Possession vs field tilt: one match, two different stories
A team can have the ball and still not have the territory. This worked example takes one match — the 2022 World Cup final — and computes possession and field tilt from the same event stream, period by period. The point is not who deserved what: it is how two numbers that sound similar can move in different directions inside one game.
In this guide (7)
The question
Possession tells you which team had the ball. Field tilt tells you which team had the ball near the opponent's goal. In a single match summary the two usually agree, which is why they are often treated as the same measure. They are not: possession is a share of the ball, field tilt is a share of territory, and a team can win the first while losing the second.
The test: take the 2022 World Cup final, Argentina against France, and compute both numbers for each period of the match from the same raw events. If the two measures were interchangeable, the gap between them would stay constant. It does not.
The data and where it comes from
The match is available in the StatsBomb open data collection, which publishes event data for selected competitions and seasons, including the 2022 men's World Cup. Every event carries a team, a possession team, a period, a timestamp and a location on a 120 by 80 grid. The full collection is published on GitHub as JSON files, one per match, and is free to use under the provider's open-data terms with attribution.
Two constraints matter for a site like this one. First, the collection covers selected competitions and seasons — it is not a live feed of every match. Second, the terms are aimed at research and non-commercial use, so what is published here is the derived aggregate table, not the raw event data. The numbers below were computed from the event file; the file itself is not redistributed.
The method, step by step
- Possession share. Count every event by its possession team field, then divide each team's count by the total. This is event-based possession, not measured time on the ball — a long passing sequence produces many events, a single clearance produces one.
- Final-third events. Keep events whose location is at or beyond the 80th unit of the 120-unit pitch length, again counted by possession team. That is the attacking third of the team in possession.
- Field tilt. Divide a team's final-third events by the sum of both teams' final-third events. Field tilt is a share, so the two values always add to 100.
- Supporting numbers. Shots by event type, touches inside the penalty area as events located beyond the 102nd unit between the width limits of the box, and the sum of the provider's own expected-goals value for each shot. That xG figure is the provider's, not ours, and it is labelled as such.
Everything is computed per period, so the half-time interval does not merge two different games into one average.
The result
| Period | Possession ARG / FRA | Final-third events ARG / FRA | Field tilt ARG / FRA | Shots ARG / FRA | xG ARG / FRA |
|---|---|---|---|---|---|
| First half | 60.5% / 39.5% | 259 / 118 | 68.7% / 31.3% | 6 / 0 | 1.29 / 0.00 |
| Second half | 46.3% / 53.7% | 211 / 214 | 49.6% / 50.4% | 6 / 6 | 0.33 / 1.16 |
| Extra time, first period | 62.0% / 38.0% | 72 / 30 | 70.6% / 29.4% | 4 / 1 | 0.41 / 0.02 |
| Extra time, second period | 63.8% / 36.2% | 85 / 36 | 70.2% / 29.8% | 4 / 3 | 0.72 / 1.09 |
Penalties are excluded. The shoot-out decided the trophy but is not part of open play, and including it would distort both shares.
How to read it
The first half is the easy case. Argentina had 60.5% of events and 68.7% of final-third events: the ball and the territory moved together, and France had no shot at all. On a match summary this looks like total control, and for that period it was.
The second half is where the two numbers separate. France moved from 39.5% to 53.7% of events — they had more of the ball. But their share of final-third events was 50.4%, almost exactly level with Argentina. In other words France's possession in that half did not translate into proportionally more territorial control: they had the ball more, but not much more of it near the Argentine box, until the two goals arrived inside two minutes.
The tilt explains the score better than possession does. France's second half produced the same number of shots as Argentina (six each) but a much higher expected-goals total, 1.16 against 0.33. Territory alone does not create chances either — the location and quality of the shots do. Two measures, two different halves of the story, and neither is the verdict on its own.
Extra time is the mirror image. Argentina went back above 62% possession in both periods and about 70% field tilt — yet the clearest chance of the second period of extra time belonged to France, and the xG totals of that period (0.72 against 1.09) show it. A team can dominate territory and still concede the better chances, which is exactly why chance quality is measured separately.
Where this method breaks
- Event-based possession is not time of possession. A team that strings together short passes accumulates events faster than a team that hits long balls; the share measures activity with the ball, not seconds.
- Final-third events include everything. Passes, carries, duels and clearances all count. A different definition — only passes, or only completed passes — produces a different tilt.
- Game state changes behaviour. A team leading by two goals will happily concede territory; a team chasing will push up. Neither is a failure of the metric, but both make a single-match number hard to generalise.
- One match is an illustration, not evidence. Nothing here says field tilt predicts results. That question needs a season of matches and a stated model.
Reproduce it
The table above is published as a CSV file next to this article, so the arithmetic can be checked line by line. The steps are the four bullets in the method section: count events by possession team, filter by pitch location, divide. No proprietary model is involved — the only provider value in the table is the expected-goals column, and it is labelled.
For the definitions behind the two measures see field tilt and progressive passes and the glossary; to see how a provider builds a model rather than a share, read how xG models are built.
Data: StatsBomb open data, 2022 FIFA World Cup, match Argentina vs France (final, 18 December 2022). Aggregates computed for this article; raw event data not redistributed. Provider terms apply to the underlying dataset.
Read next
How data defines player roles
From heat maps to role clusters: how analysts sort players by behaviour instead of by the position printed on the team sheet.
Pressing styles measured from data
High press, mid-block and low block described through measurable behaviour: start distance, intensity and recovery location.
Team style fingerprints from event data
How a handful of rate metrics become a team fingerprint, and what those fingerprints can and cannot settle in a tactical debate.