Beyond formations: Using shape to identify inverted full-backs
Written by Jon Manuel (Data Analyst, Stats Perform) — December 11, 2023
THERE'S good reason why the concept of a formation has become so popular in football: it gives a simple overview of the players involved, where they play and the team’s general shape.
However, formations are quite limited in terms of what they do. Are teams really playing a single static shape throughout an entire game? Are players really just stuck to a single position? And what about the difference in shape when teams are in and out of possession?
Arsenal manager Mikel Arteta alluded to this in September, when he said: "I think we discuss formations in a different way. The other day there were 36 different formations in the match [against Fulham]. Against Manchester City, 43, so I don't know what formation we're talking about. For me it's something very different maybe from the ways that you look."
This article is taken from Jon Manuel's presentation at TGG's Big Data 2023 Webinar on November 21st 2023. In total there were 8 presentations:
- Giles Lindsay: Squad Building In The IPL.
- Lee Mooney: Managing Decision Risk In Football.
- Jon Manuel: Using Shape To Identify Inverted Full-Backs.
- Tom Goodall: Developing Data-Driven High-Performance Workflows.
- George Gallagher: Structuring A Data Analytics Department.
- Sarah Rudd: Supervised v Unsupervised Learning With Tracking Data.
- Issa Tall: How Columbus Crew Use Analytics To Sign Players.
- Johann Windt: Aiding Decision-Making Across Performance Domains With Analytics.
To find out more and to purchase the webinar, click below.
There is clearly a need to move beyond this basic picture. In this presentation, I’m going to introduce some new outputs we’ve got at Stats Perform, which enable us to move beyond traditional formation descriptions to better identify player roles and team shapes.
Using Opta data, we have developed a new model called Shape Analysis. This captures player roles and then assigns the shape it thinks best matches them.
In total, we have identified 37 different in-possession shapes across elite professional football. A nice example is the 4-3-3 - we have seven different variants, with a false nine, narrow wingers and advanced full-backs. Then we have the equivalent out-of-possession shapes - 29 in total. These tend to be much narrower and more compact.
Below, you can see the most common in-possession shape being used by each team in the Premier League this season. You can see that Arsenal have played this 4-3-3 with a false nine for just over half of their time played in possession.
Twelve of the 20 teams use the 4-3-3 as their most common shape. Arsenal, Brighton, Fulham, Liverpool, Newcastle and Tottenham like to stick to a particular shape for a lot of their time in possession, more than 50%.
At the other end of the scale, Brentford, Luton, Manchester City, Manchester United, Nottingham Forest and Sheffield United are among the teams that vary their shape quite a lot in possession.
Then we have the equivalent for out of possession, looking at the teams' most popular shape. Here, we can see the popularity of this 4-2-2-2 shape, which nine of the 20 teams use as their most common out-of-possession shape.
ARSENAL AND ZINCHENKO
These are Arsenal’s three most common in-possession shapes (below).
Within this, I wanted to quantify and investigate Oleksandr Zinchenko’s role. Using the basic formation description, Arsenal play 4-3-3 and Zinchenko is a left back. However, there is clearly more going on than this.
Using event data, we open up a lot more detail. His touches are focused down the left-hand side and he’s taking up a lot of advanced positions. Using our shape analysis model (below), we open up even more and can quantify where he is really playing.
This shows us where he plays in Arsenal’s most common shapes. I split each game into five-minute periods and for each of those periods I utilised shape analysis to capture the player roles and the in and out-of-possession shapes. Doing this gives a better granularity and a better idea of how teams and players actually change shape within a match.
In the 4-3-3, which Arsenal play 50.4% of the time, Zinchenko is playing 80% of his minutes in the left-back role. As soon as we look at the three at the back formations, this completely changes, and Zinchenko plays the majority of his minutes in the deeper-lying left central midfield role.
Arsenal use this 3-2-4-1 in more of a build-up phase. When they win the ball back, Zinchenko steps into midfield and helps his team progress up the pitch. In the 3-4-3, Arsenal are in more of an attacking phase and Zinchenko supports the attack down that left-hand side.
Another nice example is Martin Odegaard this season. The top graphic below shows where he plays in Arsenal’s most common in-possession shapes.
In the 3-4-3 and 3-2-4-1 he is quite set on that right attacking midfield role, basically playing all his minutes there. But in the 3-4-3. his role is a lot more split across roles on that right-hand side.
In the bottom graphic, we can see the proportion of Arsenal time that Odegaard plays in each of these roles. Whenever Arsenal play the 4-3-3 shape, Odegaard is in this right attacking midfield role 66% of the rime. The rest of the 34% is taken up by other players.
In the 3-4-3, he is only playing about 50% of the minutes there, which suggests this is a shape Arsenal use without Odegaard, maybe later in matches when he is subbed off or not available.
OTHER INVERTING FULL BACKS
Now I want to focus on inverting full backs. These are players who split their time between central midfield and full-back. In this analysis, a full back is defined as any left or right full-back as well as wing-backs.
Central midfield positions are any deeper lying roles in the middle. In particular, we wanted to look for players who split their time between full back and central midfield, fluidly moving between the roles within individual matches.
I added a couple more filters: we wanted players who play more as a full-back than a central midfielder, rather than the other way round; and players who have spent at least 50% of their time playing in these positions, not those who have done it for one game.
When we apply these filters this season, these are the top 10 players (below). To give some context, 50% would mean a player splitting their minutes equally between full back and centre back, whereas 100% would mean them playing all their minutes in one of the positions.
Three of the top four are Spurs players, suggesting this is a team-wide tactic. Emerson Royal is not far off splitting his time equally between full back and central midfield. Zinchenko is second.
I also took the same data to the Bundesliga this season. Junior Ebimbe is at the top here. He is playing pretty much all his full-back minutes as a wing back and in advanced roles. Calling him a full back is a bit of a stretch though, because he is perhaps more of a wide midfielder.
Pretty much all his inverted minutes are coming from the 3-1-4-2 formation, where he is playing wing back and coming inside to play central midfield.
A nice example of the inverting full back is Danilo Soares, who has played almost 90% of his minutes across wing-back and full-back roles this season for Bochum. When we look at his most common positions played in the various shapes, we can see that in 4-2-2-2 he is spending 70% of his minutes as a left back and 30% in these midfield roles.
It’s a different look to Zinchenko, because Soares is inverting within the same shapes. In the 4-2-2-2 he is playing both left back and central midfield. In the 4–4-2 diamond he is playing full back and coming inside within the same shape.
This shape analysis can easily be extended for any roles you want to look at, for example central midfielders who drop back into centre back roles during their team’s build up, or advanced forwards who play across the whole front line.
These are other avenues you could explore with the outputs of this model:
- General player analysis: Looking at a player’s most common roles in and out of possession. Then asking whether they can they play a wide variety of roles, and linking their output and styles to these particular roles.
- Team level: Looking at shapes that teams play in and out of possession. Then whether they use certain players with particular shapes and certain shapes in particular scenarios. How fluid teams are: Do their players stick to the same roles, or do they switch between them within their shapes?
- Recruitment: You could extend this to identify any players who play specific roles or shapes. And then ask questions about their versatility, so can they play a variety of roles.