The graph is from
an athletic article I wrote, which has been posted a few pages ago, and predominately comes through multiple sources that overlap and (informally) peer reviewed each other.
"Luck" is actually fairly scientific in this case, which is natural variance not being evenly distributed plus a portion of
stochastic system. There are two very different ways that have shown exactly 38% "luck" in 82 game season standings.
(I made the graph only 33% to purposefully underestimate luck)
1)
Desjardins pointed out that even if NHL games were decided by a 50/50 coin toss, there would be some teams better than others in the standings. No team controls their destiny as hockey is a tug-a-war. Probability is just a weighted coin toss (albeit moving weighting) coin toss. So, we expect some percentage of the standings to be pushed by the coin flip as well.
2)
Wiessbock used machine learning to determine the theoretical limit to predicting the totatility of a full season.
(
If you want to see some fun on the shortened season with luck, check this out)
It's not that goals are irrelevant, just less telling than Corsi. Goals are powerful, and each goal contains about 5x the information as a shot (ie: higher correlation coefficient), but their rarity causes a weighted shot differential with weighting goals heavier have about a +0.98 correlation coefficient with non-weighted.
Raw Corsi is superior to raw goals and is entirely useful. Garnering a shot is part of the process.
You want to maximize your number of opportunities (shot volume), maximize your chance in each opportunity (shot quality), and capitalize on those chances (finishing talent). The defensive side of the sport is merely the reciprocal. You want to minimize your opponent's number of opportunities (shot volume), minimize your opponent's chance in each opportunity (shot quality), and stop their chances (goaltending). All other actions either lead up to or are part of these inputs in outscoring the opposition.