I'm having a real hard time with the XGF model this year. It doesn't seem to match A) What I see and B) line up with any of the other metrics.
Case in point Scheifele and Perfetti have the exact same XGF but Perfetti has superior CF%, FF%, SF%, SCF%, HDCF% metrics so how do they have the same expected goals metrics? It doesn't make a lick of sense.
Also I'm not sure how Cole can be in the low 40's when he's north of 50 in terms of CF%, FF% and scoring chance %. Even if you account for him being below in terms of shots (46%) and high dangerous chances (45%) how the heck do you get 42% in terms of expected goals when the lowest single indicator he has is 45%. Logically he'd be somewhere between 46 and 50 depending on the weighting of each indicator. How you can get 42 makes zero sense.
read on how xGF is calculated and it'll make sense. anything % is a share. 2 players can have xGF that are virtually the same even if you can have differing CF. CF is just a count of shot-attempts for, and xGF assigns a value based on criteria (shot distance, type etc.)
you can have high CF (high volume shooters) which will accumulate high xGF.... Ehlers is an example of this.
last season: individual shot attempts/60: 21.72 for individual xGF/60 = 1.
then you have players w/ lower attempts but higher xGF proportionally... players like PLD doesnt shoot a lot comparatively but closer to the net so score higher in quality
last season: iCF/60 14.01, but ixGF/60 0.97.
so on 7 less shot attempts, he's basically the same xGF. which is higher quality per attempt. makes sense as PLD made his living in front of the net (higher probability to score).
per 60 for vs against
Perfetti is 56 vs 54 for CF% of 51% share
Scheifele is 57 vs 66 for CF% 46%
per 60 xGF vs xGA
Perfetti is 2.61 vs 3.62 of xGF% 42%
Schefeifele is 2.45 vs 3.51 xGF% 41%
so when perfetti is on the ice the Jets give up lesser shot attempts in volume, but they are of higher quality (on a per attempt basis). think back to the PLD vs Ehlers example, but for the opposition.
their shot quality for is basically the same (on a per attempt basis)