When the Spreadsheet Goes Silent: The Discipline of the Null Result in Football Data Analysis

When the Spreadsheet Goes Silent: The Discipline of the Null Result in Football Data Analysis

11:30 p.m., Rangpur. I opened the dashboard and my eyes locked on a single column. The match log had logged 1,842 passes, 24 shots, 17 corner-related events. Yet the model output kept spinning one symbol — N/A. The same pipeline had returned 1.7 versus 0.9 xG just a week earlier. Today, nothing. The first reaction was habitual, almost automatic: "Maybe this match had nothing worth saying." That sentence is the most dangerous instinct in football analysis. Absence and missingness are not the same thing — one is the absence of an event, the other the absence of evidence. A professional analyst must judge which is which, every time. Across years of watching matches and pulling data, I have learned that most bad calls come not from wrong data — but from treating missing data as truth.

First, a methodology box, because the Data Monk never writes without one. Data source: event-level feed (passes, shots, defensive actions), sample size 1 match, model version 2.1. Glossary: xG (Expected Goals — the probability a shot becomes a goal), PPDA (Passes Allowed Per Defensive Action — lower means more aggressive pressing). Caveat: all null-result claims are provisional, subject to re-verification.

This box is not decoration, it is protection. A piece without source, sample size and model version cannot be audited — and unaudited analysis is not football journalism, it is opinion.

In 2026, when I left civil engineering for journalism, numbers were ornament to me. That changed in 2026, when I built my first xG model in an internet café in Rangpur. Logging 1,842 passes and 24 shots for Abahani Limited Dhaka versus Sheikh Russel KC, I saw the model say Abahani's 2-1 win was flattered — 1.7 versus 0.9 xG. I learned then that the real job of data is not to explain the match but to measure our confidence in the match. And confidence never begins with a silent zero. I audited that spreadsheet and found it had not lied; rather, the derby chose chaos.

After Croatia beat England 2-1 at the 2026 World Cup in Russia, I pulled the PPDA — 8.7 — and Luka Modric's distance covered, 13.8 km. Combining the two, I drew a pass-network map showing how Croatia bypassed England's press in extra time. That piece was cited by two national radio shows. But the part nobody remembers: before I drew that map, three event-windows had come back empty from my feed. I did not fill them with guesses; I re-pulled them. Modric's press became a story because every number behind it had been audited.

When COVID-19 halted sport in 2026, I built an empty-stadium model from Rangpur. In Bundesliga restart data, home xG for Bayern Munich versus Borussia Dortmund fell from 2.1 to 1.4, and home advantage dropped from 0.42 to 0.18 goals. I published daily data bulletins for 47 days straight, because I had a clear process. From then on I shifted to predictive writing — not "what happened" but "what the data expects if X happens."

But today's question is harder. When the whole pipeline goes silent, when an analytical step returns only N/A — what then? This is where most analysts, and most newsrooms, make their first mistake.

Three kinds of zero

Every null result belongs to one of three families. Fail to separate them and analysis becomes meaningless.

First — the false null. The data existed, but the pipeline failed to catch it. Event feed drops, timestamp gaps, lost shot coordinates, or outright ingestion failure. Say a team's PPDA reads 0.0. On the surface it looks like the team never pressed — passive, timid football. In reality, the defensive-action feed often simply failed to load that match. A zero press and missing press data are not the same thing, yet on the dashboard the two look identical.

Second — the true null. The event genuinely did not happen. The team really did sit deep, did not press, did not create big chances. This zero is valuable — it is evidence of a real tactical decision. But a true null is only true when your feed is complete.

Third — the ambiguous null. Partial capture. The sample is so small the zero means nothing — half a match, a ten-minute spell, an uncovered league. Here there is no number, but absence is not proof either.

Confusing these three families is the most common and most expensive error.

Zero and an empty column are different things

There is a technical subtlety here that almost nobody in football analysis states. A database holds two different things: NULL and zero. NULL means "I don't know." Zero means "I know, and the value is zero." Confusing them is one of the oldest bugs there is.

If a defender has NULL interceptions, it means we do not know how many balls he cut out. If it is zero, it means he genuinely did not cut out a single one. A midfielder's NULL progressive passes means nobody counted them in that league; zero means he genuinely did not play a single one forward. The first lesson any football analyst should learn: never read NULL as zero. Because the moment you treat NULL as zero, your whole model brands as bad a player who was merely uncounted.

How to tell them apart

The method has three layers. Layer one — metadata audit. Check ingestion logs, timestamps, and whether the event count matches the match length. If a 90-minute match's event stream suddenly stops for 20 minutes, that is not the match's story, it is the pipeline's story.

Layer two — source triangulation. Cross-check the same event against at least two independent sources. I always keep a video audit beside the PPDA. If the screen clearly shows a forward being pressed by two or three players every time he enters the opposition's defensive third, yet the model says the press was zero — the problem is not on the pitch, it is in the file.

Layer three — a confidence band. A zero should never be read as a single number. It must always carry a possible range and an error margin. Only a zero that has passed these three layers is credible.

Real examples from football analysis

The error happens almost daily in scouting. A database having nothing on a lower-league player does not mean he is bad — it means that league is not covered. Likewise, a young academy player's "zero progressive passes" is not proof of his limitation, unless his entire team's match was defensive. Missing information is not a statement; it is a gap.

The problem is sharper in international tournaments. Under World Cup pressure every match is analysed, but smaller federations have weak data infrastructure. So a team's attacking index sometimes reads zero — while on the pitch it created ten big chances. To know the difference you need both the shot map and the video.

The imported-framework trap

I was born in the UK and work in Bangladesh — so I know a Premier League model cannot simply be dropped onto the Bangladesh Premier League. In England thousands of events are recorded per match; here sometimes a few hundred. Fill that gap with English-league averages and the analysis swells with numbers but not with truth. A big league's zero and a small league's zero are not the same — one is an evidence gap, the other an infrastructure gap. Local models must be calibrated to budgets, travel, coverage and institutional limits.

The confusion of treating a zero as truth

Now to the real point. Suppose the first step of an analysis pipeline returns an empty result — no headline, no information points, no entities identified. If someone reads that empty result as "the article has no information," they are assuming the feed is a true null. But the more likely reading is that some pipeline step failed silently. A failed extraction and an information-poor article are two different diseases; so are their treatments.

Here the Data Monk's core principle applies: a null result is never publication material, it is a process signal. The first occurs when you are sure the feed is complete; the second when you are not. Publishing a null result while unsure means passing your ignorance off to the reader as information.

The limits of the spreadsheet

I am a Data Monk, but I never treat the spreadsheet as scripture. A model represents reality, it is not reality. That first Rangpur model taught me one limit — however good the model, if the feed does not arrive, the model is silent. Ten years on I know another limit: the most dangerous zero is not the one I can see, but the one I left unfilled.

The human brain cannot tolerate a gap. A player "was quiet" — that sentence is often just a polished version of "his data was missing." When a broadcaster says "the defence looked unorganised today," the question should be — is that eye-witness testimony, or a data gap covered with story? From years of watching matches I have learned that narrative fills a zero, and that filling often does more damage than the statistic itself.

Succession and process codification

Null-result discipline is not a personal skill, it is an institutional process. A newsroom or analytics team should write down a "null protocol": what to do when a zero arrives. My recommendation is four steps.

One — stop, do not publish. A null result must never be filled with guesses under deadline pressure. Two — escalate. Tell the pipeline engineer; check the ingestion log. Three — re-run. Extract again from the source; cross-check on a different source if possible. Four — flag it as an exception. Put it on the error list, not the analysed-article list. Until a valid result arrives, this is a process failure, not content.

This protocol is for speed as well as accuracy. In a crisis you can write fast — but only if it is already decided which information is trustworthy and which is not. In a crisis, speed comes from process, not emotion.

It is also for the next in line. I teach new juniors on day one — the first question is not "what is the number," it is "where did the number come from, and what did not come." A team's real legacy depends on its templates, its protocols and its training. A newsroom that never learns to recognise a zero will one day sell a zero to its readers.

The contrarian angle

An uncomfortable question surfaces here. We often use data as a witness, yet the zero reminds us — this witness has gaps in its eyes. The biggest danger is not data error but blind faith in data. Yet the reverse is also true: over-caution paralyses decisions. If at every zero you stop saying "maybe the feed is wrong," you will never reach a conclusion. The balance is — stop at a zero to verify, but do not stop forever.

A statistic can be true and still misleading. Correlation is not causation. A team's low pressing index does not mean it is bad — it may have deliberately sat deep. So learn to question the number, and question the zero most of all.

Next-round signal

When you open the dashboard next match, look at the pipeline before the pitch. If the model returns empty, that is a statement about your process, not the match. Today's question is not only for football analysts but for everyone: when information goes silent, will you listen to the silence, or build a story? Set a date for the next round — the day you will test your own null protocol again.