When Does Inquiry Help? Nonlinear Science-Instruction Associations and Changing Achievement Contexts in Southeast Asian PISA Data, 2015-2022

This study examined when more frequent inquiry-oriented science instruction is associated with higher science achievement, and whether that relationship travels across education systems. We analysed PISA 2015 student data from Thailand, whose complete-case analytic sample comprised n = 7,250 students (estimated weighted population 545,378). Singapore, Indonesia and Viet Nam served as prespecified replication systems, and Malaysia as a sensitivity case. The instructional results were then situated within 2015-2022 changes in achievement and socioeconomic gradients in Thailand, Singapore and Indonesia. Survey-weighted models pooled all 10 science plausible values and accounted for complex sampling using 80 BRR-Fay replicate weights. Restricted cubic splines showed a robust nonlinear association in Thailand (Wald χ²(2) = 37.57, p < .001); the finding survived trimmed-support and response-propensity sensitivities. The shape varied with teacher-directed instruction, teacher support and socioeconomic status, but not science self-efficacy after false-discovery-rate correction. Nonlinearity replicated in Singapore and Indonesia, but not clearly in Viet Nam; common-support curves did not indicate a universal threshold. Thailand's mean science score was 16.55 points lower in 2022 than in 2018, whereas its socioeconomic gradient showed no clear 2015-2022 change. Results support a conditional, guidance-sensitive account of inquiry; they also underscore that cross-sectional PISA associations do not identify instructional effects or explain later national trends. For practice, these results are consistent with prior guidance research in suggesting that the configuration of inquiry, rather than its reported frequency, is the more useful design focus; sequencing, scaffolding and consolidation were not measured in these data, so this study cannot estimate what they would add.