That conclusion doesn’t make any sense. Save percentage isn’t random, which is why a skater shouldn’t be penalized or rewarded based on the abilities of his goaltender. Expected goals isn’t immune to noise, but if you’re trying to assess defensive ability, there’s less noise if you take out goaltending ability.
There’s a lot of variance within a season. Certainly more than only 500 minutes would weed out. If there’s a dramatic difference in the expected numbers and the real numbers then we can try to figure out why that is and if it’s an anomaly or a trend. Someone like Erik Karlsson usually gives up more goals than expected, which suggests that the numbers aren’t accounting for his poor defense. But I think expected numbers are usually the best starting point.
The assertion from the PDO crowd is that save percentage is indeed random (which I don't agree with).
That's why, at the team level, they favor equilizing out save percentage and shooting percentage (also viewed as random) to try and see who the real team is beneath. They impute that on individual players as well.
But I also don't agree with your inference that the players in front of the goalie share no significant responsibility for save percentage or goalie performance. I think goalies often receive outsize credit and blame for the defenses in front of them. And so you have a team like Boston which had an amazing defense for over a decade, led by players who are actually good at defense (and deserve credit) and not coincidentally, this resulted in great goal tending almost regardless of who was behind Bergeron/Chara/Seidenburg/Boychuk etc.
Even Boston's backup goalies often had pretty good numbers. Even short term NHL'ers had decent stats behind that Boston defense. I think these things all work in concert.
I also think a team can play a certain type of defensive game where everything is kept to the outside and even though they give up shots, they aren't giving up lots of close up high danger chances. That could be stylistic, but it's going to be hard to account for, even with HDCF and MDCF being tracked.
So I don't think you can argue zero impact and complete randomness - especially not over an enormous sample size. Maybe you are correct that 500 minutes is too noisy. But an entire 15-year career? -I think we can draw some conclusions, for example, if a team's goaltending was noticeably worse or better with a certain player on the ice.
Or perhaps you can argue that 1 forward's impact on save percentage is so small that it is not measurable or even worth considering. Maybe that's correct, but within the context of a Selke thread, I would think, if it applies anywhere, it's here.
There are systematic problems with one-size-fits-all expected numbers. You can find certain players that are consistently underrated by xGoals moreso than others. Patrick Kane is one. So at what point, if any, do we acknowledge that xGoals is missing something in Kane's game that is so persistent that the stat is a bad fit?
