The Null Result at Major Tournaments: When Prediction Models Choose Silence

On the night of July 11, 2026, after Croatia beat England 2-1 in the World Cup semi-final on Russian soil, I stayed in the office of a sports data company in Paris until nearly dawn. Outside, the city was still awake because the final was only three days away. Inside my room, the air was as cold as a laboratory. My analysis file ran to forty pages, packed with metrics: expected goals, passes into the final third, the PPDA index measuring pressing intensity, the sprint distance of every player. I ran the prediction model for the final between France and Croatia. The output came back cold: undecidable. Every core variable sat neatly within the noise band. The machine was not broken. It simply found nothing to say.

In my industry, we call it the null result. At major tournaments, null results appear more often than anyone wants to admit.

Major tournaments like the World Cup or the EURO are the harshest environment for any data analyst. The problem is not data quality — federations and service providers now collect everything, every metre run, every touch. The problem is that the sample is too small and the variance too large. A national team plays only three group matches, at most seven games in the whole tournament. Against a club season running thirty-eight rounds, that is a set almost impossible to draw statistical inference from. When the sample is small, a single missed penalty, a single red card, a single shot off the crossbar is enough to snap the entire model.

I always tell younger colleagues: numbers do not lie, but they hide the most important thing. In the knockout rounds, what gets hidden is usually the true gap between two teams — something a single match cannot reveal.

In 2026, I was assigned to write the daily tactical bulletin on Croatia. After the semi-final against England, I wrote a two-thousand-word piece arguing that Luka Modric was no wizard, but the product of a system with three centre-backs and two deep-lying midfielders. That system let him receive the ball an average of 9.4 times per match in the zone around the centre circle — a detail that appeared in no bulletin at the time. Magic is just the name we give to what we have not yet measured.

A colleague in the office mocked me for daring to demystify a star the whole world was worshipping. Three months later, the same man asked for my file back to cross-check it against France's pressing data in the final.

That lesson stayed with me throughout my career. In 2026, while I was a research assistant at Olympique de Marseille's La Commanderie training centre, I processed the GPS positioning data of right-back Hiroki Sakai across three consecutive weeks. His high-speed running distance had fallen 18% from the start of the season, while his average receiving position had retreated 7 metres deeper. I wrote a twelve-page report, but I did not point to any technical fault in Sakai. I focused on how head coach Rudi Garcia had switched the formation from 4-2-3-1 to 4-1-4-1, leaving the right flank exposed. The report was shelved for two weeks. Only when the team lost 0-3 to Monaco did the coaching staff re-examine all my data.

At the tactical level, there is a trade-off that data routinely ignores: pressing intensity is inversely proportional to the ability to sustain it across seven matches. A high-pressing team can win its first three games, but by the quarter-finals the legs no longer obey. A model calculated match by match cannot see accumulated fatigue, and so it predicts wrongly at exactly the most important stage. At the 2026 World Cup in Qatar, Saudi Arabia beat Argentina 2-1 in the group stage, while Japan defeated both Germany and Spain in turn. No model predicted those results, because they did not come from technical quality but from defensive organisation and a decisive moment.

In 2026, when European football was paralysed by the pandemic, the editorial desk asked me to write a nostalgic series about stadium atmosphere. I refused. I submitted an alternative proposal: build a dataset comparing match tempo and the number of risky passes when crowds were present versus absent, drawing on Ligue 2 matches that continued behind closed doors. The result: match tempo rose 6%, but risky passes into the final third fell 11%. Silence did not create cautious football. It laid bare the caution that coaches already carried. Football did not die when the stands were empty. It merely exposed its true skeleton.

But to stop there would be to fool myself. A null result does not mean there is no story. It means the story lives in the dark zone — where the model cannot see. The deviation is not the machine's fault, but something people choose not to look at.

Three dark zones loom largest at every major tournament.

First, referees and VAR. At the 2026 World Cup, VAR reviews averaged more than two minutes each. Two minutes is enough to cool a goal, enough for a defence to reorganise, enough to shred the rhythm of a match. No expected-goals model accounts for this variable, yet it directly shapes the final result.

Second, psychology. The pressure of a penalty in the eighty-eighth minute rarely concerns technique. It concerns fear — the fear of losing everything in a single instant. Data cannot measure fear, and so it often overlooks the most decisive factor.

Third, fitness and system. Gegenpressing was once the weapon of the big teams, but it has now been decoded. Mid-tier sides use sheer athleticism to turn football into a track event — running more, pressing denser, neutralising technical quality. A model built on individual skill no longer works when the match becomes a running race.

So when the next major tournament arrives, I will still run the model. But I will no longer place absolute trust in null results. I do not believe in miracles. I believe in properly collected data. And properly collected data at a major tournament must include the things that cannot be measured: fear, fatigue, and the moment a coach decides to change everything in the seventieth minute.

A metric that cannot answer the question why is mere decoration. The real question still hangs over every data table: what produced this number, and what is hiding behind it?