Before proposing to present at the OptaPro Forum this year, I had considered corner-kicks (and still do) a very interesting topic in football. While fascinated by the topic, corner kicks are at times quite repetitious (inswing/outswing) with short corners thrown in occasionally, for a bit of variety. The questions I put forward for the Forum were 1) To examine if the different leagues in Europe preferred certain corner kick strategies over others and if so, which teams and 2) Were some areas more dangerous to deliver a corner kick to than others?
The Data and Process
Using corner kick event and qualifier data from the 2015/16 season of Europe’s Top 5 Leagues (Premier League, Bundesliga, Ligue 1, La Liga & Serie A), I went ahead with my analysis. After running some initial tests, I had to conclude that I would not be able to solve corner kicks in one attempt over a six-week period. So instead, I focused my attentions to the first phase of play (the delivery from the corner) and the events/actions that occurred during this phase. The sample size of corner kicks for this study was 18,425.
Using Opta’s F24 files, I first had to connect the event data with Opta definitions. On completion of that, I next used the XY co-ordinates to break the opposition half into zones. I decided on 11 zones (left), which are divided as follows:
Inside 18-yard box: 2 – 9 Danger Zone: 3, 4, 5, 8
I won’t go into the smaller details of the poster. You can read the different types of corners used by each league and number of goals scored by each corner in each league. What I wanted to bring to the attention of readers was the conversion percentage of corner kicks and zonal conversion rates for each area. For the 2015/16 season, a corner kick was converted at around 3%. Zonal conversion rates differed in each league and zone, yet averaged close to 3%.
Obviously not every team used the same corner kick strategy and the types of corners in different leagues was different. Therefore, I don’t want to give any direct take-a-ways based just on corner kick conversion rates. There was another aspect that I thought would be able to give a broader answer.
Looking at successful pass percentage rates from a corner kick, we could establish more varied results between zones and therefore give an idea of what zones are more successful to direct a corner to. A successful pass in this instance was defined: “a pass that reached a team-mate”. Naturally, I did not fully examine this per league as the sample size would have been too small.
For the Forum, I designed an interactive element through Tableau which allowed me to visualise each teams’ corner kicks and the zone the kick (first phase of play) went to. I can’t publish the visualisation because the data doesn’t belong to me, yet a few samples of the work are shown in the video. Make sure your sound is turned on (I’m narrating).
On the Day
I very much enjoyed the stand-up talks as well as listening to the other poster presentations. As for my poster, I was delighted with the interest that it received on the day. I had some interesting conversations with people from all areas of the football world, asking about my work and my take on corner kicks in general. In this next part, I will discuss some questions I received on the day and any future analysis that should be studied on the topic.
- Where was the interest from and what did they want to know?
I met so many people that day but if my memory serves me correctly, most of the interest came directly from club analysts. Nearly everyone wanted me to answer: “How does one score more goals from corner kicks?”. Unfortunately, I did not (and still don’t) have that answer on hand just yet. Corner kicks are such a difficult aspect of football to analyse, probably the most difficult due to the number of events that take place within a time frame (10 seconds for Opta coding). Yet through a first stage analysis (such as the one I undertook here) teams can use this as a method to visually analyse opponents first phase of a corner kick strategy.
- Why divide the opposition half into those areas?
Measuring out the zones from co-ordinates by hand themselves proved very tricky, especially the closer to goal I came. Therefore, the pre-measured XY co-ordinates in the F24 files from Opta played a big part in deciding how the zones were divided. While this left some zones (such as 7, 8 and 9) bigger than others, I felt it reduced the possible error count significantly.
There are a few pieces that should be undertaken as further research projects within the realm of corner kicks. Mainly being:
- Phases of play and building a model
The next steps should be to try and incorporate the other phases of play and build a model to examine corner kicks in more detail. What detail that should include will be up to the individual, yet they should incorporate attacking and defensive tendencies. For instance, not all corner kicks are scored in the same way and therefore this should be included in the next stage of analysis.
While not all in/out-swinging corner kicks are scored straight from the first phase of play, short corner kicks will need an extra phase of play to even reach the Danger Zone, which allows teams to re-group for the likes of a cross for instance.
- Tracking Data
Including tracking data such as Martin Eastwood’s poster presentation write-up last week, can help with added decision-making and optimising chance creation.
- Game State
Examine the effect that Game State has on corner kick strategies. For example, do teams have a strategy for the game and if losing, adapt that strategy to try something different (short corners due to time constraints)?
- Defensive strategy
We often seem obsessed with measuring attack before defence in any new metric. While I did the same with corner kicks, the defensive aspect of how to defend corner kicks, would also be beneficial for teams to examine. I hope that this poster/blog post is of interest for teams to examine where their next opponent delivers the ball to. From there, using video how can we stop them creating/scoring chances from a corner kick?
My thanks go out to Ryan Bahia and Tom Worville, who were extremely helpful when explaining the different data event points to me.
*My poster can be viewed below.