Modeling the PISA’s score of Indonesian students using multivariate generalized linear model

Abstract

The Multivariate Generalized Linear Model applies a lot in data science. However, this model is still less discussed in education, specifically in data analysis in the Program for International Student Assessment (PISA), which has complex structure data. This study aims to analyze the factors that influence the PISA's scores of Indonesian students simultaneously covering the three subjects of the PISA assessment, namely mathematics literacy, science literacy, and reading literacy. The complexity of PISA data which involves multivariate response variables which assumes a correlation between response variables, adds to the complexity of the analysis. One approach is the Multivariate Generalized Linear Model with the Quasi Likelihood estimation method. Took the data sources from the PISA survey was conducted by Organization for Economic Cooperation and Development in 2018. This study indicates that the factors that influence the PISA's scores of Indonesian students simultaneously are the class taken, parental education, facilities at home, student discipline, teacher feedback during learning, age of entering kindergarten, and failing a grade during elementary school. Based on the model diagnostic, it can conclude that Multivariate Generalized Linear Model produces a model that fits in modeling the PISA's scores of Indonesian students.