The Evidence on Teaching

The Impact of Enhancing Students' Social and Emotional Learning: A Meta-Analysis of School-Based Universal Interventions

Durlak JA, Weissberg RP, Dymnicki AB, Taylor RD, Schellinger KB · 2011

grade Cmeta-analysisdeveloper-involvedfailed
Sample
213 interventions, 270,034 students, kindergarten through grade 12, mostly United States
Population
K-12 school students, predominantly US.
Design
The source of the +0.27 achievement figure that underwrites the entire SEL policy edifice, and the design details are rarely quoted with it. Only 99 of 213 interventions (47%) were randomised. Implementation was not reported at all in 43% of studies. Most primary studies did not correct for clustering; the authors re-analysed and one of 24 headline findings lost significance. On the authors' own accounting, 53% of the outcome data is CHILD SELF-REPORT and only 51% of measures were rated valid. Independence is developer-involved at field level - Weissberg is co-founder and president of CASEL, the advocacy body for the intervention class under evaluation, and the paper is distributed on CASEL letterhead with him as contact.
Key findings
Grand mean effect 0.30 (95% CI 0.26 to 0.33). By domain - SEL skills 0.57, attitudes 0.23, positive social behaviour 0.24, conduct problems 0.22, emotional distress 0.24, and ACADEMIC PERFORMANCE 0.27 (0.15 to 0.39), the famous number, resting on just 35 of the 213 interventions. Within that, standardized achievement tests gave 0.27 and grades 0.33. At follow-up of six months or more, available for only 33 studies (15%), effects were attitudes 0.11, prosocial 0.17, conduct 0.14, distress 0.15 and academic 0.32 (k = 8). The often-quoted eleven-percentile gain is a U3 translation of the 0.27.
Genetic confound
Medium. Slightly over half the pool is non-randomised, and the largest effects are on self-reported constructs measured by the children receiving the programme.
Replication notes
Failed and attenuating. The achievement figure falls to 0.109 in Cipriano et al. 2023 and 0.101 in Ha et al. 2025, both produced by the SEL field itself, and it has never been reproduced by a large independent trial with an externally marked test - the two that tried returned -0.106 to +0.026 (PATHS) and 0.03 (Good Behaviour Game).
DOI / URL
10.1111/j.1467-8624.2010.01564.x

Effects

OutcomeMetricValueMeasureTimingVsHorizonClass
Academic performancemean effect size0.27 (95% CI 0.15 to 0.39), k = 35 of 213 interventionsmixedpost-interventionbusiness-as-usualend-of-treatmentdomain-skill
SEL skillsmean effect size0.57 (0.48 to 0.67), k = 68 — largely researcher-designed and child-reportedresearcher-designedpost-interventionbusiness-as-usualend-of-treatmentnon-cognitive
Conduct problemsmean effect size0.22 (0.16 to 0.29), k = 112mixedpost-interventionbusiness-as-usualend-of-treatmentbehaviour
Grand mean across all outcomesmean effect size0.30 (0.26 to 0.33)mixedpost-interventionbusiness-as-usualend-of-treatmentnon-cognitive
Follow-up at 6+ monthsmean effect sizeacademic 0.32 (k = 8), conduct 0.14, attitudes 0.11 — available for only 15% of studiesmixed6 months or morebusiness-as-usual1-2yrnon-cognitive
Proportion of outcome data that is child self-reportproportion53%; only 51% of measures rated validresearcher-designednot applicablenonenot-applicablenon-cognitive

Cited by