By Ella von Baeyer '24
Introduction
In this investigation I wanted to look at whether a history of soccer players playing together, and the subsequent interconnectedness of a team, impacted their match outcome. Like any other emergent network, I hypothesised that the coordination between players meant that their interaction results in an output greater than the sum of their skill.
As my case study, I chose the 2014 World Cup. Germany won the 2014 men’s soccer World Cup to the surprise of many pundits (it was third favourite to win - after Brazil and Argentina). In deciphering how this had happened, some fell to the conclusion that it was Germany’s mandate for Bundesliga clubs to operate training academies. Indeed, many of the players who made the German National Team came through these academies and subsequently have a shared history of playing together. Therefore, I thought this would be an ideal case study to assess whether Germany really managed to gain a winning advantage from the shared playing history of their team. To do this, I compared the team histories of the countries that made it to the quarter final, and looked at whether there was a correlation between winning and their network structure.
Methodology and Data Collection
To collect the data I first found each national team chosen for the 2014 World Cup that made it to the quarter finals. I then went through each player’s club history, by year, from about 1990 (the earliest that any player started playing for a professional club) to 2024. With these data sets, arranged by country, I then created and ran an algorithm that counted how many years players overlapped with each other. With this information, I then made an edgelist per country, weighted by the number of years the players played together. The node attributes for my graph are the positions of the players, and so this became another table for each country. In order to compare teams, I looked at their goal differences in the matches they played between each other during the World Cup.
Because goal difference is a comparative score, I could not simply plot goals conceded or won against edge density. Instead, I used the relative edge density (see example above). I plotted this in figures 2 and 3.
Using this information, I plotted the network graphs for each of the eight countries, with their edge weights corresponding to the years played together on club teams before the 2014 World Cup. The nodes are coloured corresponding to the player’s position.
By taking the matches that were played, and comparing the goal difference between the two teams with their comparative density, I found that there is a statistically significant positive correlation between comparative density and goal difference: 0.78 to 2s.f. (at a confidence interval of 0.95). This means that the more comparatively dense the team is, the more likely they are to win by a bigger goal difference.
There is also a statistically significant positive correlation between the comparative total of weighted edges and the goal difference. However, this is a slightly lower positive correlation than for the edge density, at 0.65 to 2s.f. (at a confidence interval of 0.95). This would suggest that having any tie matters more than the strength of the tie between two nodes. It is better for a team to have had lots of players play together, even if only for a year than to focus on the duration of a relationship between two players, if the ultimate aim is maximising goal difference.
In order to see how this advantage of an interconnected team might work, I wanted to look at equivalence on a pitch. In a game, there are sequences of play in order to get the ball up the pitch. Therefore, I considered whether there would equivalence based on the position of the players on the pitch, such that those who had played together before would also be passing the ball to each other: i.e. a goalkeeper to a defender, to a midfielder, to a forward. I used Concor Modelling to assess whether there was structural equivalence between the nodes based on the players’ positions. For all countries I found there not to be structural equivalence. In fact, quite the opposite. Most positions were split evenly between 1 and -1, implying a degree of homophily between the nodes on the basis of position. This can be seen in figure 4.
Given that there is no equivalence between players’ positions, it would appear that playing together as a whole team is more important than the individual relationships players have with each other.
Conclusion
To answer the question that prompted this investigation, it seems that Germany’s culture of youth academies did influence its win in the 2014 World Cup as it meant that the team was very interconnected from a history of playing together. And, as we have seen, interconnectedness is linked to a higher winning margin.
Nonetheless, whilst Germany’s culture of youth academies is therefore likely a key reason for its ability to produce great soccer players, I wondered whether this was just really symptomatic of a country’s wealth and their ability to put more money towards sports like soccer; the history of playing together did not help the team, so much as it indicated a country’s ability to have a strong home league and the advantages of great training facilities and coaches that come with that lots of funding for a sport.
In order to see whether interconnectedness could really just be used as a proxy for wealth, I ran a correlation test between these two factors. Had there been a strong positive correlation between wealth and interconnectedness, it would more likely have been true that the interconnectedness of a team was really more a reflection of how much money a country already had to fund towards their soccer program, thus giving them a winning advantage. However, given that there was not a 1:1 correlation, or any statistically significant correlation at all, it seems that the simple act of having played together before does give you an advantage. Costa Rica, for example, the poorest country by GDP per capita in 2014 in the quarter finals, has a strong domestic league, Primera División, from which it sourced many of its players. This does not mean that the country itself can put much funding into sport though, since a strong domestic league is not a signifier of how much money a country funds towards its national team. Therefore, a history of playing together seems like the more plausible reason for their successes.
Limitations and Further Considerations
A limitation of this method is that not all players are on the pitch all the time. Therefore, there may be configurations where a team was actually more interconnected than the other, based on which players were benched and which were playing. I ultimately decided to focus on the whole 23 man squad, as it would have been hard to attribute a weight to a player who subbed on at a certain point or subbed off. However, the issue of subbing on and off, does complicate the matter of equivalence.
When accounting for equivalence, whilst there may not any notable equivalence in these current models, this does not necessarily mean that the players selected on the pitch lack equivalence. For example, it might have been that the configuration of players on the pitch at a certain point actually had equivalence. Indeed, the manager might specifically choose player combinations for a game that have played together before to give them an advantage.
In a more sophisticated model, I would create a network that changes over time, to reflect the players who are substituted on and off. The density and total weighted edge score would then change, as would the equivalences of the players’ connections based on position. The goal difference would then also be measured by every time period between a player being substituted off, to account for the many unique team configurations during a single match. This would ultimately give more data points, and therefore more reliable results.
Lastly, my investigation rests on the assumption that playing together for many years creates a good relationship. Indeed, there is evidence to support that a history of working together can amplify a talented team (though it does not eliminate the need for talent). Nonetheless, these working relationships need to be positive ones. A study from Northwestern showed that a history of shared wins was a significant factor in performing well as a team: teams that had a shared history of wins had an advantage over teams that had no history together. However, for teams that had a shared history of losses, there was an adverse effect on team performance. Applying this to these national teams, it cannot always be assumed that a history of working together will lead to a better future relationship.
Works Cited
Brian Uzzi Richard L. Thomas Professor of Leadership and Organizational Change; Co- Director, and Noshir ContractorJane S. & William J. White Professor of Behavioral Sciences. “For Teams, What Matters More: Raw Talent or a History of Success Together?” Kellogg Insight, https://insight.kellogg.northwestern.edu/article/talent- versus-teamwork-for-successful-teams.
“Development of German Youth Academies: DFL Deutsche Fußball Liga.” EN - DFL Deutsche Fußball Liga GmbH, 15 July 2019, https://www.dfl.de/en/topics/youth- academies/a-brief-history-of-german-youth-academies/.
Keeney, Tim. “World Cup Odds 2014: Updated Chances for Each Country Following Draw.” Bleacher Report, Bleacher Report, 14 Sept. 2017, https:// bleacherreport.com/articles/1878805-world-cup-odds-2014-updated-chances-for- each-country-following-draw.
“World Cup 2014 Squads.” The Guardian, Guardian News and Media, 6 June 2014, https://www.theguardian.com/football/2014/jun/06/world-cup-squads-2014.