Football/ Ayush Pawar

01 / Environment Final · frozen 7 Oct 2026

Does the environment change the player?

Do weather and crowds change how much a player produces?

  • 2015/16 — 2024/25Seasons
  • 5Leagues
  • 525KPlayer-matches

Finding

Weather barely changes how much players produce: rain and humidity sit within about ±2% and wind about −1% per +10 km/h. Temperature is the one exception — about 3% more chance creation per +10 °C, mostly in Serie A. Crowds matter more: empty stadiums roughly halved home advantage, and it came back with the fans.

02Why it matters

If rain, wind or heat changed what players produce, every raw stat line would carry the weather with it, and a player's numbers would partly depend on the climate he plays in.

Crowds are the same question with people instead of weather: does a full stadium change what happens on the pitch?

Does the environment a match is played in move player numbers?

03What we did

  1. Player-match output
  2. Kickoff weather at the stadium
  3. Fans or closed doors
  4. Same player, team and season
  5. Effect of the environment

Each player is compared with himself — wet days against dry, full stadiums against empty — so differences in ability and team quality never enter the comparison.

Technical detailHow the estimate is built

Weather: on per-minute output with player × season, opponent × season, league × kickoff-hour and league × season × month ; ERA5 weather at the stadium and kickoff hour. A player × team × season version moves no estimate by more than 0.6 percentage points.

Crowds: team-level PPML with team × season and opponent × season fixed effects; 13,888 matches with fans and 1,893 behind closed doors, partial crowds excluded. Crowd status from a dated, sourced calendar per league.

Six hypotheses fixed before looking (rain and long shots, wind and headers/crosses, heat and late-game output). on cluster and p-values; with the closed-doors dates shifted 2–4 years.

04What we found

Crowds

Final

Without fans, home advantage shrank — and came back with them.

The pooled xG home edge went from +30% with fans before 2020 to +13% behind closed doors, then back to +26% when crowds returned.

Pooled team xG home edge, by period

Home ÷ away, team level, all five leagues.

The xG home edge fell from +30% (with fans, before 2020) to +13% without crowds and recovered to +26% when they returned.

Study 01 · Final Summary §3 (team level)

Every league lost xG home advantage without fans; Ligue 1 lost its whole goals edge.

Without fans, the home edge shrank on every measure.

Home ÷ away at team level: goals 1.27 → 1.11, 1.29 → 1.13, shots 1.23 → 1.13. Yellow cards went from 0.89 (home side booked less) to 1.02 — the advantage gone.

Home ÷ away, with fans vs behind closed doors

1.00 = no home advantage. Team level.

Every attacking ratio moved toward 1.00 without fans, and the home side's lighter treatment in yellow cards disappeared.

Study 01 · Final Summary §3

Weather

Final · frozen 7 Oct 2026

Weather barely changes how much players produce.

With all five leagues, rain and humidity effects on shots, xG, xA, goals and key passes sit within about ±2%. Wind is small but now detectable: about −1% shots, key passes and xA per +10 km/h. Over ten Premier League seasons alone, 0 of 20 weather × metric effects survived correction.

Weather, next to the crowd

Change in output. The shaded band is the range the five-league rain and humidity estimates sit in.

Rain and humidity move output by about 2% at most and wind by about 1% — a small fraction of what a crowd does to home advantage.

Study 01 · five-league weather run (6 Oct 2026); crowd rows team level

Six predictions, written down before looking.

0of 6pre-registered weather hypotheses survive

Rain and long shots, wind and headers or crosses, heat and late-game output: on five leagues and 445,377 non-penalty shots, none of them held up.

Temperature is the one weather effect that holds up.

Per +10 °C, across five leagues: key passes +3.2%, xA +3.2%, goals +4.2%, shots +1.2% (xG +1.2%, not significant). It survives a 1,000-permutation placebo check, stadium and team controls (which trim it slightly) and three other model types. It works mainly through whether a player registers any chance creation at all.

Serie A carries it — all five metrics survive the placebo (xA +10.7%, goals +9.3%) — Ligue 1 and La Liga show part of it, and no Premier League or Bundesliga result does. The data can't say why warmer games create more.

Temperature +10 °C, five-league pool

Change in per-90 output, same player. Red: holds up under every check. Grey: shots loses significance once the stadium is controlled for; xG is not significant.

Per +10 °C: goals +4.2%, key passes +3.2%, xA +3.2%, shots +1.2%; xG +1.2% is not significant.

Study 01 · M12 final checks (6 Oct 2026); frozen 7 Oct 2026

05What surprised us

  1. What the raw data suggested

    Forwards' xA rose 29% in the rain in the 2024/25 Premier League — a promising lead.

  2. What happened after we checked

    On three other leagues the same effect was −1.4%; on all five, −1.2% (p = 0.43). Not replicated.

  3. Lesson

    A striking result from one league and one season is a hypothesis, not a finding. It has to replicate.

06What it means

  • For rain, wind and humidity, the weather is not something a player's numbers need correcting for. Temperature is the exception: about 3% more chance creation per +10 °C, mostly in Serie A — small, but it survives every check.
  • Crowds are different — and a crowd effect is a home effect. That is where Study 2 picks up.

07Limitations

  • No mechanism: the data cannot say why warmer matches produce more chance creation, or why the effect lives mostly in Serie A. Effects are per +10 °C across normal top-flight conditions, not rare extremes.
  • The closed-doors period is mostly one season, so per-player crowd estimates are noisy; there are no attendance figures — only fans or no fans.
  • Cup matches and fixture congestion are not in the data.
  • Observational data: effects are associations within carefully matched comparisons.

SourcesUnderstat · football-data.co.uk · Open-Meteo ERA5 · Wikipedia/Wikidata · OpenStreetMap
Analysis period2015/16 — 2024/25 · frozen dataset · Methodology

08The next question

If crowds move home advantage, does it belong to individual players?

02 / Home advantage