According to Opta data, around 56% of goals in elite European football result from intentional assists, a figure that has remained much the same for the past few seasons. In this post I want to look at the distribution of assists and key passes on the field. From where do they originate? How effective are they?
In this post I am using the Opta terminology of ‘assist’ to mean a pass leading to a goal and ‘key pass’ to mean a pass leading to an unsuccessful goal attempt. I’m only going to analyse ‘intentional’ assists and key passes – events where the supporting player makes a substantial contribution to the goal attempt. So passing to a player who then dribbles out of his own half and beats three defenders to put the ball into the net doesn’t count as an intentional assist. I’ll use the term ‘final ball’ to mean either an intentional assist or key pass.
Final balls
The most obvious question to start with is from where do final balls originate on the pitch? The heatmap below (on the left) shows the origin of final balls, and the one on the right shows the origin of assists. We can see that final balls come mainly from broad zones in front of the 18 yard box and to either side of course the corners. The map for assists is similar in general shape, but shows a bit more variation, with the zones just inside the penalty area showing the highest densities.
Figure 1: Origin of final balls and assists
We can also look at the data from the receiver’s perspective. What does the picture look like if we centre the data on the end-point of the pass? In figure 2 (below) the receiver is positioned where the dotted lines cross.
The distributions of final balls and assists are quite similar. In both cases they are concentrated in four distinct zones to either side of the receiver and a little in front of him, with a less prominent zone about 10 metres behind the receiver. This kind of pattern makes some sense. First of all a pass originating from in front can be seen by the receiver, giving him a chance to position himself appropriately. We can see that the inner zones for assists are slightly further out than for final balls, and the most dangerous area is about ten metres to the side; this presumably gives the receiver enough time to react, which he can’t do if the final ball is delivered from too close. The balls coming from the wide zones are of course crosses, cutting out defenders.
Figure 2: Origin of final balls and assists relative to receiver.
Effective final balls
Next we can ask what the most effective type of final ball is; what makes an assist rather than a key pass?
Figure 3 shows the probability that a final ball is an assist (i.e. results in a goal). Once more the x/y co-ordinates are centred on the receiver. We can see that final balls are most likely to produce goals when delivered from a couple of metres in front of the receiver and about 10-12 metres to the side. Final balls delivered from behind the attacker are not as effective as those delivered from in front. There also appears to be a zone around 30 metres in front of the receiver; whether this is an anomaly I don’t know, but it may be just that there are only a few final balls originating from there.
Figure 3: Probability that the final ball is an assist.
Decision making
Finally we can look at decision making. Where do players attempt an assist in preference to attempting a shot? The next chart shows a location map for the final ball preference defined as:
A preference greater than .5 means the tendency is to attempt an assist, and a preference below .5 means the tendency is to attempt a shot.
My initial attempts to create this chart weren’t very successful; depending on the tuning parameters I selected, various straight-edged bands appeared which looked like statistical artefacts. In the end I went for a different method of tiling the display area which is why the chart below looks a bit different to the others.
The results show that the more central and the closer to goal, the lower the likelihood of attempting a final ball and the greater the likelihood of shooting. In particular, players in the area directly in front of the eighteen yard box (the so-called zone 14) will tend to shoot considerably more (around four or five times as often) as they attempt an assist. This may be because they are often the most advanced person in the attack and there is no-one better placed to pass to.
Figure 4: Preference for final ball rather than shot
Practical applications
One application of these methods is to compare different teams. In the figure below for example, I show final ball origin maps for four Premier League teams who competed in the 2013 season. We can see some quite clear differences. Newcastle’s chances were created mainly on the right, while Chelsea favoured a zone in front of the eighteen-yard area as well. Liverpool’s chance creation was nicely balanced, and Stoke created substantially fewer chances around the eighteen yard area than the other teams. Maps like these can be used to understand your own team’s performance, or even for opposition analysis.
Figure 5: Final ball maps for Four Premier League teams
Producing and interpreting these kind of maps can be quite challenging, because the same raw data can end up looking quite different depending on the methodology and tuning parameters you choose. However, I hope will find the ones I’ve produced here reveal some useful insights and that OptaPro addicts will find them interesting.
