<![CDATA[Universal social-emotional learning (SEL) is widely implemented in primary schools, yet reported effects range from negligible to implausibly large, and syntheses that separate methodological quality from true intervention effects remain scarce. This meta-analysis aimed to estimate the effect of universal, school-based SEL interventions, relative to conventional schooling, on primary-school children's social-emotional competence, and to test how that estimate depends on study design, risk of bias, and small-study effects. Following PRISMA 2020, 4,043 records were identified from Scopus (2016–2025), 882 reports were screened at title and abstract, and 93 full texts were assessed for eligibility; 18 studies met the inclusion criteria (15 controlled between-group and three single-group pre-post designs), representing roughly 11,900 pupils across 13 countries. Study quality was appraised with ROBINS-I, and effect sizes (Hedges' g) were pooled with random-effects models using the DerSimonian-Laird, REML, and Paule-Mandel estimators with Hartung-Knapp confidence intervals, complemented by meta-regression, influence diagnostics, and trim-and-fill. The naïve pooled estimate was large (g = 0.89) but extremely heterogeneous and driven by three higher-risk outliers; after excluding them, the effect stabilized at a moderate g of 0.53 under DerSimonian–Laird and 0.55 under REML and Paule–Mandel, falling to g = 0.18 when restricted to lower-risk studies. At the same time, small-study bias persisted after outlier removal. We conclude that universal SEL interventions are associated with a genuine but modest-to-moderate improvement in social-emotional competence (credible g = 0.53; conservative low-risk g = 0.18) when judged by adequately controlled, lower-risk designs, whereas the broader literature overstates effectiveness because underpowered, higher-risk studies exert disproportionate influence. Keywords: social-emotional learning, primary school, meta-analysis, effect size, risk of bias.]]>
