Gender Inequality is negatively associated with academic achievement for both boys and girls

The study investigates whether and how macro-level inequalities relate to adolescents’ academic achievement across countries. Drawing on social-ecological theory, the authors note that macrosystem socioeconomic contexts can shape academic outcomes but have been less studied compared to micro-level family and school influences. Prior work generally finds that economic development relates positively to PISA scores and that economic inequality correlates negatively with achievement. The paper also considers gender inequality as a distinct societal factor that can shape educational structures through stereotypes and institutional practices. Two recent country-level studies reported that higher gender equality associates with higher PISA scores for both boys and girls. However, those studies used only country-level analyses and did not examine the role of economic inequality concurrently, risking ecological fallacy and confounding. This study therefore combines country-level and student-level (multilevel) analyses and compares the associations of gender inequality (GII) and economic inequality (Gini) with PISA achievement. The authors hypothesize that: (1) gender inequality is negatively associated with academic achievement for both boys and girls; (2) gender inequality shows a stronger association with achievement than economic inequality; and (3) the reproductive health dimension of GII contributes substantially to the association.

Economic inequality concentrates wealth and is associated with social problems including crime, reduced cohesion, and worse health. Multiple studies using PISA and other sources have linked higher economic inequality (higher Gini) with lower average academic achievement and wider performance gaps. Proposed mechanisms include neo-materialist perspectives (resource shortfalls) and psychosocial pathways (relative deprivation, stress). Regarding gender, literature shows that gender differences in math are small or trivial in most countries, supporting the gender similarities hypothesis. Nonetheless, national gender inequality can influence education via institutional norms and stereotypes. Prior cross-national studies found that gender equality strongly correlates with PISA results for all students and that improvements in gender equality track improvements in PISA scores over time. Gaps in prior work include reliance solely on country-level analysis and not modeling economic and gender inequality together, limiting causal interpretation and understanding of their relative roles.

Data: PISA 2012, 2015, and 2018 cycles covering many countries/economies (after exclusions for missing variables), with academic achievement assessed in mathematics, reading, and science. Gender Inequality Index (GII; UNDP) measures gender inequality across reproductive health, empowerment, and labor market participation. Economic inequality measured by World Bank Gini coefficient. GDP per capita included as a covariate. Country-level analyses: (1) Unconditional Pearson correlations between achievement (overall and by subject and gender) and GII/Gini for each cycle; Fisher’s Z tests compared magnitudes of correlations. (2) Stepwise regressions: Model 1 included GDP per capita and Gini; Model 2 added GII, estimating their associations with overall mean achievement. Robustness checks ran separate regressions for boys and girls. (3) Mediation analyses tested whether GII mediates the association between Gini and achievement, with bootstrapped indirect effects. (4) Fixed-effects panel regressions using country-year panel (2012, 2015, 2018), regressing achievement (overall and by subject) on GII and Gini, controlling GDP per capita, with country and year fixed effects. Student-level analyses: Multilevel mixed-effects (hierarchical) models with students nested in schools nested in countries. Outcomes were plausible value literacy scores in math, reading, and science. Predictors: country-level GII, Gini, GDP per capita; school-level variables (location; shortages of staff and materials in 2015/2018, teacher shortages and student-teacher ratio in 2012); student-level variables (gender, ESCS socioeconomic status). PISA sampling weights at student and school levels were applied. Machine learning model selection: Compared explanatory power of GII, Gini, and GDP per capita for predicting achievement in PISA 2018 using LASSO, random forest, and XGBoost, overall and separately for boys and girls. GII subdimensions: Multilevel models replaced overall GII with its three dimensions (reproductive health, empowerment, labor market) to identify which dimension is most associated with achievement. PISA scaling and weights: Analyses used plausible values and recommended procedures (IDB Analyzer for country-level plausible values; HLM software for multilevel analyses). Data sources: PISA (OECD), World Bank (Gini, GDP per capita), UNDP (GII). Ethical approval not required for public data.

• Across all three cycles (2012, 2015, 2018), both GII and Gini were significantly negatively correlated with achievement, but correlations for GII were consistently stronger than for Gini across subjects and genders (all p < 0.001). For mean scores, |r| with GII ranged roughly 0.80–0.85 versus 0.52–0.63 for Gini.
• Fisher’s Z tests showed GII-achievement correlations significantly greater than Gini-achievement correlations: 2012 Z=1.95 (p=0.025), 2015 Z=3.53 (p<0.001), 2018 Z=2.25 (p=0.012).
• Stepwise regressions at the country level: In Model 1, Gini significantly predicted lower achievement controlling for GDP per capita (2012 β=-0.49; 2015 β=-0.44; 2018 β=-0.45; all p<0.001). In Model 2, adding GII, Gini became nonsignificant (2012 β=-0.02 p=0.915; 2015 β=-0.03 p=0.755; 2018 β=-0.09 p=0.396) while GII remained strongly negative (2012 β=-0.77; 2015 β=-0.79; 2018 β=-0.69; all p<0.001). GII explained 18%, 20%, and 15% of variance (2012, 2015, 2018).
• Robustness by gender: For boys and girls separately, GII remained the only significant predictor once included; Gini lost significance (e.g., 2018 boys β for Gini=-0.04 p=0.712; GII β=-0.69 p<0.001; 2018 girls Gini β=-0.13 p=0.183; GII β=-0.69 p<0.001).
• Mediation: GII mediated most of the Gini–achievement association. Indirect-to-total effect ratios: 2012 96.85%, 2015 92.93%, 2018 70.22%, with all paths significant.
• Fixed-effects panel regressions (country and year FE, controlling GDP per capita): GII had significant negative coefficients; Gini was nonsignificant. Approximate effects per unit increase in GII: reading -0.47 (p≈0.099), math -0.18 (p≈0.085), science -0.29 (p=0.039), and mean scores -0.31 (p=0.050). Math showed the smallest impact.
• Student-level multilevel models: GII showed significant negative main effects on student achievement across subjects and cycles; Gini was nonsignificant. For PISA 2018, a 0.1 decrease in GII associated with 16.58 to 29.36 point increases in scores, depending on subject. Similar patterns held in 2015 and 2012.
• Machine learning model selection (PISA 2018): Across LASSO, random forest, and XGBoost, GII consistently ranked at or near the top in explanatory power among country-level predictors and surpassed Gini for overall, boys, and girls analyses.
• GII subdimensions: Reproductive health inequality showed significant negative associations with achievement in 2015 and 2018 (e.g., 2018: math B=-144.61 p=0.045; reading B=-174.80 p=0.027), indicating it contributes substantially to the overall GII–achievement link; results for empowerment and labor market were mixed and less consistent.

Findings indicate that higher gender inequality is linked to lower academic achievement and that this relationship holds for both boys and girls. The authors argue that societal gender norms and stereotypes, transmitted through families, schools, peers, media, and broader culture, may dampen academic investment and performance: negative stereotypes harm disadvantaged groups, while positive stereotypes can also impair advantaged students through performance pressure (choking under pressure). Gender inequality appears more proximally tied to educational processes than economic inequality and may partially reflect a social consequence of economic inequality that then permeates schools and families. The fixed-effects panel results suggest that changes in gender inequality help explain changes in achievement over time, whereas changes in economic inequality do not. The reproductive health dimension (maternal mortality and adolescent fertility) appears particularly impactful, aligning with evidence that women’s health and opportunity constraints affect children’s educational outcomes. The intertwined nature of economic, gender, and educational inequalities implies that addressing gender norms and institutional practices may yield broad educational benefits for all students.

This study demonstrates that gender inequality, more than economic inequality, is robustly and negatively associated with academic achievement for boys and girls across PISA 2012, 2015, and 2018, at both country and student levels. Mediation and panel analyses indicate that gender inequality mediates much of the economic inequality effect and that changes in gender inequality relate to changes in achievement. Among GII dimensions, reproductive health contributes substantially to the observed association. Policy implications include prioritizing multisectoral, multilevel strategies to reduce gender inequality within societal institutions, schools, and families, alongside efforts to address economic disparities. Future research should further unpack mechanisms linking gender norms and institutional practices to student outcomes, explore within-country inequalities and performance gaps, broaden measures of achievement, and employ longitudinal and quasi-experimental designs to bolster causal inference.

• Measures: Academic achievement operationalized via PISA scores and gender inequality via UNDP GII; results may depend on these indicators.
• Aggregation: Country-level analyses use national averages, which can mask within-country educational inequalities.
• Conceptualization of gender: The study treats gender primarily as a categorical variable; alternative conceptualizations (as adjective/verb capturing relations, identities, and performances) are not modeled.
• Causality: Despite fixed-effects panel models, student-level data are cross-sectional and the design is nonexperimental; causal claims should be made cautiously.