Vietnamese Swimming and the Data Gap: When Numbers Are Not Enough to Tell the Whole Story

There was something strange about the men's 100m freestyle final at a recent national swimming championship. The winner's opening 50m split was 24.8 seconds, while his best standalone 50m freestyle time that same season was 24.5. In other words, over the first half of a race twice as long, he swam nearly as fast as over the shortest race of all. The number was not wrong. It was telling a different story from the one the stands saw. For a data analyst, that moment is where the work begins.

I sat down with the split sheet for the entire final. What I was looking for was not who won, but the structure behind the win. In swimming, what decides performance is rarely average speed — it is how speed is distributed across four lengths of the pool, how the two turns are handled, and how the swimmer breaks out after each push off the wall.

For years, Vietnamese swimming has leaned on a few outstanding individuals. The era of Nguyễn Thị Ánh Viên set a standard that most of the generations after her could not reach, and that created a dangerous analytical habit: we measure success by medals, not by the structure that produces medals. At SEA Games level, meanwhile, the gap between gold and fourth place often comes down to a few hundredths of a second. At that level, a turn that is 0.15 seconds slow, repeated four times, is enough to push a swimmer off the podium.

That is why I began treating every race as a system of three segments: the start and the underwater phase, the swim speed between turns, and turn technique. Each segment can be measured, and each has its own specific threshold. In the 100m, the underwater phase after the start usually covers the first 15 metres — the zone where the rules allow a swimmer to stay submerged, and the zone the strongest nations exploit fully with a chain of dolphin kicks. A Vietnamese swimmer can lose up to 0.4 seconds in this segment alone compared with the regional standard.

I once spent nearly a year recording the dolphin-kick counts of young domestic swimmers. The result was not in the average, but in the standard deviation. The leading group had a consistent kick count from session to session; the other group fluctuated widely — meaning the technique had not yet been fixed into reflex. This is the kind of difference that a results table never shows, yet it predicts the future better than current form does.

To picture it concretely, take the men's 200m freestyle — a distance where Vietnamese swimming once had a continental-level representative. Here a swimmer covers four lengths with three turns. If each turn is 0.2 seconds inefficient, the total loss reaches 0.6 seconds — roughly the gap between bronze and sixth at a SEA Games. Notably, that gap is not in fitness but in technique. And technique can be fixed, if we know where it is going wrong.

The problem is that we often do not know. Official result sheets at many domestic meets record only the total time, sometimes a 50m split, but almost never turn data. To get that data, you must review video and time turns by hand — a time-consuming job few have the patience for. I once hand-timed more than two hundred turns in a single season, and it taught me a simple truth: good data rarely comes from existing systems; it comes from someone being willing to sit down and do it.

Over six years of watching domestic swimming meets, I have drawn one rarely stated conclusion: most debates about Vietnamese swimming take place without data. People argue about who is better, who deserves a SEA Games place, based on memory and feeling. Memory is a poor data source — it is selective, it distorts, and it cannot be reproduced. A split sheet can be reproduced. A video clip can be replayed. That is why I always return to what can be verified.

When it comes to speed, I am often asked why we do not compare Vietnamese swimmers' metrics directly with European stars. The answer lies in context. A metric only means something alongside its measurement context: long course or short course, 25m or 50m pool, the point in the training cycle, and even water temperature. Strip away the context and you get pretty numbers that mean nothing. This is a trap I once fell into, and it led me to a rule: every number must come with at least three cross-checked sources — official competition data, training records, and video observation.

That three-source rule sounds rigid until you meet a case where the data refuses to line up. Once, a split sheet showed a swimmer accelerating over the final 25m, something almost physiologically impossible if the earlier splits were correct. Checking the camera and the timing system, it turned out a wall sensor had drifted by one cycle. Had I published a conclusion from the first number, I would have told a false story in the tone of truth.

That is why I learned to say the hardest sentence in the analytical trade: “not enough information to assess.” People tend to think that is a weak answer. To me, it is the most honest one. When a data file is empty, when the sample is only three races, when important fields are left blank, then any conclusion is just an inference dressed up with numbers. And that dressing is more dangerous than silence, because it gives the reader a feeling of certainty.

A sports conclusion is only trustworthy when it dares to point out where the data is helpless. This is the lesson I have drawn from years of working with domestic swimming datasets. Most of those datasets are incomplete, inconsistent, and recorded by people who have no habit of record-keeping. Not because they are lazy, but because the system was never designed to make data an asset.

Look at the structure and the gap is clear. Electronic timing systems at national meets produce 50m splits but rarely store separate turn times. Competition video exists but is seldom analysed against a standard time frame. Training records exist but are usually scattered across individual coaches' notebooks. The three sources do not talk to one another. The result is that every time I want to reconstruct a race, I have to start over.

That reconstruction work is time-consuming, but it gives me something the results table does not: a map of what has not yet been read. I call it the map of the unread data. Instead of trying to fill the gaps with guesswork, I mark clearly which zones are verified, which are only hypotheses, and which are completely blind. This approach makes an analysis look less smooth, but more honest.

The interesting thing is that the blind zones often hold the strongest signals. When I compared four consecutive seasons of a group of young swimmers, one pattern repeated: those who improved their times the most were not the ones with the best starting form, but the ones with the most consistent turn technique. Turn technique is rarely mentioned in the news, yet it is one of the best predictive variables I have ever measured.

Regionally, Vietnamese swimming has an edge in some short sprint events, where athletes' build and explosiveness come into play. But in middle-distance events like the 200m and 400m, which demand speed distribution and breathing management, the gap with the region's more developed swimming nations remains. Singapore and Thailand have more systematic data-recording systems, and that shows in their ability to hold form steadily across many editions of the Games.

I must, however, be careful about my own findings. Four seasons is a small sample. A correlation between turn technique and rate of improvement does not mean causation. Both could be driven by a third variable — the quality of the training programme, for instance. If I called it a cause, I would have turned a hypothesis into a certainty, exactly the habit I always warn others against.

In this trade, the most likely thing to happen is that we find one unusual number and build an entire story around it. I have done it myself. Years ago, I published a claim based on data from only two competitions, and later had to retract it when a larger sample gave the opposite result. That fall taught me that humility is not a weakness for an analyst — it is a tool.

There are times when I open a data file and find it empty. Not empty because of an error, but empty because no one recorded anything. In those moments, the reflex of a newcomer is to fill the gap with speculation. The reflex of a veteran is to stop and say: not yet. The difference between those two reflexes is the difference between an analysis and a guess dressed up to look like one.

So when someone asks me about the prospects of a Vietnamese swimmer, I usually answer with two scenarios rather than a single prediction. The first assumes current technique is kept and only fitness is optimised. The second assumes a change in the training structure, especially in the underwater phase and the turns. The two scenarios yield very different time ranges, and the gap between them is the value of data.

In other words, numbers do not lie, but people always find a way to lie with numbers. And in an environment where data is still scarce, lying with numbers becomes easier than ever — not by inventing figures, but by selecting the favourable ones and staying silent about the missing ones.

I once treated models as scripture. Now a model is only a compass — but without it, we get lost. A performance-prediction model cannot replace a coach's eye, nor the feel of the water that a swimmer has. It does one thing: it forces us to state clearly which assumptions we are relying on. And that clarity is what will let Vietnamese swimming progress systematically, rather than progressing by luck.

Looking ahead, the signal I am tracking is not next season's medals. The signal is whether youth coaching staff begin recording data consistently — turn times, kick counts, distance per stroke. If that happens, in two or three seasons we will have enough of a foundation to make genuinely trustworthy predictions. If it does not, we will keep guessing, and keep calling guesses analysis.

Swimming is a sport of small margins. One hundredth of a second can be everything. But to understand that hundredth, we need more than a stopwatch. We need a system that knows how to record, how to cross-check, and how to admit when it does not have enough data to say anything at all.