From molecule to municipality
Among >23,000 children living across Norway, we saw that the impact of children’s genes on their achievement depends on where they go to school. Higher-performing schools compensate for genetic differences between children.
To convince ourselves of the finding that children’s genetic makeup interacts with their schools, we needed to control for other differences between children in their family backgrounds and where they live. So, we built a dataset combining the Norwegian Mother Father and Child Cohort Study (MoBa) with several Norwegian administrative data sources, to capture parental genetics and socioeconomic status as well as school, neighbourhood, district, and municipality codes. We could then define ‘environment’ comprehensively, and distinguish the effects of environmental sub-components, for example making sure that school effects didn’t simply reflect socioeconomic differences between parents sending their children to each school, or wider geographical inequalities. Further, by including parental as well as child genetic data, we could set up a neat natural experiment: in effect, we could randomise children to schools regardless of their social backgrounds. Crucially, when parental genotypes are controlled for, the effects of children’s genes on their achievement reflect random genetic inheritance, disentangled from the effects of the environments which their parents selected. This randomisation comes from Mendel’s Laws, and as far as we know it hasn’t been used as a methodological tool in educational research before.
Schools moderate genetic effects
The figure below demonstrates the interplay between children’s genetic makeup and their schools. The x-axis represents genetic probability of staying in education, which we estimated using polygenic indices (PGI) for educational attainment. PGI summarise, for every individual, their ‘load’ of all education-linked DNA variants. Here, children’s PGI reflect the education-linked genes they inherited randomly, almost like a randomised controlled trial. In the figure, each line represents the unique association between children’s education-PGI and their achievement for one school. The main point is that the lines all have different slopes – showing that the impact of genes depends on the school.
In a subset of schools where genes matter the most (blue), the education-PGI explains over 8% of the differences between children in achievement. Even though PGI are far from perfect measures of genetic propensity, this is much more than the total impact of residential environments: municipality, neighbourhood and district factors combined only explain 2%. However, in some schools where genes matter the least (red), the effect of students’ education-PGI on achievement is below 2%.
Comparing the blue and red lines, we see that the red lines are all higher up. This tells us that the schools in which differences between children in terms of education-PGI have the least impact on their achievement are those with the best overall achievement. Most children are doing above average. In contrast, the schools where the education-PGI creates more achievement differences between children are the schools with the lowest overall achievement. This suggests that higher-performing schools compensate for children with lower education-PGI. They raise the achievement of those on the lower tail of the PGI distribution, whilst maintaining an overall higher achievement for everyone compared to the other non-red schools.
The second figure (below) shows the same result in a different way: not only is the impact of children's genes on achievement dependent on school, but the effect of school is dependent on genes. We see that differences between schools matter more for students with lower education-PGI, explaining 4% versus 2% of the variance in achievement for students whose PGI are 2SD below versus 2SD above the mean. In Norway, almost all students attend public (state) school, and the government has a long track record of redistributive policies to minimise school differences. Nonetheless, Norwegian schools are creating achievement differences between children who are equally unlucky in the ‘Genetic Lottery’, even when family socioeconomic background and all neighbourhood, district and municipality differences are held constant.
What does this mean for education policy? Our results suggest a need to focus on finding new ways to equalise schools in terms of opportunities for students who are similarly low on the education-PGI distribution. To this end, a useful strategy could be to identify important positive features of the ‘red’ schools (in the first figure) and implement them in the ‘blue’ schools. Since the school-level sociodemographic measures we tested in the study don’t explain the school differences in PGI effects, we need to explore in more detail what characteristics of teachers and classrooms are making the difference.
Our study shows that children’s genetic makeup interacts with their schools to affect educational achievement in non-deterministic ways, depending on social factors. We have demonstrated that a social-genetic approach delivers a new understanding of how children’s schools drive differences between them, even when other characteristics such as genes, family, neighbourhood, and region are similar. This knowledge could be valuable for future educational policy aiming to reduce inequality.