Student achievement and principal value added: the case of Brazilian schools

• Ano de defesa: 2024
• Autor(a) principal: Sarabanda, Murilo Sá Rocha
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Desempenho escolar; Efeitos de diretor; Management practices; Práticas gestoras; Principal effects; Student achievement; Valor adicionado; Value-added
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/12/12138/tde-04072024-103435/

School leadership and management are discerning factors between schools with great influence over other school characteristics. We study principal effects in Brazil using value-added models that introduce the joint estimation of principal, school and teacher effects in a parsimonious manner. First, we detail the construction of a student panel linking test scores to teacher, classroom and principal allocation by merging several datasets from different sources, paving the way for future research requiring such information. We propose two classes of models for this endeavour: a group of parametric fixed effects models, and a semi-parametric model exploring the within-school variance in effects following principal turnover, for which we propose an extension to encompass teacher effects. Additionally, we seek to investigate a possible cause of principal value-added variation by studying the association of these estimates to a management practice instrument based on principal interviews. While our parametric model estimation stumbles on a series of difficulties, rendering frail principal value-added estimates, our semi-parametric model points to a result aligned with the literature: a higher principal effect variance is associated with a score 8% to 12% of a standard deviation higher in mathematics and Portuguese. Our management practice analysis inherit the frail results from the principal value-added measure estimates, on which they were contingent. Estimation setbacks, originating both in data and in our modelling, are discussed and paths ahead in this research agenda are presented.