From 0.45 to 0.08 Goals a Match: How Home Advantage Vanished in the Season Without Crowds

In June 2026 I sat in front of a screen at two in the morning Sydney time, watching the Bundesliga return after a three-month suspension. The stands were draped in grey cloth, every seat empty. When the home side equalised in the 74th minute, no roar rolled down onto the grass, only the goalkeeper shouting at his team-mates through the pitch-side microphones, loud enough that I could hear studs grinding into turf.

My prediction model at the time, version 4.2, running on event data from the 2026/20 season, still priced home advantage at 0.45 goals per match. It was a figure I had carried for six years, almost nobody argued with it. After nine rounds without crowds, it fell to 0.08.

Numbers whisper. Whoever listens hears an entire match. That time, what I heard was the sound of my own error.

A single coefficient carrying too much

Football economics research over the past two decades typically places home advantage between 0.3 and 0.5 goals per match, depending on league and era. That range is wide because researchers are not measuring the same thing. Some measure goal difference, some measure points difference, some convert to win probability. Three definitions, three results, one name.

Before trusting a number, ask where it was born.

When I built my first model in 2026 for The Football Sack, I collapsed home advantage into one coefficient. Inside it sat at least five mechanisms: crowd pressure on referees, travel fatigue, familiarity with pitch and surface, climate and kick-off time, and the daily routine of players in the two days before a match. I folded all five into one weight and called it home advantage. That is the original sin of many football models, including models far more expensive than mine.

The trouble with a bundled coefficient is that you cannot tell which part is carrying the result. As long as crowds sit in the stands, the coefficient keeps predicting well, and you have no incentive to separate it. Only when one variable inside disappears do you discover you have been measuring the wrong thing.

For three years before that, I backtested the model on 2026 to 2026 data across five European leagues. Mean absolute error sat near 0.04 goals per match, and home advantage always swung between 0.42 and 0.48. That stability made me stop asking questions about it. A parameter that holds for three years gets treated as a law.

In March 2026, that variable disappeared across Europe.

Nine rounds and what collapsed first

I logged every round. The method was manual: pull event data after each match, isolate fixtures with a home team, calculate goal difference, xG difference, penalties awarded, yellow cards for away sides, and the home side's points rate. Nine rounds, nothing skipped.

The first thing to collapse was not goals. It was the metrics tied to the whistle.

In the first round after the restart, average goal difference still favoured home teams. By round two, near zero. By round three, negative. By round five I had split the variables into two groups: those tied to player behaviour and those tied to referee decisions. Only the second collapsed cleanly.

The rate at which home teams were awarded penalties fell sharply against the same period a season earlier. Yellow cards for away sides fell too. The xG gap between home and away narrowed but did not vanish, and the remainder came mostly from squad quality rather than from the stands. Total shots, possession share and wide attacking entries barely moved.

In other words: home teams did not get worse at football. What changed was how the match was governed.

I have written before that referees are gradually becoming the editors of a match. Those nine rounds reinforced the claim in a way I did not expect. Remove the crowd from the equation and the part of the advantage that vanishes fastest sits in decisions the eye struggles to see: a midfield duel, a yellow card for an away midfielder in the 30th minute, a penalty in the 78th.

Research on crowd effects on officials has a long tradition; in Germany and Spain the number of yellow cards shown to away teams rises as attendance rises. I could not replicate that study across Europe, but my nine rounds point the same way.

A season missing detail is like a match missing stoppage time. You still have a scoreline, but you lose the deciding part.

In June 2026 I turned down a magazine commission to explain empty-stadium football. I needed three more weeks of data, and I said so plainly to the editor. Nine rounds is a small sample, and a small sample with high variance can generate stories that are compelling and entirely wrong.

I remember sitting over a spreadsheet at four in the morning and realising I had taught the model something that does not exist. Nothing dramatic. Just a column changing colour.

When the piece finally ran, it opened with an admission: the model was wrong because I never separated the crowd variable from the home coefficient. Not because data was missing. Because I was overconfident about a parameter that had held for six years.

Misreading a single variable is like losing your bearings for an entire year.

2026: the same mistake at a smaller scale

In 2026, aged 25, I worked as a data analyst for The Football Sack, a newly founded Australian football outlet. When the A-League reached round 12, I published a 3,200-word analysis of Melbourne City's pressing metrics using GPS positional data the club released in a technical briefing.

The finding: under coach Warren Joyce, the team pressed in the wrong direction. Midfielder Luke Brattan covered 11.2 kilometres per match but produced only 1.3 successful tackles. He ran a lot, ran to the right places, and ran towards where the ball was not.

Fans mocked the piece for being dry. Three weeks later Joyce changed the pressing structure. Melbourne City won four matches in a row.

What I learned was not that data is always right. Data is right only when placed correctly in a causal chain. Had I merged distance covered and pressing effectiveness into one index, I would have missed the whole story.

World Cup 2026: when xG was treated as a gimmick

At the 2026 World Cup they laughed at my xG. This year they ask me what xG is.

That year I wrote an English-language piece for a small data blog predicting Croatia would reach the semi-finals, based on Luka Modric's chance-creation output: an average of 2.4 xG per match in the group stage. A group of amateur coaches on Reddit called me a bookworm who did not understand football. Croatia reached the final.

After the tournament a journalist from The Athletic got in touch to ask how I calculated defensive xG prevented for defenders. I spent two weeks writing Python, cross-checking against StatsBomb data, and sent back a seventeen-page analysis. It contained a section I still keep today: a list of assumptions that could be wrong.

That is why I never publish a conclusion from a single data source. Every metric has a producer, a definition, a version. StatsBomb xG differs from Opta xG in how it handles counter-attacks and blocked shots. Nobody is wrong. They are simply counting different things.

The stands are not one variable

The collapse of home advantage in the crowdless season is among the most cited findings of 2026 and 2026. It is also among the most abused.

When you take the crowd out of the stadium, you do not remove one variable. You remove a bundle: pre-match ritual, travel routine, crowd pressure on young players, gate revenue, local media pressure, and a sense of belonging to a place. I have followed football from Vietnam, where I was born, to Australia, where I live. That distance taught me that home is largely a psychological state, and a psychological state cannot be captured by one coefficient.

One research group published a conclusion that home advantage is dead. I do not believe it. Their sample is small, their measurement windows differ, and they compare crowdless fixtures with attended fixtures from different seasons, meaning two different operating systems. Correlation is not causation, and that remains the biggest trap in sports analysis.

What I do believe, on current data, is that home advantage was not destroyed but redistributed. The crowd-linked component shrank to near zero. The components tied to pitch, climate and travel schedules remain intact.

What I want readers to take away is not 0.08. That value only means something under the specific conditions of the first nine post-pandemic rounds, with a compressed calendar, more substitutions, and a summer of distancing.

VAR and millimetre lines

At the same time, another variable was changing matches in the opposite direction: VAR.

I am not against VAR. I am against how it is being operated.

Millimetre offside lines are slowly killing attacking instinct. A striker times a run well within half a second, creates space for a team-mate, and is pulled back because a toe crossed a line. Nobody in the stadium understands what is happening, and spectators learn a dangerous lesson: do not celebrate too early.

The problem is philosophical, not technological. The offside law exists to stop players camping near goal, not to determine the shoulder position of a defender moving at 30 kilometres an hour using a 50-frame-per-second camera. When the measurement error of the tool is larger than the threshold the law aims at, adjudication stops being fair. It becomes editing.

And when the referee is an editor, players learn to write for the editor. They stop learning how to score.

On another front, I think goalkeeper distribution is over-mythologised. A keeper who passes well but whose basic reflexes have declined still commands a high transfer fee, because passing metrics are easier to measure and easier to show off than reflex metrics. Transfer value is a story, but data is the signature. The market is reading the wrong page.

Three assumptions that could be wrong

After the 2026 season I added a mandatory section to everything I write.

First, my nine-round dataset comes from commercial event feeds, not full positional tracking. Off-ball situations, which occupy more than 80 percent of a match by duration, are reconstructed only indirectly.

Second, I cannot separate the crowd effect from the effect of a compressed calendar and the substitution rule. Three variables changing inside one window is an identification problem, and I have not solved it.

Third, 0.08 may still overstate the true value, because I pooled matches with limited attendance into the crowdless group.

Signals for the next round of fixtures

A major tournament cycle is approaching, and European domestic leagues have reopened their stands to differing degrees. I am tracking three signals.

One is the rate at which home teams are awarded penalties. If that rate returns to pre-2026 levels in leagues fully open to crowds but not in partially open leagues, we have stronger evidence that the crowd works mainly through the refereeing channel.

Two is the rate of yellow cards for away teams in the first 30 minutes. That indicator is more sensitive to crowd pressure than goal difference, because it is measured before the result takes shape.

Three is stoppage time. A packed stadium usually produces more added minutes. If that difference disappears, part of the pressure on referees has not returned even though the crowds have.

What is missing from this picture is full positional tracking at club level. Without it I can measure outcomes but not processes. And process is where home advantage is manufactured.

Based on my experience following matches, home advantage is on its way back, slower than I expected, and uneven across leagues. Current data does not let me say when it will touch 0.45 again.

If the noise of a crowd truly influences the person holding the whistle more than the person striking the ball, then every theory of home spirit we have passed between us for years needs to be rewritten from scratch.