Home/Soccer Analytics/Situational Expected Threat (SxT)

Situational Expected Threat (SxT)

Expected Value ModelsLevel 4 β€” Expert

Prerequisites

Representation Learning for Game StatesRecursive Expected Threat (xT) Model

What It Is

A Markov reward model treating each discovered game-situation cluster as a distinct state β€” computing expected threat per situation rather than per pitch zone. This enables evaluation disaggregated by the specific defensive and spatial context. Man City's zone-16 effectiveness comes specifically from line-breaking passes (63% of value in one situation cluster), not from being in zone 16 generically.

Correct Execution

Replace zone-based states in xT with situation-cluster states. Compute transition probabilities between situations. The result: different xT values for the same pitch zone depending on the game situation.

Edges

πŸ’Ž Elite-Only Behavior

Same Zone Swings 30+ Points in Threat Based on Game Situation

xg-modelssituational-expected-threat β†’

Man City's zone-16 effectiveness comes from line-breaking passes (63% of value in one situation cluster), not generic zone dominance. Two situations in the same zone swing 30+ percentage points based on defensive distance and teammates ahead.

What most people do
Use zone-based xT and coach to zones.
What the best do
Discover situation clusters via representation learning. Compute per-situation xT. Coach to the situation, not the zone.
Why it's an edge: Zone-level analysis averages across fundamentally different contexts. A team average in a zone may be elite in one situation and poor in another.
How to exploit: Train a multi-task autoencoder on 360 data. Identify which situations your team excels or struggles with per zone.
Zitian Tang, Tsinghua/Brown, StatsBomb Conference 2023

Sources

  • Zitian Tang, Tsinghua/Brown University, StatsBomb Conference 2023, YouTube, 2023-11-01