The impact of team quality, match outcomes, and match location on pitch-passing networks

Abstract

This study investigated the impact of situational factors (team quality, match outcomes, and match location) on pitch-passing networks in the English Premier League. We analysed 295,901 possessions across 380 matches using a 7 × 7 grid system. For each pitch-passing network we calculated the degree centrality, the closeness centrality, and the betweenness centrality. A Mann–Whitney U test was employed to examine differences in these centrality measures across situational factors. Results indicated that strong teams demonstrated significantly higher degree centrality in midfield and centre-to-forward zones, and maintained superior closeness centrality across the pitch, particularly in central zones, while weak teams showed higher betweenness centrality on pitch flanks. Winning teams exhibited small differences in degree and closeness centrality across multiple zones, with notably medium differences in forward areas, and stronger closeness centrality throughout the pitch. At the same time, home teams demonstrated dominance in forward areas through degree centrality, while away teams showed higher activity in defensive baseline zones. Consequently, these findings demonstrate the systematic influences of situational factors on spatial passing patterns, providing empirical evidence for tactical adaptation. Coaches can leverage these insights to optimise passing strategies in critical areas such as midfield and forward zones to enhance offensive efficiency.

Introduction

Network analysis has emerged as a powerful methodological framework in sports science, essential for understanding team dynamics and tactical patterns across various sports including football¹, basketball², and rugby³. This approach enables a quantitative assessment of interactions within teams, yielding valuable insights that can enhance performance and strategy^4,5. In football, early applications focused on quantifying player interactions by treating players as nodes and employing weighted passes to illuminate key tactical features such as team playing styles, offensive patterns, and defensive structures^6,7. While this player-centred approach effectively identified crucial roles, such as playmakers, and revealed team coordination strategies through network metrics, it often overlooked important spatial aspects of tactical organisation. Specifically, it neglected how teams control and utilise different pitch areas to create and exploit opportunities^1,8.

To address these spatial limitations, researchers developed pitch-passing networks by dividing the pitch into distinct areas that serve as nodes, with passes between these areas forming the network edges⁹. This analytical framework offers significant advantages as it quantifies teams’ spatial control strategies, uncovers systematic tactical patterns, and allows for consistent cross-match analysis independent of player composition. For instance, analyses of FC Barcelona’s matches have demonstrated how their possession-based style produced unique spatial patterns, evidenced by higher centrality values in midfield zones, underscoring their emphasis on controlling the central area¹⁰. Similarly, the study of teams such as Real Madrid highlights their unique spatial tactics, confirming the reliability of the method in identifying specific tactical characteristics of teams and ensuring consistency in cross-season analysis¹¹. Recent research advancements indicate that incorporating spatial multilayer networks can more effectively analyse key areas of possession transitions between teams in football matches¹².

Despite these advancements, the influence of situational factors on pitch-passing networks remains inadequately explored. Research indicates that match-related variables significantly shape team performance^13,14. For example, team quality influences tactical approaches and spatial control capabilities¹⁵, while match outcomes drive tactical adaptation during games¹⁶. Additionally, match location affects risk-taking tendencies and tactical choices¹⁷. Studies of player-based networks reveal that these situational factors significantly influence network structures, with home matches typically yielding more cohesive passing networks¹⁸. Stronger teams maintain balanced network structures¹⁵, and winning teams exhibit more efficient network organisation¹. However, the specific impacts of these factors on spatial passing patterns and pitch zone utilization remain unexplored, as existing research has predominantly focused on overall network properties rather than spatial dynamics¹².

These developments in network theory have wide-ranging application to football match preparation and coaching⁵. Understanding how situational factors influence pitch-passing networks can also help coaches create more targeted training programs and match-day routines¹⁹. For instance, coaches can alter the spatial structure and passing network of their team according to whether they are at home or away, or against weaker or stronger opponents¹⁸. When facing stronger opponents, coaches can target the construction of specific passing connections in specific areas of the pitch to win possession and create chances²⁰. Similarly, understanding how successful teams adjust their passing networks as a function of changing match situations can inform tactical decisions regarding press intensity, defensive solidity, and attacking transition⁶. This practice cross-section of network analysis offers coaches numerical means to maximise scheduling and pairing choices and thus team performance through evidence-based tactical deployments²¹.

Building on the recognition of these situational influences, this study aims to investigate the impact of three key situational factors on the characteristics of football pitch-passing networks. Although various situational factors have been identified in football research, such as weather conditions²², crowd size²³, and referee decisions²⁴, team quality, match outcomes, and location have been consistently shown to significantly influence team performance and tactical behaviour across multiple studies²⁵. We aim to construct a workable analytical framework that will provide a foundation for future applications to update data. Specifically, this study aims to: (i) quantify the differences in network centrality metrics of pitch-passing networks under different situational factors (team quality, match outcomes, and match location); (ii) identify the spatial distribution patterns of these differences across pitch areas; and (iii) assess the statistical significance and effect sizes of these differences. Through these analyses, we provide empirical evidence for understanding spatial tactical adaptation in football matches.

Methods

Dataset description

We analysed data from the 2017/2018 English Premier League (EPL), selected for its comprehensive spatial coordinates and public accessibility. The data is sourced from public datasets on GitHub, with the dataset provided by Wyscout²⁶. The reliability of data extracted from the Wyscout platform was confirmed based on results from a previous study²⁶. Specifically, researchers employed an inter-coder reliability assessment, two independent groups of operators coded the same match events, and the consistency was evaluated using Cohen’s kappa coefficient. The analysis revealed extremely high inter-operator agreement (kappa = 0.92–0.94), indicating robust quality assurance for this dataset²⁷. We used the Kloppy library in Python (3.11.7) to load and analyse the data, with its source code hosted on GitHub (https://github.com/PySport/kloppy). The dataset encompasses all 380 matches from the 2017/2018 English Premier League season, with 20 teams each playing 38 matches. In the data processing phase, we first downloaded and installed the Kloppy library, then imported the relevant modules to retrieve match and team IDs. For data filtering, we identified all events labelled as ‘pass’ and retained only those with event results marked as either ‘succeed’ or ‘completed’, ensuring that only successfully completed passing actions were included in the analysis. For each successful pass, we recorded its start and end coordinates (x, y). Wyscout provided the pitch coordinates, where all teams’ pitch x and y axes were normalised to a range of 0 to 100, including the player coordinates, with the y-axis flipped, as illustrated in Fig. 1. Both the starting and ending coordinates of the players were selected. No distinction was made between halves, with both the first and second halves included in the analysis. Each match analysed one team, with each team having its pitch-passing networks for each match. This dataset contains key information for each pass during the matches, including (i) the player executing the pass, (ii) the player receiving the pass, (iii) the coordinates indicating the positions of the sending and receiving players, and (iv) the timestamp representing the exact moment when the pass occurred, as shown in Table 1.

Fig. 1

Source: Wyscout standardised coordinate system (https://apidocs.wyscout.com/#section/Data-glossary-and-definitions).

Standardised pitch coordinate system for network analysis. Origin: Top-left corner (0,0). X-axis: Left to right (0–100). Y-axis: Top to bottom (0–100). Attack direction: Left to right.

Table 1 Standardised match event data structure. Timestamp: match time in seconds (2 decimal precision). Team id: numerical team identifier. Position coordinates: normalised to 0–100 scale. All coordinates represent successful passes only.

Study design

In particular, we first delineate the quality of the teams throughout the season, distinguishing between strong and weak teams based on the top and bottom five teams at the end of the season. The method of classifying team quality based on final league position has been widely adopted in football research. For example, previous studies have classified teams in the upper half of the league as strong and those in the lower half as weak teams¹³; similarly, researchers have used the final English Premier League rankings as a measure of team ability²⁸; furthermore, previous research analysing the Spanish LaLiga has also categorized teams into champion teams, Champions League qualification teams, Europa League qualification teams, mid-table teams, and relegated teams²⁹. Moreover, selecting the top five and bottom five teams as research subjects not only maximizes between-group differences but also has practical significance, as the top five teams typically represent European competition qualification positions, while the bottom five include relegation zone teams with specific tactical performances and coping strategies. We then proceed to delineate the outcome of the game, distinguishing between winning and losing teams. This focus on decisive match outcomes (wins and losses) enables us to identify clear tactical patterns in passing networks that differentiate successful and unsuccessful performance, following similar methodological approaches in previous pitch-passing network studies⁸. Finally, we delineate the location of the game, distinguishing between home and away teams. The subsequent step involved processing the match data using Python(3.11.7)’s Kloppy library. For each match, we extracted and matched the passing events with their corresponding temporal and spatial information. Specifically, we collected the timestamp of each completed pass (in seconds from match start), along with the spatial coordinates (x, y) of both the passing and receiving players. This data processing approach ensured that all successful passes were properly mapped to their precise temporal and spatial situational within the match. This data was subsequently visualised on the pitch, along with the passing routes between the players. The pitch is then divided into 49 blocks¹⁰ of equal size, and the number of passes in each block and the number of passes from block to block are counted and converted into an adjacency matrix, which is then drawn into a field passing network. Each team has its block adjacency matrix for each game. After constructing the pitch-passing networks, we calculated three centrality metrics for each pitch block using NetworkX library in Python (3.11.7). For each match, we computed degree centrality (DC) to measure the frequency of direct passing interactions, closeness centrality (CC) to evaluate the efficiency of ball movement, and betweenness centrality (BC) to identify key transitional zones. These calculations were performed for each team’s passing network separately. Subsequently, we conducted Mann-Whitney U tests to compare these centrality measures between different situational factors. Specifically, we compared the centrality values for each block, between strong and weak teams (based on final league positions), between winning and losing teams, and between home and away teams. Each comparison was performed independently, allowing us to investigate the impact of these situational factors on the spatial structure of passing networks. Subsequently, the eta squared effect size (ES), η², was calculated using the nonparametric ES formula provided on the https://www.psychometrica.de/effect_size.html website³⁰. According to this formula, η² = 0.01 represents a small effect, η² = 0.059 represents a medium effect, and η² = 0.138 represents a large effect³⁰.

Pitch-passing networks (PPN) construction

Based on previous empirical evidence suggesting the optimal block number for tactical pattern recognition¹¹, we employed a symmetrical grid system that facilitates systematic comparisons of attacking and defending transitions through horizontal analysis and reveals tactical preferences for left and right pitch utilization through vertical analysis. We employed three complementary network metrics to assess each area’s tactical significance, where degree centrality gauges the frequency and intensity of direct passing exchanges, closeness centrality evaluates the efficiency of ball movement through specific zones, and betweenness centrality identifies crucial areas for transitioning between defence and attack. We used a method from previous studies to construct the pitch-passing networks (PPN) for each team¹⁰. The pitch is divided into blocks, with the pitch block count specified by $\:h$ for the horizontal subdivisions (x-direction) and $\:\upsilon\:$ for the vertical subdivisions (y-direction), Use $\:N$ to represent the number of nodes. Previous studies³¹ show that pitch-passing networks with about 50 blocks can provide consistent and identifiable patterns for teams. Thus, we set $\:h$ = 7 and $\:\upsilon\:$ = 7 (respectively, representing a horizontal direction containing 7 blocks and a vertical direction containing 7 blocks), resulting in $\:N\:$= $\:h\times\:\upsilon\:$ = 49 blocks for analysis. Each block represents a node in the network. When a pass occurs from block $\:i$ to block $\:j$, a link is created from node $\:i$ to node $\:j$, with a weight assigned to quantify the total number of completed passes in that direction. This approach produces a weighted directed network represented by an adjacency matrix (as shown in Table 2). Note that each team has its pitch-passing networks. Next, we categorised the top five and bottom five teams from the 17/18 season standings into strong and weak teams, separated match outcomes into wins and losses, and categorised match locations into home and away games. We calculated the degree, closeness, and betweenness centrality for different contextual factors (as illustrated in Fig. 2).

Table 2 Network adjacency matrix for PPN. Values represent the count of successful passes between blocks. Diagonal elements indicate internal block passes.

We divided the pitch into 7 × 7 = 49 blocks for analysis, with the attacking direction set from left to right. Subsequently, each block was assigned a numerical identifier (as shown in Fig. 3), and an “x–y” symbol was used to number the blocks. Here, x represents the numerical value in the horizontal direction, where a higher value indicates proximity to the attacking zone and a lower value indicates proximity to the defensive zone. Meanwhile, y represents the numerical value in the vertical direction, with a higher value indicating proximity to the left wing of the field and a lower value indicating proximity to the right wing of the field. Each block corresponds to a node, with passes between blocks represented as edges, and the number of passes serving as edge weights. For each team in each match, we constructed a pitch-passing network $\:{G}_{i}$ and calculated the degree centrality, closeness centrality, and betweenness centrality of each node (block) in the network.

Fig. 2

Methodological framework for PPN analysis: (a) Constructing pitch block numbers. (b) Convert the PPN into an adjacency matrix for easier processing and calculation. (c) Construct the PPN. (d) Compute the DC, CC, and BC for each team. (e) Statistical analysis the DC, CC, and BC under different situational factors.

Fig. 3

Pitch division framework for network analysis. 7 × 7 grid division framework for PPN analysis showing block identification system with numbered coordinates (1–1 to 7–7) marked at each position. The pitch is divided into tactical zones: defensive third (left side, blocks 1–1 to 2–7), midfield (centre, blocks 3−1 to 5–7), and attacking third (right side, blocks 6−1 to 7–7), with centre circles and goal areas clearly marked.

Parameter definition

Table 3 Interpretation of network centrality metrics in football context. Note: practical examples are based on common tactical observations in football, provided for illustrative purposes rather than as research findings.

After constructing the pitch-passing networks with 49 nodes, we need to extract relevant information from it. We use network parameter metrics from social network analysis to obtain the necessary insights (as shown in Table 3). After dividing the pitch into $\:\text{N}$ blocks, passing between pitches can be viewed as a pitch-passing network. Each block represents a node in the network, and passing between blocks represents directed edges. We can represent this network with an $\:\text{M}$ square matrix $\:{W}_{\:}$ = $\:{W}_{ij}$, where each element $\:{W}_{ij}$ represents the number of passes from segment $\:{p}_{i}$ to block $\:{p}_{j}$. Each element denotes the number of passes from one block to another. Through this network, we can clearly understand the connections between different segments and the flow of passes, providing a foundation for further analysis and decision-making³².

$$\:{W}_{i}=\sum\:_{j=1}^{N}\:{p}_{ij}$$

(1)

Degree centrality (DC)

Degree centrality is a metric that measures the overall connectivity of passing interactions between blocks on the pitch. A high degree centrality value indicates that a block has more passing connections with other blocks, highlighting its critical position in the pitch-passing network. Out-degree centrality (ODC) measures the connectivity of passes going out from a block, while in-degree centrality (IDC) focuses on the connectivity of passes received by a block. Additionally, standardizing out-degree centrality can further aid in analysis:

$$\:{C}_{(D-out)}^{{\:}^{{\prime\:}}w}=\frac{{W}_{i}^{w-out}}{\sum\:_{i=1}^{N}\:\sum\:_{j=1,j\ne\:i}^{N}\:{p}_{ij}}$$

(2)

where $\:{W}_{i}^{w-out}$ represents the out-degree centrality value for a node $\:i$ (i.e., a pitch block), and $\:{p}_{ij}$ is an element in the weighted adjacency matrix of graph $\:\text{G}$ (the pitch-passing network). Similarly, by changing to incoming data, we can calculate the standardised in-degree centrality for each block^6,32.

Closeness centrality (CC)

Closeness centrality reflects the degree of closeness between blocks on the pitch. If a segment can easily interact with all other blocks, it is considered important in the network. Closeness centrality can be obtained by calculating the average shortest path length, which is the average distance from a block $\:{p}_{i}$ to all other blocks in the network, denoted as $\:{d}_{i}=\frac{1}{N-1}\sum\:_{j\ne\:i}\:{d}_{ij}$. Then, the closeness centrality of segment $\:{p}_{i}$ is defined as the reciprocal of this average distance $\:{d}_{i}\:$³²:

$$\:{u}_{CC}\left(i\right)=\frac{1}{{d}_{i}}=\frac{N-1}{\sum\:_{j\ne\:i}\:{d}_{ij}}$$

(3)

Nodes (blocks) with higher closeness centrality values have shorter distances to other nodes (blocks) in the network, indicating that they are more central and important in the network. Although these nodes may not have the highest number of connections, they have a wider range of influence and can better control the flow of information. Compared to degree centrality, closeness centrality is better at reflecting the overall network structure, although it has a higher computational complexity and relies more on the network’s topology.

Betweenness centrality (BC)

Betweenness centrality mainly identifies the bridge blocks that construct the pitch-passing network. Simply put, it describes a node’s influence over the shortest paths in the network and the potential traffic that the node may need to handle. The betweenness centrality for node $\:{p}_{i}$ is defined as³²:

$$\:{\mu\:}_{BC}\left(i\right)=\sum\:_{i\ne\:s,i\ne\:t,s\ne\:t}\:\frac{{g}_{st}^{i}}{{g}_{st}}$$

(4)

where $\:{g}_{st}$ represents the total number of shortest paths from node $\:{p}_{s}$ to node $\:{p}_{t}$, and $\:{g}_{st}^{i}$ represents the number of those shortest paths that passing through node $\:{p}_{i}$. In a connected network with $\:\text{N}$ nodes, the central node in a star-shaped network has the maximum betweenness centrality, reaching $\:\frac{(n-1)(n-2)}{2}$. Thus, the normalised betweenness centrality for node $\:{p}_{i}$ can be derived as³²:

$$\:{\mu\:}_{BC}^{{\prime\:}}\left(i\right)=\frac{2}{(n-1)(n-2)}{\mu\:}_{BC}\left(i\right)$$

(5)

The greater a node’s betweenness centrality value, the more network traffic it carries and the more influence it has. This also means that the node is more prone to congestion and can become a network bottleneck. Betweenness centrality can accurately identify nodes with high “traffic” in the network, but its time complexity is $\:O\left({N}^{3}\right)$, making it computationally inefficient for large-scale networks³².

When calculating centrality metrics, we normalised all centrality results, allowing for comparability of node centrality across different network structures.

Statistical analysis

Since the data do not follow a normal distribution, All data processing and statistical analyses were performed using Python(3.11.7) with multiple packages, including NetworkX for network analysis and centrality calculations, pandas for data manipulation, and scipy.stats for statistical testing. The normality of data distribution was tested using the Shapiro–Wilk test, which showed significant deviations from normality (p < 0.05) for our centrality measures. Due to the non-normal distribution of data and the potential presence of outliers caused by specific teams, we chose to use the Mann-Whitney U test for statistical analysis. As a non-parametric test method, the Mann–Whitney U test compares ranks rather than raw values, demonstrating strong robustness against outliers³³. This ensures that even if there are individual teams with exceptionally outstanding performance, they will not have a disproportionate impact on the overall between-group comparison results, thereby more accurately reflecting the systematic differences between teams of different quality. It’s safer to use non-parametric test methods to compare the pitch-passing network metrics between strong and weak teams, win and lost matches, and home and away matches. The Mann–Whitney U test does not assume any underlying distribution and compares medians instead of means. Therefore, we used the non-parametric Mann–Whitney U test to evaluate the centrality results calculated for each subdivided block, with each team’s pitch-passing network defined by a network parameter for each match. Additionally, all p-values were adjusted using a common error discovery rate procedure, with α = 0.05³⁴. This approach ensures that all statistical comparisons meet the strictest standards. We calculated the ES using an online non-parametric ES calculator. The underlying principle was to use the standardised Z-value from the Mann-Whitney U test statistics to calculate the ES η², where η² = 0.01 indicates a small effect, η² = 0.059 a medium effect, and η² = 0.138 a large effect³⁰. The specific formula for calculating η² is:

$$\:{\eta\:}^{2}=\frac{{Z}^{2}}{{n}_{1}+{n}_{2}}$$

(6)

$\:Z$ represents the Z-value in the Mann-Whitney U test. $\:{n}_{1}$ is the sample size of the first group, and $\:{n}_{2}$ is the sample size of the second group. We calculated the median difference by subtracting the latter group’s median from the former group’s median to determine if the former was higher or lower than the latter. We used red to indicate the former is higher and blue to indicate the latter is higher, with varying shades representing the ES of significant differences.

Results

We investigated the impact of situational factors (team quality, match outcomes, and location) on the English Premier League’s pitch-passing networks (PPN). Through the calculation of three complementary network metrics, degree centrality (DC, measuring passing frequency), closeness centrality (CC, evaluating passing efficiency), and betweenness centrality (BC, identifying crucial transitional zones), we examined how these situational factors influence spatial tactical patterns (see Methods Section for details on the parameter definition). The findings demonstrate systematic differences in tactical adaptation under varying match conditions, which will be elaborated upon in the following sections.

Effects of team quality on central indicators

Strong teams demonstrated significantly higher DC in midfield and centre-to-forward zones (blocks 3-(3 ~ 5), 4-(2 ~ 6), 5-(2 ~ 6), 6-(2 ~ 6), p < 0.05, η² = 0.162 ~ 0.311). The goalkeeper’s zone (block 1–4) exhibited significantly higher activity in strong teams (p < 0.05, η² = 0.169), indicating active involvement in build-up play. The left front pitch zone (blocks (5 ~ 6)-1), right central front zone (blocks (3 ~ 6)-7), and front baseline (blocks 7-(2 ~ 3), 7−5) showed moderate effects favouring strong teams (p < 0.05, η² = 0.071 ~ 0.134).

Strong teams maintained superior connectivity across the entire pitch in CC analysis. Large effect sizes were found in goalkeeper position (block 1–4), forward baseline (blocks 7-(2 ~ 3), 7-(5 ~ 6)), pitch sides (blocks (3 ~ 6)-1, (3 ~ 6)-7), and central zones (blocks 2-(2 ~ 6), 3-(2 ~ 6), 4-(2 ~ 6), 5-(2 ~ 6), 6-(2 ~ 6)) (p < 0.05, η² = 0.155 ~ 0.342). Moderate effects occurred in defensive baseline zones (blocks 1-(1 ~ 3), 1–5) and pitch sides (blocks 2−1, 2–7, 7−1, 7–7) (p < 0.05, η² = 0.072 ~ 0.137).

In BC, weak teams showed higher values on pitch flanks (blocks 5−1, 5–7, p < 0.05, η² = 0.073, 0.070). Strong teams exhibited small effects in defensive baseline (blocks 1–4, 1-(1 ~ 2)), midfield (block 4−3), and forward zones (blocks 5-(4 ~ 5), 6-(2 ~ 6), 7-(5 ~ 6), 7−2) (p < 0.05, η² = 0.015 ~ 0.056). As shown in Fig. 4, strong teams display significantly higher centrality metrics across various zones.

Fig. 4

Network centrality patterns: strong vs. weak teams. (a) PPN of the DC, (b) PPN of the CC, (c) PPN of the BC. Colours indicate significant differences (p < 0.05): red represents strong teams outperforming weak teams and blue represents weak teams outperforming strong teams. Colour intensity indicate the size of the effect: dark (η² ≥ 0.138), medium (0.059 ≤ η² < 0.138), and light shades (0.01 ≤ η² < 0.059). Attack direction: left to right.

Effects of match outcomes on central indicators

The DC analysis indicated small effects favouring winning teams in multiple zones. On pitch sides (block 3–7, blocks (5 ~ 7)-7, blocks (5 ~ 7)-1), central defensive zone (blocks 3-(3 ~ 6)), and central forward zone (blocks 4-(2 ~ 6), 5-(2 ~ 6), 6-(2 ~ 6)), winning teams showed significantly higher activity (p < 0.05, η² = 0.015 ~ 0.051). Notably, blocks 7−2 and 7−6 demonstrated medium effects (η² = 0.060, 0.059), highlighting their importance in creating scoring opportunities.

Winning teams established stronger connections throughout the pitch in CC. At the defensive baseline (blocks 1–4, 1–5) and forward baseline (blocks 7-(1 ~ 5), 7–7), winning teams showed small effects (p < 0.05, η² = 0.011 ~ 0.045). Block 7−6 exhibited a medium effect (η² = 0.066). Both pitch sides (blocks (3 ~ 6)-7, (2 ~ 6)-1) and midfield zones (blocks 2-(2 ~ 6), 3-(2 ~ 6), 4-(2 ~ 6), 5-(2 ~ 6), 6-(2 ~ 6)) displayed small effects favouring winning teams (p < 0.05, η² = 0.012 ~ 0.053).

In BC, winning teams showed superiority in forward areas (blocks 7−2, 7−6, 6−4, p < 0.05, η² = 0.013 ~ 0.018), while losing teams exhibited higher centrality in left attack positions (block 5−1, p < 0.05, η² = 0.012). As shown in Fig. 5, these findings emphasise the tactical advantages held by winning teams in various zones.

Fig. 5

Network centrality patterns: winning versus losing teams. (a) PPN of the DC, (b) PPN of the CC, (c) PPN of the BC. Colours indicate significant differences (p < 0.05): red represents strong teams outperforming weak teams and blue represents weak teams outperforming strong teams. Colour intensity indicate the size of the effect: dark (η² ≥ 0.138), medium (0.059 ≤ η² < 0.138), and light shades (0.01 ≤ η² < 0.059). Attack direction: left to right.

Effects of match location on central indicators

Home teams demonstrated dominance in forward areas through DC. The right side (blocks (5 ~ 7)-7) and left side (blocks (6 ~ 7)-1) showed small effects favouring home teams (p < 0.05, η²= 0 .014 ~ 0.050). Forward baseline zones (blocks 7-(2 ~ 6)) and additional areas (blocks 5–5, 5–6, 6–6, 6-(2 ~ 3)) exhibited small effects favouring home teams (p < 0.05, η² = 0.011 ~ 0.026). Away teams showed higher activity only in defensive baseline zones (blocks 1–1, 1–2, p < 0.05, η² = 0.010).

In CC, home teams displayed superior connectivity in attacking zones. Right pitch side (blocks (4 ~ 7)-7), left side (block 6−1), and forward baseline (blocks 7-(2 ~ 6)) showed small effects favouring home teams (p < 0.05, η² = 0.013 ~ 0.033). Limited differences appeared in defensive zones (blocks 3–4, 3–6) and central forward zones (blocks 4-(5 ~ 6), 5-(5 ~ 6), 6–6, 6-(2 ~ 3)) (p < 0.05, η² = 0.010 ~ 0.023).

The BC indicated minimal differences, with only block 2−1 showing a small effect favouring away teams (p < 0.05, η² = 0.012), suggesting similar transitional play patterns between home and away matches. As shown in Fig. 6, these findings illustrate the tactical advantages and patterns associated with home teams.

Fig. 6

Network centrality patterns: home versus away teams. (a) PPN of the DC, (b) PPN of the CC, (c) PPN of the BC. Colours indicate significant differences (p < 0.05): red represents strong teams outperforming weak teams and blue represents weak teams outperforming strong teams. Colour intensity indicated the size of the effect: dark (η² ≥ 0.138), medium (0.059 ≤ η² < 0.138), and light shades (0.01 ≤ η² < 0.059). Attack direction: left to right.

Discussion

We constructed pitch-passing networks of soccer teams^10,35 and studied the impact of different situational factors on the passing network structure. In this study, we used three centrality metrics, namely, degree centrality, closeness centrality, and betweenness centrality, to provide a comprehensive analysis of pitch-passing networks. Degree centrality helped us identify the most frequently passed zones, reflecting their active roles in both offensive and defensive positions. Closeness centrality measured how efficiently a zone can reach other areas, reflecting the overall spatial coverage and connectivity of the network. Betweenness centrality reveals which zones act as bridges connecting different regions of the pitch, indicating their importance in maintaining the flow of the passing network. Our findings largely align with previous theoretical frameworks while also revealing novel insights about spatial tactical adaptation. Consistent with earlier research on player-based networks³⁶, our results confirm that situational factors significantly shape passing network structures. However, our pitch-area focused approach provided additional insights into spatial control strategies that were not captured in previous player-centred analyses. As hypothesised based on existing literature¹⁸, we found stronger teams and home teams generally maintain more balanced network structures. However, our spatial analysis revealed that this balance manifests specifically through higher centrality in midfield and centre-to-forward zones, rather than uniform distribution across all areas. The findings regarding team quality particularly support previous research suggesting that stronger teams maintain more efficient network organisation¹⁵. However, our spatial analysis extends this understanding by showing that this efficiency is achieved through specific tactical choices, notably higher DC in midfield and centre-to-forward zones, and superior CC across the entire pitch. This aligns with but also elaborates on previous findings about FC Barcelona’s possession-based style¹⁰, providing quantitative evidence for how strong teams achieve spatial control.

Impact of team quality on pitch-passing networks

In comparing the degree centrality results for team quality, we observed that only 1–3_block and 1–7_block did not show significant differences, while other blocks exhibited significant differences of varying effect sizes. The frequency of passes by strong teams in the midfield and forward blocks was significantly higher than that of weaker teams. In contrast, weaker teams tended to rely on the flanks, indicating less control in central areas. Previous research has also found that strong teams are more likely to dominate the midfield¹². For example, Manchester City, through the outstanding performances of players like Kevin De Bruyne and Ilkay Gündogan, controls the game in midfield, rapidly transitions to attack, and suppresses opponents. This may be attributed to strong teams having powerful midfield players who can more easily organise attacks or defend from the back by controlling the midfield. Additionally, we observed that in the 1–4_block, representing the goalkeeper position for strong teams, the degree centrality and closeness centrality of goalkeepers were significantly higher compared to those of weaker teams. This indicates that strong teams’ goalkeepers are not only defensive players but also actively involved in organizing attacks. Previous research has also shown that goalkeepers can initiate attacks through long and short passes³⁷. The high connectivity of the goalkeeper’s block with other blocks suggests frequent interactions between the goalkeeper and teammates. In prior research, the goalkeeper’s position is often overlooked, and some studies even exclude goalkeeper data. Therefore, we believe that in strong teams, the goalkeeper can act as an initiator of attacks or a key passer, helping to construct a fluid attacking system.

By calculating closeness centrality, we discovered that strong teams not only excel in midfield, but they also effectively cover the entire pitch, allowing them to quickly transition the ball to different areas and maintain possession across all zones. This finding aligns with previous research, which has shown that teams like Barcelona, under a possession-based strategy, maintain high closeness centrality across all players³⁸. This suggests that strong teams effectively use every area of the pitch, initiating attacks not only from the central midfield but also from the flanks and defensive lines. As a result, they achieve broader attacking and defensive tactics, reduce the distance between key areas, and improve overall passing efficiency. In contrast, previous studies have found that weaker teams are more vulnerable to attacks from stronger teams due to less effective spatial utilization³⁹.

From the perspective of betweenness centrality, strong teams exhibit significantly higher betweenness centrality in the defensive baseline, midfield, and forward areas compared to weaker teams, showing a small effect size. Weaker teams show significant differences with small to medium effect sizes on the flanks. Betweenness centrality, as an important indicator in social networks, helps identify blocks that play a crucial role in constructing shortcuts reducing the distance of the overall pitch-passing network. Removing these key blocks could crucially damage structure of the passing network⁴⁰. The higher betweenness centrality of weaker teams’ flanks compared to stronger teams indicates their reliance on these flank blocks to establish “bridges” in their attacking and defensive strategies. Previous research has also found that weaker teams face difficulties initiating attacks from the midfield and may be more inclined to create scoring opportunities through flank attacks⁴¹.

In summary, strong teams display a more balanced and fluid passing network with higher connectivity and ball control efficiency in the midfield, forward, and flank areas. This may be attributed to their tactical advantages, team cooperation, and player synergy⁴². Weaker teams’ disadvantage in the midfield may stem from weaker awareness⁴³, tactical strategies of the coaching staff⁴⁴, or a tendency to organise attacks on the flanks. Attacking from the flanks often results in defensive players applying pressure in smaller areas, making it more challenging for defenders to reduce the space available to the ball carrier compared to central or midfield positions⁴⁵. Weaker teams’ disadvantages in the midfield hinder their ability to penetrate offensively. Therefore, they may adopt targeted defensive strategies, such as narrowing the defensive line and limiting the strong team’s activity space in the midfield. Previous research also suggests that narrowing effective passing and control areas on the pitch and strengthening midfield dominance may be necessary. This requires one or more strong players to enhance the midfield’s dominance³⁹. In conclusion, strong teams likely focus more on controlling the midfield and forward areas, resulting in denser passing networks, tighter connections, and more effective attacks. Conversely, weaker teams may focus more on breakthroughs on the flanks, leading to more dispersed passing networks, relatively weaker connections, and greater reliance on individual abilities for attacking.

Impact of match outcomes on pitch-passing networks

In comparing the degree centrality of match outcomes (win vs. lost), we observed that winning teams had significantly higher passing frequencies in the midfield, attacking midfield, forward areas, and both the left and right sides of the attacking front compared to losing teams, showing a small effect size advantage. This indicates that winning teams dominate the pace of the game by increasing the passing frequency in these areas. Previous research supports this view, suggesting that increased offensive activity contributes to a higher probability of winning³⁸. Additionally, studies have shown that winning teams focus more on attack during matches, creating more scoring opportunities⁴⁶. This is consistent with our observation that winning teams have higher centrality in several key areas. Specifically, winning teams organise their attacks through midfield control, a strategy that has been validated in multiple studies⁴⁷. Further research has also found that midfielders of winning teams generally have a higher degree of centrality, indicating that the midfield plays a crucial role in team victories⁶.

Closeness centrality results show that winning teams outperform losing teams significantly in the attacking midfield area. Previous research has indicated that winning teams achieve higher passing success rates in the midfield and attacking areas⁴⁸. In contrast to previous studies, this research finds that winning teams also establish close connections in the defensive baseline, such as in the 1–4_block and 1–5_block. This suggests that winning teams build tight connections in defence while actively pushing forward in attack.

Regarding betweenness centrality, only the 7−2 block, 6−4 block, and 7−6 block show significant differences, indicating that winning teams utilise these blocks as key attacking bridges. Losing teams, on the other hand, primarily use the 5−1_block on the left side of the pitch as their critical attacking area. The three key blocks for winning teams may reveal different angles from which teams establish passing links before shooting. Previous research has also highlighted the impact of passing network structure on team ball control, shooting attempts, and scoring⁸.

In summary, centrality indicators indeed help explain the differences between winning and losing teams. Winning teams prioritise controlling the midfield and establishing passing networks in these areas to effectively organise attacks, thereby controlling the pace of the game and creating scoring opportunities. Therefore, maintaining efficiency in ball control and passing frequency in the midfield and forward areas, combined with offensive intensity, are key factors in improving the probability of winning.

Impact of match location on pitch-passing networks

When comparing the degree centrality indicators of home and away teams, home teams exhibit a significantly higher degree of centrality in the attacking midfield, forward areas, attacking baseline, and both the left and right flanks compared to away teams. This suggests that home teams are more proficient in ball control and passing in these areas. Previous research has also found that home teams excel in ball control, passing, and creating scoring opportunities through flank attacks and crosses⁴⁹. Home advantage benefits from fan support and encouragement, which enhances player morale and confidence, leading to better performance. Additionally, home players are more familiar with the home environment, including the pitch, weather, and spectators, which helps them adapt better and perform at a higher level⁵⁰. Conversely, away teams have a significantly higher degree of centrality in the defensive baseline (1–2 block) and left flank (1–1 block) compared to home teams, indicating their emphasis on defensive strategies to restrict the home team’s attacking organisation and prevent goals. This may be attributed to the pressure of playing away, including opposition from home fans and unfamiliarity with the away environment, which may affect player performance and lead to passing errors¹³. Furthermore, coaches may adopt more conservative strategies when playing away to avoid risky attacks, potentially leading to less dynamic and creative passing in the forward areas⁵¹.

In terms of closeness centrality, home teams are significantly better than away teams in the attacking midfield block. Unlike away teams, home teams also establish close connections in the defensive midfield. This indicates that home teams have more coherent connections between the front and back areas. Previous studies have also shown that home teams can better establish connections between the front and back areas due to the home advantage, leading to more effective attacking⁴⁹.

Interestingly, the calculation of betweenness centrality did not reveal significant key blocks for home teams compared to away teams on the pitch. However, away teams have significantly higher betweenness centrality in the 2−1 block compared to home teams. This indicates that they establish bridges in this area to construct the field passing network, restrict the home team’s attacking actions, and attempt counterattacks. Away teams may use defensive counterattacks on the left flank to disrupt the home team’s defence, which may result in fewer passes and simpler passing routes.

Overall, by constructing passing networks and calculating centrality indicators, we found that strong teams, winning teams, and home teams have tighter connections between the midfield and forward areas. Through this approach, we visually observe a common feature among strong teams, winning teams, and home teams: the emphasis on strengthening connections between the midfield and forward areas. However, this study has certain limitations, regarding situational factors, our binary classification of team quality based solely on the top and bottom five teams may be overly simplistic, which might not fully capture the nuanced performance differences across the league standings. Furthermore, although this static classification method is widely accepted in football research, future studies could consider more refined and dynamic methods for assessing team quality. For example, researchers could verify the homogeneity within strong and weak team groups, or employ cluster analysis to group teams based on passing network characteristics, and then compare the relationship between these clusters and league rankings. Additionally, there are potential constraints due to the range of available data, and focusing only on specific seasons or levels of competition. Future research could expand data collection to include more seasons and various competition levels to improve the generalizability of the results. Furthermore, this study did not consider individual player capabilities and team tactical arrangements (including formations and player positions), which could affect the structure of the passing network and the connectivity patterns between different areas. Future research could extend to include data from more teams and matches while considering additional contextual factors to achieve a more comprehensive understanding of these dynamics. Additionally, analysing specific player data, such as positions and abilities, could provide deeper insights into the pitch network structure and dynamics. This aspect will be a focus of our future research. This study analyses match location and match outcome as independent factors, which is a common approach in football research. However, it’s important to acknowledge that home advantage may have some association with win rates, although this association has significantly weakened in modern football⁵². Future research could consider more detailed grouping methods, such as dividing matches into four categories: home wins, home losses, away wins, and away losses, to further distinguish the interactive effects of match location and match outcome on passing networks. This approach may provide a more nuanced analytical perspective, but would require larger sample sizes to ensure adequate statistical power. Moreover, to test the scientific and practical validity of the pitch-passing network analysis method, we used data from the 2017-18 season for case studies. It is important to note that while the data source is relatively dated, the focus of this study is on methodological exploration rather than specific tactical explanations for the latest seasons.

Conclusions

This study offers valuable insights into football spatial passing networks, providing both theoretical and practical contributions to a better understanding of match dynamics and tactical decision-making. Thus, the following conclusions can be established: (i) The study found that strong teams tend to engage in frequent passing in the midfield and forward areas, particularly in regions such as the 3-(3 ~ 6) block. This indicates that strong teams prioritise control and connectivity in these areas when organizing their attacks; (ii) The differences in the spatial passing network structure between winning and losing teams show that winning teams have significantly higher passing frequency and efficiency in the central midfield and forward areas compared to losing teams, resembling the structure of strong teams. This suggests that controlling key areas may increase the likelihood of winning; (iii) The differences in the spatial passing network structure between home and away teams reveal that home teams have significantly higher passing frequency and efficiency in attacking areas and flanks compared to away teams, who show stronger connectivity in defensive areas. This indicates that home teams may focus more on leveraging home advantage, while away teams prioritise defence; (iv) These findings provide valuable tactical insights for coaches. For example, coaches can select appropriate passing areas and routes based on different match scenarios to adapt to various opponents and match conditions. Specifically, in critical areas like midfield and forward, effective passing strategies can be designed to enhance offensive efficiency; (v) Overall, this study offers an innovative analysis of football spatial passing networks using network science methods, revealing the structural characteristics of passing networks under different contexts and providing valuable insights for tactical decision-making; and (vi) Future research will continue to expand and deepen this perspective, further enhancing our understanding of football matches. Particularly promising directions include longitudinal studies examining passing network patterns across multiple seasons, integrating physical performance metrics with network analysis, and developing practical applications that translate these network insights into specific training interventions and match strategies across different competitive levels.