Fans are worth 2.5 points a season.
That's what home teams lost when the stands went quiet. In every 100 matches, home wins fell from 45 to 40 and away wins rose from 28 to 33.
Home wins45
Draws27
Away wins28
The headlines
Five things the empty stadiums proved. Tap, drag, see for yourself.
No crowd, no bias.
With fans in, referees showed away teams about 30 extra yellow cards per 100 matches. Empty stadiums wiped the gap out, with the very same referees.
Extra yellow cards for the away team, per 100 matches
28 leagues out of 31. Same story.
From the Premier League to Brazil's Série A, home teams lost points wherever the gates were shut. It wasn't one league's quirk.
One bar per league: change in home points per match without fans. Hover for names.
5,000 fans is enough.
A few hundred people barely helped. From about 5,000, home teams were back to normal, and a full house added little on top. Drag the slider to fill the stand.
0 people
Bigger crowd, bigger fall.
Clubs that normally played in front of about 2,000 people lost nothing. Clubs used to 9,000, 16,000 or 30,000 fans lost the most when those fans disappeared.
Points lost per home match without fans. Clubs in five groups, labelled by their usual home crowd. Hover for details.
Those are the headlines. Below is the evidence for each, and then how the study was built.
The evidence
The charts behind the headlines, from 38 leagues and twelve seasons, 2014/15 to 2025/26.
Who had fans, and when
Every league closed and reopened on its own timetable. Each league-month was coded from public-health rules and, where it exists, recorded attendance. Unclear months are left out.
Crowd status, league by league and month by month
Most common status of each league's matches, January 2020 to December 2022. Blank: no matches.
Full crowds
No fans
Capped crowd
Unclear, left out
The loss, league by league
The 22 leagues added last showed the same loss as the original 16: 0.13 against 0.14 home points a match. The smallest losses came where crowds were smallest anyway: England's third to fifth tiers lost about half the average, and Scotland's lower divisions nothing.
Change in home points per match without fans
Each league against its own matches before COVID, with 95% confidence intervals.
The original 16 leagues22 leagues added later
Show the numbers
It started the month the gates shut
Home advantage was already drifting down slowly before COVID. If that drift explained the ghost-game effect, the drop would have no reason to line up with each league's own closing date. It does.
Home points, by months since each league's first match without fans
Compared with crowd months more than a year earlier. Months −1 and −2 fell in the shutdown, when nobody played.
Months with crowdsMonths without
Chance can't produce a drop this big
500 fake "ghost periods" of the same length, drawn at random from 2014–2019, give the size of shift that happens by luck alone. The largest is 0.09 points; the real one is −0.14.
500 fake ghost periods, and the real one
Change in home points per match. Each dot is one random 166-day window.
The referees, one by one
With names for 20 leagues, each of 701 referees can be compared with their own matches. Without spectators the same referee gave the away team 0.36 fewer extra cards a match, and 0.29 fewer for the same number of fouls. Experienced referees changed as much as newcomers, and those who had favoured home teams most changed no more than the rest.
The card gap closed, then came back with the fans
Change in away-minus-home yellow cards per match against 2015–19, 16 European leagues, by season.
How many fans, measured
Where leagues let in a capped crowd, home teams took 0.15 more points a match than in empty grounds in the same months. Recorded attendance for 27 leagues then gives the crowd at every match. The slider in the headlines reads from this chart.
Home points against normal crowds before COVID
Zero is a normal full house. Two separate sets of leagues, with 95% intervals.
11 leagues where the pattern was first seen16 unused leagues, test set in advance
Show the numbers
Bigger crowds, bigger losses
Across 594 clubs, one standard deviation more normal attendance meant 0.063 more points lost per ghost match, holding club strength fixed. Clubs that usually drew about 2,000 people lost nothing; above a few thousand, everyone lost about the same.
Ghost-game loss by the club's normal home crowd
Clubs in five equal groups by average attendance 2015/16–2018/19. Labels give each group's median crowd.
The bookmakers were slow to learn
Across the ghost games, betting odds cut the home team's chances by 3.0 percentage points, while home teams actually won 4.7 points less often. The gap, 1.6 points (p = 0.02), was gone by January 2021.
Home-win chance in ghost games: odds against results
16 European leagues, by month. Implied probability from bookmaker odds with the margin removed, against the share of home wins.
Home-win chance in the oddsHome wins that happened (95% interval)
The method
How the study was built, what the tests say, and where it stops.
Where it started
This began in June 2022 as an ECON 142 paper at Santa Clara University on one league, the Premier League, over two seasons. In 2026 it was rebuilt in Python: every table of the original is reproduced exactly from code, then the study was widened to 38 leagues and twelve seasons.
Compare the two teams in the same match
Spring 2020 changed football in many ways at once: five substitutes, crowded fixture lists, tired players. Any per-team before-and-after comparison mixes all of that with the crowd. A crowd, though, only cheers for one side. So every estimate here is a gap between the home and the away team in the same match, where anything that hit both teams cancels out.
The Premier League data show why this matters. Both teams pressed far more in the matches with fans than in those without, by about the same amount. A per-team comparison would credit that to the crowd; the within-match gap is close to zero.
Per-team changes, home against away
Premier League 2019/20–2020/21, per team and match: matches with fans minus matches without. Each row has its own scale.
Home teamAway team
Show the numbers
The models
The main estimate regresses the home-minus-away outcome on a ghost-game indicator, with league fixed effects and the pre-match Elo rating gap to hold team strength fixed. Robustness checks add team fixed effects, a season trend, referee fixed effects and a within-2019/20 comparison. Standard errors are clustered by league or league-season; across many outcomes, p-values are corrected with Benjamini–Hochberg.
Written down before it was run
Each analysis followed the same routine, and several patterns that look convincing in a chart ended up as "not supported".
- Look at where the useful variation lies, without looking at any outcome.
- Write the specification and the pass mark for every hypothesis.
- Commit it to the repository before running anything.
- Estimate, and report every pre-specified result, whatever it shows.
- Label anything decided afterwards as post hoc, and test it again on fresh data.
How firm is the 5,000-fan threshold?
The crowd-size pattern was first spotted in 11 leagues, after an earlier pre-specified design failed. It was then tested on 16 unused leagues with the test and pass mark fixed in advance. Crowds of 5,000 or more clearly beat empty stadiums (q = 0.002). Small crowds restored less than large ones by about the predicted amount, but q = 0.086 missed the 0.05 bar. The threshold is the most likely reading of the data, and it is not formally confirmed.
What the data can't tell us
Comparing leagues that reopened with leagues still closed rests on two control leagues and fails its own placebo test. Between March 2020 and June 2021 no league anywhere had full crowds.
A causal forest asked to rank clubs by the value of their fans could not do it: ghost games fell in only two seasons, and of 283 clubs only 4 had an estimate distinguishable from zero.
Small crowds in 2021/22 came mostly from small, lower-division clubs, so "few fans" and "small club" are hard to separate. Outside the 27 leagues with attendance, crowd status rests on documented rules, not measurement.