
Passing Networks
What are Passing Networks?
Without becoming too scientific, it is a graph which consists of dots and lines connected together. Each dot (known as nodes) are joined together through lines (also known as connections).

These types of graphs are commonly found in the science industries such as life science, cybersecurity, or intelligence and can be used to map out connections between chemical components, DNA or even to see how your Facebook friends are connected. Passing network visuals have been around football for two/three years thanks to companies such as StatsPerform and Statsbomb, but to our knowledge are new to South African football.
How might they be useful in football?
These types of graphs are especially useful for opposition analysis which allow coaches and analysts to find and break valuable connections between players on a team. In addition, some players may be in possession of the ball more than others, so an aggressive press on those players may be an option to regain possession. The most ideal method would be too look at a club’s recent games and determine which players are 1) more in possession of the ball and 2) which relationships occur most often and why.
In any given week, the video analyst of a club already has to go through numerous hours of footage to help prepare the team against the next opposition. With the help of these passing networks, it can easily give an overview of the opposition and help some time sieving through at least 10 hours of match footage.
The Visualisation
To begin with, we decided to design the pitch visualisation to go horizontal from left to right. For most people, we read a book, a screen and most other information left to right, so this was a no brainer. We might change the design to vertical in the future, who knows. As we mentioned above, too much information makes the visual un-readable, un-useable and thus a waste of everyone’s time. To keep it simple, here are the main design elements you have to remember.
● Visual contains only accurate passes
● Size of dot/node: Number of touches of the ball by a player
● Thickness of line and colour: Number of passes between nodes (players)
● Passing connections are limited to at least x number of passes between players. This helps with design and readability and signals that whilst there may be connections between players, it did not occur as frequently as others.


Limitations
Every graph that is produced in football to show some form of data has its limitations and passing networks are no different. We would encourage teams to use them together with other details in order to get the full picture. For example, with video for opposition analysis. First though, let’s go over some of their limitations.
Nodes
As mentioned above, each player is represented by a dot on the pitch. If a player were to switch sides, especially wingers, then their average position will end up being central. They may have never touched the ball in that area of the field, but their average position will be just that, the average. It is a very common flaw mentioned among coaches and that is why analysis never (or should never) give a final answer. It should always bring up a follow-on question as to ‘why’ is something as it is. That is why, we also produce individual ‘touch’ heatmaps to show where a player is control of the ball during the course of a game.
Connection lines
The passing connections (lines) between players is not exactly as shown. For example, the full-backs did not pass to the central defenders along that specific line x number of times. It is rather just a measure of this relationship happened x number of times and may or may not be as important as other connections. For that, we have separate individual passing maps (see below).


Accurate passes
Passing networks only consider ‘accurate’ or ‘completed’ passes for the simple fact that we know these were received by a team-mate and thus we have completed data for a ‘passer’ and a ‘receiver’. Unfortunately, detail is lost in the fact that if a pass was destined for a team-mate but was intercepted, that pass is not counted in this visualisation. A single player pass map will show this level of detail instead (see visuals from previous limitation).
Substitutions
Passing networks can get very busy and lose their effect if we include all players in the visualisation. Generally, only passing connections until the 1st substitution have been made are shown. After this event, shape and formation could change and thus distort the picture of events. Whilst this may limit insight after a substitution, the graph is only intended to give an overview of a game to help the overall analysis process. Data will and should never replace actual match footage as there is too much that is not currently measurable (at least at present) such as body language and positioning. The devil is in the detail and this is intended to be a mere synopsis of a match. You would not merely watch 15 minutes of a movie and conclude you know the storyline and or ending.
Application
Considering your opposition analysis process, take x number of games and use the networks and tables of the opposition and start looking for patterns of play. You may even decide to go further and divide them into home vs away games.
● Which player is on the ball the most? How dangerous is he? Where and to whom does he pass to?
● Are some player connections more valuable than others?
● Do they play short or long passes?
● What is their formation in possession?
● How high do the team play? Is one side imbalanced over the other?
● Is there space in behind for us to exploit?
In addition, for simplicity, we have also included a table to show the actual figures of passer to receiver. No filters were applied here and therefore the visual includes all players who played/touched the ball as well as all accurate passes made by the player in question.


Conclusion & Future Work
We believe passing networks can help in the analysis process that PSL clubs already have in place. Once you understand the limitations of the visualisation, it can become a vital addition in the process of opposition analysis. In future, we hope to look at incorporating a passing model into the visualisations. This would allow us to evaluate any given pass and how much ‘danger/threat’ a player’s possessions has in comparison to his team-mates.