When Data Goes Silent: The Biggest Gap in Vietnamese Sports Analytics

Late in the evening of 22 November 2026, in a small apartment on Lach Tray Street in Hai Phong, I reopened a spreadsheet I had been building for four years. One row was labelled "Argentina - Saudi Arabia". The probability column read 94 percent. The expected score column read 3-0. Three days earlier I had sent that analysis to my editors with a closing line so certain that reading it back now still makes my face burn.

Two hours later, the scoreboard read 1-2. Argentina lost.

What kept me awake that night was not the score. It was the seventh column in the spreadsheet, the one I had named "pitch surface temperature". I had created it, left it sitting there, and never filled in a single cell. Across four years of collecting qualifying data, possession figures, key passes and shot frequency, I had never once entered temperature, humidity or air pressure into the model. The column was empty. And it was that empty column that beat me, not the deep defensive block of Saudi Arabia.

Afterwards I sat down for two weeks and rewatched forty-seven matches from Gulf-region competitions across ten years. The clearer it became, the simpler the lesson: my problem was not the model. The model ran correctly. My problem was that I had built a data pipeline that ran smoothly, and that pipeline had no valve for a variable I had never thought to consider.

I used to think I was right. Qatar taught me I was wrong.

The machine behind a number

A single match in Europe's top leagues now generates between one and a half and three million data points, depending on whether the provider uses motion-tracking systems or event recording alone. That number sounds large, but what matters is the structure inside it.

The lowest layer is event data. Every pass, every shot, every duel is typed into a system in real time by one or two people sitting in front of a monitor. Opta, Wyscout and StatsBomb all operate this way. A single match can cost them two hours of manual tagging, and the disagreement rate between two independent taggers typically lands between ten and fifteen percent on complex actions such as fifty-fifty duels or attributing who created a chance.

The middle layer is positional data. Between fourteen and twenty-two cameras around the pitch record the coordinates of players and the ball at ten to twenty-five frames per second. This layer produces the metrics Vietnamese fans have heard about in recent years: space control, opportunity value, line-breaking pass rates. Without this layer, most modern tactical analysis would not exist.

The top layer is biological data. GPS vests sampling at ten hertz, accelerometers, heart-rate sensors. In Europe's top leagues, every starting player carries three to four data-recording devices for the full ninety minutes.

Those three layers flow into a pipeline with five stages: collection, cleaning, tagging, modelling, interpretation. Books and conferences devote almost all their time to the last two. Speakers talk about machine learning, neural networks, injury-prediction models. Very few talk about the first stage.

That is why I call myself a data sceptic. Colleagues notice that I never conclude from a single metric, and they assume it is a personality trait. In truth it is the result of an occupational wound, dating back to June 2026.

That year I was twenty-five, working as an analysis assistant for a young sports outlet in Hai Phong. During the World Cup group match between Switzerland and Serbia, I found that Granit Xhaka had touched the ball one hundred and twelve times but that only thirty-four percent of his passes went forward. I wrote a piece criticising an excessively safe playing style. Three days later, coach Petkovic told the press that football is not mathematics. The same day, Switzerland came from behind to win 2-1 thanks to eight decisive passes in the second half.

I sat down and took the match apart, and I found where I had gone wrong. I had looked only at pass share and ignored PPDA, the number of opponent passes allowed per defensive action. Serbia sat near the bottom of that table. They applied no pressure, so Xhaka had time on the ball, and sideways passes were the rational choice in that game state. I had read a correct number inside an incorrect frame.

Since then I have imposed one rule on myself: before publishing any analysis, check at least five underlying metrics. Never conclude from a single number.

Three kinds of silence

When people talk about data errors, they usually think of wrong numbers. A misrecorded metric, a pass attributed to the wrong player, a goal counted twice. Those errors are loud. They show up on the scoreboard and get caught within hours.

The more dangerous class is silent. And sports data goes silent in three ways.

The most common is missing data. A column left unfilled. A match with no positional record because the stadium lacks cameras. A player who forgot to wear his GPS vest. In most pipelines, missing values are automatically converted to zero. A player who ran no metres becomes the slowest player on the pitch. That is how a gap turns into a wrong conclusion without anyone noticing.

A subtler form is late data. Many live feeds carry latency of thirty seconds to several minutes. For a model running in real time, a pass arriving thirty seconds late means the model is computing on a match that has already moved on. In large betting markets, that delay is enough to give an edge to whoever receives the faster feed.

And there is the form only the people in the tagging room know about: data drift. The definition of a metric changes mid-season while the metric's name stays the same. A provider reclassifies clearances and interceptions. A defender's defensive numbers jump after round fifteen, and nobody on the coaching staff realises it is not player improvement but a change in the tagging room. Drift raises no error. It produces no anomalous value. It simply shifts the baseline, and every comparison between the two halves of a season becomes meaningless.

The empty column in Qatar

Back to my own story.

The Saudi Arabia against Argentina match kicked off in the local afternoon, in mid-November. Outdoor temperatures exceeded thirty degrees. The stadium had a cooling system, but that system did not cool a player's thigh muscles during the twenty-minute warm-up, and it did not cool the air in the technical area.

In the first half, Argentina's attack was caught offside ten times. Seven of those involved Lionel Messi or the runner behind him. That was a figure never before seen in a World Cup group match. The Saudi coaching staff had prepared that trap for weeks and adjusted it phase by phase, based on the opponent's running rhythm.

What I missed was not the offside trap. I had data on offside traps. What I missed was the effect of heat on repeated sprint ability. Players sprinting in hot, humid conditions lose speed faster, and losing speed means losing distance, and losing distance means falling into the trap more often. That variable lived in a column I had never created.

Afterwards I went back through forty-seven matches in Gulf competitions across ten years and rewrote my entire process. Every pre-match analysis I produce now contains a geography block: temperature, humidity, altitude above sea level, sunset time, and the number of days a player has already spent in similar conditions. I also report a ninety-five percent confidence interval rather than a single number.

My readers do not know that. They only notice that I stopped writing "will win" and switched to "the probability falls within this range". But that is the entire difference between someone who reads numbers and someone who is accountable for them.

Every number is a confession, if we are patient enough to listen.

The Empty Stadium Index and the value of a crisis

In 2026, at twenty-eight, I was working as a data coordinator for a club in Ho Chi Minh City. World football stopped. When European leagues returned, they returned to empty stadiums.

Our team of three selected two hundred matches from the Portuguese and Danish leagues in the post-restart period to build a new metric set. We called it the Empty Stadium Index.

The first result surprised us. Central midfielders' distance covered fell 9.7 percent in the first month without crowds. But line-breaking passes rose 13.2 percent. In other words, central midfielders ran less but passed more riskily.

Our hypothesis was that crowd noise is a pressure signal. When the stands fall silent, a player no longer hears the jeers after a misplaced pass, and social pressure drops. They dare to play into tighter gaps.

Club leadership doubted the model. I spent three weeks presenting it again, and eventually persuaded them to sign a Brazilian midfielder based on this metric profile. After ten rounds he had scored four goals and assisted three, including one from a fast counter-attack exactly as the model predicted. The club climbed six places in the table.

We did not invent that metric in an air-conditioned office. We invented it because the world was in crisis and we were forced to find a new variable to understand what was happening.

New metric sets are not born in offices. They are born in crises.

When the stadium is empty, only the data whispers the truth.

Transfer valuation and the gap between numbers

There is one field where the empty-column problem costs real money: transfer valuation.

A Vietnamese club wants to sign an attacking midfielder. They receive a profile containing goals, assists, key passes per match and pass completion rate. All four are easy to obtain. But none of them answers the question the coaching staff actually needs answered: how does this player perform when his team is trailing and the opponent has dropped ten men behind the ball.

Answering that requires data on the opponent's defensive structure in each phase, on the position of defenders when this player receives the ball, on how often he receives under high pressure. That data costs money and time to collect. So it never enters the profile, and the transfer is decided by the four easiest metrics available.

A transfer is not a calculation. It is a negotiation between people and numbers.

I once sat in a meeting where three different metrics for the same player were presented, each from a different provider, and nobody asked why they did not match. The spread between the three sources reached nearly twenty percent on key passes. Twenty percent on a three-year contract is real money, and nobody in that room treated it as a data problem. They treated it as a player problem.

Technology and the limits of technology

The 2026 World Cup introduced semi-automated offside technology. The ball carried an inertial sensor sampling at five hundred hertz. Twelve cameras tracked twenty-nine points on each player's body, fifty times per second. It sounded like a perfect solution to a controversy that had lasted a century.

But the system still needs a human to determine the moment of contact, and it still applies only to offside. Penalty-area duels, handball incidents and fouls in the build-up to a goal still depend on the referee's judgement and the VAR room. Put another way, a tiny fraction of the match is now measured by the most precise technology ever deployed, while most of the rest remains a grey zone.

This is the clearest expression of the problem I am describing. Enormous resources go into measuring the most measurable thing with extreme precision, while the harder things are left outside the system. The more precise the data, the greater the sense of safety, and that sense of safety hides the columns that are still empty.

V.League and the small-sample problem

In Vietnam, the problem has an extra layer.

Based on my experience watching V.League matches across many seasons, the data infrastructure here differs in kind from Europe's top leagues. A season has thirteen to fourteen teams and roughly twenty-six rounds. The total number of matches in a season is less than a quarter of a major league's. A small sample means any model run on this data carries very wide confidence intervals.

On top of that, most data is still tagged manually. That is entirely normal for leagues with limited budgets, but it imposes a requirement few people notice: tagging error, which is mere noise in a thirty-eight-round league, can become a false signal in a twenty-six-round league.

I once saw an internal report conclude that one team passed the ball far longer than the rest of the league, based on average pass length. On inspection, one of the sample matches was missing data from the first twenty minutes of the second half, and the system had filled the gap with a default value. An entire tactical conclusion was built on a twenty-minute hole.

Applying top-league standards to a smaller league is another category of error. The PPDA threshold considered good pressing in Europe assumes teams share the same physical baseline and the same pitch quality. In a league with uneven pitches and a denser schedule, the same pressure number can mean something entirely different. I always check the origin, the collection method and the vintage of any foreign metric before using it in writing about Vietnamese football.

There is one more thing Vietnamese clubs rarely account for. A player moving from one league to another does not just change competitive environment; he changes data-recording systems. His distance covered measured by his old club's devices and by his new club's devices can differ by several percent, not because he runs differently but because the noise-filtering algorithms differ. When a report says a player runs more than last season, the reader needs to know who measured it and how.

Basketball and the same disease

I have written about the NBA for years, so I can see clearly that this disease is not limited to football.

The NBA has the best positional-data infrastructure on the planet.

The tracking system records the coordinates of every player and the ball at twenty-five frames per second, in every game of the season. That dataset changed how people price space, how they measure the value of a screen, and how they evaluate a three-point shooter.

But load data is different. Teams track player workload using different systems, and each system defines a high-intensity minute differently. Some count by speed threshold, some by acceleration threshold, some by time sustained above a power level. When a player changes teams, his load data changes definition, and every comparison between two seasons is skewed.

For years I followed internal injury reports that leaked into the public domain. The common thread in most of them was an injury-prediction model built on non-uniform load data, with conclusions presented as if the data were uniform. No model ever reports that its own data source has drifted.

Where the blind spot sits

If I had to point at one place where sports analytics is systematically wrong, I would point at the supply stage, not the modelling stage.

A good model running on incomplete data produces confident output. That is the worst thing a model can do. It does not crash, it raises no error, it produces no absurd value. It simply returns a neat number, and that number gets printed in newspapers, inserted into reports, and used to make transfer decisions.

Conversely, a mediocre model running on complete data will at least fail honestly. It fails at the level of the model, not at the level of the data.

There is an incentive paradox here. Auditing a data source generates no headlines. Nobody writes a story about a column left blank. Meanwhile a new model with a grand name always has a place in the press. The result is an industry that pours resources into making models look smarter instead of making data more complete.

In Vietnam the gap is wider still. A club may spend money on analysis software without anyone being responsible for checking whether the input data is correct. An analysis room may have three staff, none of whom is assigned to data inventory. When nobody does that job, the gap persists forever, because gaps do not announce themselves.

Numbers do not lie, but the people who choose them do.

What to do before trusting a number

From everything I have been through, I have drawn a few principles, and I apply them every week.

Inventory the data before building the model. Write down what you have, what you do not have, and what you know you cannot have. The list of what is missing is usually worth more than the list of what is present.

Ask about definitions before asking about values. A metric is only meaningful if its definition is stable across the period being compared. If the provider changed its tagging method mid-season, every chart is meaningless.

Distinguish an empty cell from a zero. In every dataset I receive, my first action is to count the empty cells and flag them. If a system automatically fills empty cells with zero, I ask for that function to be switched off.

And state the limits inside the piece itself. I no longer present a single probability. I present a range, and I state clearly which missing data makes that range wide.

These steps are simple, and they require no expensive software. They require only a habit: suspect the structure of the data before suspecting the model's conclusion.

Data is a mirror. Do not get angry when it reflects an ugly truth.

A question to leave behind

My problem in Qatar was not a bad model. It was a good model running on a spreadsheet missing one column.

I tell this story again not to apologise once more. I tell it because I believe sports analytics in Vietnam stands at exactly that fork. We can keep racing towards ever more complex models, or we can spend part of that time answering a far simpler question: is our data source telling the truth.

I may be wrong, and this is the assumption I am working with: that the biggest gap in Vietnamese sports analytics is not in the algorithms, but in the columns nobody has bothered to fill.

Have you ever checked what your own spreadsheet is missing?