Is this the death of formations (as we know them)?

Our dark crime thriller opens…

The shadow of Pep Guardiola looms large over the bleeding body of Formations.

He covers the body with the cardigan he discarded earlier, now sodden, storing his bloodied knife in his cargo trousers. It was a methodical kill, but he doesn’t need to disguise it — everyone will know that it was he that cast Formations to the land of the dead.

As Guardiola straightens, the setting sun’s glow frames his head. A halo, or devil’s hellfire?

Yep — formations aren’t what they used to be.

The resounding reaction to Manchester City’s system against Arsenal in their latest meeting was “…what?”. After a few minutes watching, I had a go and described it like this:

After the game, I asked people to write in with what they thought it was. The answers were many and varied, with around nine different formations being suggested. Here’s the tally:

  • Assorted jokes (7)
  • 3-3-4 (4, although including formations that could be variations of a 3-3-4 it was 7)
  • Assorted points about no such thing as formations really (2)
  • (All 1) 2-1-3-4, 3-1-4-2, 4-2-3-1, 4-4-2, 3-3-2-2, 3-1-3-3, WM, 3-diamond-3

I think that consensus was about right. Out of possession (although City didn’t spend much time out of possession), it settled into a 4-4-2, but when they were on the ball it seems that ‘3-3-4’ would be the best way of describing Guardiola’s set-up.


And, for formality’s sake, this is the out of possession ‘4-4-2’:


This system — a pretty clear example of ‘in possession’ and ‘out of possession’ formations in use — raises interesting questions about formations. One is about which of these two ‘formations’ would we use to describe City; another is about what a ‘formation’ even is.

I should stress, these are not new questions. In fact, I was thinking of writing about this subject anyway, days before Guardiola shoved it into the spotlight. The day before the City-Arsenal match, I’d asked on Twitter whether people thought that the formation notation we use is closer to describing a team’s shape when defending or when attacking.

The results (as they don’t seem to embed in the newsletter) were:

  • Just show the results (voting in a poll shows the results before it closes): 9.6%
  • Defensive shape: 77.3%
  • Attacking shape: 13.1%

When people responded with theories about why the answers were leaning this way, the trend was that defensive shapes are more rigid whereas attacking ones are more fluid. That makes sense.

However, when asking people for City’s formation they settled on the 3-3-4, the attacking one*. How does that fit in?

*Some people gave multiple formations and I’ll be coming to that in a second…

Well, you could refine ‘people use the defensive shape for formations as that’s more rigid’ and refine it slightly. A tweaked version could be that people name the formation based on what pattern they see the most.

For example, if a game is fairly close, then people will notice the defensive pattern more (because defensive shapes are more rigid and regular). But with City, in games they dominate large parts of, it’s the attacking patterns they latch onto more.

An advantage in theoretical terms, but a disadvantage in practical ones, is that this means cultural baggage can have an influence. A lot of people mentioned to me that 4-4-2s are often 4-4-1-1s in practice. It feels like the reason why people still call them 4-4-2s would be cultural.

This theory that it’s about ‘what patterns do people see most’ would also explain the divide between calling a team a 3-5-2 or a 5-3-2. Teams expected to do a lot of defending will get the 5-3-2/5-4-1 treatment; teams who are higher up the league get the honour of being 3-5-2/3-4-3.

All of this being said, the Man City game makes me feel dissatisfied with the ‘formations are the pattern people see most’ explanation.

I think a more accurate description would be that the formation notation — the 4-3-3 or the 3-3-4 etc — is about giving us a starting point to know what roles everyone will be playing. This requires a little bit of football knowledge, but it’s stuff we all pick up. We might use 4-4-2 instead of 4-4-1-1 not because we’re unaware that one forward drops while the other stays high, but because we know that; using 4-4-2 signals that either forward could be dropping, whereas 4-4-1-1 would indicate a solid ‘attacking midfield’ role for one of the players.

Of course, none of this really matters. Reducing things down to a single formation is a little nonsensical, as Manchester City showed. Within actual clubs it would make more sense to talk more specifically — about in possession vs out of possession shape; about individual roles within a system. It’s kind of just for us normal fans that the ‘number-number-number’ formation notation makes sense.

For articles like this in your inbox, subscribe:

Thanks to tracking data — where the position of every player (and the ball) is collected multiple times per second — the world of stats is catching onto this.

At its most basic, we have what the Bundesliga offered during Project Restart. A few minutes into the match, they’d show a little graphic: initially it’d be the team laid out in their 3-4-3/4-4-2 ‘shape’; then the players would be moved to their average position in the match so far.

Here’s a video here showing it in action:

Outside of TV broadcasts, researchers and data companies have also been hard at work. There’s a presentation here of Laurie Shaw and Mark Glickman’s work and Ben Mackriell at StatsPerform vaguely alluded to something here. I won’t break down the research, but there’s a really good article summarising Shaw and Glickman’s work that here.

The upshot of it, though, is that instead of relying on someone to say “I think this team is playing in this formation”, you can get the actual shape a team is playing in. Shaw and Glickman’s work with tracking data also differentiates between attacking and defensive shapes, which (to my knowledge) data providers don’t currently do. [Of course, at clubs you could ‘collect’ this data yourself for your own games, and could get analysts to do it for the rest of the league]

Edit: It’s been pointed out to me that this work has been done since tracking data was first collected in the 90s. Another paper on the subject, from 2015, is here. I’ll continue to update further things noted to me here.

Separating attacking from defensive formations would open a small realm of analysis. You can — as the researchers do in their poster for the Barcelona Analytics Conference — look at which offensive formations happen alongside which defensive formations. Perhaps a team regularly look like a 4-2-3-1 in offence, but switch their defensive shape around based on personnel, for example.

With the groundwork for ‘formations as told through tracking data’ already now done, we can speculate on what could come next. Perhaps researchers will be able to create algorithms to detect the main rotations that teams use. That could be immensely useful for defences, like being able to know who they should pick up if a striker drops.

And because of the way football’s going, that information is only going to become more important. It’s not just Guardiola either — it’s Marcelo Bielsa at Leeds, Chris Wilder at Sheffield United, and that’s just covering Yorkshire.

In five, ten years time we might talk about formations in a totally different way. We might talk about in- and out-of-possession formations as standard; we might wait for the tracking data to give its verdict for us. It’s all starting now though.