and were negative till they called up Yamamoto lol)
Again, the entire purpose of underlying metrics is to find players who can generate offense. Once you have a player like McDavid, Draisaitl, MacKinnon, etc. who can produce year in and year out, then there advanced stats aren't relevant in their evaluation of their skill set. The expected results don't matter if a player can score over the long run.