1. Why Football Analytics Needed Expected Threat
For years, football data science was bottlenecked by two shot-centric metrics: Expected Goals (xG), which only evaluates the final shot attempt, and Expected Assists (xA), which only evaluates the pass immediately setting up that shot.
This created a massive analytical blind spot: what about the central midfielder who receives the ball deep in his own half, turns under intense pressure, and executes a 40-meter line-breaking diagonal ball into the half-space? That pass might dismantle two defensive lines, yet xG and xA award that player zero credit unless they personally take the shot or make the direct final pass.
In 2018, data scientist Karun Singh introduced Expected Threat (xT), a Markov chain possession value model that solves this dilemma by rewarding actions that move the ball into more dangerous pitch coordinates.
2. The Mathematical Foundation: Markov Chains & Iteration
The xT framework divides the pitch into an N × M grid (traditionally 12 columns by 8 rows = 96 cells). At each cell \( (x, y) \), a player with the ball has three fundamental choices:
- Shoot: with probability \( s(x, y) \), which succeeds in a goal with probability \( g(x, y) \).
- Move (Pass or Carry): with probability \( m(x, y) = 1 - s(x, y) \), transitioning to another cell \( (x', y') \) with transition probability \( T_{(x, y) \to (x', y')} \).
Because the equation is recursive (the threat of cell A depends on the threat of cell B, which in turn depends on subsequent actions), it is solved through value iteration across hundreds of thousands of historical match events until the values converge.
3. Interpreting xT Values Across the 5 Pitch Zones
| Pitch Zone | Typical xT Value Range | Tactical Implication | Elite Practitioners |
|---|---|---|---|
| Defensive Third (Cols 1-4) | 0.005 – 0.018 | Safe retention; low immediate goal threat. Passes here prioritize security over penetration. | Rodri, Frenkie de Jong |
| Middle Third (Cols 5-8) | 0.020 – 0.045 | Build-up corridor; vertical balls from here break the midfield press into intermediate pockets. | Toni Kroos, Joshua Kimmich |
| Flank Byline (Cols 9-11, Rows 1 & 8) | 0.035 – 0.065 | Crosses from wide areas carry moderate threat due to defensive aerial contest. | Bukayo Saka, Vinícius Jr. |
| Half-Spaces (Cols 9-11, Rows 2 & 7) | 0.085 – 0.160 | The prime assist zone; cutbacks into the penalty box have high xG conversion rates. | Kevin De Bruyne, Lionel Messi |
| Central Box / Golden Zone (Cols 11-12, Rows 4-5) | 0.220 – 0.450 | Maximum danger. Balls delivered here directly generate high-xG tap-ins and 1v1 finishes. | Erling Haaland, Harry Kane |
4. Expected Threat Delta (ΔxT): Scoring Player Contributions
When a player completes an action from cell \( A \) to cell \( B \), their individual contribution is measured by the change in threat:
If a player makes a backward pass from the attacking half-space (0.120) back to the center-back (0.015), the action has a negative ΔxT (-0.105). While tactically necessary to retain possession, xT mathematically accounts for the reduction in immediate threat. Conversely, a fullback delivering a diagonal switch into the penalty arc registers a massive positive ΔxT (+0.180).
5. Limitations of Basic xT (And How Modern Models Fix Them)
- Zero Defensive Context: Basic xT only knows the coordinates of the ball, not where the 11 defenders are standing. Passing into an empty box receives the same raw xT as passing into a box crammed with 9 defenders. Modern models (like StatsBomb's On-Ball Value or Opta's Possession Value) overlay defender pressure coordinates.
- Negative Bias on Safe Possession: Teams playing positional possession (e.g. Manchester City) frequently cycle the ball backward to unbalance a low block. Basic xT penalizes these reset passes, understating their systemic value.