Paradoxical associations between support structures and achievement: a cross-national exploratory analysis of Taiwan and Finland using PISA 2022

Educational research typically assumes self-directed learning (SDL) and family support enhance achievement. Using PISA 2022 data from Taiwan (2929 students, 186 schools) and Finland (5119 students, 211 schools), we examined associations between support structures and achievement in mathematics, reading, and science through multi-group path analysis controlling for socioeconomic status, gender, and home language. Results revealed unexpected patterns challenging conventional assumptions. In Taiwan, SDL showed consistent positive associations with achievement (β = 0.16 − 0.18, p < 0.05), but in Finland, SDL related negatively to all three domains (β = − 0.06 to − 0.10, p < 0.05), but in Finland, SDL related negatively to all three domains (β = − 0.06 to − 0.10, p < 0.05). Family support showed negative associations in both countries (Taiwan β = − 0.03 to − 0.11; Finland β = − 0.02 to − 0.15). Most paradoxically, socioeconomic status exhibited unexpected negative associations with achievement in Finland (β = − 0.11 to − 0.12, p < 0.001), contradicting decades of educational research documenting positive SES gradients. These patterns may reflect: (1) reactive support (parents/teachers intensify help when students struggle), (2) cultural differences in construct meaning (what "family support" signifies varies), (3) model misspecification or suppressor effects, or (4) measurement artifacts in cross-national comparisons. Models explained limited variance (R2 = 0.076 − 0.160), indicating most achievement variation stems from unmeasured factors. Our analysis employed simplified PISA procedures (averaged plausible values, basic weighting without clustering adjustment), underestimating standard errors by 5–15%. Cross-sectional design precludes causal inference. We present these findings as descriptive patterns warranting replication and mechanistic investigation through longitudinal or experimental designs, not as established theoretical relationships. The study contributes by documenting that relationships assumed universal may be context-dependent or bidirectional.