Early xPass models used basic geometric and contextual features from event data. For instance, StatsBomb’s original xPass used the pass’s start location, end-point distance and angle, whether it was airborne or a cross, previous play actions, and even a generic “defensive pressure” indicator statsbomb.comtheanalyst.com. However, it did not explicitly encode where which defenders were on the pitch statsbomb.com. In other words, the model knew there was pressure but not exactly where it came from. With new tracking and 360° data, modern xPass models can include every player’s position. StatsBomb’s upgraded xPass 360 model, for example, adds features for opposition presence around the passer, along the path of the pass, and near the target statsbomb.comstatsbomb.com. It creates a cone-shaped region in front of the passer and converts each defender within it into a 2D Gaussian, with weight falling off by distance statsbomb.com. It also computes continuous “soft pressure” around the ball-carrier and receiver by overlapping Gaussians, so that a nearby opponent contributes more pressure statsbomb.com. These enhancements give a “substantial uplift” in accuracy – the new model “better discriminates between easy and difficult passes” statsbomb.com.
Flipping the View: Measuring Defensive Pressure
If we can measure how hard a pass is, we can invert that perspective and ask: how much did each defender contribute to making it hard? In principle, we could take each pass and compare two scenarios: one with defenders present (the real world) and one with them “removed” (an open-field baseline). The difference in completion probability between these scenarios reflects the defenders’ collective impact. We can then apportion that difference to individual players by simple geometry. For example, intuitively each defender’s difficulty score might be larger when they are:
- Far from the receiver: longer passes are riskier to begin with theanalyst.com.
- In the passing lane: any opponent physically between the passer and target greatly increases interception risk statsbomb.com.
- At a narrow angle: a defender who sits almost in line with the pass blocks more space than one way off to the side statsbomb.com.
- Very close (applying pressure): defenders near the ball-carrier or receiver sharply lower completion chances statsbomb.com.
Concretely, one could compute a baseline xPass with defenders removed (or far away) and subtract the actual xPass. The gap Δ(xP) is the added defensive difficulty. That Δ can then be split among the nearby opponents, weighted by each player’s distance and angle to the pass. A player standing directly between passer and receiver might get a large share of credit, while distant or off-angle players get less.
The idea resembles pitch control methods. The pitch-control map above (red=home control, blue=away control) shows which team would reach each region first getgoalsideanalytics.com. Pass-prevention metrics use exactly this concept: they simulate removing a defender to see how an attacker’s control of space grows. For example, Kothari (2021) defines a defender’s pass prevention as the change in an attacker’s optimal pitch-control when that defender is taken out, normalized by the original and weighted by the area’s value analyticsfc.co.uk. In other words, if removing defender X gives the offense much more controlled area (more red), then X was indeed preventing passes into that zone.
By analogy, our inverted xPass would credit each opponent by how much they reduce the passer’s success. Marc Lamberts’ xDEF metric takes a similar approach for tackles: it measures how a defender’s action lowers opponent scoring threat, explicitly “considering spatial positioning” marclamberts.medium.com. Here we apply that philosophy off the ball. In fact, as Tom Worville (2015) observed, a great defense is often “just great positioning,” which traditional stats miss worvilleanalysis.wordpress.com. This inverted xPass would quantify those silent contributions: defenders who don’t make tackles but who constantly shade passing lanes would earn value.
Related Research and Next Steps
Our proposed metric builds on several strands of work. Many papers have already estimated pass success probabilities link.springer.comtheanalyst.com. For instance, Spearman et al. (2017) and Power et al. (2017) used physics-based models to predict completion at the pass moment link.springer.com. More recently, Fernández et al. (2020) and the SoccerMap model used deep learning on tracking data to generate full pass-probability surfaces statsbomb.comar5iv.org. (However, such methods are computationally heavy.)
On the defensive side, recent metrics also value off-ball work: Kothari’s pass-prevention (above) and Lamberts’ xDEF both credit players for their positional impact analyticsfc.co.ukmarclamberts.medium.com. Our approach would fuse these ideas: we embed xPass models into a defensive contribution framework.
A practical implementation might follow these steps:
- Data Synchronization: Align event data with positional data so that every pass event has the locations of all 22 players (a known challenge link.springer.com).
- Baseline vs. Actual xPass: Use an xPass model to compute the completion probability normally and in a modified scenario (e.g. with one defender removed).
- Compute Δ(xP): For each pass, take the difference in success chance as the total added difficulty from defense.
- Allocate Credit: Assign that Δ among the defenders based on geometry (distance, angle, pressure). One simple way is to give each defender a share proportional to the reduction in the passage of the pass they cause.
- Aggregate: Sum each defender’s credits over many passes or games to get a season-long defensive pass-pressure score.
Each step has complexities (for example, estimating intended receiver for unsuccessful passes link.springer.com), but the framework is clear. In essence, this metric would yield a stat like “expected passes prevented” or “passes completed below expected allowed” for defenders. Clubs could use this to highlight players whose positioning consistently stifles opponents before a ball is even intercepted.
In summary, flipping xPass inside-out offers a deep way to measure defensive positioning value. By borrowing tools from expected-pass and pitch-control models, we can give defenders credit for the risk they impose on opponents. This would complement traditional stats and provide fans and analysts alike with a richer picture of the game statsbomb.comworvilleanalysis.wordpress.com.
Sources: We drew on recent analytics research to guide this idea. StatsBomb’s xPass blog explains advanced pass modeling statsbomb.comstatsbomb.com; The Analyst’s xP explainer describes expected pass features theanalyst.comtheanalyst.com; academic work defines xPass formally link.springer.comlink.springer.com. We also built on defense-focused studies: Analytics FC (Kothari) on pass preventionanalyticsfc.co.uk, Hudl StatsBomb on defensive metrics statsbomb.com, and Lamberts’ xDEF on threat reduction marclamberts.medium.com. These and other sources help validate the approach and might serve as starting points for implementation.