cross-posted from: https://mastodon.online/users/hallenbeck/statuses/111293316231491706
I've made an updated graphic based on feedback. Thanks to @nooeh@lemmy.world for the critique. Updated graphic here:
https://thelastboyscout.uk/assets/img/son_xg_stats.webp
Is Son one of the best finishers EVER? Let's look at some data. 👇
We use actual goals minus expected goals (xG) as a proxy for finishing skill. Players who consistently score more than their xG means they are scoring goals other players would miss. Generally, only the most elite goalscorers *consistently* outperform their xG.
And I can find no player who consistently beats Son. It's astonishing.
Can you find anyone better at elite level?
A player's xG performance should be independent of their own team (if they play for a bad team, that would probably lead to them having less/worse chances, but it won't affect the stats as presented here because they're normalised as a percentage overperformance). What could affect it is quality of opposition goalkeepers - for example, say Allison as a top keeper, should reduce a player's goal-to-xG ratio because he saves shots that might have resulted in a goal against a worse keeper.
I'm afraid I don't think it works like that, although the specifics of xG algorithms that stats companies use are generally closely guarded secrets, they often talk about how it works in loose terms.
xG is usually produced by a huge predictive model which takes in a combo basically every possible data point you can imagine to do with a football match.
Opta talks about their model being fed by things like historic data for the ball position, limb positions, the quality of passing building up to a shot (some models talk about every pass from when the ball was out of play), the quality of the keeper in front and all sorts of other stuff they don't let on about.
Given all that I'm not sure it's possible to meaningfully normalise a player's xG because it's a product of other players on a pitch. So I think it's fair to conclude that a poorly performing team will very likely be negatively impacting an individual's xG.
The specifics of each company's algorithm is secret, but the premise of xG itself is well-understood and relatively straightforward: when taking a shot in a certain situation, what is the probability that an average player would score a goal?
The xG that a player racks up during a game is dependent on other players, but this analysis is about xG performance, i.e. how much better they are at scoring goals than that hypothetical average player. By comparing against a common statistical average player, the external factors about teammates are irrelevant.