Swimming and the Data Void: When an Analyst Must Read the Silence

INTRODUCTION: A BLANK SHEET

There is one evening in Hanoi I will never forget, and not because a record was broken, but because of a blank spreadsheet. I sat in front of my screen, opened a dataset from a national-level swim meet, and everything that appeared was empty: no reaction time off the blocks, no stroke rate, no distance per stroke, no 50-metre splits. All that existed was a list of names, a list of lanes, and a final result typed by hand from a paper scoreboard. I sat still for three minutes. Not out of confusion, but because I realized the worst thing in this profession is not a wrong number. The worst thing is a number that does not exist at all.

That was the moment that shaped how I have worked ever since. People often assume sports analysis is a trade of beautiful figures, polished models, and colourful charts. But when you open a blank file, you understand that this trade is first and foremost the trade of a record-keeper. You do not only need to know how to read numbers. You need to know what happens when numbers are absent.

CONTEXT: HOW SWIMMING IS MEASURED

Swimming is one of the most thoroughly quantified sports at the elite level. At major events — the Olympics, the World Aquatics Championships, World Cup stops — automatic timing systems provided by official suppliers record times to the hundredth of a second, alongside reaction time off the starting block, time to the first stroke, and 50-metre splits. This data is published almost instantly and becomes the reference source for every ranking of performance.

But that picture only covers the tip of the iceberg. Swimming has an extremely steep competitive pyramid: on one side, a few dozen events are equipped with state-of-the-art measurement infrastructure; on the other, thousands of competitions at national, regional, and youth level, where a simple electronic scoreboard is already a luxury. Between those two ends lies an enormous information gap, and that gap is where most real analysis actually happens.

I spent many years working as a data collector and processor for analytical units, and I learned one thing: the quality of a conclusion does not depend on how many charts you draw, but on how honest you are about what you are missing. A good analyst must be able to say two sentences: the first is "the data shows me this", and the second is "the data does not show me this, so I will not conclude". The second sentence is the hard one.

From 2026, when I began working as a swimming reporter for a newspaper, I had the chance to get closer to how national federations manage data. Most have no centralized database. Information about results, injuries, and training programmes is often scattered across coaches' notebooks, photographs of scoreboards, and phone-shot video clips. Every time I needed to analyse an athlete, the first task was not running a model, but gathering data from sources nobody would expect to have to gather from.

That context raises a question I have carried through my entire career: when data is incomplete, should we analyse at all? And if we do, how do we do it honestly?

THE CORE: DISSECTING A SWIM

To talk about swimming data, you must first understand what a swim is made of. A race, whether 50 or 1500 metres, breaks into four phases: the start, the underwater phase, the main swimming phase, and the finish. Each phase has its own set of metrics, and the irony is that exactly these metrics are the ones most often ignored in amateur analysis.

The first phase is the start. Reaction time is measured from the instant the starting signal sounds until both feet leave the block. At world level, reaction time typically ranges between 0.55 and 0.75 seconds, and although that margin sounds small, in a 50-metre event it amounts to a significant distance on the course. Many fans look only at the final result and conclude the winner was "stronger", when the real difference may have come from a reaction that was one tenth of a second quicker.

The second phase is the underwater stage. This is the most misunderstood phase. Under World Aquatics rules, after the start and after each turn, swimmers may travel underwater but the head must surface within the first 15 metres. Studies of swimming technique show that dolphin kicking underwater can be faster than swimming on the surface, which is why leading teams treat these 15 metres as an inseparable part of performance. A swimmer who holds an efficient underwater phase right up to the 15-metre mark has a clear advantage over one who surfaces early.

The third phase is the main swimming phase, where most of the race time unfolds. Here the core metrics are stroke rate and distance per stroke. The product of the two, together with technical quality, determines speed. A common mistake is to assume that a faster stroke rate always means faster swimming. In reality, when stroke rate exceeds an optimal threshold, distance per stroke falls faster than the rate rises, and total speed drops. That is why elite swimmers tend to use a lower rate but a longer stroke in middle-distance and distance events.

The fourth phase is the finish. The relevant metric here is how a swimmer adjusts rhythm in the final 25 metres and how the hand touches the wall. A mistimed touch can cost hundredths of a second, and in races decided by a hundredth, that decides the entire medal table.

The first core insight lives here: every phase of a swim is its own problem, and a conclusion that is right for one phase can be entirely wrong when applied to another. Collapsing everything into a single metric is the fastest way to produce a false analysis that still looks convincing.

NOW, TO THE QUESTION OF DATA THAT DOES NOT EXIST

At major events, all four phases have data. At smaller events, we have only the fourth phase — the final result — and sometimes a few hand-timed marks. Such a dataset has a very particular shape: a list of names, a list of finishing times, and empty cells everywhere else.

I once received a file of exactly that shape while working with a unit that analysed data for an Asian bookmaker. The initial task was to build a performance-rating model for a regional swim meet. When I opened the file, I found three problems.

The first problem was missing split data. Without splits, you cannot tell whether a swimmer accelerated in the second half or faded. In swimming, pacing strategy across the first and second halves is decisive, especially in events of 200 metres and above. Without splits, you are looking only at the endpoint of a curve without knowing what shape the curve had.

The second problem was missing condition data. Was the pool 50 metres or 25 metres? What was the water temperature? What was the altitude? These factors sound technical, but they directly affect performance. A record set in a short-course pool cannot be compared one-to-one with a long-course result, and the sport's governing bodies must maintain two separate record systems for exactly that reason.

The third problem was missing athlete context. Age, injury history, recent training load, and whether the athlete is in a growth phase. For young female swimmers, puberty can completely change the relationship between power and technique, causing performance to fluctuate in ways a linear model cannot capture.

Together, these three problems create what I call "analysis built on sand". You can construct a beautiful model, but it stands on unsound data.

The second core insight: a model based on missing data is not wrong because its mathematics is wrong, but because it has silently replaced what is unknown with unverified assumptions. And in sports analysis, assumptions are the shortest path to error.

WHAT I DELETED FROM THE MODEL

There is one line I always use with newcomers: when a model cannot explain a result, the first move is not to adjust the model until it fits, but to go looking for the missing variable.

In the case of the blank dataset, my first instinct was to strip every technical metric from the model, because they did not exist. What remained was a model based only on final results and head-to-head history. The result was a model that predicted very poorly: it could not distinguish a swimmer on the rise from a swimmer in decline, because both had the same type of input data.

Then I did something that looked counterintuitive: I deleted the "head-to-head history" variable as well. The reason is that at regional swimming level, head-to-head history is often too sparse to be statistically meaningful. Two swimmers may never have met in the same event, and comparing them based on meetings in other events is a false analogy. When I deleted that variable, the model... did not demand an explanation from me, because the model never had enough data to say anything certain in the first place. I deleted "head-to-head history" from the model and the model demanded an explanation from me — but the only honest explanation was: the data is not sufficient to answer this question.

That was a major lesson. In this profession, the hardest skill is not building a model. It is knowing when to say: "I cannot yet conclude."

SWIMMING DATA IN VIETNAM: A PICTURE FULL OF GAPS

Speaking about Vietnamese swimming, I must adopt a very cautious attitude. We have athletes who have made their mark at regional and continental level. Nguyen Thi Anh Vien is a name tied to many SEA Games and left a deep impression with a large medal collection. Nguyen Huy Hoang is a standout in distance freestyle events, having made his mark at continental level. Those achievements are real and deserve recognition.

But when I step into data analysis, I run into a systemic problem. The national swimming data system is not yet thick enough to support deep analysis. We have results, but not standardized splits. We have performances, but not continuous training data series. We have injuries, but not injury categories classified to international standards. This makes comparing a Vietnamese swimmer with their own past self, or with athletes from neighbouring countries, far harder than it appears.

I once tried to build a comparison table of leading regional swimmers in the same event. I had to abandon it, because each country publishes data differently: one country publishes splits, another publishes only final times, one publishes reaction times, another does not. When measurement systems are not aligned, every cross-border comparison carries an unstated margin of error.

The third core insight: in regional swimming, most of the difference between data systems is not a difference in performance, but a difference in measurement infrastructure — and confusing those two kinds of difference is the source of many false conclusions.

ON BEAUTIFUL NUMBERS AND THEIR PRICE

Fans love round numbers. A world record, a broken milestone, an unbeaten streak. Organizers, sponsors, and media platforms love them too, because they sell easily.

But there was a period in swimming history when beautiful numbers became an ethical problem. That was the late 2000s, when swimsuit technology advanced to the point of creating a non-technical advantage. At the 2026 World Championships in Rome, world records fell in waves, and experts realized performances were being driven by materials rather than by human technique. In response, the world swimming federation issued rules restricting high-tech swimsuits, effective from January 1, 2026.

That story teaches an important lesson: not every number that improves is progress. Some numbers improve because of training, science, and technique. Others improve because of things outside the essence of the competition. Distinguishing the two is the core task of the analyst.

AND NOW THE HARDEST PART TO HEAR: WHEN DATA GOES SILENT

This is the part I must state plainly, even knowing some people will not enjoy hearing it.

In sports analysis in general and swimming in particular, there is a habit treated as normal but in fact deeply dangerous: using the absence of data as a place to fill in with emotion. When metrics are missing, we talk about "character", "class", "spirit". When splits are missing, we talk about "race experience". When injury history is missing, we talk about "willpower over adversity". Those sentences sound wonderful, and we are used to reading them in newspapers.

But that is not analysis. It is filling a gap with an unverifiable assumption, and then calling that assumption truth.

The duty of the analyst is not to be right. It is to say what the data wants to say. When data goes silent, our job is not to speak louder, but to point at the gap and say: "I do not know this part yet."

I know that does not bring satisfaction. Readers want a firm conclusion, a clear prediction, a name circled in red. But if I hand them a firm conclusion built on data that does not exist, I am selling them counterfeit goods, and those goods will break at the first inspection.

And here is the most uncomfortable core insight of all: the absence of data is not a neutral state. It is an open space, and if the analyst does not actively mark it, someone will always fill it with something else — with bias, with emotion, or with money.

A CONTRARIAN VIEW: SHOULD WE ALWAYS ANALYSE?

At this point I must argue against myself. If I say that we should not analyse when data is missing, am I not denying my own profession?

I do not think so.

There are cases where analysing despite incomplete data is still necessary, because the goal is not to deliver an accurate prediction, but to ask the right question. For example, when a young swimmer suddenly swims much faster than before, missing data does not mean we ignore the phenomenon. It means we must say clearly: "We see a jump in performance, but we do not yet know the cause, and here are the hypotheses that need testing."

That approach turns analysis from a verdict into an investigation plan. And in many cases, the investigation plan is worth more than the verdict.

What I oppose is not analysis under incomplete data. What I oppose is erasing the trace of that incompleteness, and then presenting a conclusion as if it were built on solid ground. The difference between honest analysis and deceptive analysis is not whether there is enough data, but whether one dares to state clearly what is missing.

ON THE THINGS ALWAYS LEFT OUT OF THE MODEL

There is one category I always add to every analysis after an expensive lesson: non-quantifiable variables. These are injuries, psychology, cards, competitive conditions, and sudden events no one anticipates.

I learned this the hard way. There was a period when I was overconfident in my model, and I delivered a firm conclusion about a team based on their average performance metric before a tournament. Reality went entirely the other way, not because the model calculated wrong, but because a variable lay outside every dataset: a shocking event occurred on the field of play, changing the team's entire psychological state and turning them from an early-exit candidate into one of the most emotional stories of the tournament.

After that, I added a risk-adjustment coefficient range, usually from 0.8 to 1.2, to every model, to reflect the uncertainty of non-quantifiable variables. And I abandoned words like "certain", replacing them with "low risk" or "high risk".

In swimming, non-quantifiable variables are just as present. A swimmer can perform well in heats and fade in the final due to pressure. A swimmer can suffer a shoulder injury — a common injury among heavy trainers — and lose performance while technique stays unchanged. A young swimmer may go through physical development that changes the relationship between power and technique. No model captures all of that unless we actively bring it in.

COMPETITION DENSITY: THE INVISIBLE ENEMY OF PERFORMANCE

There is one variable I consider the most important yet the least mentioned in swimming analysis: competition density.

A swimmer may have to race multiple events in a single meet, plus regional meets, international meets, and long training camps. When the calendar is dense, the body does not have enough recovery time, and injury risk rises. In swimming, the shoulder and knee are the two most loaded areas, and injuries in those two areas can drag on, affecting an entire competition cycle.

What I mean is this: when analysing a swimmer's performance, if you do not know how many times they have raced in how many weeks beforehand, you are analysing half the story. And the other half is usually the decisive half.

I believe competition density is one of the biggest causes of injury in elite sport, and no medical team can save an athlete forced to race twice a week for months on end. That is a fact analytical models often skip, because the calendar is in the organizers' hands, and organizers have their reasons for packing it tight.

WHAT HAPPENS WHEN YOU AGGREGATE THINGS THAT CANNOT BE AGGREGATED

One of the greatest temptations for anyone working with data is to combine many metrics into a single composite index, for easier comparison and presentation. This temptation is especially strong in swimming, which has many different events.

The problem is that every swimming event has its own character. The 50-metre freestyle is almost a pure speed problem, where the start and underwater phase decide most of the outcome. The 1500-metre freestyle is a pacing problem, where a mistake in the first 500 metres can destroy the entire race. Breaststroke has very different technical characteristics, with its own arm-pull and kick cycle, and the rules have specific requirements about how the wall is touched.

If you combine all of them into a single composite index, you create something that looks scientific but has actually erased the most important differences. The fourth core insight: a composite index is only valuable when its components measure the same thing; if they measure different things, the composite index is merely a way of concealing chaos.

THE READER'S PARADOX

There is a paradox every analyst faces: readers want certainty, but data usually only allows uncertainty.

When I write that risk in an event is low, readers tend to read it as "will win". When I write that risk is high, they read it as "will lose". That binary reading compresses every honest analysis — which always contains multiple levels of uncertainty — into a single conclusion.

My solution is to force myself to write the level of uncertainty into the sentence itself. I do not write "this swimmer will win". I write "available data leans toward this swimmer, but uncertainty remains high because split data is missing". That style is longer, less attractive, and sometimes disappoints readers. But it is honest, and honesty is the only thing left after every prediction has been tested.

ON THE PEOPLE BEHIND THE NUMBERS

When speaking about swimming data, I do not want to forget one thing: behind every number is a person. A young athlete may spend their entire adolescence in the water, and a performance milestone may be the result of ten years of training. Analysis is not permitted to turn them into pure data points.

I think of female swimmers, who often face particular pressure. For years, women's sport has been spoken of as a story of ascent, but I believe the way the sports industry treats it is often instrumental. Women's competitions are included as an item in a corporate social responsibility report, promoted as a symbol, but do not always receive infrastructure and resources commensurate with men's competitions.

In swimming, this shows in the fact that women's events sometimes have uneven competition scheduling and measurement infrastructure compared with men's events at certain regional meets. When data on women's events is thinner, the capacity for deep analysis is lower, and that loop feeds itself: less data leads to less analysis, less analysis leads to less attention, less attention leads to less investment.

I do not say this to stir emotion. I say it because the data shows it. And when data shows a gap, my job is to point at it, not to fill it with a pretty story.

A LESSON FROM EVENTS NO ONE ANTICIPATES

There are events in sport no model can calculate. An athlete falls ill before a final. A medical incident occurs right on the course. A referee's decision changes the entire picture. These events remind us that sport is a human activity, and humans do not operate by formula.

My way of handling such events in analysis is to separate them into their own section, called non-quantifiable variables, and assign them an impact range rather than a fixed value. I do not try to predict them, because that is impossible. I only make sure that when they happen, my model does not collapse entirely from having ignored the possibility of their existence.

That is the biggest lesson I have drawn from years in this trade: humility is not a sign of weakness in analysis, but a sign of maturity.

CONCLUSION: WHAT A BLANK SHEET TAUGHT ME

Back to that evening with the blank spreadsheet in Hanoi. At the time I thought I was facing an obstacle. Later I understood it was one of the most important lessons of my career.

A blank sheet is not a failure of data. It is a reminder. It reminds us that every model has limits, that limits must be spoken rather than hidden, and that the value of an analyst lies in how honest they are about their own ignorance.

In swimming, where every hundredth of a second can decide a title, honesty with data is the most valuable thing there is. And in a season when the calendar grows ever denser, when athletes must move between continents, when regional meets still struggle with measurement infrastructure, I believe that value will only grow.

Fans will always want a firm answer. But I choose to keep for myself the right to say "I do not know yet" — because that is the most loyal friend of any analysis. Every course sends us a signal. The analyst does not decode that signal, but listens to it, even when the signal is only silence.

And if there is one thing I want to send to young people entering this profession, it is this: learn to read data, but also learn to read the gaps. Because in swimming, as in every data-driven field, what defines who you are is not what you know, but how you behave toward what you do not know.