Digital equity and socioeconomic gradients in PISA 2022 mathematics: an integrated equity-diagnostic framework for Latin America and the Caribbean

Source abstract

This study develops and applies an integrated equity-diagnostic framework to examine mathematics achievement in PISA 2022, with Latin America and the Caribbean (LAC) as the focal comparative case. Using public-use PISA 2022 microdata covering 613,744 students across 80 participating countries and economies, the analysis organizes evidence into complementary diagnostic layers: baseline regional performance, socioeconomic dependence of achievement, ESCS-linked point gaps, multilevel variance structure, and model-based interpretation of socioeconomic, digital, and school-level correlates. Design-consistent estimates were computed using final student weights and Balanced Repeated Replication with Fay's adjustment, while multilevel models and LightGBM/SHAP analyses were used as complementary robustness and interpretation layers. Results show that LAC has the lowest regional mean mathematics score among the classified regions, 56.9 points below the global descriptive mean. However, this disadvantage is not reducible to a single regional average or a uniform socioeconomic gradient. ESCS gradients vary substantially within LAC, from comparatively steep gradients in Chile, Uruguay, and Argentina to flatter gradients in Guatemala and Paraguay. ESCS-linked point gaps further show that socioeconomic dispersion changes the magnitude of inequality expressed in score points. Variance decomposition indicates that achievement differences are structured across country, school, and within-school levels, supporting a system- and school-level interpretation of equity. Across weighted and multilevel models, socioeconomic background, home possessions, disciplinary climate, language background, and digital-resource indicators show relevant but specification-sensitive associations with achievement. SHAP results are compatible with this interpretation by showing that digital-resource indicators contribute to prediction variability within broader socioeconomic and institutional contexts, rather than providing evidence that these indicators function as stand-alone equalizers. Overall, the study contributes a reproducible, survey-aware framework for monitoring educational equity and interpreting LAC's mathematics disadvantage as a multidimensional equity profile.