Missing data in educational assessment: a comparison of data treatment methods

• Autor(a) principal: Vinha, Luis Gustavo do Amaral
• Data de Publicação: 2018
• Outros Autores: Laros, Jacob Arie
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Estudos em Avaliação Educacional
• Texto Completo: https://publicacoes.fcc.org.br/eae/article/view/3916
• topic: Treatment of Missing Data; Education Assessment; Academic Performance; Simulation.; Tratamiento de Datos Ausentes; Evaluación de la Educación; Desempeño Académico; Simulación.; Tratamento de Dados Ausentes; Avaliação da Educação; Desempenho Acadêmico; Simulação.; Métodos para avaliações educacionais

Missing data are common in educational assessments. Using the appropriate methods has, therefore, become essential to reduce the impact of the loss of information. The present study aims to compare the performance of four methods for dealing with missing data (mean imputation, listwise deletion, maximum likelihood and multiple imputation), all based on regression models applied to the educational assessment of data collected in the State of Ceará. Information about 7,000 students was used, simulating various scenarios according to the percentage and the type of the missing data. The mean imputation method showed the worst performance in all simulated scenarios and the other methods showed similar results among themselves. Moreover, the use of auxiliary variables in the estimation by maximum likelihood and multiple imputation proved to reduce the bias of estimates of some important parameters of the model, when the simulated missing data is not random.