The Empty Data Frame: The Core Sickness of Modern NBA Analysis

On the night of February 1, 2026, I sat in front of my screen at 2 a.m. Hanoi time, waiting for the official announcement about the Luka Doncic trade. When the first line appeared from Shams Charania, the first thing I did was not open Twitter, but open my dataset — an Excel file I have maintained for seven years, tracking every transaction since the 2026-19 season with full salary figures, contract terms, and cap status. I typed the names of the two teams, Dallas Mavericks and Los Angeles Lakers, into the search box. The dataset returned a blank frame. Not a single row. My filter had not been designed for the scenario in which a team trades away its 25-year-old superstar mid-season with no prior signal.

That blank frame was the moment I realized something I had avoided for years: most modern NBA analysis is built on empty data frames, and worse, our analytical systems have no mechanism for detecting this — they simply fill the gap with plausible-sounding assumptions. This sickness is not only in writers like me. It lies in how teams evaluate players, how media construct narratives, and how fans form beliefs about the game.

I am not writing this piece to retell the Doncic saga. I am writing to address what analysts call "null handling" — dealing with empty values — and why, in basketball, the failure to handle empty data is the most deadly flaw in any model.

The necessary context for understanding this lies in the data infrastructure the NBA has built over more than a decade. Since the 2026-14 season, the league introduced the Second Spectrum system — six cameras mounted on ceilings tracking the movement of the ball and every player at 25 frames per second. Before that, NBA data consisted only of the box score, meaning the final outcome of every action. After Second Spectrum, we knew a player's top speed, distance traveled per game, number of touches, pass height, and hundreds of other micro-metrics. By the 2026-18 season, the NBA released PBP Stats — play-by-play data labeled by action type (pick-and-roll, isolation, spot-up, transition). And EPA (Expected Points Added), provided by companies like PBP Stats and Synergy Sports, became the standard measure of value per possession.

The problem is that when the data catalog expands, analysts tend to believe their system is complete, while in reality the gaps have merely migrated from the obvious to the invisible. If in 2026 you knew you only had the box score and were humble about it, then in 2026 you have two hundred metrics and you have lost that humility. I call this phenomenon "the illusion of total coverage" — when a data system is large enough to appear complete, its operators stop asking themselves what is missing.

This has haunted me since the 2026 World Cup — an event I often describe in talks with colleagues at VnExpress. I built a model predicting Germany would advance from the group stage based on the highest accumulated xG in the group. Germany was eliminated. Looking back, I realized my model lacked data on Japan's defensive pressure — they registered a PPDA of 6.8 in their matches against Germany and Spain, a metric outside my collected dataset. My data frame was not empty. It was empty only where I did not know it was empty. And so I filled it with confidence.

In the NBA, the same mechanism of error appears at three distinct layers: the game-data layer, the player-data layer, and the transaction-data layer. I will walk through each, because I believe how we handle the blank frame at each layer determines the quality of the entire analytical field.

The first layer is game data. Take a concrete example from the 2026-25 season: the Oklahoma City Thunder finished the regular season with a 68-14 record, best in the NBA, and the league's best defensive rating (DefRtg) at 106.6 points per 100 opponent possessions. But when you look at the box score of a specific game the Thunder won by 20, you see nothing but the final number. You do not see that Shai Gilgeous-Alexander ran 4.2 kilometers in that game, or that Chet Holmgren altered the direction of seventeen shots while blocking only three. The box score is an empty data frame disguised as a complete one. And if you only read the box score, you are building conclusions on a blank frame — not because the data does not exist, but because you have not pulled it in.

I tested this myself in mid-March 2026. I took ten random Thunder games and recorded Shai's box score stats next to Second Spectrum metrics. On average, Shai scored 32.7 points in that sample. But "points created" — points he generated directly or indirectly through passes — was 47.1 per game. The gap of 14.4 points per game is the invisible portion of the box score. If you evaluate Shai only by the box score, you evaluate him at 68% of his true value. That is not a small margin. That is misreading a superstar.

The second layer is player data. This is where I see the most blank frames in my own analytical files. Take the "empty stats" debate — data that amateur analysts often use to diminish players on weak teams. The typical example: a player averaging 25 points on a team losing 60 games. By empty-stats theory, these points are meaningless because they contribute nothing to wins. But I asked myself: if his data is not placed in the team context, how do I know he could not carry a stronger team? This is a blank frame about context, not about numbers. And when I pull in usage rate (USG%) and efficiency (TS%), I often find that players labeled "empty stats" typically have a TS% equal to or higher than the league average — they are just forced to shoot more because nobody around them can. For example, in the 2026-24 season, Tyrese Maxey of the Philadelphia 76ers averaged 25.9 points with a 57.7% TS and 27.3% USG while Joel Embiid missed many games to injury. If you only read the 76ers box score without data on how Maxey had to carry the offense in that context, you will rate him below his real value.

This is where I want to spend a few lines on a principle I learned from the collapse of my own model: every empty data frame needs to be clearly marked as empty, rather than being auto-filled with assumptions. In NBA analysis, this means every time you lack data on a player — injury, psychology, tactical role, coach relationship — you must write "insufficient data" in your report instead of automatically inferring from the numbers you have. I add a "risks and gaps" section to every piece after the 2026 World Cup for that reason. I do not believe in intuition. But I believe in what intuition confirms when the data supports it — and when the data is absent, I need to know that I am standing on a gap.

The third layer is transaction data — where the blank frame does the greatest damage. Back to the Doncic deal. The February 1, 2026 trade sent Luka Doncic from the Dallas Mavericks to the Los Angeles Lakers, in exchange for Anthony Davis, Max Christie, a 2029 first-round pick, and a second-round pick. I built a simple model to value both sides of the trade based on estimated production value (EPM) over the previous three seasons and player age. The Lakers' side acquired a 25-year-old with three consecutive All-NBA selections and an EPM of +6.8. The Mavericks' side received a 31-year-old with a significant injury history, an EPM of +3.4, and three remaining years worth 160 million USD. By every asset-valuation model in professional sports, the Lakers won this trade overwhelmingly.

But where was my data frame blank? It was blank where I had no data on the meetings between the Mavericks' leadership and Doncic's representatives about fitness — the weight and training-regimen concerns that were rumored. I had no data on Doncic's feelings about the team's future after reaching the 2026 NBA Finals and losing. I had no data on whether the Mavericks genuinely believed they could retain Doncic in the 2026 free-agency period. Those are three important fields — arguably the three most important — and I did not have them.

That night, I posted a short line on my personal Twitter: "The value based on public data is reasonable for the Lakers. But I am missing about 60% of the information needed to judge." A colleague replied publicly: "Why so humble? You are the expert." I did not respond. Because the real answer is: epistemic humility is not a moral virtue in analysis. It is a structural requirement. If you do not know which blank frame you are standing on, you are not analyzing. You are storytelling.

When I said this at a seminar with young sports journalists in Hanoi in November 2026, one asked me: "Are you ever afraid that if you admit your data is incomplete, readers will lose trust in your entire piece?" I answered that I fear the opposite. I fear readers trusting my piece completely, then discovering that I led them to a conclusion built on a blank frame. Reader trust is the writer's asset. Losing it because of an overconfident piece is far worse than admitting the gap upfront.

I have witnessed this from the receiving end of criticism. In 2026, when I wrote that Hanoi FC deserved to win 3-1 against Quang Nam rather than winning 1-0 by luck, I was mocked for "football is not mathematics." But what I learned was not that data is right. What I learned was that I presented my conclusion too confidently while my data frame only had xG and possession — no data on how deep Quang Nam defended, no data on whether Hanoi deliberately slowed the tempo. A week later, coach Chu Dinh Nghiem admitted he had reviewed the tape and changed tactics based on that analysis, and I could have been happy. But I reflected on my presentation method, not the conclusion.

An empty data frame is not a defect. It is a fact. The issue is how we respond to that fact. There are two failure modes. The first is "fail-open": when the frame is blank, you continue analyzing with reduced confidence, fill the gap with plausible assumptions, and deliver a conclusion that sounds good. The second is "fail-closed": when the frame is blank, you refuse to conclude, you mark the gap, and you request more data. I was in the first mode for years. I switched to the second after the 2026 World Cup, and I lost some readers because of it.

So what makes holding the second mode difficult? Time pressure. In modern sports media, the window to deliver post-game takes is most valuable in the first 12 to 24 hours. After that, readers move to the next game. If you spend 24 hours gathering enough data for a proper analysis, you have missed the window. If you respond immediately, you are delivering a conclusion based on a blank frame. This is a real paradox every data-driven sports writer must live with. I resolve it by clearly distinguishing two types of responses: "initial observation" — clearly marked as preliminary — and "complete analysis" — published once the data is full. My readers know the difference, and they have gradually accepted that a serious analytical piece may take 48 hours to appear.

All of this leads me to an angle I believe is counterintuitive to most contemporary analysts. When I talk to younger colleagues, the most common worry I hear is whether they have too much data. They fear being overwhelmed. They install software to filter metrics. They keep only ten to fifteen core metrics per piece. And I understand that — I used to do the same. But I argue the real problem of modern NBA analysis is not too much data. The problem is that we believe we have enough data, while in reality we have enough data about the things easy to measure and missing data about the things that decide.

The paradox of data collection is: the data easiest to collect is often the data least decisive, and the data hardest to collect is often the data most decisive. Points are easy to collect. Man-to-man defense is hard. Usage rate is easy. The relationship between player and coach is hard. A player's mental state after losing a family member is hard. And because the sport-media incentive structure rewards volume — more metrics is better, more articles is better, faster is better — we tend to drown in easy data and ignore hard data.

From that angle, one thing I consider important that modern NBA teams are beginning to publicly acknowledge: their models also have blank frames, and they are shifting from fail-open to fail-closed, just as I did. For example, in the 2026-24 season, when the Oklahoma City Thunder decided not to chase deals at the trade deadline despite a strong record, they did not explain it as "we are strong enough." They explained it as "we do not have enough data on whether a new player could integrate within four weeks." This is fail-closed in practice. The team admitted the blank frame and refused to act on it. The result: they reached the 2026 playoff semifinals, and in 2026-25, they finished with the league's best record. I am not saying patience was the sole cause of that success — I am not repeating the old correlation-causation mistake. But I am saying that their acceptance of standing on a blank frame at a decisive moment did not harm them, and to me, that is enough evidence to believe fail-closed can work in practice.

On the other side, I have seen cases where fail-open caused clear damage. The typical example is how many teams handled load-management data. Before detailed Second Spectrum movement-intensity data existed, teams relied mainly on coaches' intuition and medical-department reports. When tracking data arrived, many teams went fail-open: they continued to use intuition alongside data, combining two types of information without a mechanism for deciding when one outweighs the other. The result is injury rates that remain high and controversial, and by the 2026-25 season, the NBA was forced to introduce stricter resting rules, including financial penalties for teams resting stars in nationally televised games. This does not solve the blank-frame problem — it merely shifts responsibility from teams to regulation.

So what can we do about the blank frame in NBA analysis? I propose three practical principles that I personally apply and have found effective.

Principle one: every important metric should come with a confidence interval, not just a single number. When I write that a player has a 60.2% TS, I must also write that with a 47-game sample, the 95% confidence interval ranges from 57.8% to 62.6%. This changes how readers understand the number. It is no longer an absolute truth. It is an estimate with a margin of error. And when the margin is wide enough, readers understand for themselves that a strong conclusion is inappropriate.

Principle two: every analysis should include a risks-and-gaps section, written clearly, not just a footnote. I began writing this section in the 2026-23 season after the collapse of my World Cup model. At first it was only three lines. Now it is often as long as a quarter of the piece. I told my editor that if he cuts this section to save length, he is converting the article into a blank frame. He no longer agrees to cut it.

Principle three: clearly distinguish between what the data says and what I infer from the data. This is the hardest principle because it demands linguistic discipline. When I write "Doncic's EPM of +6.8 over the past three seasons ranks third in the league," that is data. When I write "therefore Doncic is the third-best player in the league," that is inference. These two sentences need to be written differently and marked differently. I use the simple present tense and specific sources for the first, and the conditional for the second.

These are small principles. But I believe the accumulation of small principles determines the quality of the whole analytical field. Numbers never need us to defend them. On the contrary, we need them so we do not lie to ourselves. And the worst thing we can do with a number is believe it is the whole truth.

I still keep my seven-year Excel file. It still has blank frames where I have not pulled data in — especially in the columns on players' mental states and locker-room relationships. I have tried to build proxy metrics for those factors — disciplinary violations, missed practices, social-media interaction frequency — but I know I am only measuring the shadow of the truth, not the truth. And that is something I accept. Some things in basketball will remain blank frames forever. How we treat that blank frame — with structural humility or with substitute confidence — is what distinguishes analysis from narration.

On the night of February 1, 2026, I did not write about the Doncic deal immediately. It took me three days to gather enough data from public sources, from subsequent reports in American media, and from informal conversations with people working in the industry. When my piece appeared on the 4th, it was not the first, and it was not the most read. But it was the only one I could reread a year later without seeing that I had built conclusions on a blank frame. To me, that is enough.

The signal I am tracking in the next cycle is not about the Doncic deal, but about a broader trend: whether NBA teams will begin publicly acknowledging the blank frames in their data. In the 2026-25 season, I heard at least two team executives say in interviews that they "did not have enough data" to evaluate certain situations — a phrasing I had not heard in ten years prior. If this trend continues, we may be witnessing a shift of the entire sports-analytics industry from substitute confidence to structural humility. If it stops, we will keep seeing good analyses, strong conclusions, and ever-larger blank frames beneath them. And once a blank frame is large enough, no analysis — however elegantly written — can bear its weight.

A contract is only truly right when the number signs alongside the signature. An analysis is only truly right when the writer knows what he is standing on. And when the writer does not know what he is standing on, the best way to respect the reader is to say exactly that.