Predictors of Math Achievement Among Students with Low Economic, Social, and Cultural Status Attending Schools with Large and Small Within-school Variance: An XGBoost approach with School-level random effects

This study explored the key predictors of mathematics achievement among students in the lowest quartile of the Economic, Social, and Cultural Status (ESCS) index, attending schools with either high or low within-school variance in math achievement. Across both groups, shared predictors, identified through XGBoost with school-level random effects and SHAP analysis, included students’ preference for mathematics, achievement motivation in mathematics and science, and time spent on after-school learning. Meanwhile, for students in schools with larger achievement variance, important predictors included instructional time, preference for science, number of digital devices at home, frequency of part-time work before school, parents’ expectations for their child’s career in mathematics, and teachers’ ICT instructional strategies. For students in schools with smaller achievement variance, key predictors included perceived instructional quality, time spent on math homework, digital self-efficacy, classroom disruption factors, and the number of math teachers. The results stress the need to tailor support policies to the contextual characteristics of learning environments for disadvantaged students.