Education inequalities persist globally, particularly in Italy, often influenced by factors beyond student ability. This paper aims to study the pure territory effect on education inequalities controlling for other factors such as gender, socioeconomic status, and high school performance. As in a clinical study, we evaluate the effect of multiple treatments represented by territorial clusters on the students’ enrolment choices. The inverse of propensity score estimates obtained with a gradient-boosted model (GBM) procedure are used as weights of a multinomial logistic regression model to evaluate the probability of enrolling at the university inside or outside their own place of residence. The study highlights the importance of considering the complex interplay between territory and economic variables in explaining inequalities in the transition from high school to university. Especially in northern regions, we show that the territorial effect on mobility choices, was covered by socioeconomic status and previous educational attainment. In the southern regions, the results highlight a more pronounced territorial heterogeneity regarding the choice to move to another region compared to northern regions.

Several studies in the mathematical education literature show the effect of students’ high school skills in maths on their success at higher levels of education and work. In particular, the importance of maths course taking in US high schools is highlighted to be important for college enrollment and completion. The choice of taking additional maths courses or, as in Italy, of choosing a high-school curriculum with more maths, is not random: it depends on several substantial factors such as gender and socio-economic status. This selection bias implies that the differences in the academic outcomes might be traceable not only to mathematics ability and knowledge. In this paper, the aim is to estimate the treatment effect of attending a relatively new high school curriculum in Italy with more maths, with respect to the traditional track of the scientific “liceo”, on two academic outcomes: university enrollment and first-year university performance. After having reduced the selection bias using a caliper multi-level propensity score matching procedure, a multi-state Markov model is used to study the treatment effect on the joint educational outcomes.