Imagine being a keen football player, looking for a topic for a thesis to conclude your Master in Econometrics. You figure that with your skills, you should be able to come up with a model that increases the level of sophistication in the analysis of the game by measuring each player’s effective contribution to the outcome of the match.
The model turns out to be so successful that your thesis is published in a scientific journal and you are offered a job at a company that provides data intelligence for professional football. This is exactly what happened to alumna Lotte Bransen.
A dream come true
"It’s a dream come true", admits Lotte. "After obtaining my Bachelor’s degree in Mathematics at the University of Utrecht, I decided to do a Master in Econometrics, specialising in Quantitative Logistics and Operations Research. I chose Erasmus School of Economics because here you learn methods to solve real-life problems. In these cases data are never perfect; they are incomplete, contaminated or both. The techniques presented during my premaster and my master were indispensable for the kind of research I did."
She conducted her research in cooperation with SciSports, the company that hired her after graduation. How did she manage to land such a coveted job? "I knew SciSports from the days when I was a bachelor student. Back then they were just starting and didn’t employ any paid staff yet. It all fell into place when it became time to work on my Master’s thesis. Originally located in the East of the country, they were by now one of the fastest growing sports analytics companies in the world and planned to open a second office somewhere central. That made the commute more doable for me. I knew Giels Brouwer, one of the founders and in a joint effort with Jan Van Haaren, Head of Data Analytics at SciSports, we came up with the idea for my research project, which Jan agreed to supervise together with professor Michel van de Velden at Erasmus School of Economics."
In the footsteps of baseball
Statistical analysis in sports has received more attention since 2003 when Michael Lewis wrote Moneyball, a book that was also the basis for a movie. It tells the story of the efforts of the Oakland Athletics baseball club to bring together a successful team despite modest means by using statistical analysis. How come basketball and baseball were so much quicker to adopt these techniques?
Initially, it was an American phenomenon, limited to sports that are popular in the states. Baseball lends itself especially well to this type of analysis because the actions are to a large extent predictable: someone pitches and the opponent hits the ball. That’s easier to measure than football, where scoring doesn’t happen as often as in many other sports and is usually preceded by a series of passes.
Not all passes are equal
What made your research different from other projects that analyse football? "First of all, our research distinguishes itself by the use of a large dataset. We used input from over 9000 matches. Other researchers focused on general data about whether a pass reached a team mate for example, but we have taken into account the circumstances in which a pass was given. A pass between two defenders is not the same as a pass from a midfielder to an attacker in the penalty area.
We have compared each pass to a historical database of passes that were given in the same part of the pitch covering a similar distance, using a domain-specific distance function. This enabled us to determine how good the pass really is. By aggregating such data for each player over 90 minutes, we can conclude what the Expected Contribution to the Outcome of the Match (ECOM) is for each individual.
To give a few examples: based on these data we have calculated that during the 2017/2018 season Mesut Ӧzil of Arsenal was the most impactful player in terms of passes. David Silva (Manchester City) came second and Lionel Messi (FC Barcelona) third. We had also identified that Frenkie de Jong would be the most suitable replacement for Barcelona’s Andres Iniesta who left for Japan. De Jong was indeed signed by Barcelona (after our research was published), but we were not involved. They had probably been following him for a long time already.
It’s also possible that a player will take the initiative to ask us for input. When Dutch international Memphis Depay was leaving Manchester United he asked us to analyse for which clubs he would be best suited. Our advice was part of his decision to accept an offer from Olympique Lyonnais."
Focus on midfielders
It appears the research is most valuable for the assessment of midfielders. Attackers pass relatively seldom and defenders are not often involved in setting up a goal.
That’s correct. We are working on models that would be suitable for these other groups as well. Developing a new model takes several months. We like to come up with a beta version in a few weeks, so that we can show others what the idea is and get feedback to improve it.
Computer game
Lotte’s research has been incorporated in SciSports’ platform, called Insight, where clubs can keep track of players around the world that interest them. The algorithm calculates the so-called SciSkill, making it possible to quantify a player’s current quality and potential.
Clients pay a monthly fee to use Insight. Forbes Magazine recently quoted Giels Brouwer as saying that the company has taken inspiration from computer games such as Football Manager and FIFA: "The idea was to build a real-life Football Manager using data by helping football clubs build an index on all players to aid the recruitment process."
"Females are also still a minority in the world of data science"
Attainment gap
Football is increasingly dominated by big money, making it harder for less affluent clubs to keep up. Will this type of service increase that gap? Lotte doesn’t think so: "Our website shows that the packages start at EUR 750 a month, that is cheaper than having a player on the bench. Among our clients you will find names like Ajax, Vitesse, Heracles, SC Heerenveen, Wigan Athletic, Olympique Lyonnais, Club Brugge, but also the KNVB. Some of the giants like Liverpool and Barcelona have their own departments where they employ teams of data scientists. Most of these clubs are quite secretive about their methods, although Barcelona recently presented at a conference in Boston where I also showed our work."
When asked if she encounters many women at such events, she laughs: "Sports professionals and researchers are predominantly male. Females are also still a minority in the world of data science, so if you combine these two it’s not surprising that the answer is no. However, the amount of girls that study econometrics nowadays makes me think that this is about to change."
Women’s football
Is data science also used to analyse the qualities of female players? "It’s still very rare, but data from the European championships in 2017 have been used for a collaborative project by several universities. However, the comparatively small amounts of money going around in women’s football hamper a further roll out. A lack of historical data might also play a role."
Does she still enjoy watching the game or is it difficult to switch off the analytical mode? "I’m a fan who isn’t actively thinking about work when watching football, but it does happen that an idea pops up in my head when I see something occurring in the game or when I’m playing myself."
It sounds like Lotte has scored a special hattrick, being successful as an attacker on the field, coming up with interesting research and maintaining her love of the game.
- More information
Read more about the stories of our alumni in Backbone Magazine here.
