Contribution of machine learning to the analysis of grade repetition in Spain: A study based on PISA data.

Abstract

Introduction: The rate of grade repetition is excessively high in Spain despite being a controversial measure. In order to obtain evidence to contribute to reducing it in compulsory education, the present work is an in-depth study of the PISA 2018 context indices that are most closely linked to this phenomenon. Method: With the sample of Spanish students (n = 35 943), we used an automatic machine learning method to select and order the predictors, and multilevel logistic regression (students and centres) to quantify the contribution of each one. Results: For each educational stage we obtained the 30 most significant contextual variables, which explain 65.5% of the grade repetition variance in primary education and almost 55.7% in secondary education. Conclusions: The main indicators are principally at student level, which suggests the suitability of psychoeducational interventions based more on individual support than general policies. This gives rise to potentially more efficient and equitable measures than grade repetition, aimed at, for example, the management of learning time or academic/professional guidance, and predictors with specific differential significance at each stage. Methodologically, the study contributes to improving the specification of predictive models.