Source-linked AI summary
A network theory analysis of football strategies
Javier López Peña, Hugo Touchette
TL;DR
The paper asks how football team style and player performance can be described more richly than by conventional statistics. It constructs weighted, directed passing networks from 2010 World Cup data and analyzes them visually and with centrality measures. The resulting networks distinguish passing styles and support analysis of player importance, popularity, and vulnerability to player removal.
Problem
Conventional football statistics such as assists and goals are insufficient for measuring team and player performance, while newly available match data permits more detailed analysis.
Method
The paper represents each team as a weighted, directed passing network, positions nodes according to tactical formations, and computes network invariants including centrality measures.
Results
The networks provide visual summaries of team style, reveal pitch-use patterns and potential weaknesses, and quantify player importance, popularity, and removal effects.
Takeaways & Limitations
Passing networks and centrality measures provide useful information about team strategy and individual player importance for coaches and sports journalists.
Takeaways & Limitations
Passing networks are static representations even though players do not remain in fixed positions during matches.
Abstract
from arXiv · showhide
We showcase in this paper the use of some tools from network theory to describe the strategy of football teams. Using passing data made available by FIFA during the 2010 World Cup, we construct for each team a weighted and directed network in which nodes correspond to players and arrows to passes. The resulting network or graph provides a direct visual inspection of a team's strategy, from which we can identify play pattern, determine hot-spots on the play and localize potential weaknesses. Using different centrality measures, we can also determine the relative importance of each player in the game, the `popularity' of a player, and the effect of removing players from the game.
1 Introduction
The paper applies network theory to football passing data to quantify team style and player performance. This approach addresses the limits of conventional football statistics by exploiting newly available detailed match data.
- Football’s continuous play and low scores make assists and goals insufficient measures of team and player performance.
- Publicly released match data creates opportunities for more detailed football analysis beyond traditional statistics.
- The paper constructs weighted, directed passing networks whose players are nodes and whose arrows encode successful pass counts.
- Positioning network nodes according to tactical formations makes team style visually inspectable, including overused or underused pitch areas and potential player problems.
- Centrality measures quantify team performance, player contributions, popularity, and the effects of removing players from play.
2 The network of a football team
A football passing network represents players and successful passes as a weighted graph, positioned to visualize tactical structure. Network features then support quantitative analysis of passing patterns, pitch use, and robustness.
- A passing network uses players as nodes and directed arrows weighted by the number of successful passes between them.
- Fixing nodes near players’ tactical positions turns the topological network into a visual representation of team strategy.
- The network is an oversimplification because players move during matches, although arrow thickness and hue still provide immediate tactical insight.
- The network can reveal favored or neglected pitch areas, passing-distance tendencies, under-involved players, weak spots, and problematic teammate connections.
- The weighted adjacency matrix A records pass counts, while the unweighted matrix E records whether at least one pass occurred.
- Edge connectivity measures robustness as the smallest number of passes whose interception disconnects the team’s passing flow.
3 Player performance
The paper uses local network invariants and centrality measures to characterize individual players’ roles in passing networks. Closeness captures reachability, betweenness captures dependence on a player, and PageRank captures recursive passing popularity.
- Local network invariants, especially centrality measures, estimate player relevance or popularity according to different parameters.
- 3.1 Closeness: Equal weighting of incoming and outgoing passes is used for closeness, although arbitrary weights can adjust that choice.
- 3.1 Closeness: Closeness measures how easily a player can be reached within the team, with higher scores indicating smaller average network distance.
- 3.2 Betweenness: Betweenness measures how much ball flow between other players depends on a given player rather than how well-connected that player is.
- 3.2 Betweenness: A player’s betweenness indicates the potential impact of removing that player, while evenly distributed low scores indicate a balanced passing strategy.
- 3.3 Pagerank: PageRank recursively defines popularity: a player is important when receiving passes from other important players.
- 3.3 Pagerank: PageRank scores are jointly computed because each player’s score depends on all teammates’ scores.
- 3.3 Pagerank: PageRank’s pass-retention probability is a heuristic parameter that may be replaced by player-specific probabilities when players differ in ball-holding tendencies.
4 Clustering and communities
The paper uses clustering and clique structure to quantify how tightly football players interact and how connected a team is. However, community detection is uninformative in these small, highly connected networks.
- Clustering measures the degree to which players in a network tend to cluster together.
- The analysis uses a slight modification of the originally defined weighted-network clustering coefficient.
- A player has a high clustering score when many passes follow the path j → i → k, making that player a well-used middle-man.
- The team’s global average clustering coefficient is the average of its players’ clustering coefficients, while unbalanced passing lowers the coefficient.
- A maximal clique is a group of players connected by direct passes in every pair, with larger cliques indicating better team connectivity and smaller ones indicating fragmentation.
- Community detection usually yields one community containing every player because football teams have high connectivity and few nodes, so the initial approach provided little useful information.
5 Results and analysis
The authors analyze knockout-stage 2010 World Cup teams using passing networks and centrality measures, finding distinct patterns of connectedness, balance, and player dependence. Spain exhibits the strongest passing cohesion, while other teams show contrasting distributions and tactical structures.
- Data and scope: The study analyzes knockout-stage World Cup teams, focusing on Spain–Netherlands and Germany–Uruguay, using FIFA passing data and network measures.Passing networks were constructed from aggregate FIFA data and analyzed computationally.
- Teams in the last 16: Spain records the highest number of passes, clustering, and clique size, alongside high edge connectivity and low betweenness.These measures are interpreted as reflecting Spain’s well-connected total-football or tiki-taka style.
- Teams in the last 16: Paraguay shows low degree connectivity and high betweenness, appearing to rely disproportionately on a few players.The paper contrasts this pattern with teams whose passing structures are more connected and distributed.
- Spain vs the Netherlands: Spain’s players combine low, evenly distributed betweenness with consistently high clustering, indicating a balanced and highly connected passing strategy.Pedro is an exception, while forwards generally have the lowest closeness, betweenness, and pagerank scores.
- Spain vs the Netherlands: The Netherlands has clustering scores close to Spain’s but lower pass density, greater player separation, more evenly distributed pagerank, and a left-side pitch preference.The results suggest no Dutch player has a predominant role in the passing scheme.
- Germany vs Uruguay: Germany is more connected than Uruguay, but its less-distributed pagerank indicates greater dependence on a few players, especially Lahm and Schweinsteiger.Their central roles are compared with Xavi’s role for Spain.
6 Further work
The paper identifies extensions that would make football network analysis more detailed, including goal-directed shots, pass accuracy, and defensive events. It also notes that centrality measures complement static passing-network snapshots by quantifying player importance and connectedness.
- Scope and extensions: Passing networks are static snapshots, so centrality measures are needed to quantify individual players’ importance and connectedness.The paper notes that this information may benefit coaches, sports journalists, and their readers.
- Scope and extensions: A richer attack model could add the opponent’s goal as a node and represent shots rather than passes directed toward it.The proposed extension distinguishes shots on target from wide shots.
- Scope and extensions: Pass accuracy could be modeled by weighting each player according to the probability that passes from that player succeed.A first approximation would use the percentage of completed passes as the accuracy measure.
- Scope and extensions: Defensive strength could be incorporated by tracking passing interceptions and recovered balls.