The Empty Split Column: The Limits of Data Analysis in Swimming

The stands went quiet after the final horn. The scoreboard cycled through its numbers: first place, time, nationality. The next column — the 50m split — was a single dash.

I was sitting in the third-tier press row, reopening the official results portal, refreshing a second time. The dash stayed exactly where it was. Below me, a coach looked up at the board, then turned and said something to his assistant. I could not hear a word of it, but I could guess the content: he was describing something neither of them had actually seen in numbers. The swimmer in that final might have produced the fastest heat of her career, or she might simply have been the only one who hit her rhythm. Two scenarios, one final number on the board.

I have written hundreds of split tables in my career. This time, what I had to write about was the empty space on the board.

The two data layers of a swim meet

A major swim meet runs on two separate data layers. The first is the scoreboard layer: finishing order, times, nationalities, records. It is almost always complete, published within minutes of the touch, and it is what most media use.

The second is the decomposition layer: reaction time, 50m splits, turn times, stroke rate per 25m segment, the 15m underwater mark. This layer determines nearly every tactical conclusion — and it is frequently absent.

In preliminaries, organisers often publish only total times, sometimes as PDFs scanned from sheets taped to a hallway wall. At junior and regional meets, splits barely exist in public form. In finals they exist, but with different lag times depending on the meet, on whether the pool is 25m or 50m, and on the timing-system provider.

That asymmetry matters because swimming is a sport where pacing distribution determines the meaning of a result. In a 25m pool, a stronger turner can gain 0.2 to 0.4 seconds per turn. Across a 200m race, that compounds to roughly one to one and a half seconds. Converting a short-course result to long course without splits is therefore an inference, not a measurement.

In the Paris 2026 cycle, Léon Marchand won four Olympic gold medals and Katie Ledecky reached nine Olympic golds. Both milestones were recorded by media through final times, not through split structure — and that is precisely the blind spot.

Every transfer deal is a problem waiting for its solution. Swimming has no transfer window, but exactly four things can move after each Olympic cycle: coaches, training bases, sports-science staff, and sporting nationality. When all four shift at once, the second data layer disappears first — the receiving programme has no baseline on the athlete yet.

Five things inside the empty cell

One, pacing. In a 200m freestyle final, two swimmers who both touch in 1:57 can carry opposite physiological profiles. A swimmer who goes 27.5 – 29.2 – 29.8 – 30.5 held rhythm and faded late. A swimmer who goes 29.1 – 29.4 – 29.5 – 29.0 negative-split, meaning the back half was faster than the front. Same line on the scoreboard, two different forecasts for the next round. Without splits, all that remains is one number and a story of our own choosing.

Two, reaction time. The column is small, sits off to the right, and is routinely skipped. But in the 50m freestyle, a gap of 0.08 seconds is enough to reorder four places. Over that distance, a reaction-time deficit cannot be recovered mid-pool. I once tracked a swimmer who won her heat with a 0.60-second reaction and lost the final with a 0.72, while her swim time barely changed. Reading the scoreboard while ignoring the reaction column leads to a verdict of “loss of form” — a verdict that is mechanically wrong.

Three, turn economy and the underwater phase. The 15m underwater mark is almost never published. But in butterfly and backstroke, the underwater segment accounts for most of the margin between elite swimmers. Someone who holds 12 metres under water saves roughly half a second against someone who surfaces at eight metres, and that saving does not live in the arm stroke — it lives in the kick.

Four, relay splits. This is the most common error in quick-turnaround reporting. A relay leg is measured from a flying start: the swimmer already carries momentum when entering the water. Comparing that figure directly with an individual flat-start time means comparing two measurements with different origins. In the 4×100 freestyle relay, the systematic gap typically falls between 0.5 and 0.7 seconds. Every season still produces lines built on “personal bests” derived from exactly such splits.

Five, the provenance of the number. Automatic timing does not err. Data entry does. I once cross-checked a public results sheet against slow-motion footage and found a 0.03-second discrepancy in a final. Three hundredths does not change a medal. But it changes the value of an entire analysis if that analysis is built on the assumption that the number is untouchable.

A natural experiment and the small-sample trap

2026–2026 left behind a rare dataset: meets held without spectators, compressed calendars, reduced preliminary rounds. Empty stands, but the numbers still knew how to score — they simply scored differently.

When I compared average final reaction times between spectator meets and spectator-free meets in that period, the difference I found sat inside statistical noise. The honest conclusion is that the sample is far too small to assert anything. I raise it not to prove a point about crowd noise, but to show a principle: when the denominator is small, silence is a scholarly option.

The counter-intuitive angle

More data does not automatically make better sports journalism. It makes sports journalism more confident — and sometimes confident in the wrong place.

The biggest temptation is to fill the empty cell with adjectives. No splits, so we write “she ran out of gas in the last 50.” No splits, so we write “he timed his surge perfectly.” Those sentences read like direct observation, but they are guesses dressed in a confident voice.

I do not argue with emotion; I present a data chain. But a data chain has limits, and those limits must be stated. A swimmer who negative-splits three consecutive finals has not established a trend. That is a denominator of three. And correlation is not causation: elegant pacing may signal a strong aerobic base, or it may simply be the consequence of rivals in the same heat swimming slowly early.

Being right too early is also a form of rejection. In 2026, I built an xG model for a professional football league and concluded that an expansion side owned the best attacking index in the competition. My editor declined to publish for fear readers would not follow it. I posted it on my personal blog, and the piece only drew attention after the season ended. The lesson I keep is not about whether the model was right, but about the fact that I failed to explain its uncertainty from the outset.

On the other side of that table, every number is a person sweating. An athlete does not control whether organisers publish splits, nor whether a reporter reads them correctly. She controls only the rhythm of her kick at the seventeenth metre. Writing about her without data is one thing. Writing about her and inventing data is another.

When an editor says no, I learn to listen to the data. But when the data says no, I have to learn to listen to its silence as well.

Signals for the next round

The match is over, but the data is still playing stoppage time. In swimming, that extra time can run for weeks, until a PDF goes up at two in the morning and someone discovers the preliminary splits were there all along.

What to watch next round is very specific: whether organisers publish the decomposition layer at the same time as the scoreboard layer, and whether the 15m underwater mark appears in the raw data file. If both are present, a new analytical tier opens. If not, readers should ask one question of every analysis they read: is this built on measurement, or on inference written in the voice of measurement?

Amid the noise of the stands, I choose to sit with the numbers. But I have learned that sitting with the numbers also means accepting that sometimes the table is empty — and that my job is to say so rather than fill it with a good story.