How event data is collected and coded

Before a metric can be calculated, a match must be turned into events. A pass, a shot, a tackle, a foul: each is recorded with a location, a time and a set of attributes. Event data is the foundation of almost every football metric, and how it is collected decides how much those metrics can say.

This guide explains how event data is collected, from live manual coding to automated detection, why definitions matter so much and how the resulting dataset feeds the metrics readers see on a dashboard.

Manual coding of events

The traditional method is manual coding. Trained analysts watch the match, live or from footage, and record each event as it happens: who did it, where, when and with what outcome. The result is a structured log of the match, and the coder's judgement is embedded in every entry.

Manual coding is flexible and can capture ambiguity that automation misses. A coder can note that a pass was made under heavy pressure or that a clearance was a deliberate long ball. That judgement is valuable, and it is also the source of variation between coders and between providers.

Automated and semi-automated detection

Automated systems derive events from tracking and video data. Using player and ball positions, software can detect when a pass occurs, when possession changes and when a shot is taken. Machine learning models are trained to classify these events, and human reviewers often check and correct the output.

Automation scales and reduces coder variation, but it inherits the limits of its input data. If the ball is occluded, the event may be misread. If a model was trained on one league's style, it may need recalibration elsewhere. Every automated event set carries its detection accuracy with it.

Why definitions decide everything

Event data is only as good as its definitions. What counts as a tackle? Is a blocked shot a defensive action or a shot? Does a cross into an empty area count as a pass or a delivery? Each answer is a choice, and each choice changes the metrics built on top.

This is why two providers can publish different pass totals for the same match without either being wrong. They defined the event differently. For a reader, the lesson is to treat a metric's definition as part of the metric, because the number alone cannot tell you what was counted.

From events to metrics

Once events are recorded, metrics are computed on top of them. Possession, progression, pressing intensity and chance quality are all built from the same log of events, combined with positional context where tracking data is available.

This layering is what makes event data so valuable and so easy to misuse. A single metric sits at the top of a stack of definitions, and each layer in the stack adds assumptions. Understanding the stack is what separates an analyst from someone who merely quotes a number.

What an event record contains

A typical event carries a set of attributes that allow it to be aggregated later.

• The player and team responsible.
• The event type: pass, shot, tackle and so on.
• A location on the pitch, usually as coordinates.
• A timestamp within the match.
• The outcome of the action.
• Context attributes, such as pressure or body part.

Where event data comes from

Different collection approaches trade coder judgement against scale. The bars indicate relative volume handled by each method.

Collection methods side by side

Each method has a distinct trade-off between judgement, scale and consistency.

The analyst's currency is attention. Every report competes with a coach's time, so the value lies in selecting what matters rather than in exhaustively describing what happened.

Football analysis sits between data and judgement. The data supplies a pattern; the human supplies the context, the priorities and the decision that no dataset can make alone.

A workflow is a set of definitions repeated. Agreeing those definitions in advance is what allows results to be compared across matches, players and seasons.

Automation frees time for the parts of analysis that need a human: interpreting context, choosing what to show and communicating it clearly.

The best analysis is falsifiable. A report that states its sample, its method and its uncertainty can be checked and challenged, and that is what makes it trustworthy.

Tooling matters less than process. Clubs that agree on what they measure, how they tag it and how they present it outperform clubs that simply buy more software.

Everything a workflow measures is a decision about what mattered. Those decisions deserve to be made deliberately rather than inherited by accident from whatever tool was installed first.

Definitions live inside the number

Every metric on a dashboard rests on a definition of an event, chosen by whoever recorded it. Knowing how event data is collected, and where judgement enters, is what lets you read a football number with your eyes open.