Empty Data, False Conclusions: The Weakest Link in Sports Analytics

Summer 2026, I sat in front of a screen and realized: the ball is not the most readable thing. I was building a small model for the English Premier League, and one night my data table came up empty. Not a few missing rows — completely empty. The pressing column had nothing, the passing column had nothing, even the team names were gone. I stared at a blank spreadsheet and asked myself: if I just kept writing, what would happen?

The answer came faster than I expected. I would invent a team. I would invent a player. I would build a story that sounded reasonable, convincing, and entirely false. That was the moment I understood that in this industry, the greatest enemy is not ignorance — it is confidence built on an empty block of data.

Modern sports analytics runs on an almost religious belief: if there is data, there is truth. Clubs set up entire data-science departments. Bookmakers price odds in real time. Broadcasters put xG, PPDA and plus-minus metrics on screen as if they were gospel. But in the middle of that chain — between the camera filming the match and the number appearing on the odds board — sits a link few want to mention: the data-entry stage.

This stage is not glamorous. It is the work of people retyping every event: who passed, to whom, in which minute, under how much pressure. A small error here does not explode immediately. It flows downstream, is reshaped through dozens of layers of calculation, and steps out into the world as a number that looks highly professional. That is why I always tell colleagues: the most dangerous thing in sports analytics is not bad data, but empty data dressed up as good data.

To understand why this link matters, look at how an odds line is born. A match ends. Event data is pushed into the system. The model calculates attacking strength, defensive strength, home advantage, recent form. The bookmaker adds a margin. The final number appears on screen. The bettor looks at that number and believes it is the distillation of all knowledge about the match. They do not see that at the head of the chain, someone may have mistyped a play, or worse, left a data column blank because they could not collect it in time.

I did not enter this industry out of love for beautiful plays. People enter the industry because they love football. I entered because I wanted to prove that luck is just a form of data poverty. But precisely for that reason, I learned something contrary to instinct: when data disappears, an honest analyst must stop, not fill the gap with imagination.

Take an example I once dissected very carefully. In the 2026-2026 season, Burnley's xG model produced an actual figure of about 36.2 goals, against an expected 44.8. That gap of nearly 8.6 goals said Burnley scored far fewer than the quality of chances they created. Read only the table, and you see an ordinary survival team. Read xG, and you see a team whose attacking structure was better than its position — the basis for predicting it would not collapse the following season. A single number cannot say that. Every isolated number is a lie. Only when you place them side by side does the truth begin to vomit itself out.

But the xG story is only credible when the underlying event data is credible. If the entry stage records a shot as a pass, the model still runs, still produces numbers, still looks real. It is just wrong. And that error gets packaged, sent to the bookmaker, sent to the bettor, and becomes a skewed odds line.

I once built a small test for myself: take a dataset, deliberately delete 5% of events, then rerun the model. The result did not collapse. It only drifted. One team was rated stronger than reality, another weaker. And the drift sat within a range the naked eye cannot detect. Empty data is more dangerous than wrong data, because wrong data often exposes itself through contradiction, while missing data stays silent.

This is where I want to raise a deeper variable. The live data that betting companies collect is becoming the darkest side effect of sports digitization. Every pass, every breath of a player, every second of muscle strain can be turned into a data point and sold. Fans think they are watching a match. In reality, they are watching a mass data-collection session, and they themselves are part of the data.

When I talk to people in the trade, I always ask the same question: where did this data come from, who entered it, when, and who double-checked it? Very few can answer all four parts. Most only know it sits in some file, provided by some system. They trust the shell of the data, not its origin.

In daily work, I learned to separate two things that are easily confused: available data and valid data. A file can be packed with numbers and still be useless, if those numbers were collected the wrong way or inconsistently. Another file may be smaller but absolutely trustworthy. Between the two, I always choose the smaller one. Because in analytics, it is better to know little and be right than to know much and be wrong.

This problem is not confined to football. I follow basketball too, and there it is even clearer. A player returning from an anterior cruciate ligament injury may have numbers that look fine on paper, but those numbers cannot measure the fear in his head when he has to change direction. The stat sheet records the jump, not the hesitation before the jump. Rushing a player back after ACL reconstruction is destroying the second phase of their career, and psychological fear is far harder to repair than the body. But if your model has no variable for fear, it will assume fear does not exist. Empty data, once again, is read as no problem.

At Euro 2026, I was once tasked with assessing Denmark's potential after Christian Eriksen's incident. Their injury data and pressing history — an average PPDA of 8.7, the lowest in the group stage — showed their proactive defensive structure remained intact. The model recommended Denmark to advance from the group at odds of 4.75, and they reached the semifinals. The key point was not the correct prediction. The key point was that I only dared to draw a conclusion because their structural data was thick enough to back it.

I also look at esports, where the speed of digitization is so fast that the verification stage is almost left behind. Professionalization is turning players into the product of an assembly line, where every action is measured, every error logged, and individual style is sanded smooth to fit the digital training model. When everything becomes a metric, what cannot be measured is treated as if it does not exist — even while it is still happening on the field.

I am not against digitization. I am against digitization that forgets the verification stage. A model is only as good as its input data, and a process is only as good as its weakest link. When the whole industry races forward for more metrics, very few turn back to ask: are these metrics real, and are they enough.

Here I have to bend myself back. For years I was the one shouting "bring data into every corner." But a lesson from the strange summer of 2026 made me revisit that entire belief.

When European football leagues returned in empty stadiums, I spent six months processing data. Home advantage collapsed. The average figure, once 1.32 points per home match, fell to 1.08. Borussia Mönchengladbach lost 7 of 12 available points at home after the ball rolled again. Empty stadiums, yet there had never been so much clean data. The pandemic was a toxic gift. It pulled the crowd out of the equation and exposed a variable long hidden: how many points the roar of a crowd is worth.

But if I rushed to conclude that no crowd means no home advantage, I would fall into the very trap I always warn about. Correlation is not causation. An empty stadium is one variable, but it came bundled with a packed schedule, more substitutions, and unsettled player psychology. I cannot isolate a single cause from a bundle of variables tied together. A weak analyst picks one number, builds a story, and calls it truth. A clear-headed analyst must be able to say the hardest sentence: I do not know yet.

That is also why I do not watch a match the usual way. I do not watch the match. I watch the crowd betting on the match. And the crowd usually does not read data; it reads the story built from data. When an analysis sheet steps out with beautiful numbers but empty roots, the crowd still believes. That belief itself becomes a flow of money, and that money flows back into the market. Euro 2026 taught me one thing: no one pays to predict correctly. They pay to believe they are predicting correctly. An analytics system that generates empty belief is more dangerous than one that generates nothing at all.

Back to the night of the empty data table. The right thing I could do was not to keep writing. The right thing was to stop, mark it clearly as insufficient information, and send the data back to where it came from for checking. In an industry where everyone wants speed, the act of stopping is the hardest and most valuable act.

The signal I am tracking for the next cycle is simple: the share of analyses with verified data provenance. When that number rises, the market gets cleaner. When it falls, we will see more beautiful conclusions built on empty data columns — and those conclusions, sooner or later, will be refuted by the match itself.

The question I leave for readers today: the last time you looked at a sports number, did you know where it came from?