Abstract
As artificial intelligence, science, and technology advance, mathematical self-efficacy for 21st-century reasoning tasks has become increasingly important. This study aimed to explore student- and school-level predictors of mathematics self-efficacy among 15-year-olds in South Korea and Singapore—two countries that rank high in mathematics achievement but differ in student self-efficacy as measured by the Programme for International Student Assessment (PISA) 2022. This study focused specifically on the newly released PISA 2022 self-efficacy scale, which assesses students’ mathematical reasoning and 21st-century mathematical tasks. To identify potential predictors of mathematics self-efficacy, we applied a variable selection approach using glmmLasso, a machine learning method that accounts for multilevel structure. The selected predictors were subsequently tested for statistical significance using multilevel linear models. A total of 108 predictors were considered, including 52 student- and 56 school-level variables. The results identified 14 student-level predictors associated with mathematics self-efficacy in both countries. Regarding country-specific predictors, seven student-level predictors were identified in South Korea, whereas in Singapore, two student- and two school-level predictors were identified. This study examined common and country-specific predictors of mathematics self-efficacy across the two countries and provided implications for its improvement.