The modern football analytics revolution began with a simple question: How likely is a shot to become a goal? This gave birth to understanding Expected Goals (xG) . But as the game evolved, so did the data. Analysts realized that orchestrators who dictate the tempo were being undervalued, leading to the creation of possession-chain metrics. However, even advanced frameworks like the StatsBomb xGChain and xGBuildup models have a fatal flaw: they only reward actions that actually happened. They cannot evaluate the dangerous passing lane a midfielder recognized but rejected under pressure. When evaluating how clubs scout midfielders today, the paradigm has completely shifted. Elite recruitment departments increasingly value passes that raise possession probability even when they never appear in assist statistics. Welcome to the bleeding edge of football data science: Expected Possession Value football models and Pitch Control . Fig 1: A spa...
The modern football analytics revolution began with a simple question: How likely is a shot to become a goal? This gave birth to understanding Expected Goals (xG). But as the game evolved, so did the data. Analysts realized that orchestrators who dictate the tempo were being undervalued, leading to the creation of possession-chain metrics.
However, even advanced frameworks like the StatsBomb xGChain and xGBuildup models have a fatal flaw: they only reward actions that actually happened. They cannot evaluate the dangerous passing lane a midfielder recognized but rejected under pressure. When evaluating how clubs scout midfielders today, the paradigm has completely shifted. Elite recruitment departments increasingly value passes that raise possession probability even when they never appear in assist statistics. Welcome to the bleeding edge of football data science: Expected Possession Value football models and Pitch Control.
Fig 1: A spatial mapping of Pitch Control, illustrating which team commands specific zones of the field based on player velocity and positioning.
1. Pitch Control and the Geometry of Space (Voronoi Diagrams)
If you pause a football match at any given second, who controls the space on the pitch? Traditional scouting relies on the "eye test" to answer this. Modern data science relies on Voronoi diagrams.
Leading researchers such as William Spearman helped formalize modern pitch-control frameworks as highly probabilistic models. A basic Voronoi diagram partitions the pitch into geometric regions based purely on static distance. However, elite Pitch Control models ingest 25-frames-per-second tracking data to account for player velocity, momentum, and ball trajectory.
Interestingly, the frame-rate physics governing these advanced tracking cameras is identical to the mathematical reality we exposed when analyzing why VAR offside geometry is often scientifically flawed. By embracing dynamic, predictive tracking, algorithms can map the geometry of passing lanes in real-time, instantly identifying where a midfielder can safely distribute the ball without interception.
2. EPV vs Expected Threat (xT): Why Continuous Models Matter
Pitch Control tells us where a player can pass safely. But it doesn't tell us why they should pass there. Passing safely back to the goalkeeper yields 100% Pitch Control but creates zero attacking threat.
Many analysts use Expected Threat (xT) to value territory, but xT operates on a coarse, grid-based system focused solely on zone progression. Expected Possession Value (EPV), pioneered by data scientists like Javier Fernández and Luke Bornn in their groundbreaking MIT Sloan Sports Analytics paper, is now one of the most sophisticated possession metrics in football analytics. It calculates a singular, continuous value: the exact probability that a team's current possession will end in a goal, fluctuating dynamically with every micro-movement.
| Metric | What It Tracks | Primary Limitation |
|---|---|---|
| xG | Shots & finishing probability | Ignores build-up play entirely |
| xA | Final pass before a shot | Ignores earlier progressive actions |
| xT | Zone progression (Grid-based) | Coarse grid; lacks micro-spatial detail |
| EPV | Full continuous possession value | Highly data-intensive (requires tracking) |
Fig 2: A breakdown of how tracking data evaluates the incremental value of a midfielder's passing decisions.
3. The Ultimate Metric: Incremental Possession Value
When you overlay the EPV map onto the Pitch Control model, you unlock the holy grail of midfield scouting: Incremental Possession Value. An analyst can mathematically evaluate every passing option a midfielder had at the exact moment the ball was at their feet.
Expected Value of Pass = Pitch Control(Target) × EPV(Target)
In practice, full models also weight interception probability and subsequent transition risk.
This formula penalizes players who constantly take low-risk, backward passes (high Pitch Control, low EPV). Conversely, it highly rewards creative visionaries. Take Kevin De Bruyne as a concrete example: when he identifies and executes a line-breaking pass into the half-space, he is actively choosing an option with moderate Pitch Control but massive EPV.
By utilizing this framework, data-driven clubs can scout a player's raw footballing intelligence and spatial awareness, completely divorced from whether their teammate actually finished the chance.
Frequently Asked Questions (FAQ)
What is a Voronoi diagram in football analytics?
A Voronoi diagram partitions a football pitch into spatial regions based on which player can reach that specific area of the field first. Advanced versions factor in player velocity, momentum, and reaction time to create dynamic zones of Pitch Control.
How does EPV differ from Expected Goals (xG) and Expected Threat (xT)?
While xG only measures the probability of a specific shot becoming a goal, Expected Possession Value (EPV) calculates the probability of a possession ending in a goal from any location on the pitch, even before a shot is taken. It functions similarly to Expected Threat (xT) but utilizes deeper, continuous spatiotemporal tracking data rather than a coarse grid.
Why are Pitch Control models essential for scouting midfielders?
Pitch Control models allow analysts to evaluate a player's decision-making. By mapping the best available passing lanes, scouts can see if a midfielder consistently identifies and executes the pass with the highest incremental value, rather than just settling for the easiest completion.
Watch: Friends of Tracking’s EPV Framework Demonstration
Friends of Tracking breaks down how Expected Possession Value algorithms assign credit to complex on-ball actions:
Comments
Post a Comment