Abstract

This contribution examines the role of gender as a moderating factor within the education production function, focusing on students enrolled in different faculties at a major Italian university. Specifically, the study investigates how gender shapes the relationships among academic performance, time-to-degree, and other relevant covariates, with particular attention to whether such relationships vary in direction, magnitude, or even existence across gender groups. In this perspective, gender is not treated merely as an additional explanatory variable, but as a structural dimension capable of redefining the network of dependencies underlying students’ academic trajectories.
To capture these complex interdependencies, we employ graphical models that allow the identification of context-specific independencies and relational structures that differ between male and female students. This approach enables us to detect situations in which the association between two variables changes sign or vanishes depending on gender, thus providing a nuanced representation of the education production process.
A specific focus will be devoted to the enrollment of the students in the different faculties and in their heterogeneities of the gender-specific education production function. Given the hierarchical structure of the data, we adopt multilevel modeling techniques to appropriately account for unobserved heterogeneity both at students and degree program levels. In particular, being our main feature of interest the time-to degree, we extend regression models for discrete-time survival analysis by incorporating random effects that reflect the multilevel nature of the data. This is achieved by leveraging a Cox frailty modeling perspective, adapted to discrete time.
Within this framework, we propose the inclusion of additional parameters designed to detect context-specific gender effects within the multivariate system. These parameters allow us to test whether gender moderates not only direct effects but also conditional relationships among variables, thus integrating moderation analysis within a graphical and multilevel survival framework. Furthermore, we jointly model time-to-degree and academic performance, recognizing that these dimensions are interrelated outcomes of the same educational process. The joint specification enables us to assess how performance dynamics and graduation timing co-evolve and how gender differentially shapes this interplay.
The work also develops tailored inference strategies to identify and quantify gender-specific moderator effects. By comparing results across STEM and non-STEM degree programs, we aim to highlight differences in the mechanisms generating academic success and persistence. Ultimately, the study seeks to provide evidence-based insights into the factors that sustain or mitigate gender disparities in higher education.