Determinants of Academic Dropout in Higher Education in the Southeast Region of Brazil

Academic dropout represents a critical barrier to building inclusive and efficient higher education in Brazil, where cumulative rates have recently reached 59%, with 53% in public universities and 61% in private ones. This study analyzes the Southeast region, which holds strategic importance as the nation's economic hub and concentrates the largest proportion of student enrollments, providing a favorable environment for robust comparative analysis. Despite advances in democratization policies like REUNI and SISU, significant obstacles to student retention persist. Utilizing microdata from the 2023 Higher Education Census (INEP), this research examines a representative sample of 6,930 course observations. The methodology integrates binary logistic regression and linear discriminant analysis (LDA) to identify factors influencing a 'Relative Dropout' indicator, which classifies courses according to the national average of 24.3%. The validation of the discriminant function utilized Wilks' Lambda and Box's M tests to ensure statistical consistency. The independent variables encompass sociodemographic proportions, including gender, ethnicity, and various age groups, alongside institutional factors such as course shift, admission through quota systems, participation in the National Student Assistance Program (PNAES), and access to financial aid. The findings reveal that dropout propensity is significantly influenced by student profiles and institutional support. A higher presence of female students reduces the chance of abandonment by approximately 45%, and the 18–24 age group demonstrates an 80% reduction in dropout probability. Conversely, enrollment in evening shifts increases the relative risk of dropout by roughly 67% due to work-study conflicts. Crucial retention mechanisms, including admission through quota systems, social support, and financial aid, show powerful effects, reducing dropout propensity by between 45% and 76%. The logit model correctly classified 61.9% of courses, while the LDA achieved a 60.1% accuracy rate. Discriminant analysis confirms that younger age, daytime enrollment, and institutional support are the primary variables separating low-dropout environments from high-risk scenarios involving older age groups and lack of assistance. These results underscore that affirmative action and student assistance are vital for academic efficiency. Policy interventions should prioritize the strengthening and continuity of PNAES and financial aid programs, particularly for vulnerable students in evening shifts. Institutional strategies must integrate quota systems with robust psychosocial support and longitudinal monitoring to maximize the social return on public investment and foster sustainable economic and social development.