When the Data Table Goes Blank: Notes From a Night Without a Single Number

2:47 a.m., Miami time. On my second monitor, the tracking file for the game is still open on its first row. The play-by-play column is empty. The shot chart column is empty. The touches column is empty. The minutes column is empty. The only row with text is the domain tag: basketball.

I sat like that for nearly an hour. In my head I already had four very fluent explanations for that game. First, the defense lost its structure in the third quarter. Second, the coach made his substitutions seven minutes too late. Third, a key player was on a minutes restriction because of a fitness issue. Fourth, the classic excuse anyone can reach for: a congested schedule.

Four explanations. Not one of them backed by a single number. And all four sounded entirely reasonable.

I sat there and asked myself: if I published those four sentences, how many readers would believe them? The answer, I knew, was almost all of them. Not because they were true, but because they matched what viewers had just seen on television. They needed no evidence. They only needed agreement.

That was the moment I understood my profession carries a temptation bigger than any other: when the data goes silent, writing a plausible story is always easier than writing three words — insufficient data.

Every number I touch carries a scar. That night, my entire data table was one large scar.

The pipeline nobody sees

Fans see the box score. They do not see the pipeline that delivered it.

A modern professional basketball game generates millions of data points across three layers. The raw layer holds video, optical sensors mounted in the arena, the 24-second clock, and the officials' control panel. The middle layer is the automated tracking system, translating the movement of ten players and the ball into coordinates, then into discrete events: who held the ball, who set the screen, who rolled, who shot, from where, in what situation. The final layer is the writer, taking that chain of discrete events and turning it into a story.

When the pipeline runs clean, the three layers lock together and we have a measurable game. When the pipeline breaks in the middle layer, the final layer still has to publish. And the final layer always publishes. That is precisely the problem.

Based on my experience tracking games live for more than two decades, I can say most errors in sports analysis do not come from miscalculation. They come from filling a gap with an assumption, then forgetting the gap was filled.

Before you watch the game, watch how the data breathes. If the data is not breathing, it is dead, and every conclusion drawn from it is the writer's interpretation, not the game's.

There are three states a writer must learn to tell apart, and it took me years.

No data: the feed is down, the sample is blank, the table has not synced. The only correct handling is to publish the emptiness.

Bad data: numbers exist, but they measure the wrong thing. A familiar basketball example — a player's touches spike after a schematic change, and commentators immediately conclude he is "being given more of the ball." In reality the team simply played more overtime, or a lower shooting percentage produced more rebound opportunities, and the stat sheet counted touches with no offensive meaning at all.

Contradictory data: the numbers exist but say the opposite of what the writer wants to say. This is the most ignored state. Nobody deletes the numbers. They simply do not cite them.

An evidence chain that cannot be invented

Let us talk about how a decent analysis is actually built, and why it cannot be built from an empty table.

At the deepest layer, a basketball game is a problem of space and time. Offensive rating, defensive rating, pace, shot quality, assist generation — those are macro variables. But macro variables almost never explain anything. They only describe. To explain, the writer must descend to the micro layer.

The right question is not how many points a team scored. It is what structure produced those points, and whether that structure repeats.

The box score is a lossy compression format. It keeps the outcome and throws away the process. A player who scores twenty points may have had a brilliant night or a terrible one, depending on whether those twenty points came from high-quality shots inside a system or from eighteen misses rescued by four lucky makes. The box score cannot tell those two apart. The writer must.

Defense is where this compression does the most damage. No column on the box score tells you who guarded whom, across how many possessions, and how many points the opponent scored when he was the primary defender. A guard can give up twenty-five points and be called a disaster, when in fact he spent thirty minutes on the opponent's best player because his teammates kept getting screened out of the play. To know the truth, you need matchup assignment data — something that exists only at the tracking layer.

A few years ago, tracking a team in the middle of a miserable run that the media blamed entirely on its defense, I did the opposite: I removed defense from the equation and went looking for the missing variable. It sat in the least glamorous position on the floor — a ball-handling hub in the middle of the park. His average touches per game had fallen nearly forty percent from the start of the season. No highlight package mentioned it. The standings did not reflect it. But when a hub's entry points collapse, the entire attacking system collapses with them, and the defense is thrown into more transition situations than it can absorb. Three weeks later, that team dropped him lower in the formation.

A hidden variable can only be found if micro-layer data exists. And it can only be protected if the writer accepts losing time in order to look.

The same logic applies to the reverse problem: valuing potential. I have never written the sentence "this player has huge potential." That sentence is meaningless. Instead, I place a player inside a historical percentile band: same age, same position, same workload, same usage rate. If a twenty-two-year-old sits at the eightieth percentile for shot-quality-adjusted shooting efficiency but only the thirtieth percentile for chances created for teammates, I know exactly where the break point is — and I am obliged to publish the sample size.

Sample size is the conscience of this profession. A player shooting forty percent from beyond the arc across five games is a statistical event, not a skill. If I fail to state that it is five games, I am selling an illusion. If I state it, readers can decide for themselves whether to believe it. That is the difference between analysis and propaganda.

The chaos on the court always has a hidden order. The data writer's job is not to retell the chaos. The job is to find the order, and when it cannot be found, to say so.

There was a stretch when I tracked a systemic change that inverted every familiar variable. When stadiums closed and leagues returned in silence, I planned a three-month tracking project across five major leagues, logging every shift. The result I published then: home win rate fell from roughly forty-six percent to roughly thirty-two percent, and average goals per match dropped from about three point one to about two point four.

The meaning of that finding was never in the number. It was in the question: when the crowd disappears, what disappears with it? The answer turned out not to be skill, but pressure — pressure on officials, on players in decisive moments, on the tempo of the game.

That model maps directly onto basketball. When games were played inside a pandemic bubble, home-court advantage all but evaporated, and teams whose systems were built on stirring a crowd suddenly lost part of their weaponry. When an external variable shifts abruptly — attendance, schedule, travel distance, time-zone change, defensive workload — the writer must immediately ask: what is changing systematically? Random variance and systemic change look identical on a box score, but only one of them is predictable.

Back to that Miami night. My data table contained not a single event. Which means I could not know any team's offensive structure, could not know who guarded whom, could not know which shots were good and which were bad. Everything I could write would be speculation dressed in technical vocabulary.

When the reward goes to whoever fills the blank

There is a paradox I have to state, even though it is not flattering.

The sports news economy does not pay for accuracy. It pays for clarity. A piece saying "this team lost because its defense lost its structure" will out-read a piece saying "I do not have enough data to identify the cause." Algorithms do not reward caution. Readers do not share caution. Nobody quotes the line "insufficient information."

During the transfer window, the paradox gets worse. Noise completely drowns the signal. A rumor posted at eleven at night can become a "source close to the situation" after three shares, and "all but certain" after five. Nobody traces back who the original source was, what the leak motive was, and who benefits if the information spreads.

My handling is to rank sources by evidence rather than by appeal. A report from a journalist with direct access to a club's leadership is tier one. A post from an agent is tier two, and must be read on the assumption that the agent is negotiating for his client. An unsourced aggregation is tier three, which is to say no tier at all.

But the most interesting part of the transfer window is not the rumor. It is the contract structure — release clauses, extension options, performance bonuses, and how those fit against the salary cap. A team can be sitting right against the tax line, forced to choose between two players, while the coverage talks only about their "interest" in all three. Clause structure and the salary cap are the real story; the rest is noise with a player's name attached.

Correlation is not causation, and in sport this is the most common error and the hardest to detect. A team changes coach and wins four straight. Everyone calls it the new-manager effect. But in those same four games it faced only the weakest defensive opponents, and its shooting percentage ran twenty points above its season average. Two variables are enough to explain the entire run, without any new coach at all.

A closer example for Vietnamese fans: a player scores heavily in three straight games and is instantly called a "future star." Nobody checks his usage rate across those three games, nobody checks shot quality, nobody notices the team was missing two starters so every possession funneled to one place. The points are real. The story drawn from them is not.

Being counterintuitive does not mean doubting everything. It means separating what we measured from what we inferred, and placing the two in different parts of the article.

Signals to keep tracking

Every one of my analyses ends with a signal list, because a conclusion that cannot be tested is a worthless conclusion. That night my list was empty — but the emptiness itself was a signal.

For readers, there are three things worth checking whenever you read a basketball analysis. How was this number measured, by whom, and across how many games? Which variables were left out of the story — schedule, injuries, travel distance, rest days? And most important: is the writer describing what happened, or predicting what will happen? Those two kinds of sentences require completely different standards of evidence.

For me, the open signal is a question: if the data pipeline of a major league went down for a week, what percentage of the analysis published that week would contain conclusions with no basis at all?

I do not know the answer. But I know I will go looking, because that is the kind of question data can answer — unlike the four fluent explanations sitting in my head at 2:47 a.m.

Basketball is never empty. Only our way of looking at it is empty. And if one night your data table holds not a single number, the right thing to do is not to hunt down a substitute number. It is to say honestly that you have nothing.