Event Definitions and Their Effect: a data reference

Every metric on this site is built from events, and every event is a decision about what happened. A tackle, a dribble, a pass and a shot are all defined objects, and the definitions change the totals.

The event log

The foundation of every advanced metric is an event log: a timestamped list of everything a player did with the ball, with coordinates and a type. Producing that log requires an analyst to watch the match and to decide, for every touch, which category it belongs to.

Those decisions are constrained by a written taxonomy, and the taxonomy is where disagreement begins. Whether a heavy touch that a defender pokes away is a dribble lost or a tackle won, whether a clearance under pressure is a pass or a defensive action, and whether a shot that hits a teammate is a block or a wayward effort are all judgement calls made at speed.

Why this matters for metrics

A metric is an aggregation over events, so a systematic difference in event definition produces a systematic difference in the metric. If one taxonomy counts more tackles, every tackle-based rate is higher under that taxonomy, and the difference survives aggregation because it is not random.

This is the mechanism behind most of the disagreements in public analytics. Two summary figures that appear to contradict each other frequently turn out to be aggregations over different event definitions, and the argument about which is correct is usually an argument about a taxonomy neither party has read.

The most contested event types

Four event types generate most of the disagreement. Tackles and interceptions overlap with each other and with recoveries, and different taxonomies draw the boundaries differently. Dribbles and carried progress overlap similarly. Passes and clearances overlap for defensive players under pressure. And shots overlap with crosses for deliveries that travel towards goal.

Each overlap produces a set of events that can legitimately be labelled either way, and the size of that set varies from match to match. A match in which many defensive actions were ambiguous will produce totals that move more between providers than a match in which few were.

The coordinate problem

Events carry coordinates, and coordinates carry their own definitional choices. Where exactly a shot was taken from, whether the coordinate is the position of the player or the position of the ball, and how the pitch is oriented in the data are all decisions that shift model outputs.

The orientation problem is particularly consequential for models that value positions, such as expected threat, because a coordinate that is logged from the wrong end of the pitch produces a nonsensical value. Data hygiene checks exist largely to catch that class of error, and they are not decorative.

Why the definitions cannot simply be harmonised

There is a genuine argument for a single global taxonomy, and it has not happened for a reason: each provider's taxonomy is embedded in a historical archive. Changing the definition of a tackle would invalidate the comparison between the matches recorded before the change and those recorded after it.

The consequence is that the disagreement is stable and will persist. The practical response is not to seek a single definition but to state which one a figure uses, which is what this site does in every caption that carries a table.

Reading a figure whose definition is unknown

Where the taxonomy is unknown, only ordinal and internal comparisons survive. Within one table produced by one provider, the ranking of two players is meaningful. Across two tables produced under two taxonomies, the absolute values are not comparable and the rankings may not be either.

That is a real limitation and it is worth stating plainly, because it means a great deal of published comparison is weaker than it appears. A reference that marks which comparisons are sound is more useful than one that presents every figure as equally firm.

Key reference points

• Every metric is an aggregation of events, and events are defined objects.
• Taxonomy differences are systematic, so they do not cancel out in aggregation.
• Tackles, dribbles, passes and clearances are the most contested event boundaries.
• Coordinates carry definitional choices that shift position-based models.
• Taxonomies will not be harmonised because each is embedded in a historical archive.
• Without a known taxonomy, only comparisons within one table are sound.

The event log is a set of judgements made under a written taxonomy, and a metric is those judgements summed; understanding that is the entry ticket to reading any of them.