Niana Analytics · football data hubMetrics · tactical data · formats · video analysis

Home › Metrics

Expected threat explained: turning pitch zones into value

Niana Analytics desk · Metrics ·
Expected threat explained: turning pitch zones into value

Expected goals answers one question well: how good was the shot? It says nothing about the pass that split two lines, the carry that pulled a defender out of position, or the switch that turned a crowded side of the pitch into an empty one. Expected threat, usually shortened to xT, was built to price exactly that missing part — the value of where the ball is, before anybody shoots.

In this guide (8)
  1. Why a shot-only metric leaves a gap
  2. The pitch divided into zones
  3. Two probabilities per zone: shoot or move
  4. Why the grid has to be solved, not counted
  5. From zone value to action value
  6. An illustrative ledger
  7. What the grid cannot see
  8. Reading an xT total without overclaiming

This guide describes the mechanics of the model: how the pitch is divided, how each zone receives a value, how those values are solved rather than counted, and how a single pass or carry is scored. It also sets out what the grid cannot see, because the limits explain most of the arguments the metric provokes.

Why a shot-only metric leaves a gap

An expected-goals model evaluates an attempt at the moment it is struck. Everything that produced the attempt sits outside the ledger. Two teams can arrive at a similar shot total by very different routes, and xG will describe both routes as if they were equally difficult to construct.

Progression metrics close part of that gap by counting forward passes and carries. Counting, however, treats every completed forward pass as one unit, whether it travelled five metres into space or twenty metres through three defenders. A count measures volume; it does not measure the change in danger.

Expected threat replaces the count with a price. It asks how much a team's chance of scoring changes when the ball moves from one part of the pitch to another, and it expresses the answer in the same unit as xG: goals.

The pitch divided into zones

The model overlays a grid on the pitch. Columns run across the width, rows run along the length, and each cell becomes a zone with its own coordinates. Every event is assigned to the cell where the ball was at the moment it was played.

The resolution of that grid is a design choice. A coarse grid gives every zone plenty of events and produces stable but blunt values; a fine grid separates the edge of the box from the six-yard area but leaves some cells with so few actions that the estimate becomes noisy. There is no universal grid, only a trade between detail and sample size.

The grid is normally learned from one competition and one period of play. Zone values computed on one league's football describe that league's football; carrying them into a different context carries the assumptions with them. The pitch is also normalised for direction, so both teams attack the same way and a zone near the opponent's goal has one set of coordinates regardless of who is in possession.

Two probabilities per zone: shoot or move

For each zone the data yields two shares. The shot share is the proportion of actions in that zone that are shots; the move share is the proportion that are passes or carries. In the simplest formulation the two shares account for every action in the cell.

The shot share is multiplied by the conversion rate of shots taken from that zone. That product is the direct value of the cell: the probability that a possession ends in a goal immediately, without another move. It is the only part of the model that touches a shot.

The move share opens the recursion. A move does not score by itself; it transfers the ball to another zone whose value is already defined. The value of moving is therefore the probability-weighted average of the values of every destination zone, with the weights given by how often moves from this cell actually go to each of them.

Why the grid has to be solved, not counted

Written out, the definition is circular. The value of a zone depends on the values of the zones it can move into, and those values depend on this zone in turn. Counting actions cannot resolve the circle, because every count refers to a number that is not yet known.

The model resolves it by iteration. Every zone starts at zero, the formula is applied across the whole grid, and the output becomes the input for the next pass. Values creep outward from the goal, because the cells where shots are taken are the only ones with direct value at the start.

After enough passes the numbers stop changing beyond a small tolerance, and the grid is said to have converged. The fixed point is stable under normal conditions because each move multiplies the next zone's value by the move share, a factor below one, so the influence of distant zones decays with every step.

A closed-form solution also exists: the same equations can be written as a linear system and solved directly. Iteration is common in practice because it is easy to audit — you can watch the values spread from the goal outward and check that the order makes sense.

From zone value to action value

Once the grid is stable, every pass and carry can be scored. The value of an action is the value of the zone where the ball ends minus the value of the zone where it started.

A pass from a deep zone into the final third earns the difference between those two cells. A backward pass earns a negative number, because it moves the ball to a less dangerous zone. A move that stays inside a single cell earns zero by construction, which is the model's way of saying that circulation without progression adds no threat.

Shots are handled separately. A shot's value belongs to the expected-goals ledger, and folding it into xT would count the same event twice. That is why xT totals usually describe passes and carries only.

Summing action values across a match gives a team total; dividing by minutes gives a rate. The total measures progression, not chance quality: a team can accumulate threat and never take a shot, and the model will still record the accumulation.

An illustrative ledger

The two tables below use round numbers chosen to show the arithmetic. They are not a fitted grid and they do not describe any team or match.

ZonePosition on the pitchIllustrative value
Deep own thirdBuild-up, far from goal0.01
Middle thirdBetween the boxes0.03
Wide final thirdAdvanced, outside the box0.08
Edge of the penalty areaCentral, just outside the box0.20
Central six-yard areaImmediate shooting range0.45

Illustrative zone values only. A real grid is produced by solving the model on a large sample of actions; the numbers above exist to make the subtraction visible.

A sequence of five actions shows how the values combine. Each row is one action, and the last row sums the column.

ActionStart valueEnd valueValue added
Pass from deep third to middle third0.010.03+0.02
Pass from middle third to wide final third0.030.08+0.05
Carry from wide final third to the edge of the box0.080.20+0.12
Pass from the edge of the box into the six-yard area0.200.45+0.25
Backward pass from the edge of the box to the middle third0.200.03−0.17
Sequence total+0.27

The ledger makes the accounting explicit. Four forward actions build value and one backward action removes part of it. The net figure belongs to the sequence, not to a player, and the same sequence replayed with different zone values would produce a different total.

What the grid cannot see

The limits follow directly from the design. A zone is a rectangle, and a rectangle contains no information about the players inside it.

  • Defenders and teammates. A pass into a crowded cell and a pass into the same cell with space are scored identically, because the model knows the cell, not the bodies in it.
  • Pressure and orientation. A difficult pass under pressure and an easy pass to the same destination share a value; body shape, foot and time on the ball never enter the grid.
  • Off-ball movement. A run that drags a defender away and creates space without receiving the ball earns nothing, even when it is the reason the next pass works.
  • Defensive actions. A tackle or interception that stops a dangerous move is not credited, because the model prices progression with the ball.
  • Context of the values. Zone values depend on the league, the period and the sample used to fit them, so comparisons across implementations are unsafe.
  • Resolution inside a zone. The most valuable metres in football sit immediately in front of goal, and a coarse grid can fold them into a single cell with much less dangerous positions.
  • Prediction. The model describes the value of positions in the data it learned from; it does not forecast the outcome of a particular match.

Reading an xT total without overclaiming

An xT total is a sum of many small numbers, and every sum grows with the number of actions. Compare rates per 90 and state the minutes, because a raw total rewards whoever played more. A player with twice the minutes can post twice the threat without being any more effective per action.

The number is also model-specific. Two grids of different resolution, fitted on different competitions, produce different zone values and therefore different action values for the same pass. Compare within one implementation and one definition, the same rule that applies to expected goals.

Role and team style shape the total. A player who receives the ball in advanced zones starts from high-value cells and needs fewer actions to add value; a deeper player must progress the ball much further to earn the same number. Read xT beside progressive passes and field tilt rather than alone, because volume, territory and value answer different questions.

Finally, treat a single match as noise. Over a run of matches the total describes a tendency; over a season it describes a level. Any claim about a player should name the sample, the grid and the model behind it — and should still be read as a description of what the ball did, not as a verdict on how good the player is.

Common questions

What does expected threat measure in football?

Expected threat prices the position of the ball rather than the shot. Each zone of the pitch receives a value equal to the probability that a possession starting there ends in a goal, and an action earns the difference between its destination and its origin.

How does the xT model assign a value to each zone?

A zone's value combines two parts: the chance that an action there is a shot multiplied by the conversion rate of those shots, plus the chance that the action is a move multiplied by the values of the zones it can reach. Because the parts refer to each other, the grid is solved by iteration.

Why can a single xT action value be negative?

The value is a subtraction: destination zone minus origin zone. A forward pass into a more dangerous area adds value, while a backward pass moves the ball to a less dangerous cell and therefore subtracts. A move that stays inside one zone is worth zero by construction.

Can xT totals from two different models be compared?

Not safely. Grid resolution, the training sample and the period of play all change the zone values, so the same pass can earn different numbers in two implementations. Compare within one model and one definition, and use rates per ninety minutes rather than raw totals.

About our numbers. Every figure on this site describes a method, a formula or a competition rule. We deliberately do not publish invented player statistics, match scores or rankings, and we add no links to outside sites. Read every metric as a tool with a definition, not as a verdict.

Read next

Metrics

What is xG and PPDA? A football metrics primer

Expected goals and passes allowed per defensive action explained from first principles: what each number measures and how to read it.

Metrics

Expected goals (xG): how the metric is built

Shot location, angle, body part and defenders: the ingredients of an xG model and why a single number never tells the whole story.

Metrics

Expected assists (xA): valuing the final pass

How xA scores a pass by the quality of the chance it creates, and how to use it without overrating a creative midfielder.