Article by Ben Griffis

I’ve been reading a bit on “flow motifs” recently, which were introduced in 2014 by Laszlo Gyarmati, Haewoon Kwak, and Pablo Gonzalez and expanded upon by articles like Håland et al. (2020) (which I based this work off of), Peña & Navarro (2015), and Bekkers & Dabadghao (2019) to name a few. I’ve finally cracked the code (literally) on how to make these myself, so want to introduce them to those who haven’t read about them yet and show some interesting visuals of team passing styles for not only UEFA Top 5 leagues, but also a larger sample of 18 global leagues.

Flow Motifs

Passing flow motifs are the orders of players involved in a sequence of three passes. In a passing chain with more than three passes, you will get multiple flow motifs. The method only looks at unique players (or “nodes”), and uses A, B, C, and D to represent the unique players. This loses information about what specific players are involved, but it allows us to get at a team’s specific style of passing.

Overall, five motifs are possible: ABAB, ABAC, ABCA, ABCB, and ABCD.

An ABCD motif would mean player A (always the start of the motif) passes to player B (the second unique player in the motif), player B then passes to player C (the third unique player), and finally player C passes to player D (the fourth unique player). Again, each motif is capturing the order of passing in a three-pass sequence.

Thus, ABAB is something you see when, for example, center backs pass between each other. Dier to Romero, Romero to Dier, and Dier back to Romero.

ABAC, ABCA, and ABCB involve three players, and as noted in Gyarmati et al. (2014), are very much characteristic of the Pep Guardiola/Barcelona-style passing (see the box plots on the 2nd & 3rd pages of their article). Conversely, ABCD is a much more direct style of play, and Håland et al. (2020) find that the worse teams in the 2017 Eliteserien typically use this motif much more frequently than other teams.

My Method

I followed the methods laid out in the papers I’ve already mentioned here, stemming from the original Gyarmati et al. (2014) method. I have a lot of full-season leagues with data from Opta. Overall I’ve calculated motifs for the Top 5 UEFA leagues (Serie A, Premier League, La Liga, Bundesliga, and Ligue 1), as well as top divisions in Poland, Slovakia, Australia, Denmark, South Africa, Thailand, Saudi Arabia, the UAE, Greece, and second divisions in Germany, Denmark, and Turkey. Just to get a large sample from leagues and teams across the world.

Since motifs are chains of three completed passes, I note all instances of three completions amongst the same team and then turn the players involved into our five motifs. This, as noted above, takes out the info of which players were involved, which is an avenue I want to explore in the future. Beyond overall team style like this article, questions like are some players more involved in a specific motif than others, could be interesting. Peña & Navarro (2015) and Bekkers & Dabadghao (2019) go into some individual player analysis if you’re curious.

Finally, after noting each motif in the dataset, I calculate the internal percentages of each of the five motifs for all teams in the league.

The image below are the results from Håland et al (2020). They show the internal percentages of each motif for the 2017 Eliteserien, with the three largest percentages in bold. the “Obs” column is the number of total motifs, and the “SpG” column is the shots per game the team took.

Compare those results to my results for the 22/23 Premier League, shown below. It’s best to compare my results to a published piece to ensure that my calculations are sound, and that the relative percentages (<5% for ABAB, ~16% for ABAC, ~10% for ABCA, ~16% for ABCB, and >50% for ABCD) are somewhat similar. In a 20-team league, I shouldn’t see massively different results, although some differences can happen given the leagues will have different styles overall.

I didn’t bold the top 3 percentages for each motif, nor did I add the total motifs or shots per game, but I hope you can see that overall, the internal percentage results are very similar. Thus, I’m confident that my calculations are sound and that I can continue.

Results of the Top 5 UEFA Leagues in 22/23

Now I’ll combine all of the top 5 leagues and share some results. This ends up giving us a sample of 98 teams for the 22/23 season. I’ve turned each percentage into z-scores relative to the other percentages in the motif to allow us to see any abnormal teams, either in a single motif or across several.

Below is an image with all teams in the T5 this past season, with their internal percentages converted to z-scores (i.e. their place on the distribution of that motif). Here is an interactive Tableau dashboard of this data.

Christian Streich’s Freiburg immediately stand out here. They are abnormally high for both ABAC and ABCB, while being very low for ABCD. They appear to be the most unique team in the UEFA top 5 leagues in terms of passing sequences.

Other interesting things when digging into the motifs are similarities between teams. For example, here are Pep Guardiola’s Manchester City and Mikel Arteta’s Arsenal. Arteta is very much a Pep Guardiola student, and that shows up in their similar flow motifs as well.

This season, Roberto De Zerbi’s Brighton really stamped their brand into people’s minds. Looking at their motif z-scores, there are two other teams who come out as relatively similar: Erik Ten Hag’s Manchester United and Imanol Alguacil’s Real Sociedad.

There are many more similarities we can make, but I’ll only show one more: Barcelona and Bayern Munich exhibited pretty similar z-scores in most motifs this season. Compared to Gyarmati et al. (2014), it appears that Barcelona have lost their abnormal nature in most of these motifs. Instead, it is Freiburg whose z-scores more closely represent Barcelona of 2012/13.

Expansion: 18 Global Leagues

Overall, I have collected data on 18 leagues from around the world. In addition to the UEFA T5 leagues, I have:

• 2. Bundesliga
• Allsvenskan (as of July 10th, 2023)
• Brasileirão (as of July 10th, 2023)
• Danish Superliga
• Ekstraklasa
• Eliteserien (as of July 10th, 2023)
• Saudi Pro League
• Slovak Super Liga
• South Africa PSL
• Super League Greece
• Thai League 1
• Turkish 1. Lig
• UAE Pro League

This represents a large sample (305 teams) of many different styles of leagues. Below is the full visual with z-scores for all teams. It is too large to use badges or to see most teams, so please follow this link to the interactive Tableau dashboard.

There are, however, two teams I’m very curious to see. Fernando Diniz’s Fluminense in the Brasileirão and Henrik Rydström’s Malmö in Allsvenskan. These are two of the exemplary “relationist” teams in the world (for more info on this, see Jamie Hamilton‘s growing corpus). I’m curious to see a) if they exhibit similar motif z-scores, and b) what Diniz’s relationism looks like in these motif distributions.

Perhaps initially surprising, Fluminense and Malmö are not all that similar. Malmö actually come out as having the lowest z-score for ABCD motifs and the highest z-score for ABCB motifs. Where Malmö, except for the ABCA motif, tend to be near the extremes, Fluminense are relatively less extreme.

Although, the key to relationism is having players fluidly rotate, move, and link up in non-rehearsed, non-patterned ways. So these clubs being more different than, say, Brighton and Manchester United is probably telling that there is a difference in the passing sequences of “relationist” vs “positionist” teams. And further, there’s not one “pattern” that might define this style. But I digress, as this was a curiosity of mine and not a central component of this article. This could be an avenue for further work.

League Motifs: Do Leagues Have Unique Styles?

With all this data, the final thing I want to see is if leagues differ in their average z-scores for these motifs. We should assume that leagues will have relatively distinct styles compared to other leagues, so it will be interesting to see which leagues are abnormally high or low for different motifs.

Norway’s Eliteserien immediately jumps out. Freiburg appear to be in the wrong league! Eliteserien is, by a decent margin, the most abnormal league for ABCD and ABAB motifs. In fact, they have the highest z-score for ABAB, ABAC, ABCA, and ABCB motifs, and the lowest for ABCD. Now that is interesting. You can play around with the interactive Tableau dashboard here.

Conclusion

Pass flow motifs are a good way to begin looking into a team’s passing style and perhaps how they may play in possession. We can find teams who may play similarly in possession, as well as compare different managers or teams to see if there are any unique “signatures” of a system. Further, we can see if any leagues might exhibit interesting or unique passing characteristics.

I’m not going to perform a lot of other tests like some of the academic papers I mention here at this time, but it’s an avenue for more work. As is performing something similar for individual players to see if they are involved in any motifs more or less frequently than others, and perhaps their positions in the motif (such as, is a player usually the B in an ABCB motif? The C? And more).

Extra Fun: Percentile Radars for Selected T5 Teams

More of an appendix, here are a few percentile radars of some teams mentioned here. The bars are percentiles compared to all other UEFA T5 teams, and the info in the box for each motif are the raw percentage for that motif for that team, the z-score, and then the exact percentile.