Internal Vs. External Drivers of Adolescent Violence: Evidence from Brazil

Two programs, each eight weekly activities of about one class period (~50 minutes), delivered over roughly eight weeks to 9th-grade students by trained facilitators in treated schools:

MEN (social norms constraining men and women). Activities on gender stereotypes and masculinities (e.g., "what men are taught to show," debunking gender myths, protecting rather than covering up). Targets the norms and beliefs that sustain violence against girls.
CBT (emotional self-regulation and conflict resolution). Activities on identifying and naming emotions, anger management, relaxation/breathing, and structured problem-solving. Targets the self-regulation and conflict-resolution skills that reduce interpersonal aggression.

Note: both MEN and CBT have one common activity at the end, which discusses violence against women.

Comparison schools receive no activities during the study period and are offered the program in 2027.

2026-09-01

2026-11-30

1) Self-reported violence

We build an index of the following violent acts toward schoolmates in the last 30 days, each measured on a 6-point frequency scale (No times=0, Once=1, Twice=2, Three times=3, Four times=4, Five or more times=5). We construct the index separately for incidence (the student's own perpetration) and for victimization, and, within each, separately by the sex of the other person — an index of violence toward boys and an index of violence toward girls. Each index combines three act types: physical (hit/slap/push), verbal-psychological (insult/threaten/mock), and bullying.

2) Social norms

We measure second-order beliefs in two forms: what peers actually do (descriptive norms) and what peers find acceptable (injunctive norms).
- Descriptive norm (perceived prevalence): what the student believes their peers actually do — "Out of 10 boys/girls in your class, how many do you think have [done the behavior] in the last week?" This captures the empirical expectation of how common a behavior is.
- Injunctive norm (perceived approval): what the student believes their peers consider acceptable — "Out of 10 boys/girls in your class, how many do you think agree it is acceptable for a boy/girl to [do the behavior]?" This captures the normative expectation of whether a behavior is socially approved of, independent of how common it is.

The behaviors are: (a) sharing fears/insecurities with friends (emotional openness), (b) swearing at or hitting a girl, (c) swearing at or hitting a boy, and (d) doing something to please a girl/boy against one's own will (coercive peer pressure). Each is asked separately about boys and about girls in the class.The behaviors are:
(a) sharing fears/insecurities with friends (emotional openness)
(b) swearing at or hitting a girl
(c) swearing at or hitting a boy
(d) doing something to please a girl (for boys)/boy (for girls) against one's own will (coercive peer pressure).

Two-weeks point:
- Self-regulation
- Mental health
- Adherence to masculinity and femininity

We randomize N=40 public schools in Salvador, stratified within each of 10 school regions (Gerências Regionais), into three arms: MEN (15 schools), CBT (15 schools), and comparison (10 schools). Randomization is stratified within each of the 10 school regions, following:

- 8 regions of 4 schools each are randomly assigned to Type A (2 MEN + 1 CBT + 1 comparison) or Type B (1 MEN + 2 CBT + 1 comparison)
- 1 region with 3 schools is split 1 MEN / 1 CBT / 1 comparison
- 1 region with 5 schools is split 2 MEN / 2 CBT / 1 comparison.

This yields 15 MEN + 15 CBT + 10 comparison. 9th graders participate in the activities, and 8th graders respond to the baseline and endline surveys.

Note: one treated school has no 9th grade, so 8th graders will receive the activities instead.

Within-school spillovers. 8th-grade students receive no activities but are surveyed in all 40 schools at baseline and follow-up. We estimate spillover from treated 9th graders to untreated 8th graders in the same school with a difference-in-differences design (comparing 8th-grade outcomes in treated vs. comparison schools, before vs. after).

Network. We collect a rich set of student peer-network measures (friendship, and popularity nominations against a school roster).

Heterogeneity cuts.
- Student sex
- Social-desirability score
- Baseline self-reported violence index (incidence and victimization)
- (Mis)perceptions of the norms (descriptive and injunctive)

Not available

Randomization done by computer in Stata (randtreat), stratified by school region (Gerência Regional), using a fixed seed (the University of Melbourne IRB approval number) for reproducibility.

School (cluster), stratified within each of the 10 school regions. Within treated schools, the intervention is delivered at the 9th-grade level.

Yes

40 schools, across 10 school regions.

≈3,779 students (8th grade: ≈2,119 · 9th grade: ≈1,660) --- assuming ≈75% response rate among all the N = 5,039 enrolled students. 9th grade by arm (treated grade / main sample): CBT ≈710 · MEN ≈563 · Control ≈388 8th grade by arm (spillover diff-in-diff sample): CBT ≈905 · MEN ≈699 · Control ≈515

MEN: 15 schools
CBT: 15 schools
Control: 10 schools