The Empty Cell Does Not Lie: The Discipline of Missing Data in Professional Sport

In June 2026, in an editing room in Mapo, Seoul, I opened a spreadsheet with 64 rows, one for each World Cup match. At the eleventh column I stopped. That column recorded the conversion rate of set-piece situations into goals, and 23 cells were empty. Not because I was lazy. The cells were empty because the video sources did not allow me to classify a phase as a direct free kick, a corner, or a long throw-in that an opposing defender headed clear.

The producer asked a very simple question: "So where is South Korea weak?" I could have answered in four seconds. I already had a line about character, a line about mentality, a line about physicality. But the correct answer at that moment was: "I don't know yet."

It took fourteen days of re-watching the footage before I dared to state the real figure: 1.9 percent against a tournament average of 4.1 percent. That "I don't know" cost me two weeks and nearly broke the schedule. It was also the only thing in the entire project that made the final segment stand up.

Sports analysis has travelled a long road in fifteen years. In 2026, a post-match press conference could still revolve around "spirit" and "character." Today, the analysis room of any professional club has at least one person who can write a query, one who can build a chart, and a camera system recording every player's position twenty-five times a second.

Esports is ahead of football here. A single LCK match generates tick-level data: positions, gold totals, item timings, movement vectors inside a fight. No other sport lets an analyst stand that close to the action.

That abundance creates a very specific trap. The fuller the spreadsheet, the harder it is to say "I don't know." An empty column looks like laziness. An incomplete conclusion looks like incompetence. Meanwhile, a confident wrong conclusion looks like expertise.

I follow LCK and K League at the same time, in two different time zones of the same profession. This regular season has made the problem sharper than usual: esports patch cycles are shortening while football calendars are thickening. An LCK team plays roughly thirty regular-season matches, while Riot ships a balance update about once every two weeks. A champion's win rate over the last ten games can be obsolete before the report is printed. In the K League the problem runs the other way: plenty of data, but slow-changing context, so people easily mistake a three-round trend for a law.

One hundred and forty-one matches nobody watched

In 2026, when the pandemic closed stadiums, I proposed a project tracking the K League. There were 141 matches played with no spectators in the stands. I had nothing special beyond time and patience, so I sat down and logged every one of them.

When I finished, three numbers surfaced. Home win rate fell from 46.3 percent to 34.7 percent. The draw rate rose by 7.2 percentage points. And at Seongnam FC, sponsorship income dropped 23 percent in a single season, because what the club sells to sponsors — the presence of a crowd — simply no longer existed.

What matters is how long it took me to believe those numbers. For the first fourteen matches I kept a competing hypothesis in my head: maybe away teams that year were simply stronger than usual. Only when I isolated matches between teams that had finished the previous season in the same tier could I discard it.

After every cluster of numbers, I force myself to write exactly one sentence about what the numbers cannot say. For these 141 matches, that sentence is: home advantage in the K League does not live in the grass, or in travel, but in noise. Referees whistle against the home side less, not because they are biased, but because they hear twenty thousand people reacting to every decision.

In an empty stadium, the goalkeeper's shout rings out like a tactical manifesto. And COVID-19 taught football that noise is not a crowd, and a crowd is not noise.

This is where I part company with most analyses I read. They use empty-stadium data to talk about fitness or motivation. Both are reasonable. But that data speaks loudest about something few want to hear: the referee is a variable influenced by the stands, and any prediction model that ignores this variable is measuring the wrong thing.

Forty-two goals and one empty column

Back to the 2026 spreadsheet. The World Cup in Russia produced 42 goals from set-piece situations. I once wrote a line I still use: the 42 set-piece goals at the 2026 World Cup are not about technique, they are about how a team reads the match. That line is true, but it is not enough.

When I broke the data apart, another correlation appeared: teams that scored the opening goal from a set piece won 78.2 percent of those matches. A figure of 78.2 percent is suspicious enough that it should have triggered a question about sample size. And it did. With only 42 matches in that group, the confidence interval is so wide that the true value could sit anywhere between 62 and 89 percent.

I still put it in the film, with one line of narration: this figure may be wrong. The producer initially wanted to cut it because it diluted the climax. We kept it. After broadcast, it was the single detail coaches quoted back to us most often.

The South Korea section was far clearer. South Korea converted 1.9 percent of set-piece situations into goals, against a tournament average of 4.1 percent. This is not a wide confidence interval. This is a gap of more than double, sustained across several matches, verifiable on video.

But stopping there would have repeated the exact error I had just warned against. A low conversion rate does not say the team takes poor set pieces. It says nobody on the team reads the space before the ball is struck. Reviewing South Korea's fourteen set-piece situations at that tournament, the common thread was that the receiving player was always standing where the defender had already guessed he would be, half a second earlier.

No index measures that half second. That is why one column in my spreadsheet had to stay empty, and why the finished segment ran ten minutes instead of three.

Forty-eight thousandths of a second

In 2026, while studying for a master's in sports management, I spent twenty days analysing 100m video of a sprinter, Kim Ji-hoon, who ran 10.24 seconds. I measured the left elbow angle across six starts. The average deviation was 14.2 degrees, and it cost him roughly 0.048 seconds each time.

The report ran fourteen pages, with data tables and a stride-cycle chart. A documentary producer read it and offered me a job.

What I learned was not about running technique. Six starts are six data points. With six points you can draw a trend line, but you cannot claim that line is a law. I wrote exactly that in the report: at least thirty starts are needed before any conclusion.

People usually think analysis is a profession of drawing conclusions. For me, the first half of my career was a profession of learning to refuse them. A 0.05-second slower start is sometimes the way to finish earlier. The best sprinter is not the strongest one, but the one who understands his own limits most clearly.

One transfer and a before-and-after test

In 2026 I was the first to report the loan of defender Park Ji-soo from Gwangju FC to a J-League club. I had no special inside source. I had a framework I had used many times before.

The framework said that if the new club pushed its defensive line higher, Park's interception count would rise, because he would be defending in more space instead of dropping deep. The result matched the calculation: average interceptions per match rose from 1.8 to 3.2, and pass accuracy from 72 percent to 85 percent.

There is one detail I never put in the film. Before publishing the prediction, I tested three other scenarios. In one, the new club kept a deep block, and in that case the projected numbers barely moved. I published the scenario that came true, but I had prepared for all three.

The transfer market resembles a 100m race: a successful deal is one that starts at the right moment, not the earliest one. This profession rewards people who guess right. It does not reward people who say they are only 60 percent certain.

Tick-level data and the trap of completeness

Esports is where this problem is most visible, because the data there is so full that people forget it still has holes.

An LCK team plays around thirty regular-season matches. When a team like T1 loses three in a row, analyses sprout like mushrooms, each with a chart and a conclusion. Very few of them say that three matches is too small a sample to conclude anything about the meta.

The problem gets worse at patch level. A champion can be buffed in an early-March patch and nerfed in a late-March patch. That champion's win rate between the two patches is a real number, but it does not measure the champion's strength. It measures the lag between a design change and a change in player behaviour.

I once saw a scouting report with a column headed "meta fit" for twelve players. All twelve cells had a score; none was blank. That is deeply suspicious. A report with no empty cells is usually a report filled in by the author's biases rather than by data.

Tick-level data measures behaviour, but not intent. When a mid laner retreats to defend in the eighteenth minute, the data logs a disengage. It does not log that he was holding position for a fight that would break out in the bottom lane twenty seconds later. The analyst has to choose: write "disengage," or leave it blank and wait. Most choose the first, because nobody pays for a blank.

The price of confidence

This industry rewards confidence, and that is why the empty cells disappear. An analyst who says "I need more data" gets filed under not good enough. An analyst who says "I believe" gets invited on air. That incentive structure does not come from audience ignorance. It comes from the nature of broadcasting: a show needs a conclusion before the clock runs out.

There is a paradox few in the industry will say out loud. A wrong conclusion costs more than a missing one, but only over the long run. In the short run, a wrong conclusion costs nothing. It can even pay, because it generates argument, and argument generates views.

I once sat in a meeting where an editor suggested cutting the entire section on sample size because "the audience doesn't care." He may have been right about a general audience. But I write for people who will use that conclusion to make a decision: a fitness coach, a scout, someone weighing whether to change a player's position. For those people, an honest empty cell is worth more than a pretty number.

When you see a sports analysis table with no empty cells, read it like the report card of a top student: beautiful, but possibly edited. The question I ask myself on every new project is no longer "what conclusion did I find," but "did I leave the right things blank." If the answer is no, I am probably writing an article, not an analysis.