The Empty Table: The Limits of Data-Driven Basketball Analysis

It was 2:47 in the morning in Hanoi. The street outside was still wet from the night rain, and on my screen was a spreadsheet that opened with every column header in place but not a single row of data. The match name was correct. The date was correct. Everything else was empty.

I had been assigned a breakdown of a youth national basketball game, with a 7 a.m. deadline to make the morning edition. By eleven o'clock the previous night, two data sources had still not returned: the arena's automated stat system and a positional tracking file a partner was supposed to send. By two in the morning, both were silent. No connection error. No corrupted file to open. Just nothing.

In my trade, this is the most dangerous moment. The disaster for a writer who works with numbers is never having too much data. It is having too little, while the deadline does not move. My phone buzzed. The editor asked: can you file tonight, readers are waiting.

And I sat for a long time in front of that empty table, asking myself a question that eighteen years in the profession has never fully answered: when there is no data, what do you write?

The most honest answer is: nothing. Almost nobody has the nerve to send an empty file to the newsroom.

That night I did not send an empty file. I wrote a different piece. But that blank table stayed with me for weeks, and it forced me to look back at how I had worked for nearly two decades.

Vietnamese basketball and its information gap

Basketball in Vietnam carries its own paradox. The professional league has been running for nearly a decade, audiences keep growing, teams have sponsors and media coverage, and some nights sell out. But the data infrastructure behind the court is far thinner than the noise in the stands.

Consider the sheer volume of information a professional game generates abroad. Every shot is logged with floor coordinates, shot-clock time, the passer, the situation that produced it. Every player carries efficiency figures by minute, by matchup, by play type. Every team gets reports on the effectiveness of each five-man combination.

Here, most of what the public sees stops at points, rebounds, assists and shooting percentage. Those four numbers say almost nothing about how a team won. A player scoring 22 points on 8-of-20 and a player scoring 22 on 7-of-12 are entirely different stories, yet in the next morning's coverage they look identical.

The gap produces three consequences.

First, writers are pushed to describe with adjectives instead of numbers. Second, readers grow accustomed to being told stories rather than shown evidence. Third, and most seriously, anyone with a sliver of data can manufacture an impression of precision, even when that data does not measure what it claims to measure.

It was the third consequence that brought me back to that blank table in the rainy night.

Empty and zero are two different things

In statistics, missing data and a value of zero are fundamentally different, even though on screen they can look alike.

When a player goes 0-for-5 from three, we know something: five attempts, no makes, in a specific context. When the data file returns no shot attempts at all, we do not know whether that player took five or none. The distinction sounds academic, but it determines the entire value of an analysis.

I once watched a painful case at professional level. A young team published its internal metrics, and one player was listed with the best defensive rating on the roster. Looking closer, that rating was computed over exactly fourteen minutes across two games. Fourteen minutes. A game and a half of basketball.

Nobody lied. The number was correct under its own definition. But placed next to numbers computed over seven hundred minutes, readers had no way to tell signal from noise.

How wide is the noise band for small samples in basketball? A simple calculation. A player whose true three-point rate is 35 percent takes ten attempts. The probability he shoots 50 percent or better over those ten is not small. Over twenty attempts, it remains substantial. In other words, a four or five game hot streak says nothing about a shooter's level.

I often put this question to younger writers: if a player shoots 12-of-20 across three straight games, do you write about him as exploding or as fluctuating? Most choose the first word. That is a choice of language, not of data.

How narrative fills the void

Human instinct cannot tolerate a vacuum. Where data is missing, a story appears. Cognitive psychology has a name for this, but for people in my trade there is a simpler one: the laziness of imagination.

A team loses three in a row. There is no data on workload, on shot quality, on how opponents changed their defensive scheme. Media reaches for the cheapest explanation available: they lost their spirit, they lack character, the coach has lost the locker room.

I have written pieces like that. At twenty-eight, I was heavily criticised for arguing that a team deserved to win 3-1 rather than scraping a lucky 1-0, based on an expected-goals figure of 2.87 against 0.45, 68 percent possession and fourteen shots inside the box. The piece was mocked for treating football as mathematics. A week later, the head coach admitted he had reviewed the tape and changed his approach based on that analysis.

That night, the media called them soulless. xG said the opposite, and I chose to trust xG.

But that story did not teach me that data is always right. It only taught me that data beats a story assembled out of thin air. Those are very different lessons, and for years I confused one with the other.

The trap of round percentages

There is a category of data more dangerous than wrong data: data that is correct but severed from context.

A team shoots 41.7 percent from three for the season. That sounds strong. But that 41.7 percent may come from four games in which opponents left them open, and twenty more at an average clip. When that four-game stretch falls early in the season, media calls the team a phenomenon. When it falls late, they call it playoff momentum.

The same number, two stories, both drawn from the same thin fact.

In professional basketball abroad, this problem is well studied. Point differential per game varies enormously, far more than the true quality of a team. A mid-tier team can win six of seven on a soft schedule and above-average shooting, then lose eight of ten when both factors reverse. Team-rating metrics need at least twenty games before they stabilise, and for many advanced measures the threshold is higher.

After enough mistakes, I set a rule for myself: no verdict on a team based on a run shorter than fifteen games, unless a tactical factor explains the cause rather than merely describing the result.

The difference between describing and explaining is the entire border between a data writer and someone reading a box score aloud.

Transfer value and belief in what has not happened

There is one domain where missing data does the most damage, and few notice: player valuation.

When a young player has a good season, the transfer market does not price that season. It prices the next season, and the four after that. The whole valuation rests on projection, while the data covers a single year.

I call this the bubble of certainty. In European football, clubs paid enormous sums for players who had not played fifty top-level matches. Basketball does not escape the rule; only the currency changes. A player averaging eighteen points in a minor league over two good months can land a contract larger than one earned by a player who has proven stability across four seasons.

A contract is only truly right when the number is signed alongside the signature. The number can be right for the sample, but the signature binds both sides for years, once the sample is old and new variables have appeared.

In basketball, I check three things before trusting a big contract. First, the denominator: how many minutes the player has logged in the current system. Second, the dependency: how much of his production comes from chances created by others. Third, transferability: which skills retain value when the defence is better and the teammates are weaker.

None of these three questions requires a complex model. They require the patience to tolerate a gap rather than fill it with belief.

When the stands emptied, my model collapsed

One year I built a dataset on home advantage stretching back to 2026. When leagues had to play without crowds, I bet that home performance would fall from roughly 54 percent to below 50. Early results backed me: the home win rate in a major league dropped to about 48.7 percent, and one club won only three of its eight remaining home games.

Then my model collapsed during the recovery phase.

When the stands emptied, my model collapsed. I knew I had forgotten the human factor. I had accounted for the loss of crowd pressure on referees and the psychological edge for the home side. I had not accounted for how much training-ground quality, disrupted travel and individual mental states would differ from club to club.

The lesson was not that the model was wrong. It was that I had trusted a single variable, and that variable happened to be right early, so I believed I was right throughout.

This is the mistake I see repeated constantly in Vietnamese basketball debate. Someone offers a number. The number is correct. They draw a conclusion from it. The conclusion sounds reasonable. Nobody asks what that number left out.

Numbers show trends, not prophecies

At thirty-three, I was invited to work as an analyst for a major sporting event and built a prediction model on expected goals, actual goals and control metrics. The model concluded that one national team would advance from its group because it carried the highest accumulated figure in the group.

That team went out in the group stage.

Looking back, I found the cause. My model lacked data on the pressing intensity of direct opponents, particularly a metric measuring how many passes a side is allowed before pressure arrives. That team hit 6.8 across two decisive matches, a figure outside the dataset I had assembled before the tournament.

It took weeks to steady myself. I then spent three months building a system that integrated more unconventional sources, and more importantly, I added a mandatory section to every analysis: risks and gaps.

Numbers show trends, not prophecies. I put that line at the foot of every model I publish, not to protect myself, but so readers know exactly what they are reading.

In basketball, the gaps tend to sit where nobody looks. We have data on shots but not on the decisions that produced them. We have data on individual defence but not on how teammates rotated to cover. We have data on fitness but not on sleep, nutrition, or what a player is dealing with away from the court.

Every such gap is an invitation for a story to walk in.

Silence is also a form of data

Only now can I return to that blank table in the rainy night.

After years, I understood something I did not understand at 2:47 that morning. A model with no input is not a failed model. It is a model telling the truth about itself: that it does not yet have grounds to speak.

My job was not to fill the table with my own guesses. My job was to write about the gap, and about why the gap existed. In that particular case the answer was simple: the arena had no automated stat system, and the partner suffered a technical failure. But behind that simple answer sits a larger question about the data infrastructure of Vietnamese basketball.

Numbers never need us to defend them. It is we who need numbers, so that we do not deceive ourselves. When I was forced to admit I had nothing to analyse, I held something more important than any metric: the knowledge that a place exists in the system where data does not flow.

A good data writer is not someone who always has a number ready. A good data writer is someone who knows precisely when their number has no standing to speak.

The contrarian angle: humility does not sell advertising

Now comes the uncomfortable part, the part I must state plainly because I recognised it in myself.

Sports media runs on emotion. A piece arguing that one team is better than another will draw more readers than a piece admitting we lack the data to know. A confident headline gets shared more than a headline acknowledging limits. This is a market fact, and anyone denying it is deceiving themselves.

But here is the contrarian part: precisely because humility does not sell advertising, it becomes scarce. And what is scarce holds long-term value.

Over years of watching basketball debates in Vietnam, I have noticed a pattern. The people who state the strongest conclusions tend to have the shortest lifespans. They are right a few times, get attention, and when they are wrong they lose all accumulated credibility. Those who state conditional conclusions, hedged with limits, are dismissed as bland in the short run, but within five years become the people others quote.

I do not say this to praise myself. I say it because I have stood on both sides. I wrote very strong, very certain pieces, and I paid for them.

There is another temptation worth naming. After several successful pieces built on opposing the crowd, a writer easily turns contrarianism into a reflex. Whatever the majority says, say the reverse, regardless of what the data supports. That is an intellectual addiction, and it destroys honesty faster than anything else.

Before every piece I ask myself one question: if this year's data agreed with what the majority believes, would I dare write it that way? If the answer is no, I have no standing to dissent.

What I kept from that night

I did not send an empty file to the newsroom that night. I wrote a different piece on a different subject, and it made the edition on time. But I changed three things in how I work.

First, every analysis I publish now carries a section stating which data is missing and why. Readers have a right to know before they reach the conclusion.

Second, I stopped using absolute claims about metrics. No more this metric proves, no more data definitively shows. Metrics describe, models predict, people decide.

Third, I began writing about data infrastructure as a sports subject rather than a dry technical matter. Because when a league lacks good data, the damage is not that writers struggle. The damage is that players are misjudged, teams are misinvested in, and fans are told stories built on nothing at all.

When the stands emptied, my model collapsed. I knew I had forgotten the human factor. That blank table taught me a similar lesson in a different way: I had forgotten that behind every data cell sits a person, and behind every empty cell too.

What to watch

Going forward, there are three signals I will spend time observing and recording.

The first is the arrival of positional data in domestic games. When shot coordinates begin to be logged at youth level, we will be able to judge players by shot quality rather than shot volume. That turning point matters more than any single contract.

The second is how teams handle small samples. When a young player explodes across four games, will the club immediately hand him a primary role, or wait patiently for twenty more games? The answer will reveal a great deal about the maturity of basketball management thinking in Vietnam.

The third, and to me the most important, is whether newsrooms allow pieces that say we do not know yet. If they do, it signals that readers have matured. If not, every beautiful number published will remain decoration for a story already written in advance.

That night I sat for a long time in front of a blank table. I still keep that file on my machine, undeleted. Sometimes I open it, look at the column headers with nothing beneath them, and remind myself that the honesty of a data writer is not measured by how many rows he has, but by how many times he dares to let the data stay silent.