Team Chemistry

Is it possible to find out who plays more efficiently with whom?

About 15 days after Mesut Özil was transferred to Fenerbahçe, Irfan Can Kahveci, who was playing for Başakşehir at that time, also joined the team. On the one hand, the world star, Özil, who is among the top “No10” of all time, and on the other hand, Kahveci, one of the most talented names in Turkey, would play together.

So what happened?

About 1 year and 2 months after that day, I shared the following viz from my Twitter account.

Yes, the legendary number 10 was out of the squad.

From my own football point of view, the biggest reason why Mesut couldn’t perform the expected performance was that he played with Kahveci, who has a similar playing style.

They’re both kind of playmakers. They are skilled in passing. They do not often receive the ball in the opponent’s penalty area. They are not ball carriers.

After Özil was excluded from the squad, Fenerbahçe won 6 of the 8 matches and was never defeated.

The subject of this blog post is whether it is possible to detect such situations in advance. Moreover, I will use match sheet data to do this because match sheet data is easily accessible to everyone.

The title of the blog post is team chemistry. I think it is possible to build a squad based on the duos. This post will go beyond Ozil.

Let’s consider the problem in detail

Sometimes the chemistry of two players just doesn’t match. They don’t play well together. Their styles may be similar to each other. The “Mesut-Irfan” example is just one of them. Or maybe they can be very different types of players. In the end, with 11 players completing each other, a good performance is displayed.

So the goal is to first detect this situation with data. -with some descriptive analysis-

The second step is to build a model, but that will be a continuation of the next post.

Where did the inspiration come from?

The incompatibility of the Mesut-Kahveci duo was always in the back of my mind. Fortunately, the working principle of the brain is a natural wonder. In an instant, newly learned things can be matched with backed-up problems.

I had one of these moments after watching a SciSports webinar about Player Indexes.

Is it enough to use the match sheet data to measure the strength of a player? Incredible!

That’s how the first spark turned into fire.

Let’s focus on the work

I wanted to start from the team level first. Thus, I would know better what I wanted to do and I would have some code practice.

At this stage, the first thing that came to my mind was to compare the playing time of the teams according to the score difference with those of the players and get something out of it.

If a player is usually on the pitch when his team is winning, that’s a positive indicator for him. If we look at it through two players at once, it would be more wonderful.

So why was this method troublesome?

The score difference when players sub in the game can be quite different. For example, Fenerbahce was never behind when their wonderkid Arda Güler was on the pitch because he was usually brought on to the pitch when the score difference was +1 or +2.

There was another issue here. A player who subs in the game when the score difference is +2 will have a positive indicator even if his team concedes a goal.

Next method!

So, in the first second that a player is in the game, whatever the score difference is, I take it to be 0. For example, let’s say our team was behind by 3 goals. Messi subbed in and scored 2 goals. With the new method, the team scored 2 goals in the time Messi was in the game. The minutes before that are not important to us.

When you look at the numbers in this light, Fenerbahce’s wonderkid is once again showing his quality.

This time, if we sort the figures in order from worst to best, we see two familiar names.

Mesut Özil and İrfan Can Kahveci being on this list don’t surprise me personally.

At this point, you may ask me. There are goalkeepers and defenders on the list. What is the connection between them being on the pitch and the team scoring goals? You are right. In fact, scoring goals is not the only evaluation criterion. Conceding goals is also particularly important. For example, last season, Berke Özer guarded FB’s goal for a while after Altay Bayındır’s injury, and his performance was criticized a lot.

After this stage, I started experimenting with two players. A total of 55 duo combinations come out of an 11-man squad. When we apply this to all the players on the squad, the number is quite high.

In other words, to put it more simply, how did the team perform when two players were on the field at the same time.

The first results were quite interesting.

Fenerbahçe has a total of 170 different duos who have played together for a minimum of 500 minutes. If we rank them in order of FB’s goal-scoring rate per 90 minutes from the highest to the lowest, I am sure you will be surprised to hear the 170th duo.

The Mesut-İrfan duo played together for 542 minutes in the league and during this time the team scored 5 goals, that is 0.83 per 90 minutes.

So we are back to where we started. This time we also have data.

I want to take things even further. Let’s look at the percentage ranking of all the Süper Lig’s duos for the 2021–22 season according to this method and get an offensive and defensive chemistry score between 0 and 100. A graph just like in FM and FIFA would be nice.

After Mesut Özil’s suspension, the choice of Mert Hakan and the series of victories that followed reflected on our team chemistry. Even more valuable, it is possible to conclude that the problem is not with Irfan but with Mesut.

Moreover, this situation reflected positively on the team not only offensively but also defensively. Based on my personal observation, I can say that Mert Hakan’s play maximized the performance of Crespo and Zajc.

Let’s take a step away from Turkey and look at Ronaldo’s chemistry with his partners, whose performance has been frequently discussed in the last 2 years.

Ronaldo’s connection with his teammates is lower compared to the duo in the whole of Manchester United. Especially in the minutes when the Ronaldo-Bruno Fernandes duo, which was often talked about on social media and in the press, played together, MANU’s goal-scoring rate was mediocre, while the goal-conceding rate was below mediocre.

For those who want to analyse, players from the EPL who played a minimum of 500 minutes together in the 2021/22 season can be found below, along with their chemistry points.

Final Word

I have a personal interest in squad building. It’s not so difficult to find good players anymore. It’s a small world now. Everyone can follow every league. Companies are collecting data from even the most unknown league in the farthest corner of the world. This has actually caused soccer player inflation. Certain players stand out, but I wonder how right it is to make those names a transfer priority. Because the main issue is to find the most suitable player for the squad, not the best player.

Whether this method is right or not is debatable. For example, I think it would be much more effective to do this calculation based on xG rather than goals. Since I don’t have xG data for the Turkish league, I chose to go by goals, but that was only the first step.

I have a dream to develop a metric where teams can find the most suitable player for their squad, using match sheet data which is much cheaper than event data. I’m sure every team or company would want to use it. Maybe it’s hard to achieve this goal on my own. I’m open to working with anyone who comes up with an idea for this project.

So what do I have in mind for the future? I actually want to develop a machine-learning model using player roles. For example, the compatibility of a false 9 and an inside forward is not the same as the compatibility of a classic winger and a false 9. I think it would be interesting to use their team’s xG/goals (p90) as output when they both are on the pitch.

Thank you for your time and reading :)