01Study 02
Everyone gets better at home.
+27% xG
But no reliable individual home specialists: the lift is about the same for every player.
Football research · 01—05
What actually belongs to the player?
Five studies investigating how much football performance belongs to the player — and how much belongs to the context.
- 5Leagues
- 10Seasons
- 525KPlayer-matches
- 18,011Matches
- 8,348Players
Premier League, La Liga, Serie A, Bundesliga, Ligue 1 · 2015/16–2024/25 frozen study data · live seasons tracked separately
The thesis
Context moves the numbers.
A player’s output doesn’t exist in isolation. Venue, crowd, opposition and league can all change what the numbers look like.
+27%
Home
The same player, in the same team and season, produces this much more xG per minute at home.
Study 02 · 95% CI +25–28%
−14%
xG per +1
Every step up in opposition quality cuts output by about the same share — for almost everyone.
Study 03 · −13.8%
97%
Average output on transfer
After a league move, compared with . Averages hide wide individual swings.
Study 05 · 95% CI 93–101%
So we tried to remove the noise.
Five studies, one investigation
Each study builds on the one before.
-
01 / Environment 2015/16 — 2024/25 Weather freeze pending
Does the environment change the player?
Mostly no.
Rain, wind and humidity had surprisingly small effects. Temperature remains the open question. The five-league models are done, but the final checks and freeze are still pending, so this answer is provisional.
Read the studyWeather, next to home advantage
Change in a player’s xG. The shaded band is the largest weather effect the data still allows.
On all five leagues, rain and humidity move output by about ±2% at most (wind about −1%). Playing at home lifts the same player’s xG by 26.6%.
Study 01 (five-league run, 6 Oct 2026; final conclusion pending) · Study 02
-
02 / Home advantage 2015/16 — 2024/25 Complete
Does playing at home change output?
Yes.But not because some players are simply “home specialists.”
Everyone gets about the same lift at home. Not one of 4,665 players has a reliable personal home edge.
Read the studyEach player’s personal home edge
4,606 players in this export (the reliability test reports 4,665), xG at home vs away. Each row is scaled to its own peak; the outer bars collect everything beyond ±200%.
As measuredspread ±77 pts
After removing noisespread ±2 pts
−200%−100%0+100%+200%Measured naively, players’ home edges look wildly different. Once match-to-match noise is removed, 4,390 of 4,606 land between +20% and +30%.
Study 02 · empirical-Bayes shrinkage
-
03 / Opposition 2015/16 — 2024/25 Complete
Do some players thrive against elite teams?
We couldn’t find reliable evidence.
Strong opponents cut everyone’s output by about the same amount. No flat-track bullies, no big-game players.
Read the studyxG per 90, by opponent strength
All players. Opponents split into ten equal groups by pre-match strength rating.
0.180.10Weakest opponentsStrongestxG per 90 falls from 0.18 against the weakest tenth of opponents to 0.10 against the strongest — and by about the same share for everyone.
Study 03 · sequential market rating from pre-match odds
-
04 / Similarity Profiles to 2025/26 Complete
Can we find players who play alike?
Yes — within limits.
Context-adjusted profiles find sensible matches across leagues. But similarity describes a player; it doesn’t forecast one — and some players have no real equivalent.
Read the studyChance creation, above role average
Bruno Fernandes (Premier League) against Serie A’s top 8 attacking midfielders and wide forwards. In standard deviations.
Bruno 3.6Best in Serie A 1.901234 SDBruno sits 3.6 SD above his role average. The best in Serie A’s pool, Martin Baturina, is at 1.9. His closest statistical matches all create less.
Study 04 · Bruno case study, profiles to 2025/26
-
05 / Transferability Moves to 2025/26 Complete
Does similarity predict what happens after a move?
No.Player history does better.
A model of the player’s own history, age and both clubs beats “he’ll do what he did” on seasons it had never seen. Knowing who he resembles adds nothing.
Read the studyPrediction error on 157 league moves
Mean absolute error in xG + xA per 90, first season after the move. Lower is better.
Similar players alone barely beat the naive guess. The player’s own history cuts the error by 17%; adding similarity on top moves it by less than 0.001.
Study 05 · comparison on transfers with Study 4 profiles
Findings
Four things the data changed my mind about.
02Study 03
There may be no such thing as a “big-game player” — at least in this data.
0 / 23,022
player × metric estimates showed a reliable resistance to strong opponents after .
03Study 04
Bruno Fernandes doesn’t have a statistical twin in Serie A.
3.6 SD vs 1.9 SD
Bruno’s chance creation above his role average, against Serie A’s best comparable profile.
04Study 05
Similarity doesn’t predict transfer performance.
+0.000
Adding Study 4 similarity changed Study 5’s prediction by essentially nothing.
Live
Does the research still hold?
The research isn’t finished when the paper is published. New seasons are added weekly without rewriting the historical results.
Loading the live tracker…
Methodology
How do we know?
- Data18,011 matches
8,348 playersEvery player in every match of Europe’s top five leagues, 2015/16–2024/25: 525,328 player-matches and about 452,000 shots.
- ContextPlayer × team × season controls
Every comparison is a player against himself — same club, same season — so ability and team quality cancel out.
- Leakage155,282 checks
No rating is allowed to peek at the future. Scrambling later results changed none of the ratings checked.
- ValidationLocked unseen transfer seasons
The transfer model was frozen before the last three seasons of moves were opened, then scored once.