During the last couple of weeks, the heavyweights of football analytics created the channel Friends of Tracking in order to share insights into their work. Among many great lectures, they also published a considerable amount of code and tracking data — in particular one dataset which covers 19 goals from the Liverpool team in 2019.
Finally, they set a challenge to analyze how Liverpool create danger in their attacks and — ultimately — score goals. In this article, I want to write about an approach to evaluate player decisions, in particular for attacking runs. In the case of Liverpool, we can see with such instruments that their players create danger by running into spaces which are not yet occupied.
The idea for this was mainly inspired by a statement of Borussia Mönchengladbach’s assistant coach Rene Maric in a recent interview:
He says:
“ In my mind, tactics describe the sum of a team’s decisions about how they’re going to solve a particular situation. […] Ultimately, it’s a very simple process: on the pitch, you’re either protecting the ball, demanding the ball, or creating space.”
Simplifying, Maric sees his task as a coach to enable players making better and/or faster decisions on the pitch. From an analytics perspective, this opens up the following question: can you identify a player’s decision in the data? And if so, how can you measure the success of his decision?
Obviously, this is a very wide and difficult question, so I will focus on a very small sub-topic: we want to look at attacking runs of Liverpool players and measure their success in terms of pitch control.
We will start with a simple assumption. If a player makes a decision to run into a space in order to receive the ball there, he will (in most cases) change two variables: his acceleration and/or the direction of his run.
Tracking data gives us insight on both of these variables. Thus, we can identify the time of a player’s decision and analyze its effect on the game. We want to measure the direction with the angle of the player’s current velocity.
In order to create danger, attackers need to identify spaces which are currently not occupied and try to gain control over such spaces. This in turn will force the defense to adapt and move constantly which can — ultimately — lead to a delayed reaction or an error and thus to a chance/goal.
Pitch control is the perfect instrument to evaluate this, as it models — for each location on the pitch — the probability of both teams to receive the ball in the respective location after its arrival.
Thus, we can say that a player made a good decision for an attacking run, if he increases pitch control in the space he is running into. Of course, players could also generate an advantage in other space by dragging out defenders, but we will ignore this for now.
Salah goal vs. ManCity
Let us look at an example. It covers the goal of Mohamed Salah against Manchester City in their 3–1 win in November 2019. You can watch the goal here:
We show three graphs in order to describe Salah’s movement and decisions in the scene leading to the goal:
- Acceleration: high acceleration indicates that a player just made a decision.
- Angle/direction: a change in the direction (measured by the angle towards the goal) could also indicate a player’s decision.
- Pitch control ahead: we measure the probability of receiving in the zone into which the player is moving (one second ahead). Keep in mind that this is somewhat a forward-looking measure (see also the technical details below).
We can see that Salah — after running in from behind — decides to move more vertically instead of running into the direction of defender in front of him (which was Fernandinho). This is the phase between frame 90–120 where the angle decreases. This change turns out to be necessary in order for the ball to reach Salah without being intercepted by Fernandinho and lead to the goal.
The increase of pitch control in this period reflects the (positive) outcome of Salah’s decision. Interestingly it reaches its maximum around frame 120 which is exactly 1sec before Salah receives the ball (20 frames = 1sec).
Salah goal vs. Porto
As a second example, we want to have a look at a goal of Salah in the 4–1 away win vs. Porto, originating from a classical counter-attack. Unfortunately, I have not found a video that captures the whole scene on Youtube.
Here are the same plots for Salah’s run as above. The dashed lines are the moment where he decides to change direction and when the pass (to Salah) is played.
Again, we can see that Salah first decides to run into a different direction — maybe because he spots that Mané already covers the space where he is running towards. After that first decision, his pitch control decreases as he turned more into the direction of the defender. However, he accelerates for some time and is able to run past the defender. His run created a second passing option (aside from Mané). In the moment when the pass is played, he has increased his pitch control ahead from around 0.5 to 0.6. When the defender is not able to intercept the pass, the pitch control makes another jump upwards. Salah is left with an 1vs1 against the keeper.
Conclusion
Combining their speed with run-ins from behind the defender, Liverpool players are able to create danger by running into previously unoccupied spaces.
When a player makes a decision of such a run, we can often spot the time of his decision using direction and acceleration data. With a pitch control approach, we try to evaluate if his decision was beneficial.
Aside from the obvious fact, that our approach has several limitations and is only tested on a small (and very specific dataset), it could help to visualize a subset of player decisions in a simple way.
Technical details
We want to explain more details of the methodology.
- The angle we plot is measured between -90 and 90 degrees. It is always measured towards the goal the player is running into. Thus, if he switches from moving towards the opposition goals to running towards his own goal, this will produce an incontinuity. However, for attacking runs this should happen seldomly.
- The third plot shows the pitch control of the region the player is running into. More precisely, we take the velocity of a player and compute the point where he will be within 1 second (if he would continue moving in the same direction in the same speed). We also rotate the velocity vector by 15 and -15 degrees and evaluate at those two point as well. Then we take the mean of the pitch control at these three point - weighted with their distance to goal (closer points have higher weight).
It is important to note that the pitch control model already looks into the future itself, so the pitch control for a point can be high even if the player has not arrived yet.
