Since its introduction in the major European leagues, VAR has been presented as a refereeing revolution. It would put an end to gross errors, make the game fairer, substitute objectivity for human fallibility. Seven years later, a meta-analysis covering 15,000 matches from the four major European leagues provides a more sobering answer: of ten game variables studied, only one changes significantly thanks to the technology. Biases have not disappeared. They have changed form.

This result is not an indictment against video in football. It is a more general, and more uncomfortable, lesson: technology rarely corrects what it promised, and almost always something other than what was expected of it.

The Essentials

  • A meta-analysis published in 2026 in Frontiers in Psychology covers 15,000 matches from the Premier League, Liga, Serie A, and Bundesliga since the introduction of VAR.
  • Of 10 variables analyzed (goals, penalties, red and yellow cards, offsides, possession, shots, corners, fouls), only offsides decrease significantly with VAR.
  • The effect on penalties and red cards varies by league, suggesting that technology redistributes refereeing power rather than eliminating it.
  • At a time when several federations are considering AI systems for automated refereeing, this interim assessment redefines the terms of the debate.

Only One Variable Out of Ten Really Changes

The meta-analysis published in Frontiers in Psychology rests on a rigorous protocol: 15,000 matches, four leagues among the most scrutinized in the world, a period covering the years before and after the introduction of VAR in each championship. The researchers tracked ten quantifiable variables: the number of goals scored, penalties awarded, red and yellow cards distributed, offsides flagged, and several raw play indicators such as possession, shots on target, corners, and fouls.

The main result is uncomfortably clear. Of these ten dimensions, only one experiences a statistically significant variation: offsides. Their number decreases measurably after the adoption of VAR. This is not nothing: for decades, contentious offsides calls have been one of the main sources of frustration for players, clubs, and supporters. Technology has solved this precise problem, the only one for which it offered a quasi-perfect mechanical solution: the offside line is traceable to the millimeter, the stoppage time is measurable, the decision lends itself to automated verification.

But for the rest? Nothing, or almost nothing. Goals scored per match remain stable. The number of penalties awarded does not change significantly across the four leagues. Same for red cards, whose evolution falls within expected statistical margins.

It is not that VAR is useless. It is that it is useful differently than expected.

Biases Do Not Disappear, They Migrate

The most unexpected effect of the meta-analysis concerns what researchers call the “redistribution” of refereeing power. When the results are broken down by league, a more nuanced picture emerges: in some championships, penalties increase after the introduction of VAR; in others, they decrease. Red cards follow the same heterogeneous pattern.

This dispersion is not a statistical artifact. It reflects something structural: VAR does not replace the referee’s judgment, it modifies the conditions under which that judgment is exercised. Before VAR, a referee would sometimes refuse to whistle an obvious penalty because he doubted his positioning or anticipated difficult protests to manage. With VAR, the same referee can instead whistle what he would not have dared to, knowing that a video check will serve as his safety net. In other cases, the presence of VAR encourages restraint: the referee hesitates to decide, implicitly waiting for VAR to intervene, which produces the opposite effect.

Refereeing bias is not a correctable error. It is a set of cognitive mechanisms, social pressure, institutional context. Changing the tool modifies the context, but the mechanisms rebuild themselves around the new tool.

This observation is not unique to football. The literature in behavioral psychology has long documented this phenomenon: introducing an external control system tends to displace individual responsibility rather than eliminate it. The human actor adapts to the presence of technology, often in unanticipated ways. It is the same phenomenon observed with driver assistance systems in automobiles, which reduce certain frontal collisions while promoting increased inattention in other dimensions.

The Promise of Objectivity Was a Political Promise

It is worth asking why VAR was presented and adopted with such unanimity. Decision-makers in global football, FIFA leading the way, sold the technology as a tool of objectivity: the camera does not lie, the still image does not make mistakes, the slow-motion replay enables a decision based on facts. This rhetoric is what silenced skeptics and convinced reluctant leagues.

Yet the objectivity in question was always partial. Offsides, yes, lend themselves to mechanical measurement. But fouls in the penalty area? The notion of intentional or unintentional play? The intentionality of a gesture that warrants a red card? These questions are not reducible to a pixel on a screen. They involve interpretation, a norm, a culture of play that varies from one country to another, from one championship to another, from one generation of referees to the next.

The promise of total objectivity was therefore a political promise before it was a technical one. It allowed institutions under pressure to demonstrate that they were acting, that they were taking the problem seriously, that they were modernizing the game. This is not without value: restoring confidence in refereeing has real utility, even if the technology does not keep all its substantive promises. But confusing the restoration of confidence with the elimination of biases is to take a risk: when biases persist in a new form, disillusionment is all the more violent for the promise being so high.

The Leagues That Get the Most Out of VAR Have Changed Their Human Protocols

What distinguishes championships where VAR produces the most consistent effects is not the technological quality of the system. It is the quality of the human protocol surrounding it.

German Bundesliga, for example, has invested heavily in training VAR referees and standardizing intervention criteria. The Köllner Keller, the centralized video refereeing room in Cologne, operates with written procedures, precise decision trees, and a system of regular feedback between field referees and VAR. It is not the technology that produces consistency: it is the governance surrounding it.

In the Premier League, by contrast, the first years of VAR were marked by opaque decisions, failed communication with supporters, and persistent variability in intervention thresholds. Audience studies show a decline in satisfaction among English supporters on the question of refereeing fairness between 2019 and 2023, despite the adoption of technology. The Premier League has since revised its approach, particularly on real-time communication of VAR decisions in stadiums. But the experience illustrates that technology without governance produces discontent, not fairness.

This is a lesson that federations considering even more automated systems should contemplate. AI in refereeing, should it arrive, will need to be accompanied by a reform of decision-making structures, not merely a technological upgrade. It is not very different from what we observe in other sectors where AI is inserted into complex human processes: gains depend more on the organization adopting the tool than on the tool itself.

What This Assessment Says About the Next Step: Automated Refereeing

Several federations are considering going further than VAR. The semi-automated offside detection system, already tested during the 2022 World Cup and adopted for Euro 2024, pushes the logic to its conclusion on this specific variable: exit the assistant referee who raises the flag, the decision is algorithmic. The results are there: offside errors have nearly disappeared in competitions using it.

But the same federations are reflecting on extending this logic to fouls, dives, and contact in the penalty area. This is where the meta-analysis data becomes valuable as a warning. If VAR, which keeps a human referee in the loop, has not succeeded in homogenizing decisions on penalties and red cards, it is because these decisions are not reducible to an algorithm. They require judgment about intention, context, and the culture of play. A fully automated system will need to encode these judgments as rules, which amounts to displacing the debate: instead of discussing the referee’s decision, one will discuss the parameters chosen by engineers and governing bodies.

This is not neutral. Who decides that a foul at 30% force is punishable? What training database defines what constitutes a “dive”? These choices are as much political as technical. Automation does not erase them, it makes them less visible, therefore less contestable. Which, from a democratic perspective, is not necessarily progress.

The question that the data pose is therefore not “did VAR work?” It partially worked, on what it could resolve. The real question is: before deploying the next generation of systems, do we know precisely what we are seeking to correct, and what we accept not to correct?

What Football Teaches Other Sectors

Football is a useful laboratory because its data are public, its rules are stable over time, and its effects are measurable match by match. What this meta-analysis documents goes beyond sport.

When a technology is inserted into a complex human process, it does not eliminate biases: it corrects some, displaces others, and sometimes creates new ones. This is true of AI in candidate selection for employment, which has sometimes reproduced discriminations encoded in training data. It is true of algorithms in bank lending, which have redistributed credit decisions without necessarily making them more equitable. It is true of medical recommendation systems, which improve diagnosis for certain well-documented pathologies while remaining blind to populations underrepresented in the data.

In all these cases, the initial promise was the same: to substitute objectivity for human fallibility. And in all these cases, the empirical result is the same: objectivity is partial, conditioned by available data and encoded rules, and human biases rebuild themselves around the new tool rather than disappearing.

This is not a reason to reject refereeing technologies or decision-support systems. It is a reason to deploy them with rigorous evaluation protocols, success criteria defined ex ante, and correction mechanisms when actual effects diverge from promised effects. The meta-analysis on VAR is exactly this type of evaluation. That it is published seven years after the introduction of the technology illustrates a problem: we could have started earlier.

Tomorrow’s Refereeing Is Decided in Protocols, Not in Pixels

Global football institutions are not starting from scratch. The IFAB, which governs the Laws of the Game, has undertaken since 2023 a revision of VAR protocols in several member associations. FIFA is working on a standardized system for communicating decisions between VAR referees and stadiums, drawing lessons from experiences in the Premier League and MLS. These adjustments do not make headlines, but they matter more than the next technological upgrade.

The next frontier is not the pixel: it is governance. Who controls VAR decisions? How are referees trained in human-machine interaction? How are intervention criteria revised when data show drift? These institutional questions will condition the effects of future technologies far more certainly than their technical sophistication.

The 2026 meta-analysis gives decision-makers an empirical basis for posing these questions. It also gives them, implicitly, a success criterion for the next generation of systems: if in ten years, a new meta-analysis shows that biases on penalties and red cards remain as heterogeneous across leagues as today, it is because governance, once again, will have failed. Technology, meanwhile, will probably have done its part.

Sources

  1. Meta-analysis on VAR in major European championships, Frontiers in Psychology / ScienceDirect, 2026: https://www.sciencedirect.com/article/pii/S3050544526000174
  2. AI Predicts Weather Better Than States, Journal d’un Progressiste: https://journaldunprogressiste.fr/lia-predit-le-temps-mieux-que-les-etats/