A1169
Title: A novel BPPIT model for educational forecasting: Projecting SAEB scores via school-level data
Authors: Alessandro Jose Queiroz Sarnaglia - Federal University of Espirito Santo (Brazil) [presenting]
Marcelo Pereira Antunes - Secretary of State for Education (Brazil)
Ricardo Moreira Sant Anna - Secretary of State for Education (Brazil)
Eduardo Alves Araujo - Secretary of State for Education (Brazil)
Anderson Lucas de Paula - Secretary of State for Education (Brazil)
Luciene Dellaqua Bergamin - Secretary of State for Education (Brazil)
Diego Roberto Colombo Dias - Federal University of Espirito Santo (Brazil)
Bartolomeu Zamprogno - Federal University of Espirito Santo (Brazil)
Abstract: The Basic Education Assessment System (SAEB) is essential for public policy design and monitoring in Brazil, particularly as a core component of the Basic Education Development Index (IDEB). However, its biennial frequency hinders continuous monitoring and delays potential interventions. A methodology is proposed using School-level proficiencies from the Espirito Santo Basic Education Assessment Program (PAEBES) to forecast the SAEB score of the State School System. The approach involves forecasting both SAEB outcomes and the number of students assessed per School. A linear regression model was employed to fit the SAEB-PAEBES behavior, while a novel model called Binomial Permanence plus Poisson Immigration with Truncation (BPPIT) was proposed to fit the number of students assessed based on enrollment data. Forecast intervals were implemented using bootstrap. Data from PAEBES (2022) and SAEB/enrollment (2023) were used for model estimation and in-sample forecasting, while PAEBES (2024) and enrollment (2025) data were applied for out-of-sample SAEB 2025 forecasting. Results demonstrate precise in-sample forecast performance, as the interval encompasses the actual SAEB 2023 value. Additionally, the 2025 forecast indicates a high probability of score improvement relative to 2023. This work was supported by the Espirito Santo Research and Innovation Support Foundation (FAPES) Grant No. 922/2026.
