Hierarchical Modeling of Municipal Contextual Influences on Brazilian National High School Exam (ENEM) Performance: An AI Approach for Educational Policies

• Felipe Gomes Paradas UFRJ
• Carla Amor Divino Moreira Delgado UFRJ
• Carolina Gil Marcelino UFRJ

Resumo

This paper applies Hierarchical Linear Modeling (HLM), a technique often employed in machine learning and statistical analysis, to investigate educational inequalities in Brazil, focusing on the understudied influence of contextual factors at the municipal level. Although HLM is established to analyze school-level effects, this research expands its application by integrating individual student performance data from the National High School Exam (ENEM) with municipal-level indicators from the Social Progress Index (IPS) for more than 5,000 Brazilian municipalities. We develop a multilevel model that quantifyes how the local social context impacts educational performance beyond individual and school characteristics. Our analysis reveals that a significant portion of the variance in ENEM scores is attributable to differences between municipalities (Intraclass Correlation Coefficient – ICC reported in the results). Beyond established individual predictors (e.g., parental education, family income proxy, school type), the model demonstrates that specific municipal social progress components significantly predict performance variations, even after controlling for individual factors. In particular, the analysis indicates that the performance advantage associated with attending private schools varies significantly between municipalities and is inversely related to the overall level of municipal performance, suggesting contextual moderation of educational inequalities. This research contributes interpretable models capable of informing targeted educational policies tailored to diverse municipal contexts, demonstrating how AI-related methodologies can address complex social challenges like educational equity.