Football Data Journal: 1,204 Shots, 81 Empty-Stadium Matches and a Forgotten Right Flank

Marseille, August 2026 — when the spreadsheet started to speak

"In the summer of 2026, I learned to trust something nobody had named yet: xG."

I remember that afternoon in Marseille. Sunlight poured over the old harbour while I sat in my office with two monitors. One showed the Opta dataset freshly released for Ligue 1 in 2026-18; the other was an empty spreadsheet I had built myself. I was 57 that year, had spent more than two decades as a transfer market administrator, and had signed or rejected hundreds of player files based on what my eyes saw plus a handful of scattered numbers sent by scouts.

The new xG table did not excite me. It annoyed me. A column called "expected goals" could claim that my team was winning through luck. It could claim that a striker I had just recommended for the watchlist was really only scoring from positions anybody could score from. Nobody had shown me verification. So, born with a chronic suspicion of data, I decided to do what no colleague wanted to do: count for myself.

Over six weeks I logged 1,204 shots from all 20 Ligue 1 clubs across the first half of the season, recording position, distance, angle, the situation leading to the shot, and the type of play. Then I compared Opta's aggregate xG against actual goals. The correlation coefficient came out at 0.84. That did not prove xG was true. It proved something more useful: the signal was strong enough to use, provided I understood its limits.

Colleagues said my reaction was slow. Some laughed and told me France had been using data for years, and I was merely confirming the obvious. I did not argue. I added one line to my notebook: "Sample size 1,204; confidence interval narrow enough for striker valuation; not usable for defenders or defensive midfielders at this stage."

That was the origin of a private dataset I still maintain today. Not because it is perfect, but because I know exactly where it is weak.

Context: how football learned to count, and where it learned wrong

Football's data history does not begin with Opta or StatsBomb. It begins with men sitting in stands with notebooks and pencils, counting passes while the world watched only the ball. Charles Reep is the name English analysis circles mention with hesitation: credit for logging thousands of matches, alongside the memory that his conclusions pushed English football in the wrong direction for decades.

The lesson from Reep is simple and deeply uncomfortable: record-keeping does not create truth. It creates raw material. Truth comes from placing raw material in the right frame of reference.

The 1990s brought another turning point. Data companies began encoding every on-pitch event: passes, shots, tackles, positions. The event became the unit of measurement. By the 2010s, when computers were powerful enough to run probability models over hundreds of thousands of shots, xG arrived as a natural consequence of event data plus statistical modelling.

But here is what few people tell audiences: every xG model is a philosophical choice. One provider decides a penalty is worth 0.76 goals. Another decides a shot with the weaker foot is worth 12 percent less. These differences are not wrong. They are simply different. And when you blend two xG sources in a single analysis without stating the source, you are mixing two languages and calling it one sentence.

I have watched this happen hundreds of times in scouting meetings. A sporting director opens a dashboard, points at a column and says: "He is 0.08 xG per 90 worse than last season." Nobody asks which data provider. Nobody asks the sample size. Nobody asks which system he played in, which opponents he faced, how many minutes he got.

"I am 66 years old, old enough to know a number never tells a story unless we ask it to."

For the European transfer market, this is a life-or-death issue. A club spends 30 million euros on a 21-year-old striker based on accumulated xG in the Dutch league, then grows disappointed eighteen months later. The story is so familiar it is boring. The fault is not xG. The fault is skipping the hardest part of the problem: league context, teammate quality, the space the tactical system grants, and the ability to react when that space is taken away.

Ligue 1 and the lesson of frame of reference

When Opta published xG data for Ligue 1, the French league occupied a peculiar position. It was the league producing the most young talent in Europe, and simultaneously the league the biggest European clubs treated as a supermarket. Paris Saint-Germain had shifted the financial balance of the entire competition from 2026 onward. But the rest of Ligue 1 still operated on the old logic: buy cheap, develop, sell high.

In that logic, data is not a tool for understanding football. It is a tool for valuation.

I work in the transfer market, so I see everything through one question: what is this player actually worth, and where is the market mispricing him? That question forces me to separate two kinds of numbers that analysts routinely blend.

The first kind describes process: touches inside the box, line-breaking passes, successful presses in the final thirty metres. It shows how often a player participates in high-value situations.

The second kind describes outcome: goals, assists, points. Easy to see, noisy to trust.

For a 27-year-old with four Ligue 1 seasons behind him, outcomes have stabilised and can be used directly. For a 19-year-old with 900 minutes, outcomes say nothing, and you are forced to read process. But reading process in a small sample is packed with risk, because the process of a young player at a club that is constantly pinned back does not resemble process at a possession-dominant club.

The most common error in modern youth valuation models is assigning too much weight to potential and too little to structure. A 19-year-old with strong process metrics at a weak club is typically priced on an ideal scenario: at a stronger club, space will be wider, teammates better, numbers will rise. Reality runs the other way. At a strong club, space disappears. Opponents sit deep. The young player must relearn every decision in a thousandth of a second.

This is why I always split home and away data before proposing any file. A 21-year-old striker who scores nine at home and two away in a season is an entirely different profile from one who scores six at home and five away. Same eleven goals. Different substance.

Le Havre once used this principle to negotiate down the price of a young striker another club was chasing. They said nothing poetic. They simply presented the home-away split, highlighted the gap, and waited. The negotiation closed at a fee below the seller's opening expectation. I tell that story not to boast, but because it proves data only has power when it answers one specific question.

2026: Croatia, PPDA, and the limits of a single letter

World Cup 2026 pulled me out of my Marseille comfort zone. My dataset caught the eye of a sports newspaper, and they invited me to collaborate on all 64 matches.

I was 58. I had no VIP seat. I sat in front of a screen, logged every match, and counted a metric that was barely known in Vietnam at the time: PPDA, the number of passes an opponent is allowed before a defensive action.

The principle is simple. Lower PPDA means higher, more aggressive pressing. Higher PPDA means sitting deeper and waiting.

In the semi-final between Croatia and England, the table showed Croatia allowed England only 8.2 passes per defensive action, while England allowed Croatia 12.5. I filed a short note predicting Croatia would win through their pressing in extra time, once England's midfield lost structure.

Croatia won 2-1 after extra time.

I did not shout in celebration. I reopened the spreadsheet and went looking for outliers, because a correct prediction does not prove a model correct. It only proves the model has not yet been falsified in that case.

And the outlier I found came from Croatia itself. That team reached the final with a PPDA in the highest bracket of the tournament, meaning they did not press high for most of the competition. They pressed selectively, by match phase, with intensity varying by opponent.

"Croatia won a tournament of low PPDA? Then PPDA is only one letter."

I wrote that into my notebook right after the final, and it has been my working principle ever since. A metric is never a story. It is one letter in an alphabet. To read a sentence, you need the other letters.

If you read only Croatia's PPDA, you conclude they were a proactive defensive side. If you add the opponent's possession minutes, how often Luka Modrić received the ball between the lines, and how many long passes Croatia were forced into when pinned back, a different picture emerges: a team that chose its moments proactively, not its space.

That distinction matters for Vietnamese followers, because we tend to copy metrics without copying the circumstances that produced them. A V.League side cannot have Croatia's PPDA unless it has a central midfielder capable of turning a long ball into a short one in two touches.

2026: when empty stands turned football into a laboratory

"Empty stands are the finest laboratory for anyone who loves data."

In May 2026 the Bundesliga restarted after the shutdown. My editor assigned me to cover it, because I had already used World Cup 2026 data to write about pressing and structure.

I sat in Marseille and analysed 81 matches played in empty stadiums over the remainder of the 2026-20 season. The result made me reread the spreadsheet three times.

The home win rate fell to roughly 26 percent, against a long-term Bundesliga average of 43 percent before the pandemic. Home goals dropped clearly. Yellow cards for away teams fell. Referee decisions favouring the home side fell.

I wrote a report with a plain title: "Empty stands kill home advantage." The report argued that home advantage in professional football does not come mainly from the pitch or travel fatigue, but from social pressure: crowd noise nudging referees toward the home side in contested calls, away players hesitating in decisive challenges, home players running harder in the final fifteen minutes.

Remove the crowd and you remove that pressure. The match becomes a pure test of tactics and fitness.

For me this was a rare natural experiment. Football seldom lets you isolate a variable. The pandemic did it for free.

But I also had to state the limits. Eighty-one matches is a short sample. Clubs were unprepared, the schedule was compressed, injuries rose, and crucially there was no normal recovery time. The conclusion "empty stands kill home advantage" holds under those conditions; it is not automatically true for a normal season.

This is why I never publish a table without its limitations. Readers deserve to know where my model is weak.

Since that report, every statistical table I build separates home and away metrics, including when analysing Vietnamese players. When assessing a young V.League striker, I always split home and away goals before comparing against any xG figure.

Vietnamese football and the data gap

Talking about Vietnamese football in numbers is far harder than talking about Ligue 1. Not because Vietnamese football has less to record, but because the recording infrastructure is not yet synchronised.

When I follow V.League matches through streams, the first thing I notice is uneven event data quality between rounds. Some rounds I get enough data on touches by zone; other rounds I have only passes and shots. To analyse pressing, I have to count myself.

That does not discourage me. It reminds me of 2026, when I counted 1,204 shots myself because I did not yet trust Opta's xG table.

The bigger problem sits in league context. The V.League has features that strip value from many European metrics. Pitches vary in size and turf quality. Weather swings sharply between regions. Travel distances for away fixtures create physical strain unlike anything a German player experiences going from Munich to Dortmund.

This means that if you copy a European PPDA model wholesale and apply it to a V.League team, you are measuring something else while naming it with a familiar term. I have seen that error in many youth analyses across Southeast Asia.

One example I use when talking with younger colleagues: the trajectory of a striker who graduates from an academy to the first team. His numbers in youth football always look good, because space is wide, opponents lack physical maturity, technical errors are not punished. In the first team, numbers fall. The player did not get worse. The context got harder.

A good data analyst does not see falling numbers and conclude the player is not ready. A good analyst asks: which categories fell, over how many minutes, when starting versus coming off the bench, and against which tier of opponent.

Those three questions separate a scouting report with value from a decorative spreadsheet.

Transfer valuation: where data meets people

"Players are variables, the market is a function, but most of my life has been a constant."

I have been in the transfer market long enough to know the final decision is rarely made by data. Data narrows the option set. People choose within it.

One file I handled had every good number: a 22-year-old, top-bracket xG per 90 in his league, high touches inside the box, strong conversion of good chances, no serious injury history. Every dashboard agreed.

We did not sign him.

The reason sat outside the spreadsheet. Through three independent sources we learned the player had a long-running conflict with the senior group in the dressing room, and had once refused to join a collective recovery session for personal reasons. No metric captures that. But everyone in the industry knows a fractured dressing room can sink a season, and the cost of repairing it usually exceeds the cost of buying a different player.

This is the largest blind spot in modern valuation models. They measure potential well, technique well, physicality well, but they barely measure dressing-room chemistry. And chemistry is the single variable with the greatest effect on the real value of a contract.

I say this not to dismiss data, but to place it correctly. Data is a noise filter, not a decision machine.

There is another mispricing pattern I watch closely: overvaluing youth potential and undervaluing experience in difficult circumstances. A 30-year-old defender who has survived two relegation battles, who knows how to stay calm when his team concedes in the 80th minute, is usually priced below a 23-year-old with better passing numbers who has never played a decisive match.

In a cup tie, that difference can be an entire season.

2026: Morocco's right flank and the art of reading compensating variables

My empty-stadium report reached Canal+, and in 2026, at 62, they sent me to Qatar for the World Cup.

That tournament delivered Morocco to the semi-finals, one of the finest tactical stories I have covered live. It also taught me the biggest lesson of my career about reading individual metrics.

Achraf Hakimi received universal praise. Dashboards showed more than 142 sprint efforts across the tournament, roughly 2.3 chances created per match, and involvement in nearly every Moroccan wide attack.

I dug into positional data and found something that made me stop. The corridor behind Hakimi was empty for about 34 percent of match time. In other words, for more than a third of the match, the zone he vacated when advancing had nobody covering it.

Morocco did not collapse because their centre-backs covered at speeds above 31 km/h. That is a technical detail statistical commentary usually ignores, because it does not sit in the full-back's column. It sits in the centre-back's column.

I wrote a cautionary note arguing that a high full-back model only holds if the defensive structure behind it is fast enough to compensate. That note did not deny Hakimi's talent. It simply said his talent was standing on a foundation not every team possesses.

Against France, the French side attacked Morocco's right flank relentlessly. The final result is known to the world. But what I want to stress is not that my prediction was right. It is the method: when evaluating a new tactic, list the compensating variables before listing the benefits.

A high full-back needs three compensating conditions: centre-backs quick enough to cover, a defensive midfielder able to read situations and fill the corridor, and a wide forward diligent enough to track back on turnovers. Miss one of the three and the system breaks.

Every tactical trend has necessary and sufficient conditions like these. Analysts have a duty to state them, even when doing so makes the story less exciting.

The contrarian angle: correlation is not causation

This is the part I repeat most often in club workshops, and the part that gets me labelled difficult.

A metric correlated with success does not prove that metric creates success. For years people observed that champions score a lot, then concluded that scoring a lot causes championships. The conclusion is true and useless, because it does not tell you how to score more.

More dangerous is the intermediate case. A possession-dominant team usually generates higher xG. But if you chase possession by passing sideways more, you raise the possession share while lowering xG. You have optimised a correlated metric and damaged the underlying variable.

"There are matches won on the pitch but lost on the spreadsheet — I choose the spreadsheet."

That does not mean I dismiss victories. It means I want to know whether the victory can repeat. A team that wins 1-0 with 0.4 xG against 2.1 xG for the opponent can win this time, and will usually lose the next meeting if everything else stays the same. Fans have the right to enjoy three points. Professionals have the obligation to remember the spreadsheet before making personnel decisions.

Three hypotheses I always write down before concluding anything about a match:

First, the result came from execution quality. Players did exactly what the system demanded, and the opponent had no answer.

Second, the result came from cyclical luck. A player finished from a difficult position and the ball went in, or the opposing goalkeeper made a positioning error.

Third, the result came from measurement error. Your metrics failed to capture something important that happened on the pitch.

Of the three, the third frightens me most, because it means I am missing data, not that the match lacked logic.

Sports business and the pressure of financial reporting

Over more than thirty years working in Europe, I have watched many clubs move from traditional ownership to public company structures. That shift carried a consequence fans do not immediately notice: quarterly reporting pressure began to weigh on sporting decisions.

A listed club must publish revenue, costs and profit on a cycle. Management must answer to shareholders. Shareholders want growth. Short-term growth usually comes from selling young players or cutting the wage bill.

I do not oppose clubs making money. I simply observe a pattern: when sporting decisions are subordinated to financial decisions, squad quality tends to be traded away over a two-to-three-year window. Talented youngsters are sold before maturity, defences are not reinforced at the right moment, and managers are judged on short-term results while the squad they inherited reflects financial decisions made two seasons earlier.

In the V.League and Southeast Asian competitions, this trend is only emerging at a smaller scale, but the logic is identical. When a club moves from sponsor-funded to self-balancing, the wage bill becomes the pivotal variable, and squad quality depends on whether that club's academy is good enough to fill the gaps.

For analysts, this means sporting data no longer stands alone. To assess a contract, you must understand the club's financial statements. To understand why a manager was sacked, you must read the cash flow, not just the league table.

Esports: the same data heartbeat

There is a subject I came to late but now pursue seriously: esports.

"A single mouse click on an esports screen carries the shape of a football pass."

When I first followed a professional competition, I was struck by the degree of datafication. Every action is logged: position, reaction time, hit rate, distance travelled on the map, number of engagements, number of angle openings. Nothing escapes.

The problem lies elsewhere. Heavy datafication pushes coaches to optimise players toward templates. A metric showing a strong reaction time leads to training a player to operate on a preset rhythm. Individuality, the thing that produces actions nobody anticipates, tends to be smoothed away inside that system.

I see the parallel in football. When academies develop players through data models, young players learn to choose the safe option with good metrics rather than the risky option that might create the decisive moment. Over the long run, this reduces the number of individuals capable of determining a match.

This is not a complaint about technology. It is a warning about using data to optimise rather than to understand. Optimisation and understanding are different goals, and confusing them produces consequences that take years to surface.

What data cannot measure

I want to use this section to list what fifty years of observation have taught me that data cannot reach.

Data measures running speed. It does not measure the decision not to run in order to hold position.

Data measures completed passes. It does not measure the pass that forces an entire team to change its movement pattern.

Data measures ball recoveries. It does not measure the recovery that makes 60,000 people stand up.

Data measures a goalkeeper's goals conceded. It does not measure that goalkeeper's effect on how the defence organises in front of him.

And data cannot measure one simple thing: whether a player wants to stay at the club.

In transfer work, that question usually matters more than any technical metric. A player who wants to stay trains differently. He pushes harder in recovery sessions. He talks to younger teammates. None of this appears in a dashboard, yet it spreads through a dressing room like a kind of temperature.

Clubs that understand this pay lower wages and keep their squads. Clubs that do not pay higher wages and sell players after eighteen months.

Sixty-six years old, and still counting

At 66, I still keep the habit of writing by hand. Not because I distrust software, but because writing by hand forces me to slow down and think.

Every week I spend two mornings updating my Ligue 1 dataset. Every season I pick one topic for deep study. This season my topic is compensating variables in high defensive systems, after watching matches in Europe and Southeast Asia show me that young teams are copying the model without building the compensation capacity.

I still follow the rule of counting a sample myself before using any externally supplied metric. With xG, I verified. With PPDA, I verified. With newer metrics such as action value models or controlled space models, I am still verifying, and I am not ready to conclude.

That is why I never praise a new metric simply because it is new. Novelty is not evidence. Novelty is an invitation to verify.

Takeaway: signals for the next round

If you follow football through data, here is what I consider worth watching in the period ahead.

First, watch clubs shifting from a low block to mid-block pressing. This is the most dangerous transition zone in modern football, because it demands continuous compensation that not every club can build. The metric to track is not PPDA but the average distance between the defensive and midfield lines at the moment of turnover.

Second, pay attention to young strikers with a large home-away split. They are the most mispriced group in the transfer market, in both directions.

Third, watch how Southeast Asian clubs handle wage bills over the next two to three seasons. When the financial model changes, squad quality changes about a season and a half later, and that window is where good transfer decisions carry the highest value.

And finally, if you work in analysis, keep a notebook. Not to record what the model says, but to record what you counted yourself and what you have not finished verifying. That notebook will be more honest with you than any dashboard.

"A cancelled match is not a loss of points; it is a lost page of the journal."

I have lost quite a few pages in my life. But the pages I kept are enough to make me believe one thing: data does not replace the eye. It only forces the eye to be more honest.