Synthetic Methods to Deal with Big Data in Soccer

The FMH Social Networks Project

Ricardo Duarte, Sérgio Tomás and Daniel Baião

CIPER, Faculdade de Motricidade Humana, Universidade de Lisboa, Estrada da Costa, Cruz Quebrada, Lisboa, Portugal

Keywords: Complexity, Synthesis, Context-Dependence, Centrality, Passing Distribution, Soccer.

Abstract: Big data is a controversial issue in performance analysis. Social networks have been adopted as a synthetic

method to uncover the web of interactions captured from long data sets. Here, we describe the research

findings of two studies. First, we investigated the influence of the ball possession characteristics in the

competitive success of Spanish La Liga teams. We found that competitive performance was influenced by

the density and connectivity of the teams, mainly due to the way teams use their possession time to give

intensity to his game. In the second study, we developed and validated a multiple context-dependent social

networks method for applied performance analysis. Face and quantitative content validity were assessed

using panels of subject-matter experts. Sensibility was also measured, suggesting the multiple context-

dependent networks are sensitive enough to capture differences in the way players interact with each other

in different game contexts.

1 INTRODUCTION

Today, ‘big data’ is a controversial issue in

performance analysis. While there is a need to have

sophisticated technologies compiling all the

statistical information from match performance,

some research has highlighted the urgency to find

‘synthetic’ methods to enhance the usability of such

‘big data’ (Lames and McGarry, 2007; Travassos et

al., 2013). Recently, social networks were adopted

as a synthetic method to uncover the structure and

organization of the web of interactions captured

from long data sets, such as the multiple players

passing distribution tendencies (e.g., Duch et al.,

2010). For example, an investigation using this

method revealed that English Premier League teams

characterized by high intensity (higher passing

work-rate) and low centralization (distributed work)

are associated with better team performance (Grund,

2012). Moreover, other authors claimed that social

networks is a promising approach once it allows

integrating multiple statistical information in a

simple visualization mode, with the advantage of

extracting objective individual and team

performance measures (Cotta et al., 2013). Here, we

describe the research findings of two different

studies developed at the Faculdade de Motricidade

Humana, Lisbon-Portugal.

2 BALL POSSESSION

CHARACTERISTICS AND

TEAM PERFORMANCE

In this first study, we investigated the influence of

the ball possession characteristics in the competitive

success of a representative sample of the Spanish La

Liga. Based on OPTA passing distribution raw data

from 380 matches involving all the teams of the

2011/2012 season, we calculated three team

performance measures to assess ball possession

tendencies: graph density, average clustering and

passing intensity.

Findings showed that bottom-ranked Spanish

teams tend to have less number of connected players

and triangulations than intermediate and top-ranked

teams. However, all the teams significantly diverged

in terms of passing intensity. Top-ranked teams were

the teams with higher number of passes per

possession time, followed by intermediate and

bottom-ranked teams.

These findings suggest that competitive

performance was influenced by the density and

connectivity of the teams, mainly due to the way

teams use their possession time to give intensity to

his game.

Duarte R., Tomás S. and Baião D..

Synthetic Methods to Deal with Big Data in Soccer - The FMH Social Networks Project.

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

3 MULTIPLE CONTEXT-

DEPENDENT SOCIAL

NETWORKS

Since the social networks approaches typically

employed in research were not designed to provide

practical performance analysis insights for soccer

practitioners, in the second study we developed and

validated a multiple context-dependent social

networks method for applied performance analysis.

We used OPTA passing distribution raw data

from 32 teams obtained from 16 English Premier

League matches. Beyond the team global network

for the entire match we developed a set of specific

social networks capturing different game contexts:

goalkeeper’s distribution, defenders’ distribution,

midfielders’ distribution, forwards’ distribution,

shooting opportunities path analysis, 1st phase area

distribution, 2nd phase area distribution and creation

and finishing area distribution. Topological graph

visualizations and three individual players’ metrics

(betweenness, closeness and eigenvector) were

obtained from each specific network to objectively

quantify the individual influences in team

performance. This methodology was developed

under the advice of a panel of 5 experts, who assured

its face validity. Feedbacks from the experts were

integrated during the method developmental stage.

In a second step we determined the Content Validity

Ratio (CVR) (Lawshe, 1975) of a panel of 8 subject-

matter experts, who were asked to indicate whether

each context-dependent network is “essential” to

capture the essence of each corresponding game

context.

The CVR data of every context-dependent

network was superior to 0.5, which ensured the

content validity of the methodology. In a third step,

we tested the sensibility of the methodology

comparing the mean and variance of players’ metrics

in each match between the different context-

dependent networks, using repeated measures

ANOVA. The significant differences we found, with

large effect sizes attributed to eigenvector (ɳ

=.570)

and betweenness (ɳ

=.590) measures, suggested the

multiple context-dependent networks are sensitive

enough to capture differences in the way players

interact with each other in different game contexts.

4 CONCLUSIONS

In sum, social networks seem to be an appropriate

synthetic method to deal with large data sets

quantifying the number of connections between

team players. Here, we used passing data as

informational variables linking the network nodes.

However, future research should explore the

usefulness of positional tracked data from multiple

players’ movement trajectories to assess, for

instance, the positional stability of teams.

ACKNOWLEDGEMENTS

The authors wish to acknowledge OPTA statistics by

providing the data sets used in these studies.

REFERENCES

Cotta, C. Mora, A. M. Merelo, J. J. And Meelo-Molina, C.

(2013) A network analysis of the 2010 fifa world cup

champion team play. Journal of System Science and

Complexity. 26. p. 21–42.

Duch, J. Waitzman, J. S. And Amaral, L. A. N. (2010)

Quantifying the performance of individual players in a

team activity, PLoS ONE. 5(6). p. e10937.

Grund, T. (2012) Network structure and team

performance: The case of English Premier League

soccer teams. Social Networks. 34(4). p. 682–690.

Lames, M. And Mcgarry, T. (2007) On the search for

reliable performance indicators in game sports.

International Journal of Performance Analysis in

Sport. 7(1). p. 1-18.

Lawshe, C. H. (1975) A Quantitative Approach To

Content Validity. Personnel Psychology. 28(4). p.

563-575.

Travassos, B. Davids, K. Araújo, D. And Esteves, P. T.

(2013) Performance analysis in team sports: Advances

from an Ecological Dynamics approach. International

Journal of Performance Analysis in Sport. 13(1). p.

83-95.