The Empty Stat Sheet in Basketball: The Discipline of the Data Analyst

It is 2:17 a.m. in a small apartment in Ngo Quyen district, Hai Phong. The scouting report for the weekend game is open on the screen. Ten data fields. Not a single number in any of them. Player name: empty. Efficiency metrics: empty. Matchup context: empty. Data source: empty. All that remains is the file title the system generated at midnight — a meaningless string of text.

Eighteen years ago, I would have picked up the phone immediately and snapped at a colleague: "The system is broken." That night, I sat still. Because I had already paid for a lesson: when the arena is empty, only data whispers the truth. And sometimes the truth it whispers is — there is nothing to say.

To many people, that empty table is a failure. To me, it is one of the most honest reports I have ever read.

What my job actually is

To understand why an empty table can be more honest than a full one, we need to be clear about what a team data consultant does.

We do not count points. Anyone can count points, and the arena scoreboard already does it for us. Our job is to answer the questions the box score cannot: Why did this team win despite shooting poorly? Is that player genuinely good, or just playing next to someone better? Is a hot quarter a signal, or just the echo of luck?

To answer, we use a baseline layer of metrics: OffRtg and DefRtg — points scored and allowed per 100 possessions; Pace — the speed of play; TS%, true shooting percentage; USG%, usage rate; plus composite measures such as PER or EPM. At team level, Net Rating — the gap between OffRtg and DefRtg — is the cleanest single strength proxy, because it is pace-neutral. Every number is a confession, if we are patient enough to listen. But the condition sits in the second clause: we must be patient, and we must be honest about what we hear.

In 2026, I learned the first lesson about that honesty while working as an analysis assistant for a new sports outlet in Hai Phong. World Cup 2026, Switzerland against Serbia. I saw Granit Xhaka touch the ball 112 times but send only about a third of his passes forward, and I wrote a piece criticising an overly safe style. Head coach Vladimir Petkovic replied bluntly that football is not mathematics. Three days later, Switzerland came from behind to win 2-1 through decisive passes at the right moment.

I had missed a key baseline metric: PPDA — pressure on the ball carrier — where Serbia ranked near the bottom. I looked at possession and ignored pressing intensity. Since then, I force myself to check at least five baseline metrics before any conclusion: PPDA, xG chain, pass progression, conversion rate, and pace context.

But a bigger question remained: when should we stop and say "I do not know"?

The anatomy of a null result

In data science, one concept is most often misunderstood: missing values and null results. Beginners treat them as errors to delete. Veterans understand they are data in their own right.

Picture a pre-game scouting report. Normally it contains: opponent name, projected lineups, shooting efficiency by zone, conversion trends, defensive weaknesses, and a few personnel notes. If the system returns a table whose fields are all structurally correct but substantively empty, what does that mean?

Three possibilities, in order of probability. First, the source failed to load — a page was blocked, a video had no captions, or a file was truncated. This is an operational failure, and it accounts for most cases. Second, that game has no tracking data because the league has not deployed camera systems. Third — rarest of all — you are asking the wrong question.

The key point is this: an empty table is not a "safe" table. It is an unverified one. If you read a report full of "no data" and sigh with relief that there are no risks, you have read it completely wrong. No risks being named does not mean no risks exist.

That is why, in my workflow, every report must carry a status flag: empty, partial, or complete. Without that flag, an empty table is more dangerous than a wrong one — because a wrong table gets caught immediately, while an empty table is quietly ignored.

The laboratory called Orlando 2026

If any modern basketball event exposed the limits of data when context changes, it is the Bubble.

In 2026, when the pandemic forced the NBA to build a "bubble" in Orlando, 22 teams played from July 30 to October 11 without fans. Technically, it was a rare natural experiment: the same rules, largely the same players, but with the crowd variable removed entirely.

Our prediction models at the time — built on thousands of games with crowds — were nearly useless for the first two weeks. Not because they were mathematically wrong, but because they rested on an assumption that no longer held: that crowd noise was a fixed background constant.

What is interesting lies elsewhere. The Miami Heat, the fifth seed in the East, eliminated Giannis Antetokounmpo's Milwaukee Bucks and then the Boston Celtics to reach the Finals. The Denver Nuggets, led by Nikola Jokic and Jamal Murray, twice came back from 1-3 down — first against the Utah Jazz, then against the LA Clippers. None of those scenarios were in any pre-Bubble model.

But this is where I must be careful. I cannot claim that "an empty arena made team X stronger." I can only say: when the context changes, old correlations lose predictive value. Correlation is not causation. A team that won in the Bubble may have won because it was fresher after a long break, because opponents lost key players, because the schedule was easier, or simply because it adapted faster to a new rhythm. The empty arena is one variable, not the sole cause.

This is the line the analyst must draw in white chalk: describe what is visible, do not judge what is unproven.

When a beautiful number misleads

In the 2026 NBA Finals, LeBron James averaged 29.8 points, 11.8 rebounds and 8.5 assists per game, winning Finals MVP as the Los Angeles Lakers beat the Miami Heat 4-2.

That stat line is almost unarguable. But if I dropped it into a meeting without saying anything more, I would be deceiving the coaching staff. For three reasons any honest analyst must state.

One, a six-game sample is small. Over six games, random variance is large enough to make an average player look like a superstar, and vice versa — which is why composite metrics such as PER or EPM must be adjusted for sample size.

Two, the Bubble context was special: no fans, a long prior break, dense scheduling. The numbers are not wrong, but their meaning differs from those of a normal Finals.

Three, numbers do not lie, but the people who choose them do. If I pick only the three most flattering metrics and ignore shooting efficiency by zone, turnover rate, and how opponents defended, I have built a story rather than presented a fact.

That is why I require every report to include a short methodology section: where the data came from, how many games the sample covers, over what period, and which assumptions are in play. Careful readers deserve to know those things before they believe.

The Empty Arena Index

In 2026, when global football and basketball shut down, I and a small team built what we called the Empty Arena Index, based on hundreds of matches in Portugal and Denmark after their restart, plus NBA games in the Bubble.

We measured something notable: central midfielders' running distance dropped by roughly 9-10% in the first month after restart, while line-breaking passes rose by about 13%. The simplest reading: without fans, players run less off the ball but dare to play more riskily. The leadership of the club I advised doubted the model. I still persuaded them to sign a Brazilian midfielder based on it. After 10 rounds, he had scored 4 goals and assisted 3 — including one from a fast counterattack the empty-arena data had predicted. The club rose six places in the table.

But I must be clear: that success was partly luck. If I told it as absolute proof that an empty arena makes midfielders run less, I would have turned an observation into a law. A sample of a few hundred matches across two leagues is not enough to say that for all of world football.

The number is right. But its scope is far narrower than it looks.

Team operations and the price of missing data

In modern basketball, there are constraints outsiders do not see but which decide almost every move: the salary cap, the First and Second Aprons, Bird rights, the Mid-Level Exception, and the Supermax. A team that crosses the Second Apron loses nearly every tool for flexible roster-building — from salary-matching in trades to the ability to sign buyout players.

The paradox is this: the more constrained a team is, the more precise its decisions must be — and that is exactly when data is often thinnest. When valuing a contract, you need to know how much Net Rating that player contributes on and off the floor, how his efficiency shrinks in a playoff series, and how many prime years he has left.

You rarely have all three answers. For a young player with one season, the sample is too small. For a recently injured player, the data is interrupted. For a player arriving from another league, the conversion factors between the two leagues are not fully reliable.

In those cases, the honest professional must say: this is an uncertainty band, not a firm number. A transfer is not a calculation; it is a negotiation between people and numbers. And both sides can be wrong.

Vietnamese basketball and the missing-data problem

This story is more current for Vietnamese basketball.

The Vietnam Basketball Association — VBA — has grown quickly since its founding in 2026, with teams such as the Saigon Heat, Cantho Catfish, Hanoi Buffaloes, Danang Dragons, and Thang Long Warriors. But the data infrastructure here is far from the NBA. Not every game has tracking cameras to record every step, every gap, every off-ball contest.

That means much of the data an NBA analyst takes for granted — real-time player positions, the distance between shooter and defender, shooting percentages under pressure — simply does not exist here.

Many outsiders see that as a pure disadvantage. I see the opposite: it is a chance for Vietnamese basketball analysis to be more honest than analysis elsewhere. Because when data is thin, you are forced to say exactly what is missing. In the NBA, people easily forget that, because the data is so abundant that we think we know everything.

I once watched a young colleague build an entire prediction model for a domestic league using advanced metrics that league never collected. The spreadsheet was beautiful, the formulas were correct, but the input column was empty. He filled it with figures from another league. The result: a forecast that looked highly professional and was entirely meaningless.

That is the most dangerous kind of error, because it does not expose itself.

The counterintuitive angle: an empty table is more honest than a full one

In this profession, there is a temptation greater than lying: the temptation to fill in the gaps.

When data is missing, the beginner's reflex is interpolation: use the league average, estimate from last season, assume it is roughly that. Every time they do it, they create a number that looks reasonable but is not real. And that number spreads — into slides, into meetings, into transfer decisions. A fabricated number lives longer than a correct one, because it is never questioned.

Take Nikola Jokic. He was selected in the second round of the 2026 NBA Draft, at pick 41. No model at the time placed him among the stars. Yet Jokic later won three MVP awards and a Finals MVP in 2026 with the Denver Nuggets. That does not prove every model is wrong — it proves models have limits, and the limits usually lie in the input data, not the mathematics.

So I argue that an empty table is more honest than a fabricated full one. An empty table says: there is nothing here yet. A fabricated full one says: there is something here — when in fact there is nothing.

In 2026, I paid for forgetting that. Before Saudi Arabia met Argentina at the World Cup, I used a model combining four years of qualifying data and declared Argentina would win with a very high probability, with a minimum score of 3-0. The result: Saudi Arabia won 2-1, using an almost perfectly executed offside trap in the first half that repeatedly caught Argentina's forward line. I had missed two variables the model did not have: temperatures above 30 degrees Celsius and altitude, which affect players used to lowland conditions differently from those from high ground.

I once thought I was right. Qatar taught me I was wrong. The most concrete lesson is not "add a climate variable to the model." It is: when the model lacks that variable, I must say the model lacks it. I did not. I chose confidence over honesty.

After 2026, I never write "will win." I write "the 95% confidence interval falls within..." and let the reader decide.

Takeaway

The question I now carry into every report is not "what does this number say" but: "if this table is missing a field, what will I do?" The correct answer most of the time is: stop, flag it, and tell the decision-maker plainly that there is nothing here yet.

Basketball is generating ever more data. Cameras track every step, sensors measure every heartbeat. In that world, the scarce thing is no longer the number. The scarce thing is the person willing to say "I do not know." And when the arena is empty, only data whispers the truth — even when that truth is silence.