Expected Threat (xT): Valuing Positions — explained

Expected threat abandons the shot entirely and asks a different question: how dangerous is the ball's current position, and how much danger did this action add? It is the cleanest model of territorial value in public use.

The model in outline

The pitch is divided into a grid of zones. Each zone is assigned a value: the probability that, from this position, the team currently in possession will score before losing the ball. Those values are estimated from historical data by working backwards from shots scored through the chains of moves that produced them.

The value of an action is then the difference between the value of the position the ball reached and the value of the position it left. A pass from a low-value zone to a high-value one adds a great deal of threat; a pass sideways between two low-value zones adds nearly none, which is the entire point of the metric.

Why the grid is shaped the way it is

Zone values rise steeply as the ball approaches goal and as it moves towards the centre. The penalty area is worth more than anywhere else, the zone immediately outside it is worth a good deal, and the wide areas of the defensive third are worth almost nothing. The shape of the grid is a quantitative map of where football is dangerous.

That map has a virtue that shot-based models lack: it assigns value to positions where shots are rarely taken, so a progressive pass into a dangerous area counts even when nobody shoots. This makes the model usable for describing play in midfield, which is where most of a match happens.

Where the model is thin

The model is blind to everything that is not position. It does not know how many defenders stand between the ball and the goal, whether the receiver is marked, or whether the player on the ball is capable of using the space. Two passes to identical coordinates receive identical credit irrespective of the difficulty of either.

It also assumes that the value of a position is independent of how the ball arrived, which is false in the same way that a shot model without service information is false. A pass into the box after a switch of play reaches a defence in a different state from one played into a set block, and the zone model cannot tell them apart.

What it is genuinely good for

Its strength is scale. Because it needs only coordinates, expected threat can be computed from any historical dataset of passes and carries, which makes it one of the most portable advanced metrics in existence. It can be applied to old seasons and to leagues where tracking data has never existed.

Its second strength is that it describes the build-up. Players who never shoot and rarely assist acquire a measurable contribution, which is exactly the gap that goals and assists leave. For defenders who break lines and for deep midfielders who progress the ball, threat models are often the only statistics that record their value at all.

Reading it against shot models

Expected threat and expected goals answer different questions and should never be added together. Threat measures the value of moving the ball; goals measure the value of the chance it produced. A side can accumulate threat all match and create one poor shot, and the two models will disagree in a way that describes the side rather than contradicting each other.

The productive comparison is between the two over a season: a team with high threat and low expected goals is circulating the ball into good areas without delivering the final pass, which is a specific and actionable description of an attacking problem. That is the kind of finding the model was built to produce.

Using it in a table

A threat table is most useful when it reports threat per ninety minutes alongside the count of actions, because the rate separates efficiency from volume. A player with a great many low-value actions and one with few high-value actions can post similar totals, and only the components distinguish them.

As always, the caption must state the grid resolution and the historical sample used to estimate zone values, because a fine grid trained on a small sample produces unstable values and a coarse grid loses detail. Both choices are legitimate and both must be declared.

Key reference points

• Expected threat values positions by the probability of scoring from them.
• An action's value is the rise in threat between the position left and the position reached.
• The grid is a quantitative map: central and near-goal positions dominate.
• The model is blind to defenders, marking and the identity of the player.
• It needs only coordinates, so it is portable to old and un-tracked competitions.
• Never add threat to expected goals; they measure different things.

Expected threat is the best available answer to the question of what a pass is worth when nobody shoots, which is most of the time in most matches.