Can algorithms improve how children are assigned to schools?

Oliver Dlabač, founder of the startup VILLE JUSTE, speaks with Nora Marketos about the promises and challenges of data-driven decision-making in assigning students to schools.

Nora Marketos: What impact does school composition have on learning outcomes?

Oliver Dlabač: Over the years, the PISA studies have demonstrated that the cultural and socioeconomic composition of schools has a strong impact on student performance, particularly in countries like Switzerland, the Benelux countries, Germany, and France, which have large and diverse migrant populations. In those countries and others, migrants and individuals of lower socioeconomic status are increasingly concentrated in certain neighbourhoods of urban centres. Students attending schools in such areas tend to have worse academic outcomes, for a number of reasons: They are unable to benefit from the positive effects of interacting with advantaged peers who have more positive attitudes towards education, they are subject to more disruptive classroom conditions, and their teachers have lower expectations.

“The cultural and socioeconomic composition of schools has a strong impact on student performance.”

NM: How are children currently assigned to schools in Switzerland?

OD: In Switzerland, local authorities are responsible for school assignments, and this process is completed manually. Families are not allowed to choose their children’s schools, and children are usually assigned to the school with the shortest and safest route home; in Switzerland, it is common for children to walk to and from school on their own. This system often fails to produce a good mix of students from different backgrounds, as we show in a nation-wide study conducted at the University of Zurich. At VILLE JUSTE, we have refined an algorithm developed in the study, which is designed to enable school authorities to adjust school catchment areas to generate a better mix of students and thereby lead to more equitable outcomes. The algorithm assigns the right number of students to each school, but it also balances the students’ socioeconomic, cultural and academic backgrounds.

In the canton of Zurich, local law already explicitly requires that primary school classrooms be balanced with respect to students’ socioeconomic background, native language, ability, and gender. At the moment, however, information on socioeconomic background is completely unavailable during the manual assignment process, so this requirement cannot be met at all.

NM: The VILLE JUSTE algorithm has been met with controversy. Why is that?

OD: Some people have expressed concerns about the algorithm, but they are largely based on misconceptions. Headlines like “The computer reassigns pupils” are misleading. They give the erroneous impression that individual pupils will be assigned to support a disadvantaged school, or that pupils will have to be transported across the city.

“The algorithm adjusts area boundaries to generate a better mix of students in each school.”

The algorithm, which has been fully developed and adapted for pilot implementation in the city of Zurich, is very conservative and leans strongly towards the current approach. School commutes remain short and safe. Rather than reassigning individual pupils, the algorithm adjusts area boundaries to generate a better mix of students in each school. Catchment areas are already adjusted manually every year – although many parents and politicians are unaware of that fact – and an algorithm can assist the administrative staff by automating the iteration process and including different types of data. This allows for data-based and strategic decision-making at the school administration level.  

NM: What does a good mixture of students look like?

OD: The aim is to make sure that disadvantaged pupils – pupils who are not fluent in the local language and those from a lower socioeconomic background – never make up more than 30-40% of the school population. Earlier research – confirmed by our own replications as reported in our study – shows that in this scenario, there are only winners. No student’s school performance will deteriorate. Achieving a mix of students in each school is not a zero-sum game. There is no need to take something away from one student to put another in a better position. It is possible to improve the situation of disadvantaged pupils without having to disadvantage anyone else.

Seventy years of research in the United States clearly shows that many positive effects are associated with an optimal school mix. Students in well-balanced schools learn social skills and gain intercultural competence. They are more creative and able to engage in productive discussions. These skills are in high demand in the workplace, especially in a country like Switzerland, which has a large tertiary sector.

“Students in well-balanced schools learn social skills and gain intercultural competence.”

NM: What do you say to those who worry that unpredictable algorithms will determine children’s socioeconomic futures by deciding who they go to school with?

OD: Our algorithm follows a clearly defined procedure to systematically check combinations for possible catchment area shifts. It acts just as a human being would, were that person given sufficient time to calculate the various options accurately. Our algorithm is not based on machine learning; if it were, it might be impossible to determine which factors affect a decision. Processes like these must be fully documented, easy to understand, and accessible to researchers and the public. This is why we intend to publish our algorithm; it will be very clear exactly how each factor influences decisions about school assignments. I hope that we will be able to offer this resource to other school communities starting next year. We’re already in discussions with partners in Finland and Canada who are interested in bringing it to their schools. Our goal is to make this option available to as many schools as possible, in the interest of greater educational equity in Switzerland and globally.