Accelerometer-based network analysis in female soccer: performance levels

This dataset examines the complexity of network structures in professional and collegiate women’s soccer teams using directed network analysis based on tri-axial acceleration data. The study involved one professional team and one university-level team, with data collected from matches during their respective seasons. Directed network analysis identified dyads and triads, representing cooperative interactions among players, while movement entropy quantified the influence of individual movements within the team. Network diversity, defined as the variability in activation probabilities of dyads and triads, was calculated to evaluate the tactical dynamics and cooperative behaviors of the teams. Data were collected using GNSS devices equipped with tri-axial accelerometers, ensuring precise measurement of movement intensity. The findings provide insights into the structural and functional differences in team coordination between professional and collegiate levels. The dataset is anonymized and adheres to ethical guidelines, enabling reproducibility and further exploration of team dynamics in sports science.

Methods
Participants
Prior to participant recruitment, we calculated the minimum required number of matches using G*Power 3.1.9.4 (Heinrich Heine Universität Düsseldorf, Germany). This study employs a two-way analysis of variance (ANOVA) to primarily examine the interaction effects between the period of the match (the first half and second half of the match) and three team groups (professional teams during the first half of the season, professional teams during the second half of the season, and collegiate teams). Thus, the calculation for the F-test with ANOVA was conducted a priori, given an effect size of 0.40, an α error probability of 0.05, a power of 0.80, and a numerator df of 2 with six groups. The effect size (0.40) for this analysis was set based on findings from a previous study that examined changes in team coordination states during matches and reported a large effect size (η² = 0.240 to 0.263) for differences influenced by the level of the opposing team. The total required sample size was calculated as 64 matches across six groups, with 11 matches per group.

Based on this analysis, we recruited one professional-level women’s soccer team and one university-level women’s soccer team, both playing 11 matches per season. However, data were missing for four matches in the first half of the season and three matches in the second half for the professional team. Similarly, four matches were missing for the university team. Ultimately, data were collected from seven matches in the first half of the season (hereafter referred to as Pro1) and eight matches in the second half of the season (Pro2) for the professional team. Pro1 included 17 players, while Pro2 included 21 players. For the university-level team (hereafter referred to as Amateur), data were collected from seven matches in the second half of the season, with a total of 21 players included in the analysis.

The selection of teams ensured data consistency and allowed for in-depth comparisons of competitive characteristics and tactical features at each level. Focusing on a single team per level minimized variability in tactical approaches and playing styles, enhancing result comparability. All participants provided written informed consent prior to data collection. The study adhered to the principles outlined in the Declaration of Helsinki and was approved by the Ethics Committee of Waseda University (Approval Number: 2023-044). Data were anonymized to ensure confidentiality, and all acceleration data were securely stored to prevent unauthorized access.

Measurement Methods
Each player wore a Global Navigation Satellite System (GNSS) device equipped with a triaxial accelerometer (PlayerTek, Catapult, AUS) on their upper back (bottom of the neck) to record acceleration data during matches. Data collection spanned from immediately before match entry until the end of play or substitution. Recorded data were segmented into 45-minute halves, excluding stoppage time. For substitutions, data were split to include only active periods for each player. For example, if Player A was substituted by Player B in the 25th minute of the second half, Player A’s data for the first half and the first 25 minutes of the second half were analyzed, while Player B’s data from the 25th to the 45th minute of the second half were included.

Network Analysis Methods
Analysis of Dyads and Triads
In directed network analysis, nodes represented players (sources and targets of connections), while edges denoted directed connections. A dyad consisted of two nodes and one edge, whereas a triad involved three nodes with directed edges among them. For example, if four nodes (A, B, C, D) were connected, and an additional edge linked B and D, the network would contain two triads and five dyads.

This analysis followed procedures similar to those described by Tanaka et al. (2021). Dyads and triads were identified by computing movement entropy based on acceleration signals captured in the same spatial and temporal context. All analyses were performed using a directed network analysis system (Hitachi Ltd., Tokyo, Japan).

Calculation of Movement Entropy
Movement entropy quantifies the influence of one player’s movements on another. The calculation involved three steps:

Acceleration data were collected at 10 Hz and converted into movement intensity scores.
Intensity data were normalized using histograms based on Sturges’ formula.
Movement entropy was calculated to evaluate how much one player’s movements predicted another’s, with results ranging from 0 (no influence) to 1 (strong influence).

A basic model was used to simplify the analysis, considering only the previous movements of both players (k=l=1k = l = 1k=l=1). The entropy value represented the degree of uncertainty reduction when predicting Player B’s next movement based on Player A’s current movement. This method provided a clear measure of directional influence between players in the network.

Complexity of Network Structure

In this study, we adopted a method to calculate information entropy based on the occurrence probabilities of Dyads (2 nodes) and Triads (3 nodes) and evaluate the difference from the maximum information entropy. This information entropy serves as a measure of the complexity of the network structure and is defined as "network diversity." Specifically, the complexity is calculated using the ratio of the maximum entropy (Hmax​) when nodes are randomly activated with equal probability to the actual entropy (H) based on observed activation probabilities. A value closer to 1 indicates the presence of fixed cliques (Dyads or Triads), while a value closer to 0 reflects a more even distribution.