科研速览 · Science Skim继续刷下去 · Keep skimming →
◆ Frontiers in psychology2026-01-01

What predicts adolescents' mathematics anxiety in the Middle East and North Africa? An interpretable machine learning analysis of six Arab education systems in PISA 2022.

Omar Allohibi

一句话结论 · In one sentence

XGBoost achieved the best cross-validated performance (CV R 2 = 0.275; test R 2 = 0.266, RMSE = 0.978), outperforming linear benchmarks by approximately 10% in explained variance. Mathematics self-efficacy was the dominant predictor, followed by effort and persistence in mathematics, self-efficacy for mathematical reasoning, and mathematics achievement. Growth mindset ranked fifth despite a near-zero bivariate correlation, indicating a conditional predictive association that marginal analyses conceal. Girls reported systematically higher predicted anxiety, and while the protective core (self-efficacy, persistence) operated near-identically across genders, the anxiety penalty associated with bullying victimization was almost twice as steep for girls. The predictor hierarchy replicated across all six systems (mean pairwise rank correlation = 0.88).

原始摘要(英文原文)· Original abstract
INTRODUCTION: Mathematics anxiety is among the most consequential affective barriers to mathematics learning, yet large-scale evidence from the Middle East and North Africa (MENA) remains scarce. Existing research on the region rarely moves beyond linear models, and no study has applied interpretable machine learning (ML) to identify which psychological, instructional, and contextual factors jointly predict mathematics anxiety among Arab adolescents. METHODS: Drawing on the Programme for International Student Assessment (PISA) 2022, we analysed 47,652 fifteen-year-old students from the six participating Arab education systems: the United Arab Emirates, Jordan, Morocco, Palestine, Qatar, and Saudi Arabia. Six algorithms (linear regression, elastic net, random forest, AdaBoost, XGBoost, and LightGBM) were tuned with randomized search under 5-fold cross-validation to predict the PISA mathematics anxiety index (ANXMAT) from 50 predictors spanning competence beliefs, engagement, instructional climate, social relationships, achievement, and demographics. SHapley Additive exPlanations (SHAP) values, computed exactly for the selected gradient-boosting model, quantified the magnitude, direction, and heterogeneity of each predictor's contribution, complemented by gender- and country-stratified decompositions and five robustness analyses. RESULTS: XGBoost achieved the best cross-validated performance (CV R 2 = 0.275; test R 2 = 0.266, RMSE = 0.978), outperforming linear benchmarks by approximately 10% in explained variance. Mathematics self-efficacy was the dominant predictor, followed by effort and persistence in mathematics, self-efficacy for mathematical reasoning, and mathematics achievement. Growth mindset ranked fifth despite a near-zero bivariate correlation, indicating a conditional predictive association that marginal analyses conceal. Girls reported systematically higher predicted anxiety, and while the protective core (self-efficacy, persistence) operated near-identically across genders, the anxiety penalty associated with bullying victimization was almost twice as steep for girls. The predictor hierarchy replicated across all six systems (mean pairwise rank correlation = 0.88). DISCUSSION: The findings are consistent with control-value theory (CVT): perceived competence, not socioeconomic status or resources, is the primary correlate of adolescents' mathematics anxiety in MENA. These cross-sectional findings are predictive, not causal; within the fitted model, the gender-asymmetric social stressors and the region-wide consistency of the predictive architecture identify efficacy-building instruction, anxiety-sensitive homework design, and anti-bullying policy as priority candidates for intervention research across Arab education systems.
读原文 · Read the paper ↗

AI 追问PRO

登录后使用 AI 追问

讨论区

登录后参与讨论

相关论文 · Related

What predicts adolescents' mathematics anxiety in the Middle East and North Africa? An interpretable machine learning analysis of six Arab education systems in PISA 2022. — 科研速览 Science Skim