Nirmal GhimireNakhchivan State University2026-06-172026-06-172026-09-19https://doi.org/10.1016/j.stueduc.2026.101620https://rims.khazar.org/handle/123456789/1377This study examines multilevel associations between student characteristics, school contexts, and academic achievement across reading, mathematics, and science using PISA 2022 U.S. data. Hierarchical linear models with 4552 students nested within 154 schools reveal substantial between-school variance, highest for mathematics (24.5%), followed by science (22.4%) and reading (18.9%). Each of the 10 plausible values per domain was treated as an imputation of the latent achievement construct, fit on every PV x covariate-imputation combination (100 fits per model), and pooled via Rubin’s rules. Gender demonstrates domain-specific patterns: males score 22 points lower in reading but 13 points higher in mathematics, with marginal advantage in science. Home language and parental education show consistent positive associations across domains. School-mean parental education and home ICT resources are associated with roughly 3.7 and 11 times the corresponding within-school effects. Cross-level interactions are significant though random-slopes reveal substantial school-level heterogeneity in demographic associations.en-USMultilevel modeling Educational disparities Ecological system theory Post-pandemic education Cross-subject comparison Compositional effectsjournal-article