View attachment 982414
Ever since 2015-2016, Scheifele has never failed to convert less than 15% of his shots.
According to most stat nerds, even converting 15% is already a bit high and unsustainable, yet Scheifele has almost an entire decade of performing well above 15%. He's even flirted above 20% twice, and looking like he'll finish this season shooting above 20% as well. What exactly makes him so efficient in converting his chances or being the PDO king to say the least?
15-16: 29g, 22.85 ixG, 1.27g/xg
16-17: 32g, 22.53 ixG, 1.42g/xg
17-18: 23g, 16.82 ixG, 1.37g/xg
18-19: 38g, 27.33 ixG, 1.39g/xg
19-20: 29g, 24.56 ixG, 1.18g/xg
20-21: 21g, 15.94 ixG, 1.32g/xg
21-22: 29g, 19.24 ixG, 1.51g/xg
22-23: 42g, 31.57 ixG, 1.33g/xg
23-24: 25g, 22.94 ixG, 1.09g/xg
24-25: 32g, 19.22 ixG, 1.66g/xg
You could argue that this year is somewhat unsustainable. However, it's not that far out of the line.
Because you have good shooters, and bad shooters. Not everyone converts xG at the same rate. You have two components: xG generation, and xG conversion.
PDO is a team statistic, because of the assumption that it's highly unlikely that an entire team could have especially higher or lower shooting percentage. For individual players, this is different.
xG conversion of 1.00 assumes the population average. However, no individual player is the population average.
It's expected that the actual xG conversion rates follow a normal distribution centered around the population average. Scheifele's conversion parameter might be +2 for 2 standard deviations above average(top 10% or so). That's still well within the expected parameter range.
Keep in mind that xG as a stat is most meaningful for population-level analysis. It doesn't hold for any individual player. Ideally, every player would have their own xG models, but there's not enough sample size for that. So you need to adjust the parameters in relation to the population mean.
Another interesting trend is that xG has become worse and worse as a predictive tool as the seasons have passed. This suggests that the current models are largely out of date, and could use a refresh. Perhaps players are playing more for high-danger chances than they used to, which causes the xG parameter to always underestimate the number of goals scored.
