The Empty Data Column in a Major Tournament Season: The Analyst's Discipline

At 21:47 on a late June night in Nha Trang, my spreadsheet held 4,118 rows of event data from a knockout round and exactly one empty column: the one recording when a player left the pitch injured. Without that column, the schedule-density model I had spent three weeks building was just a handsome table of numbers. I shut the laptop and published no prediction for the following round. The next morning, an acquaintance published a piece with a decisive conclusion about that same match. He had a conclusion. I had an empty column. After forty-eight years of watching sport, I have learned that an empty column is usually more honest than a rushed conclusion.

Two people watch the same match and look at two different things. One looks at the scoreline, and the scoreline always answers. The other looks at the structure of the information: what data exists, what data is missing, and whether the gap is large enough to collapse the whole conclusion. My work belongs to the second group, and most of the time it is unglamorous. It is counting, cross-checking, and, most often of all, saying: not enough information to assess.

In sports analysis, data gaps are not distributed evenly across categories; they cluster on the exact variables that decide matches. Over more than a decade in this trade I keep meeting three forms of the same gap.

The most common form is total absence. Injury status is the clearest case. News that a player left the pitch in the 63rd minute with a hamstring problem typically surfaces hours after the match, sometimes days later, and rarely with a severity grade. For anyone building a schedule-density model, that is the single most important column and also the emptiest one. When the calendar compresses in a knockout round — a match every three days — I need no further data to know that injury risk spikes. No medical department, however good, rescues a six-week stretch of two matches per week. That is arithmetic, and arithmetic does not negotiate.

More dangerous is mis-measurement, because it leaves no empty column behind. Instead it leaves a figure that looks entirely plausible. Gegenpressing is the example I have tracked longest. For roughly a decade, PPDA became the default measure of pressing intensity. When mid-table sides learned to press, they did not copy the system; they copied the physical threshold. PPDA improved while the quality of chances created barely moved. A team runs four extra kilometres per match and generates 0.1 additional xG. The spreadsheet says they press better. The pitch says they have turned football into athletics.

Hardest of all to catch is mis-attribution. The numbers are complete, the numbers are correct, but the cause is attached to the wrong thing. A side that wins four of five can be credited to a switch to a back three when the real driver is an easier fixture list. This is the line between analysis and storytelling, and it is where I have to be strictest with myself.

In 2026, on matchday 14 of the V-League, I calculated xG by hand in a spreadsheet after a game in which Hanoi FC held 71% possession and took 22 shots, then lost 1-2 away to Sanna Khanh Hoa. My numbers came out at 1.8 for Hanoi FC and 2.1 for the hosts. The article titled Possession Is Not Victory grew out of that night and was shared more than 3,000 times. What I took from it was not that xG beats the scoreline; it was that Vietnamese fans do not lack emotion — they lack a frame of reference. From then on, every piece of mine has carried at least two advanced metrics as its spine, not as decoration.

Three years later, when the pandemic stopped every league, I did work I normally have no time for: I collected 3,100 matches from Europe's top five leagues in 2026-2026 and calculated average home advantage at 0.42 xG. When the Bundesliga resumed in May 2026 in empty stadiums, I predicted the home win rate would fall from 43% to 27%, drew the chart and published it. Reality followed exactly. When football died, I realised my home-advantage model had grown roots in an assumption — and that assumption was called a crowd. Correct data resting on a wrong frame is the worst class of error.

The 2026 World Cup gave me the opposite lesson. I analysed Croatia's qualifying campaign by counting from video myself: the Modrić – Rakitić – Brozović trio completed 4,321 passes, and Modrić alone hit 87% pass accuracy while under pressure. I published a prediction that Croatia would reach the final; they did, and lost 2-4 to France. The piece drew 120,000 reads. Croatia 2026 taught me that a pass under pressure is a manifesto written in technique — and that this kind of data only exists if you count it yourself, because no off-the-shelf statistics package contains it.

A goal is only the verdict. xG is the testimony. Between those two things lies my entire trade, and its entire risk. An off-the-shelf stats table records events; it does not record the conditions that produced them. That shot, from that position, in that game state, under that pressure — three different variables yield three different conclusions from the same column of numbers.

Table tennis gives me a cleaner illustration. A player can hold a good ranking position on points defended at smaller events while performing poorly against the leading group. Read the ranking alone and you will misjudge that player's chances at a major. To assess properly you must separate three layers: accumulated points, the quality of opponents faced, and results in deciding matches. That last layer is almost always the least populated, because it needs a large enough sample before it means anything statistically.

This is where I part company with most people writing on the same subjects. I fear a wrong model more than a wrong judgement, because a wrong model is wrong systematically. A wrong judgement costs you one match. A wrong model contaminates every match it touches and, worse, leaves its users convinced they are standing on solid ground. When I read an analysis that uses three matches to assert a tactical law, I do not argue with the conclusion — I ask about sample size.

There is a paradox that makes data discipline hard to sell. Readers want conclusions; algorithms reward conclusions; and a piece saying not enough information to assess will almost always travel less far than one saying this team will win it all. But correlation is not causation, and a major tournament season is the perfect environment for the two to be blended. Six knockout matches is far too small a sample to separate nerve from luck. A missed penalty in the 88th minute tells you very little about a player's technique and a great deal about the fact that he was playing his fourth match in ten days. Data forgives no emotion. That is precisely why I converted.

I also have to own a professional temptation of my own. When an earlier analysis of mine is stress-tested by fresh numbers, my first reflex is to defend the old conclusion. The only way to last in this trade is to treat updated data as evidence that the system is working rather than as an indictment. I was wrong about home advantage in a season without crowds — wrong in a direction that was at least useful.

There is one more trap I see often in domestic analysis: bolting European templates directly onto Vietnamese football data. Tempo, fixture density, pitch quality, travel distances between rounds — all of it differs. Before comparing any metric with a European league, I set aside a block of work to recalculate that metric from source data gathered in Vietnam. Without that step, every comparison is translation, not analysis.

What I am waiting for in the next round is not a result box but whether the injury column gets filled. If it does, the schedule-density model will produce the first figure I am willing to publish. If it does not, I will sit back down with the spreadsheet, add a time column, and wait. In this trade, waiting at the right moment is a measurable skill — it is simply never measured in a stats table.