Qwiss This thing that over performing you xG is bad but underperforming it is good needs to go away.
It's neither, isn't it? There's some kind of categorical distinction that's missing in the commentary. If you model a player's G stat as something like:
G := xG + ΔG
where Δ is the overperformance relative to xG (ΔG = G – xG), then Δ should be further carved up:
Δ := ΔEnvironment + ΔBaseline
The ΔEnvironment aspect of Δ is the aspect of overperformance which is due to the ordinary variation of the expectation of scoring a goal in relation to whether or not the goal is scored, which includes blind luck, anomalies of circumstance, how wet the pitch is, the imperfectly measured slickness and ability of classy team-mates in picking out assists, etc.
This ΔEnvironment can be pretty high relative to G and it could be a relatively bigger factor for a player with fewer minutes and goals (if an unusually high proportion of those fewer cases where the player was "expected" to score were actually fucked up in some unmeasured respect).
It's the job of the wonks designing xG models to try to engineer out as much of ΔEnvironment as possible but we can be sure they don't succeed.
The ΔBaseline aspect of Δ is a posited aspect of overperformance due to the player's own scoring ability relative to the abstract player modelled by xG who is expected to score the goal. Is the real player here a natural finisher, as opposed to Gabriel Jesus (who definitely has a negative ΔBaseline)?
If you've got a way of estimating or interpreting ΔEnvironment and ΔBaseline for a player, then you want to go after players:
- undervalued with respect to their environmental challenges in scoring (minimise ΔEnvironment and find the goalscoring talents flying under the radar due to their difficult conditions)
- undervalued with respect to the way their finishing ability increases their end product (maximise ΔBaseline and find the natural finishers)
All relative to budget of course …
