Quote:
Originally Posted by joejoe1337
In the view of the markets and xg adherents, Brighton are a historic outlier this season. Underlying numbers look really solid. 538 has them as 8th best in their model, fb ref has them 4th best.
Clearly an eye test tells you they've got a lot of forwards who are not good at putting ball in net, but they're clearly really unlucky too.
yeah i get that
unfamiliar with their soccer stuff but iirc the 538 model loses money each year when it paper trades and it's a running joke in the betting forum back when people posted there, not saying i can do better or anything
i think i mentioned this last year, but xg calculations are extremely subjective and have some massive holes in them, it's closest equivalent to US analytics would be defensive ratings in baseball, which everyone agrees are meaningless but nice to have as guidance
they don't account for the quality of a player, so a free kick taken by myself or by david beckham in his prime will will have the same xg so long as we kick it in the same general region
futhermore, by default, most off foot shots are dramatically reduced, to the point where it could have been .6 xg on the normal foot but .1 on the other, yet we know many players are excellent from both feet - this is a big reason why salah always outperforms his xg by significant margins
there's also a lot of "let's make the math work" done after the fact to normalize the data, this is why if you add up the individual player xG contributions, it does not equal the team xG
it's good stuff, i'm a big fan of it, it's a step in the right direction, but sadly the really good and actionable data is gathered by hd cameras who measure where everyone is on the field and that's all proprietary and none of the teams are letting that go public out of fear they may be ahead of the curve
i'd much rather just look at posession, corners and shots taken and find that more useful
i'm by no means an expert, just while doing my own modeling, i've found xg to not be all that useful if i was already accounting for the more traditional data, it was more a way smooth out the edges than anything else