Abstract
This study evaluates the causal effect of artificial intelligence (AI)-driven adaptive learning platforms on science, technology, engineering, and mathematics (STEM) academic performance among Grade 10 secondary school students in the National Capital Region, Philippines. Addressing the long-standing challenges of heterogeneous student readiness and large class sizes in developing educational systems, AI-driven platforms provide personalized instruction tailored to individual learning rates. To address self-selection and non-random deployment biases inherent in educational interventions, we employ a rigorous Propensity Score Matching (PSM) framework utilizing data from 1,280 students across 24 public and private secondary schools during the 2023–2024 academic year. Baseline covariates including prior academic achievement, socioeconomic status, teacher qualifications, and digital literacy were balanced using nearest-neighbor matching without replacement. The empirical results reveal a statistically significant positive Average Treatment Effect on the Treated (ATT) of 8.42 percentage points (p < 0.001) on standardized STEM post-intervention assessment scores. Subgroup analyses indicate that students from socioeconomically disadvantaged public schools experienced higher marginal gains compared to their private school peers, underscoring the equity-enhancing potential of adaptive educational technologies. Rosenbaum bounds sensitivity analysis confirmed that these estimates remain robust against potential unobserved confounding. The findings provide strong empirical justification for integrating adaptive learning ecosystems into national educational reform agendas, offering actionable insights for policymakers and educational administrators in the Asia-Pacific region.
