Take the ball, pass the ball — a simple method for analyzing passing behaviour
Years ago, we only looked at pass completion rates of teams or players as a statistic. Recently, many statistic sites also report the amount of progressive passes. However, the definition of progression is often unclear or it includes only the x-axis, i.e. the length of the pitch.
In this post, we want to show how passing data can answer interesting tactical questions. It also proposes a simple measure for the value of a pass — namely by the distance to goal of its destination. This is also in line with most of the expected goals models.
I want to point out that there are many research papers on that topic which are far more general, e.g. by including other event types or sequences of passes. For example, see the paper below or references therein.
However, as we will see in the following, our simple approach still opens up many interesting ways of analyzing passing data.
We start with splitting the football pitch into equally sized rectangles, which are called zones from now on (as in the picture above). For each pass, we know its start and end point as well as if it was succesfull. Hence, for each pair of zones i and j we can count all passes that went from i to j. This gives a matrix of pass counts.
From there on, many analyses can be carried out: we can for example look at the probability for a pass attempt (or a succesfull pass) to zone j, given that the player in possession is currently in zone i. Similarly, we can also look at the probability for a pass to originate from zone j, given that it landed in zone i. These can be aggregated on player or team level.
What is this good for?
Only by looking at the distribution of passing destinations/origins, some interesting questions for coaches or scouts could be answered:
- Which zones are targeted from a team in a certain phase of buildup or attacking play? On the other side, does a team have a weak spot, i.e. a dangerous zones where it concedes (above average) many passes.
- On a player level, which are the favourite target zones of a player? Which zones does he pass very seldomly, maybe because of worse perception capabilities? Are the success rates of players evenly distributed or does he have a weak side.
- If we give a value to the zones (which we will do in a second), is a player rather risk-averse in his passing or not? Only looking at passing rates may be misleading in that aspect.
An FC Barcelona showcase
To make this a more practical, I used (again) the Statsbomb Messi dataset for analyses of the FC Barcelona passing behaviour.
Looking at Barcelona as a team, we can clearly see how their passing behaviour evolved over the years: in the last seasons, the volume of passes — especially high up the pitch — decreased. The clear (horizontal) structure — wings, halfspaces and the center lane — from the years under Guardiola and Luis Enrique slowly faded out into a more uniform distribution.
Looking at the most influential players in the last years, we get the following picture:
I want to point out two patterns that I already perceived while watching Barcelona and which got confirmed by the data: Messis’s sweet spot of passing seems to be the left corner of the penalty box. Coming from the right side and being left-footed, this is naturally a very convenient area. With Jordi Alba running in behind, this has almost become a signature move in the last years.
On the other hand, the pattern of Busquets stands out compared to Xavi or Iniesta. Busquets seems to favour passes into the central lane or half-spaces rather than onto the wings. With Messi in the right half-space as a receiver, this creates immediate danger.
Valuation of passes
It is natural that we may want to assess the passing abilities in a more numerical fashion — i.e. with a single number. We use a very simple approach: the closer we get to the opposition goal with a pass, the better it is for the attacking team. Hence, we assign values ranging from 0 (our own corner flag) to 1 (oppositions goal). Visually, the value is distributed like in the first picture of this article.
Now, we can calculate the value provided by a set of passes with simple multiplication:
Value = sum_i(number of passes into zone i * value of i)
Naturally, a player’s score depends mainly on the his passing volume and his position. But if we want to compare teams or players of the same position this could serve as a measure of risk-aversion in passing.
Another question we can answer with this metric is the following: from which zones does a player or a team advance into dangerous zones?
In order to do so, we have to turn the table, and instead of aggregating by passing origin, we aggregate by passing destination (which we can do easily with our matrix of pass counts). Moreover, we weight each pass with its just described value.
Doing this for all successfull passes of Barca from 2008–2019, we get the following picture:
Surprisingly (or maybe not, because we look at FC Barcelona), very few value/danger is created from the wings, i.e. with crossing. On the other hand, half-spaces seem to be very good spots in order to approach the oppositions goal.
Conclusion
Only taking pass origin and destination data can be useful for many tactical aspects of the game. Even without the context of positional data, it reveals patterns of a team or a player’s behaviour on the pitch.
