What the data can support
Event data records selected actions such as passes, shots and defensive events. Tracking data records player and ball locations at repeated time intervals. Contextual event products may add visible-player locations around an action. These sources answer different questions and should not be described as interchangeable.
Metrica Sports' sample repository demonstrates synchronized event and tracking files. StatsBomb describes its 360 product as contextual visible-player locations around events. Neither source means that a model has complete off-camera intent or perfect identity tracking.
Match the question to the input
| Question | Minimum input | Output | Main limitation |
|---|---|---|---|
| Where did possessions progress? | Located events | Zones and action sequences | No continuous off-ball movement |
| How compact was a block? | Tracking | Team length, width and spacing | Identity and ball-state quality |
| Which actions increased threat? | Event locations and an xT model | Change in expected threat | Model and provider dependence |
| When did pressure occur? | Defined pressure events or tracking rule | Pressure counts or states | Definition and observer dependence |
| Which formation was visible? | Tracking or lineup plus locations | Shape estimate over a time window | Shape is dynamic, not one label |
Opta's definitions show why terms such as possession, pressure and PPDA need provider-specific definitions. Contextual xT research describes one model construction and its limits rather than a universal tactical score.
A defensible workflow
- Define the tactical concept in observable terms.
- Document provider fields, pitch orientation and missing frames.
- Segment phases with reproducible rules.
- Compute the metric without looking at the desired narrative.
- Review high and low examples against video.
- Test agreement between analysts or rule versions.
- Report match state, opponent and sample size beside the result.
Worked interpretation
Suppose a team’s median defensive width is 34 metres in one defined out-of-possession phase and 39 metres in another. The measured difference is five metres under that segmentation rule. It does not by itself establish whether the coach ordered a narrower block, whether the change improved performance, or whether opponent positioning caused it. The numbers are illustrative.
Human review remains part of the method
Video can reveal marking assignments, camera exclusions, player swaps and tactical transitions that a derived metric misses. Human review should be structured: sample cases before reading the model’s conclusion, record disagreement, and revise the operational definition when errors repeat. The differing views in Metrica's synchronized sample data and StatsBomb's contextual 360 data support retaining source and visibility boundaries.
Define the phase before measuring the shape
Terms such as high press, settled block and transition need operational rules. One implementation might define a settled defensive phase only after the opponent completes two passes and the defending team has most players behind the ball. Another may use speed and field position. Both can be defensible, but they will produce different samples.
| Phase rule | Sensitivity to test |
|---|---|
| Possession start and end | Provider event definition |
| Transition duration | Two, five or ten seconds |
| Team shape | Mean, median or convex-hull positions |
| Pressure | Event label or distance-speed rule |
| Ball location | Continuous coordinate or zone |
Opta's football definitions show why event-derived tactical terms need provider context. Metrica's sample data support inspectable event and tracking alignment for learning and method checks.
Validate the interpretation, not only the calculation
Sample clips before and after setting the threshold, then record analyst agreement without revealing the model label first. Investigate disagreements: the segmentation may be wrong, the metric may miss a player outside the camera view, or the tactical concept may simply be ambiguous.
Publish the observable result separately from the interpretation. “Average team length decreased under this rule” can be checked; “the coach intended a low block” needs additional evidence.
Continue the workflow
Use defensive metrics explained for a denominator-aware example of converting a tactical concept into measurable event rates.
Continue learning
- Next guide: Betting Market Data in Football Models
- Related guide: Bias in Football Prediction Models
Assumptions and limitations
Tactical labels are abstractions over continuous play. Provider coverage, coordinate systems and definitions vary. This page explains analytical controls; it does not claim that any model can recover intention or produce a complete tactical verdict automatically.
GambleAware
GamCare
Gambling Therapy
GAMSTOP