Peer Effects in Socio-Emotional Learning

Within each class, non-seed students are randomly assigned to small groups, and each group is assigned one seed student.
Treatment: Groups in the treatment condition are assigned a high-achieving seed student
Control: Groups in the control condition are assigned a median-achieving seed student.

2026-04-27

2028-12-31

social emotional skills, mental health, student's academic performance.

The study uses data from the trial registered as AEARCTR-0018314. Academic performance is measured using administrative test records. Social-emotional skills and mental health are measured using standard instruments, including the CES-D and WCSD, complemented by student reports on relationships with parents, teachers, and peers.

1. compliance of group assignment, group interaction quality,students’ reflections on role models.
2. other non-cognitive skills, and relationship between parents, peer and teachers, parental time inputs

Part 1 is new, not avaliable in baseline. We collect measures of group interaction quality at follow-up to explore potential mechanisms. This scale will cover dimensions such as support, group cohesion, task orientation, interpersonal trust, and role modelling.
Part 2 uses data from the SEL program survey.

It is embedded in an SEL program, whose full details are available in a separate registry (AEARCTR-0018314). The curriculum naturally involves group-based activities, such as role play, discussions of emotions, and peer sharing.
We leverage this setting by using baseline administrative test scores to identify 601 seed students, comprising approximately one quarter of students in the treatment schools. Treatment seeds are drawn from the top 12.5% of the baseline test-score distribution, while control seeds are drawn from around the 50th percentile.
The remaining non-seed students are randomly assigned to 601 small groups of approximately four students each. Within each class, groups are then randomly assigned to either the treatment or control condition: treatment groups are joined by a high-achieving seed student, while control groups are joined by a median-achieving seed student. This design creates experimental variation in peer composition and allows us to identify how exposure to higher-achieving peers shapes students’ engagement and social-emotional skill formation during the intervention.
When the required number of seed students is odd in a class, the additional seed is selected from the high-achieving seed pool. As a result, non-seed groups in these classes have a slightly higher probability of being assigned to the high-achieving seed condition.
The design is one-sided blinded. Neither teachers nor students are informed about the seed selection or group assignment process. Teachers are only told during training that the SEL curriculum involves group activities, that students have been organized into groups, and that they should seat students according to the provided group lists before each course session.

Not available

Randomization done in office by a computer

at the group level, with four students per group

Yes

74 schools, 601 peer groups

about 1750 non-seed students, and 601 seed students

The sample includes 279 control groups and 322 treatment groups.
Because some classes contain an odd number of groups; in these cases, the group is randomly assigned to the high-achieving seed condition with slightly higher probability.

With 322 treatment groups, 279 control groups, 3 non-seed students per group, and an ICC of 0.07, the minimum detectable effect is approximately 0.13 SD at 10% significance and 80% power. With baseline controls explaining 20% of outcome variation (non-cognitive outcomes), the MDE falls to 0.11 SD; with controls explaining 50% of outcome variation (test scores), it falls to 0.09 SD.