The Empty Cell Does Not Lie: Three Lessons in Reading Football Through Missing Data

Manchester, Monday morning. I opened the spreadsheet. Forty-six rows, twelve columns, and every cell in the xG column blank. The match had not been goalless; the parser had failed. From the next desk came the question every analyst dreads: "So what's the story?" The only honest answer was that there was no story yet, because the evidence had not arrived. He nodded and said, "Fine. At least write what the match was like." That was the trap.

"What the match was like" is the most dangerous question in this trade, because the answer is always pre-built in the mind. A blank scoreline means cautious, flat, both sides afraid. What the eye sees, what memory keeps, and what the report wants to say all push toward the same sentence. And there the empty column sat, a warning: you do not know yet, so do not write yet.

The distance between a zero and a missing value is where the ethics of football analysis live. A 0.00 xG means the team created nothing. An empty cell means I do not know whether they did. The first is a finding; the second is a confession. A model is a promise you keep to the future with the data you have today, and promises break in two ways — by lying, or by going quiet.

My own route folds into this oddly. In 2026 I finished a civil-engineering degree and went into journalism, first at Ajker Kagoj, later as founding managing editor of The Daily Star. Engineering installed a habit: when the numbers do not reconcile, you do not fill the gap with a plausible estimate, you go and find which cell is failing. Fifteen years of club analytics later, when I joined StatsBomb's Manchester office in 2026, that habit was all I had to offer.

That same year I consulted for Huddersfield Town's Championship play-off run. I built a standardised xG/PPDA dashboard across forty-six league matches — same columns, same definitions, same time windows for every fixture. The point was that the number should be in front of the eye before the decision, and should speak the same language every week. I built the xG template before Huddersfield made the numbers breathe; the structure existed while the story did not.

Aaron Mooy's numbers stood out on that dashboard: 2.8 shot-ending passes per 90, 0.18 xGChain per pass. Read together, they describe a specific role — he does not score, he manufactures the precondition for scoring. In the play-off final against Reading, the scoreboard read 0-0 after 120 minutes and went to penalties; Mooy completed seven progressive passes in that match. The scoreboard wrote goalless. My sheet wrote a different sentence, one the scoreboard never writes.

From that came a rule I have not broken since: every match report opens with numbers, and the prose arrives afterwards. Editors found it strange at first, then began demanding the same structure everywhere. The reason is simple. When the number is already on the page, the sentence can no longer hide behind "it felt like"; it has to explain why the number looks the way it does.

In 2026, on a World Cup data desk in Russia, that structure earned its keep. After Germany's 0-1 defeat to Mexico the stands talked about shock and bad luck. The dashboard showed something else: Germany's PPDA had been 7.8 in qualifying and stood at 12.4 against Mexico. They were being forced to string together far more passes before winning the ball back; the press was no longer arriving where it used to. Germany did not collapse in ninety minutes; the PPDA line had been rising for months.

In that same match, twenty-six shots produced 1.3 xG. Against South Korea, in a 0-2 defeat, field tilt was 68 per cent while open-play xG was 0.9. The ball was with them; danger inside the opponent's box was not. Across the tournament, eighteen German high turnovers led to zero goals. Eighteen is not a single-match story, it is a structural ledger. Since then I do not write the word "dominant" without field tilt and xG beside it.

A confession is due here. There is a gap between what was knowable in 2026 and what hindsight now reveals. At the time I could say the pressing tempo was dropping and shot quality was poor. I could not say they would exit in the group stage. It is satisfying to write "we told you so" after the fact, and it is not honest. Separating what was knowable then from what hindsight reveals is the whole discipline; retro-fitting today's knowledge into yesterday's note is self-promotion.

In 2026, consulting for Brighton & Hove Albion through Project Restart, another compulsory experiment arrived: empty stadiums. Auditing ninety-two Premier League matches behind closed doors, home advantage fell from 0.35 goals per game to 0.12. For Brighton's 2-1 win over Arsenal on 20 June 2026 I built a crowd-adjustment model that lowered Arsenal's expected home pressure by 18 per cent and raised Brighton's xG from 1.1 to 1.6. The empty stadium was a control group I never wanted, but it answered the question.

That is also where my own worst trap sat, and I do not hide it. Emptiness alone is not a cause. Fitness, motivation, fixture congestion, the effect of crowd noise on referees, even travel time all shifted at once. A model that cannot separate those variables is not science, it is a story with columns. So I published the confounder list alongside the model. Readers were irritated. In retrospect, the irritation was the most useful part of the exercise.

Back to the empty cell. Missing data in an analytics workflow has three possible meanings. One, the event never happened. Two, it happened but was not recorded — a coverage gap. Three, it was recorded but never reached my pipeline — a processing failure. Each has a different remedy, and the first has nothing in common with the other two. An analyst who cannot tell them apart is not an analyst; he is an author of estimates. So every report of mine now ends with a short line: coverage note, sample size, uncertainty range.

The industry's real problem is not bad data. It is confident data. Bad data gets caught, disproved, corrected. Confident data survives, because it is neatly formatted, carries two decimal places, and sits beside a credible name. Filling a blank is usually done politely — "approximate", "roughly", "in the region of". The outcome is identical: the reader mistakes an estimate for a measurement.

This is where possession statistics belong. Sixty per cent of the ball is not an achievement if it is mostly travelling sideways. The broadcast graphic shows possession and omits xG, yet xG is what decides matches. Possession is the most deceptive number in the game because it displays effort and hides progress.

One more reminder is needed periodically: correlation is not causation. Pressing intensity and results are related, but claiming pressing intensity produces results requires controlling for opponent quality, match state and sample size at minimum. When the press breaks, the pass map bleeds before the scoreboard does. That is my most quoted line and my most misused one.

A related point, not the subject of this piece. We celebrate lower-league runs like Huddersfield's, consume them, and forget them. Who asked about structural redistribution of resources after that season? Almost nobody. The story sells; the system stays as it was.

Something from esports is relevant here: reaction time is currency, and football is still learning the exchange rate. Pressing, recovery, second balls are all reaction-time accounting, and we still lack a decent measure of it. And to prevent misunderstanding — I do not hate football. My suspicion is reserved for those who treat numbers as a substitute for it.

So here is what I will be watching in the next round of fixtures. Which broadcaster or outlet publishes only the result, and which publishes its coverage note, its sample limits, its uncertainty band. That difference will set the standard of analysis over the next two years. The question is not complicated: are you looking at a number, or at a promise of one? And what evidence would change my mind? A complete, method-published, verifiable open dataset — one where an empty cell is labelled unknown, never zero.