Childhood Trajectories of Neighbourhood Disadvantage and Educational Attainment: Towards a Longitudinal Machine-Learning Approach

Abstract

Prior research has shown that deprived neighbourhoods negatively impact educational attainment, including when individual and family characteristics are controlled for, but still remains unclear on the temporal dimensions of this effect. Using Swedish administrative data over 20 years, and drawing on bespoke neighbourhoods, we examine two questions on the relationship between residential context and high school graduation: 1) Does enduring exposure to disadvantaged neighbourhoods matter more than brief exposure? And 2) During which childhood period does it matter the most? We address the challenge of dynamic residential selection using inverse probability weighting within a marginal structural model of high school graduation and neighbourhood exposure. To flexibly model the complex, non-linear processes underlying residential selection, we use Bayesian additive regression trees to estimate propensity scores, providing a more flexible alternative to conventional parametric models. Findings suggest that cumulative exposure to disadvantaged neighbourhoods linearly reduces the likelihood of high school graduation, with effects varying across childhood stages. The negative effect is strongest during late adolescence and, to a lesser extent, in early childhood, with prolonged exposure reducing graduation rates by up to 34 percent. These findings indicate that both the duration and developmental timing of neighbourhood exposure are consequential for educational attainment, even in the context of Sweden’s strong welfare state. More broadly, the study highlights how longitudinal approaches that integrate flexible machine-learning methods with causal inference can improve our understanding of how residential trajectories contribute to educational inequality.

* This article belongs to a special issue on “Migration Trajectories Across the Life Course”.