Event data and tracking data answer different questions

Two different recordings of one match

Event data captures discrete actions involving the ball, each with a type, a location, a time and the participants directly involved. Tracking data captures the position of every player and the ball at a fixed sampling rate, producing a continuous record of the whole field. The first is sparse and interpretive, since somebody decided that an action occurred and assigned it a category from a fixed list.

The second is dense and largely uninterpreted, describing where bodies were without any claim about what they were attempting to do. Because they record different things, questions answerable from one are frequently unanswerable from the other.

What event data is good for

Event data supports anything concerning the ball, including where actions occur, how sequences develop and how often particular action types happen. It is the foundation of most published analysis because it is comparatively cheap to collect and has been gathered for far longer than tracking. Its main limitation is that everything away from the ball is invisible, so nine tenths of what determines an outcome leaves no trace.

It also inherits the interpretive layer discussed wherever collection is manual, since categories are applied by an operator under time pressure. Analysis built purely on event data therefore tends to overweight the ball carrier and underweight the structure that created their situation.

What tracking data is good for

Tracking data supports questions about space, shape and movement, such as how compact a defensive block is or how quickly a structure reorganises. Because it samples continuously, it can describe the situation before an action as well as the action itself, which is where most causal explanation lives. Its weakness is that it contains no intent, so a player standing still may be perfectly positioned or completely disengaged and the data cannot distinguish them.

It is also expensive to collect and store, which historically limited availability to well-resourced competitions and to teams rather than to the public. Interpretation therefore requires domain knowledge to a greater degree than event data, because the raw record makes no claims of its own.

Joining them is where the value is

Most genuinely useful analysis combines the two, using tracking to describe the situation and events to mark the decision taken within it. That combination allows a question about options, since the tracking record shows which alternatives existed at the moment the event data says a choice was made. Joining is technically awkward, because the two streams are collected by different systems with different clocks and different notions of when an action occurred.

Alignment errors of a fraction of a second are enough to associate an event with the wrong frame, which corrupts everything built on top. Much of the unglamorous work in a modern analysis department is spent making the two records agree about time.

Why the distinction matters to readers

Published claims frequently sound like tracking analysis while resting on event data, particularly anything describing pressure, space or off-ball movement. A claim about how much space a player creates cannot be supported by a record that only knows where the ball went. Asking which recording a claim came from is therefore a quick and effective filter on how much weight it can carry.

It also explains why some widely quoted measures behave oddly, having been constructed from a data source that could not see the thing being described. The distinction is not a technicality, since it determines the entire class of questions a given analysis is capable of answering.