The Empty Cell in Badminton Statistics: Where Matches Are Decided and Nobody Measures

On a domestic badminton night, the arena screen lights up with the post-match stat sheet. Fastest smash. Longest rally. Points won on serve. Unforced errors. Net winners. The eleventh column is empty.

Nobody in the stands knows what that column is for, and nobody asks. It is reserved for the distribution of rally lengths by game.

Fans leave the arena feeling the match has been fully explained. But the hardest questions in badminton sit outside the stat sheet: why a player who dominates game one collapses in game three, why a pattern that works for thirty minutes becomes a burden in the final twenty, and why two broadcasters covering the same match publish two different numbers for the same metric.

Nine years ago, in Guangzhou, I was stuck with exactly that empty cell.

The error is never in the scoreboard. It is in the place nobody bothers to check.

In 2026 I was twenty-three, a reporter for a young sports outlet in Guangzhou. Chinese Super League, round 15, Guangzhou Evergrande versus Shanghai SIPG. I pulled public GPS tracking data and calculated Paulinho's distance covered: 12.8 km. The figure the club published for the same match was about 15% lower.

I wrote the story. A male commentator said in front of the whole newsroom: "What does a girl know about data?" I asked for a face-to-face review. I brought time-series charts, a phase-by-phase comparison, and three independent sources. By the end of that meeting, the club admitted its statistical system had flaws.

The lesson was not that I was right. It was how close I came to being wrong: with a single source, I would have had nothing to defend.

So I built a three-layer protocol for every number before it enters a piece. Layer one, provenance: which device produced it, who entered it, under which definition. Layer two, reliability: has it been cross-checked by at least two independent sources. Layer three, context: what does it measure, under what conditions, and what does it fail to measure.

That sounds simple. But most sports arguments I have witnessed over nine years trace back to someone skipping layer three.

Badminton multiplies the problem. The public data ecosystem is far thinner than football's. The world federation publishes a tidy post-match set: points won, longest rally, top smash speed, serve-point conversion. The things that decide a third game — rally tempo, recovery time between rallies, actual movement distance, changes of direction — almost never appear in any report a Vietnamese audience can access.

As a result, most badminton analysis here runs on three inputs: a commentator's memory, a feeling about "form", and phrases like "this player is peaking". All three have value. None can be verified, and therefore none can be wrong. An argument that cannot be wrong cannot improve.

In football they call it luck. In data, I call it an uncontrolled variable.

The empty cell is never neutral

Statistics recognises three kinds of missing data. The first is missing completely at random: a dead sensor, a corrupted file, unrelated to the phenomenon. The second is missing depending on variables you can observe. The third is missing depending on the value that is missing: nobody measures the thing whose measurement would be inconvenient.

In sport, the third kind accounts for almost everything.

Clubs publish distance covered but not high-intensity sprint counts, because those numbers speak directly to injury risk. Broadcasters publish top smash speed because it is beautiful, but not average recovery time between rallies, because that shows who is fading in game two. Organisers publish total rallies but not the rally-length distribution, because that distribution shows where a match was steered.

The key point: missing sports data almost always depends on the missing value itself. It is missing because someone decided not to measure it.

Once you see that, an empty cell stops being empty. It becomes a decision.

I once compared two datasets from the same domestic badminton event. One counted "rallies" as every shuttle crossing the net. The other counted "rallies" as every sequence ending in a point. Two definitions, numbers more than 40% apart, both labelled "rally count". Nobody was wrong. Nobody had written down the definition.

That is layer three, and it is the most skipped layer of all.

The evidence chain: four times a number changed the verdict

In Guangzhou in 2026, the equipment was not the problem. The GPS devices produced stable raw data. The problem was the definition of "distance covered". The club counted light jogging and hand-corrected off-ball movement. The public dataset counted only movement above a set velocity threshold. Same player, same match, two definitions, a 15% gap.

It took three days to find. Three days of technical documents, phase-by-phase comparison, rebuilding the time series. When I presented it, I did not say the club was wrong. I said both sides were measuring different things and calling them by the same name.

Numbers do not lie. The people writing them down do.

In 2026, during the World Cup group stage in Russia, I worked with a different metric: PPDA — the passes an opponent completes per defensive action, measured in the opponent's 60% of the pitch. Lower means more aggressive pressing.

I tracked Germany across the group stage. In the opener their PPDA sat at 9.2 — high pressure still intact. By the final group game against South Korea it had drifted to 11.5. The pressing front had weakened sharply, and the space behind the midfield was widening.

I predicted Germany would lose to transition counterattacks. The editor waved it away: "Women cannot read tactics." Germany lost 0-2, conceding from two transition sequences. After the match, the channel put me on air for a special.

Germany left the 2026 World Cup before the ball rolled. We simply refused to look at the data.

But stopping there would have been self-deception. PPDA only measures pressure in one zone. It does not measure the dressing room, the commitment to the game plan, or the individuals who had lost motivation. The metric gave me a signal. It did not give me a cause.

In 2026, at the World Cup in Qatar, I followed Morocco from the knockout rounds. Their profile read quickly: 38% possession, more than forty successful tackles in their own third, a back line that refused to be stretched. The media called it luck. I called it a model of spatial allocation.

What caught my attention was not the low possession figure. Precisely because possession was low, they had to optimise a different variable: the quality of every transition. They did not hold the ball. They held the moment.

That was also when I recognised a professional trap. Many analyses of Morocco used PPDA as a label, and that metric measures one zone, one time window, under a definition chosen by whoever supplies the data. Applying that label to a team that deliberately sits deep misreads the entire structure. I nearly made that mistake myself. It pushed me to rewrite my analytical vocabulary: space control instead of possession, transition quality instead of counterattack count.

In 2026, with global competitions suspended, I started a small project that redirected my career. I assembled five volunteers and split the collection of performance and injury data on 120 players across three leagues: the J-League, the K-League and the Chinese Super League. Four months later we had our first report.

Result: 68% of the sample averaged a 12.4% drop in distance covered across their first five matches after the restart. Hamstring injury rates doubled year on year. A peer-reviewed sports analytics journal cited the report.

What I learned was not in the numbers. It was that nobody published that data, because nobody has an incentive to publish a metric showing their own competition is draining its players.

A pandemic does not create problems. It exposes what we never measured.

A good data system is not born from technology. It is born from the pain of those who lacked it.

Applying the method to Vietnamese badminton

Badminton has no luck. It has unmeasured variables.

Take the most common question after any three-game match: why does player A win game one and lose the next two. The intuitive answers are "lost focus", "ran out of gas", "the opponent changed tactics". All three may be true. But only one is testable, if you build the right variable.

Based on my experience watching matches both domestically and across the world circuit, the most neglected variable in badminton is the interval between rallies, combined with the rally-length distribution by phase of game.

Neither requires expensive equipment. Both require someone timing with full attention.

The construction is concrete. Sit where you can see both sides clearly. Log the start and end of every rally. Log the moment of the next serve. The gap between those two points is recovery time. Accumulate by game and you get a recovery curve. Put that curve on the same axis as the score, and things the scoreboard never tells start to appear.

In many matches I have tracked, the recovery curve does not move with decision quality. When average recovery time drops below a certain threshold, unforced errors late in the game rise before movement speed declines. In other words, decisions collapse before legs do. And that only shows up if you measure tempo rather than outcomes.

This is where badminton diverges sharply from football. Football gives you thousands of discrete events per match to anchor on. Badminton gives you a few hundred rallies, and every rally is a meaningful unit of time. In badminton, the tempo variable carries more weight than the event variable.

But tempo alone is only half. The other half is the context of each rally: at what score it happened, how long after the mid-game interval, in which game, and how many consecutive rallies had already passed.

Stack those three — recovery time, rally length, score — and the picture is far clearer than any commentary. A third game at elite level usually shows one of two structures: rallies shortening with more early winners, or rallies lengthening with more errors. The first signals a player accelerating to finish before the tank empties. The second signals both players are spent and the match is being settled by error margins.

Those two structures demand opposite advice. Without measurement, we give the same advice to both.

The cross-border lens: one variable, two rulers

I was born in Vietnam and work in China, which gives me an odd advantage: I can watch the same match through two data systems.

What I found after years is not which system is better. It is that most of the differences people call "inexplicable" are the same variable measured two different ways.

A small but telling example. "Distance covered" in badminton is sometimes calculated as total displacement of the body's centre of mass, sometimes as total displacement of the feet in contact with the floor. Same player, same game, the two numbers can differ by 8% to 15%. And so two opposing analyses of the same match appear, each citing a real figure.

After enough of these comparisons, I stopped believing claims like "Vietnam's physical base is weaker" or "Asian styles are more durable". I believe in process: if two parties define a metric differently, do not compare. If they define it identically, compare to the end.

That is why I spend hours on something colleagues find pointless: rewriting the definition of every metric before using it, with source name, version and update date.

I do not trust intuition. I trust intuition that has been verified by ten thousand rows of data.

The contrarian angle: the closed room of the data analyst

There is one risk I have to state plainly, because it is the biggest risk in my own method.

A self-built data system can become a closed room. I choose the variables, I choose the definitions, I choose the thresholds. Every conclusion drawn from it is consistent with the system, and that consistency creates a false sense of correctness.

A model never argues with the person who built it.

In 2026 I nearly fell into that trap with Morocco. Had I used a single metric to describe them, I would have had a tidy, wrong story. What saved me was being forced to reread that metric's definition and realising it covers only one zone of the pitch in one time window.

The antidote is concrete. Regularly put outlier data points in front of your own model. Once a month I pick the case my system explains worst and write about it alone, with my favourite variables banned.

The second risk is turning every story into a quantification exercise. A data background makes it easy to forget that every number was entered by a person, on a shift, under pressure. I keep one rule: every piece must carry at least one layer of context behind the number — who measured it, how, and why.

The third risk is the illusion of causation. Two metrics moving together does not mean one causes the other. Germany's pressing weakened and Germany lost to South Korea. But PPDA did not beat Germany. It recorded a symptom of an illness I had no data to diagnose.

And there is one more thing I learned later than I would like to admit. Sometimes the emptiness itself is the finding. When a tournament does not publish its rally-length distribution, that does not only say it lacks the equipment. It says nobody is being held accountable for the workload placed on its players.

In those cases, pointing at the empty cell is worth more than any model I could build.

Signals for the next cycle

A major season is coming, and the pressure will fall on what happens on court. That is correct. But there are three signals I will track, and none of them live in the scoreboard.

The first: which tournaments start publishing raw data instead of processed data. When an organiser releases an aggregate figure without a definition, they keep the right to tell the story.

The second: which federations start hiring data specialists for badminton at national-team level. When that happens, the talent-development cycle shifts axis from "find good players" to "build players suited to a measured model".

The third, and most important: who in Vietnam will be the first to publish a self-built badminton dataset with a stated method, defined terms, an error margin, and public accountability.

Such a dataset does not require large capital. It requires one person willing to be mocked for a few months, and patient enough to let history speak.

If you want to know whether a sport is rising or falling, do not read the medal table. Read the list of what it chooses to publish, then compare it with the list of what it chooses not to. The gap between those two lists is the answer.

And if you are wondering whether to start building your own dataset, even with just a notebook and a stopwatch — my answer is yes. Not because you will be right immediately. But because the only person who can prove you right is you, a few years later, with an evidence chain nobody can dismantle.

I was once mocked for a number. Three years later, history spoke for me. It will do the same for anyone who sits long enough with the empty cell.