The purpose of this study was to construct a model of student performance using multilevel analysis, a type of multiple regression that takes into consideration the hierarchical structure of the data. The study used the data of the third grade students of secondary school who participated in the SAEB 2001 assessment. After controlling the effect of the variables of selectivity and school composition, an intraclass correlation of 0.17 was found, indicating that 17% of the variance in student performance can be attributed to the school level. In the final multilevel model seven variables of student level and nine variables of school level were included. The variables with the strongest effect on student level were: school delay and comparison of the student with his colleagues. On school level these variables were: aggregated cultural resources and aggregated school delay. The final model elaborated in this study will hopefully be used to elaborate public policies with a view to improve the Brazilian educational system.
student performance; educational evaluation; multilevel regression analysis; SAEB
