Advanced Football Metrics: The Numbers Scouts Use

Advanced football metrics are sold as a way to see what the eye misses, and mostly they do the opposite: they quantify what the eye already noticed. The value is in scale rather than insight, because a model can apply the same rule to four thousand players in an afternoon.

For recruitment work, the question is which of those metrics survive a change of shirt. The ones that describe the player travel; the ones that describe the team do not.

Choosing a small set and sticking to it

Departments that use fifty metrics usually rely on five. The others are consulted when the five disagree, which happens less often than the size of the dashboard suggests, and the discipline of a short list is what keeps a model usable by people who are not analysts.

A workable short set for recruitment covers minutes and availability, one progression measure, one defensive or duel measure, one possession value measure and an age curve. Everything else is a check on those five rather than a rival to them.

Football performance metrics: possession value

Football performance metrics have moved beyond counting events towards pricing them. Possession value models estimate how much a single action changed the team's chance of scoring, which lets an analyst compare a safe sideways pass with a difficult one into the half space.

The appeal is obvious and so is the trap. Possession value is built on a model of an average team, so a player in a side that attacks in transition can look wasteful for attempting the passes that his own team needs him to attempt.

Used carefully, the metric is an excellent screen for playmakers. Used as a ranking, it rewards players in possession-heavy sides, which is a statement about their clubs.

Expected threat model: what it adds to expected goals

An expected threat model assigns value to positions on the pitch rather than to shots. Every completed pass or carry that moves the ball into a more dangerous area earns credit, even when no shot follows, which fills the gap that expected goals leaves between the halfway line and the penalty area.

That makes it useful for the players who create chances indirectly: the midfielder who plays the pass before the assist, the full back whose progression forces a defence to shift. Those contributions are mostly invisible in goals and assists, and they are exactly what a mid-block defence has to cope with.

The model also has a distinctive failure mode. Because it values movement towards goal, it can reward a player for carrying the ball into trouble when a simple pass out wide was the better decision, so it should never be read without a turnover figure beside it.

• Possession value: prices individual actions for and against the team.
• Expected threat: prices the position of the ball, following passes and carries.
• Counterpressing: measures how quickly a team — or a player — wins the ball back.

Football performance metrics: counterpressing and duels

Counterpressing metrics attempt to measure how quickly possession is recovered after it is lost, usually inside a fixed number of seconds. The team-level version has become standard; the player-level version is noisier but still useful for screening forwards who press and forwards who do not.

Duel win rates are the other half of this family. A centre back winning 70 percent of his aerial duels is meaningful, but only alongside the volume of duels he contests, because a defender who contests very few headers is not being tested by the same standard.

Expected threat model limits and data quality

Every one of these measures inherits its provider's event data. Different companies record a duel, a tackle or a pass under pressure differently, and the differences are large enough to change a shortlist. The practical rule is to keep comparisons inside one data source and to document which one it is.

Historical depth matters as well. A model trained on five seasons across four leagues will behave differently on a player from a division it barely observes, and the honest output in that case is a note saying the data cannot support a view.

Using advanced metrics without losing the eye

The most reliable use of these numbers is to build a filter and then verify it on video, in that order. A midfielder flagged by expected threat contributions is worth watching twice; a defender with outstanding duel numbers but no minutes against top-half sides is worth watching for a different reason.

That verification step is what a live viewing is for, and it is where the metrics hand over to a person who writes a scout report in plain language. The numbers start the argument; a role profile such as our box to box midfielder breakdown usually settles it.