Combining multiple and boosting regression analyses to identify factors predicting achievements on mathematics national assessments

Numerous factors are related to students’ performance on standardized mathematics tests, including teacher-assigned math grades, gender, socio-economic status (ESCS), origin, geographic location, and school type. Typically, the influence of these factors on students’ outcomes on standardized mathematics tests is analyzed using classical regression analyses, which might not be able to handle complex, high-dimensional datasets and capture non-linear relationships. Boosting regression, a machine learning approach, was employed to overcome these limitations, allowing for more robust estimation of variable importance, detection of interaction effects, and provision of insights that are directly relevant for educational policy and intervention. Therefore, in this paper, we investigate the significance of these factors in predicting students’ achievements on the Italian national assessment of mathematical knowledge, INVALSI, through both multiple regression and boosting regression analyses. Results indicate that school type and teacher-assigned grades emerged as the most influential predictors, contributing approximately 42% and 30% of the relative importance within the predictive model, respectively, followed by macroregion (19%), gender (6%), and ESCS (3%). Students’ origin accounted for less than 1% of relative importance. These values reflect predictive contributions rather than causal effects and should be interpreted as indicators of the strength of association within the model rather than direct proportions of explained variance. This study demonstrates that the combination of multiple regression and boosting regression analyses can offer educators and policymakers clearer insights into the determinants of students’ performance in the national mathematics knowledge assessment.