Does a team have a distinctive playing shape, and is it held consistently from game to game? I decided to construct some heat maps to find out, and I’m reporting my findings in this post.
If there is no consistency from match to match, we might expect to see a meaningless smear when we average all the touch positions over a season. On the other hand a consistent geographic pattern of ball contact should produce a clearly delineated patchwork of touch zones on the field.
I chose to examine the Premier League 2013-14 season, and plotted all ball touches recorded by Opta. I didn’t distinguish between attacking or defensive actions, or home and away matches, or even between formal formations, but I just lumped everything together. I had to experiment a bit with the colouring, and scaling, but - perhaps surprisingly - some clear patterns emerged.
In the first gallery below, the blue areas indicate a low volume of touches and the red areas a high volume. Each map is drawn to its own heat scale, which is shown on the right hand side of the image when you expand it.
Touch Zones EPL 2013 [Individual heat scales]
Individual scales are great for highlighting shapes within a team, and we can see some very clear differences. For example, we can see Manchester United’s neat formation of four evenly-spaced touch zones either side of the centre line, as compared to Chelsea who have a hotspot on the left side of their midfield, and Swansea who have a symmetrical pair of streaky zones in their own half. We can also see that teams like Fulham, Hull, Newcastle, Norwich and Swansea tend to attack predominantly down the right, while Everton, Manchester City, Southampton and Spurs are more evenly balanced
While this way of scaling the data certainly reveals team shapes, it can’t show the relative volumes between teams; for example it looks as if Spurs (3rd image along on the bottom row) have more of the ball than West Brom (the image next along). But that is exaggerated by the individual scaling; if you look at the scales, you can see that red for Spurs is a density of .00020, while red for West Brom is .00025 - so there is naturally more red in the Spurs map.
So next I mapped the data onto a common heat scale. The gallery below shows the exact same data as the previous one, but all the heat mapping is now on the same scale.
Touch Zones EPL 2013 [Common heat scale]
In fact, it was a bit difficult to find a scale that worked well for all teams. A scale that captures the very highest intensity zones tends to lose the differentiation at the lower end, making all low intensity zones look the same. The one I finally settled on didn’t quite capture the highest densities, and you can see a little overload in the Manchester United map and one or two others. Together with the individualized scales it does give us quite an interesting visualization overall.
Obviously this kind of map could be refined - for example by mapping different types of ball touch.
Another thing we can do is to compare teams by using ‘difference’ maps. Here the colouring is designed to emphasise areas of the pitch where one team is more active than another. The difference maps below compare Arsenal, Manchester United and Manchester City two at a time.
In the first map, the red areas show where Manchester United have more ball touches than Arsenal and the green areas show where Arsenal activity dominates. Areas of the pitch where the two teams are equally active (or equally inactive) are light or colourless. We can clearly see the way Manchester United play down the touch-lines, while Arsenal play is concentrated in three narrower channels. The muted area towards the opposing goal indicate that both teams are equally active here. The comparison between Manchester City and Manchester United looks similar - again, United favour the wide midfield, while City play more centrally. Finally, the map comparing Arsenal to Manchester City reveals a patch of high Arsenal activity on the left of mid-field, while Manchester City are more active over a broad area in their own third and on the right and middle beyond the half-way line.
Given the rough and ready nature of the analytical process, and the a-theoretical data selection, I was pleasantly surprised to see these intriguing patterns emerge. But that’s the serendipitous fun of exploratory work.