Modelagem do crescimento da aprendizagem nos anos iniciais com dados longitudinais da pesquisa GERES

Abstract

This article compares two value-added approaches to data from the longitudinal education survey called Estudo Longitudinal da Geração Escolar 2005 (GERES – Longitudinal Study of 2005 School Generation), which followed a cohort of students from more than 300 public and private schools over the first four years of primary education. Both approaches use hierarchical linear models, allowing the natural grouping of educational data from three levels: student, class and school. The first value-added approach builds models whose dependent variable is the student proficiency in each year evaluated. With a separate model for each year, it is possible to detect factors of the student, class and school associated with student performance. The second approach creates models to show the effect of the covariates of student, class and school on progress curves of proficiency throughout the study period. When comparing the two types of value-added models, the first one was the most efficient in diagnosing the effects of the environment and the teacher's pedagogical practice, but only in certain grades. The second type of model was able to identify progress curves of proficiency of different formats according to certain characteristics of schools and students, but was less sensitive to identify variables associated with the group formation process and the teacher's pedagogical practice. The two types of value-added models offer indications of differentiated learning processes for the disciplines of Portuguese and Mathematics, which deserve further study.