Euro 2022 Review, Part One: Visualizations 101
I didn't plan for this to be a series, but it was getting too long so here we are
There are many aspects of the modern world that the aristocrats of Victorian England who developed the sport of football would find jarring, but perhaps none more so than the notion that passing is one of the sport’s most important elements. We of the enlightened modern age, however, are far removed from those dark times when the big man tried to dribble the length of the pitch, never turning his head from the goal in front of him. Instead, we’ve grown up watching teams that actually value the ability of a player to kick the ball to a teammate.
Football’s pioneers may have looked askance at passing, but the other subject of this series, women playing football, would have been a little more familiar (albeit unwelcome) to them – at least until the FA banned it in 1921. Nowadays, however, both of those elements are more accepted in the football world, though we should acknowledge that there is still a long way to go on the second point.
What I’m going to do here is look at Euro 2022 – which wrapped up a few weeks ago – through the lens of passing. My aim is to try and tell the stories of individual teams (probably four or five of them) in this tournament using their passing patterns, visualized in some neat graphics that are popular in football analytics. I’ll be releasing an individual piece for each team over the next few weeks (hopefully that’s the timeline, but life comes at you fast), and in this introductory article I’m going to explain the visualizations so you’re ready to go when we get to the good stuff.
All the data I’m working with in these pieces comes from StatsBomb, which provides free event data – information about every on-ball action from a match – from a number of tournaments and competitions for public use, including World Cups and Euros.
The first type of graphic I’m using is a pass network. In this visualization, we lay out a team’s players on the field in the average location from which they each played their passes in a match. Then, we connect them with lines that indicate how often each player passed to each teammate. This can give us an idea of the team’s shape in possession and can also provide tactical insight.
Here’s an example network from Sweden’s match against Switzerland where you can clearly see their shape and how it aligns with their nominal 4-2-3-1 formation (I say nominal because formations are a restrictive construct that often limit us from talking about player roles):
Brighter lines with greater width indicate more passes between players, while the size of each player’s “node” indicates the total number of passes they made relative to the rest of the team (the numbers are their kit numbers).
There are a number of additions or variations that you may see in other pass networks. Some color the nodes with a gradient based on some sort of possession value model. Others split the pass connections into two separate arrows between players. You may also see more simplified networks that filter for a certain number of passes, thereby emphasizing the higher-volume connections. These features can enhance the story that the network is telling, but I haven’t gotten around to implementing these in my own code yet.
Of course, pass networks have their limitations. Consider this one, showing England’s passing from the opening game of the tournament against Austria:
I want to focus on Beth Mead (No. 7) and Lauren Hemp (No. 11), whose average locations are fairly central despite being the wide attackers in this setup. Sometimes, this can be an indicator of a tactical plan – maybe the wingers come narrow and combine while the fullbacks push up the wings. In this match, however, Mead and Hemp actually swapped sides partway through the first half. The passes played on the left side versus the right side ended up averaging to a location that’s more central than their passes on each individual side would indicate.
So, pass networks in this form can’t really account for changes in team shape or formation. Attempting to display separate networks for individual periods of the game in which player roles stay constant could address that, but we’d then run into the issue of using fewer passes to build the network –often not enough to build a coherent narrative.
The sample size issue can crop up in another way: substitutions, which often affect team shape as managers change systems and player roles shift. Since we’re trying to use these pass networks to understand team shape and tactical plans, including substitutes and their pass connections obfuscates the graphic more than it adds information. Unfortunately, this means that we can sometimes end up with networks such as this one:
The Dutch goalkeeper, Sari van Veenendaal (No. 1), had to be substituted due to injury after 21 minutes; she doesn’t appear on the graphic at all because she didn’t complete a pass from open play in that time. The network, therefore, only covers that early period of the match, and the smaller number of passes that each player completed limits the conclusions we can draw from this graphic.
Because Van Veenendaal’s involvement here was so limited, I could probably build a network using data up to the next substitution without disrupting the network – I’d basically be swapping her out for her replacement, Daphne van Domselaar (No. 16), like Indiana Jones in Raiders of the Lost Ark (hopefully with a little more success). This wouldn’t work, however, in the network below – Marie-Antoinette Katoto (No. 9) did connect a couple passes before her injury.
Her position is extremely skewed because of the low number of passes – I promise you guys that she really does play striker and isn’t some weird combination of sweeper and right back. The other players also look a little out of place – again, small samples – but it’s more extreme for Katoto because strikers play fewer passes than other outfield players, as a general rule.
While pass networks can provide a lot of overall team information, the second type of graphic I’m working with, called a pass sonar, focuses on individual players. This diagram groups a player’s passes by the direction in which they were played, with each directional slice colored according to the average length of all their passes in that direction. Pass sonars can speak to what a player actually did in a match, throwing some light on the role they played. Here’s an example showing Giulia Gwinn’s passes for Germany in the final against England:
Because sonars focus on one player at a time, they can cover a player’s entire time on the field, extending beyond the scope of pass networks. Like pass networks, though, they can be limited by small numbers of passes – I’m not going to bother showing you a sonar for a player that came on as a substitute and only played one pass in the four minutes she played.
I do think that these pass sonars demand some caution when we attempt to talk about player effectiveness and roles because they don’t include passes that weren’t completed. Passes that don’t reach their target can often tell us as much about a player’s performance as those that do. For example, creative players are often tasked with playing riskier passes than some of their teammates; seeing that a player had a large number of incomplete forward passes in a match could indicate that they were tasked with a creative role. On the flip side, it could be because they just had a bad performance, but that would be insightful in and of itself. Compare the sonar above with this one, which adds Gwinn’s incomplete passes:
This looks pretty different and tells a somewhat different story from the original sonar. One idea I want to play around with in the future is superimposing a sonar of completed passes over one that includes incomplete passes; maybe that will add some useful context.
I didn’t include incomplete passes for the pass networks either; in both graphics, I also chose to use passes only from open play. I think passes from set pieces and other irregular phases of play can distort the graphics, obscuring the in-possession dynamics and tactical plans of teams and players.
If I feel that it’s necessary, I may occasionally throw a pass map in there to show an additional level of detail. These maps show each individual pass on the field; I think they’re pretty self-explanatory, so I haven’t bothered including one here. I want to limit my use of them because a lot of the same information is already presented in the networks and sonars, but every so often I may need to use one to demonstrate some point.
Hopefully this exposition has helped you understand more about the pass networks and pass sonars that you’ll see in this series, including how to read them and some of the limits on what they can tell you. I guess I’ll leave it there, and I’ll get to work writing about the first team in the series. I’m not going to tell you who it is, though. That’s definitely not because I don’t yet know where I want to start, and it’s definitely because I’m having some fun at your expense and keeping it secret. I’m devious that way.