Admission Test to University in Italy: A Performance Comparison of Regression Models for TOLC-S

Abstract

In Italy, the transition from secondary school to university is a critical phase, with approximately 20% of students experiencing dropout or course changes. Admission tests, such as the TOLC-S, aim to assess students’ preparation and support informed decision-making for both students and institutions. This study investigates the predictive power of the TOLC-S in forecasting university success, measured by the number of credits earned in the first year. To facilitate the analysis, a matching procedure was implemented to merge the CISIA admission test database with the National Student Archive (ANS), overcoming the absence of a common student identifier through a derived key. The study compares three regression models: a Generalized Linear Mixed Model (GLMM) with the university as a random effect, an Elastic Net, and a Random Forest. Model performance was evaluated in terms of accuracy, implementing a repeated 10-fold cross-validation to get more robust results. The results offer insights into the extent to which admission test scores influence long-term academic outcomes, contributing to the improvement of university predictive models and informing policy decisions on student selection and support.