Gender norms, peer beliefs and socio-emotional education: a multi-site study in Italian lower secondary schools

At each site, an external organization (a municipality, an association or a foundation) delivers a program on emotional and socio-relational competencies to lower secondary school students, in or outside school hours. Programs differ across sites in content, format, setting and dosage, and each is documented in a standardized site sheet (provider, content, number and length of sessions, calendar, enrollment rule) filed with the site amendment before its baseline. The common feature is the aim: promoting socio-emotional competencies, positive relationships between girls and boys, and critical reflection on gender norms and stereotypes.
Site 1 (Collepasso and Tuglie, Lecce). The program “Fuori Bordo. Laboratori per scoprirsi liber*” is promoted by the Municipality of Collepasso with the associations L’Albero delle Parole and Donato Metallo APS, in collaboration with the Istituto Comprensivo Collepasso-Tuglie. It is extracurricular: three laboratories (theatre, photography, mindfulness) held outside school hours in September and October 2026, [number] sessions in total, ending with a public exhibition and performance. The program is presented in the Collepasso unit of the school and students enroll individually and voluntarily, independently of their class. The comparison unit is the Tuglie unit of the same comprehensive school (same head teacher, neighboring town), where the program is not offered during the study period; the program will be proposed to that unit in the following school year. Program period: [first session date] to [last session date].
At every site, all students of both units complete the same printed questionnaire in class, before and after the program, whether or not they enroll. Questionnaires are administered by a research assistant, never by program staff, and are linked across waves by a self-generated anonymous code. Half of the students in each classroom receive the version of the questionnaire with female protagonists and half the version with male protagonists (see Experimental design). To encourage accurate reporting of second-order beliefs, one prediction item is drawn at random after data collection and the four students of each classroom with the most accurate predictions receive a 5 euro voucher.

2026-09-18

2026-10-31

Component 1, baseline survey experiment (all sites, wave 1).
• P1. First-order beliefs: the student’s own agreement (1 to 5) with each character’s action in the two scenarios and with seven statements on control, jealousy, image sharing and physical aggression among peers.
• P2. Second-order beliefs: the student’s prediction of the modal answer of male classmates and, separately, of female classmates, for the same items.
• P3. Misperception: the gap between the student’s second-order belief for a given peer group and the actual modal (and mean) first-order belief of classmates of that gender, computed within classroom excluding the respondent.
Component 2, program evaluation (all sites, wave 2 minus wave 1).
• P3b. In-group and out-group misperception: P3 computed separately for the four respondent-by-target cells (boys about boys, boys about girls, girls about boys, girls about girls). In-group misperception is the diagonal (pluralistic ignorance); out-group misperception is the off-diagonal (beliefs about the other gender).
• P4. Change in first-order beliefs (P1) and in the Gender Stereotypical Traits and Roles scale (Global Early Adolescent Study, 11 items).
• P5. Change in misperception (P3).
• P6. Change in behavioral intentions: agreement with the defending character versus the mocking or controlling characters, and the coded content of the message the student would send in the group chat (coded blind to treatment as challenging, joining, neutral or silent).

Items are standardized within outcome family using the control group and baseline mean and standard deviation, and combined into family indices (equally weighted average of standardized items, following Kling, Liebman and Katz, 2007). Families: (a) vignette judgments, (b) acceptability of controlling and aggressive behaviors, (c) gender stereotype scale, (d) second-order beliefs, (e) misperception. Hypotheses are tested on the family index first and on single items second, with correction for multiple testing within family (Romano and Wolf step-down).
Hypotheses for component 1. H1: judgments of the same behavior differ by the randomly assigned gender of the protagonists; we expect mocking of a boy who cries and control of a girlfriend to be judged more acceptable than the female-protagonist counterparts. H2: second-order beliefs differ by the gender of the peer group referenced, with norms attributed to male peers more permissive of control and aggression. H3: students overestimate the acceptability of controlling and aggressive behaviors among their peers (positive misperception), more so for male peers. H4: out-group misperception is larger than in-group misperception; in particular, girls overestimate the acceptability of controlling and aggressive behaviors among boys more than boys overestimate it among their own group. H4b: within group, misperception is larger among boys than among girls.
Hypotheses for component 2. H5: exposure to the program reduces agreement with stereotype and violence-justifying statements (P4). H6b: the reduction in misperception is larger for out-group beliefs (beliefs about the other gender) than for in-group beliefs, consistent with the mixed-gender setting of the laboratories. H7: exposure increases the intention to challenge mocking or controlling behavior (P6). H8: effects on non-participating students of the treated unit are positive and increase with the share of their baseline friends who participate (spillover).

• S1. Socio-emotional competencies: five items on authenticity, emotional expression and autonomy; sense of control over one’s life.
• S2. Gender identity and self-perception: felt femininity and masculinity, importance of conforming, free associations with “feminine” and “masculine”.
• S3. Appearance-based self-evaluation and beliefs about the role of appearance for women and for men.
• S4. Peer networks: friendship, collaboration, personal-disclosure, participation and admiration nominations; network density, gender homophily and cross-gender ties; change in the position of participants.
• S5. Educational aspirations: desired level of education, ranking of upper secondary tracks, reasons for the choice including same-gender peers.
• S6. Program exposure and awareness in wave 2: participation and number of sessions attended; for all students, having heard of the program, friends who participated, conversations about it (used to measure contamination in the comparison unit).
• S7. Digital and AI tool use (descriptive).

S1 to S5 are analyzed with the same models as the primary outcomes. S6 is used to define exposure and to test the integrity of the comparison unit; if more than 20 percent of comparison students report friends who participated, results are reported with and without this subgroup. Family background (parental education and occupation, division of domestic and care work) is used for covariate adjustment and heterogeneity.

The study is a multi-site, two-wave, paper-based survey in lower secondary school classrooms. Each site consists of a treated school unit, where the program is offered, and a comparison unit of the same school or a closely matched school where it is not. All students of both units are surveyed before and after the program with the same instrument.
Component 1 (randomized). Within each classroom, questionnaires alternate between two versions that differ only in the gender of the protagonists of two group-chat scenarios: in scenario 1 a group mocks a classmate for a gender-atypical reaction (a boy crying at the cinema; a girl shouting at the referee) while one member defends the classmate and another stays silent; in scenario 2 a jealous partner controls the other (a boy checking his girlfriend’s messages; a girl told by her boyfriend not to go out with a male friend) while the chat members justify or challenge it. The version is assigned at the individual level by alternating the pre-sorted pile of questionnaires along the seating order, independently of student gender and of program enrollment. The version is recorded on the class roster and each student receives the same version in wave 2. For each scenario and for the seven common statements, students first predict the modal answer of male and of female classmates (second-order beliefs, incentivized), then report their own answer (first-order beliefs). Classmates are identified by roster numbers read aloud, so that network items can be collected without names.
Component 2 (non-randomized). The program is not randomly assigned. At site 1 the treated unit was determined by the promoters (the Municipality of Collepasso and the associations) before the study; within the treated unit enrollment is voluntary and individual. Participation is defined as enrollment in at least one laboratory; the number of sessions attended (self-reported in wave 2) is used as a dosage measure. Identification relies on a difference-in-differences comparison between units, on parallel trends supported by baseline balance, and on the within-unit contrast between participants and non-participants for spillovers. We state this limitation explicitly and pre-specify the analyses below.

Not available

Questionnaire version: assigned in the classroom by alternating a pre-sorted pile of the two versions along the seating order; no computer randomization. Program: not randomized. The treated and comparison units are pre-existing units of the same comprehensive school; the treated unit was chosen by the program promoters.

Individual student, for the questionnaire version. The program is offered at the school-unit level with individual enrollment.
Questionnaire version: no. Program: yes, at the school-unit level, with individual take-up within the treated unit.

No

Site 1: 2 school units, [8] classrooms (4 per unit). Multi-site target: [5] sites, [10] units, about [40] classrooms.

Site 1: about 180 students (about 90 per unit), each surveyed in two waves. Multi-site target: about [450] students in treated units and [450] in comparison units.

Questionnaire version: about 90 students per version at site 1 (about 45 per version in each unit), about [450] per version in the full sample. Program: about 90 students in the treated unit, of whom about [30] are expected to enroll, and about 90 in the comparison unit at site 1.

All figures are for two-sided tests at the 5 percent level with 80 percent power, in standard deviation units of the control-group baseline distribution, and assume a pre-post correlation of 0.5 in panel outcomes. Component 1 (wave 1, individual randomization, no clustering): 0.42 SD at site 1 (90 versus 90) and 0.26 SD in the full sample (450 versus 450). Component 2, intention to treat at the unit level, allowing for classroom clustering with an intra-class correlation of 0.05 and 22 students per class (design effect about 2): 0.60 SD at site 1 and about 0.27 SD in the full sample. Effect on participants (about 30 enrolled versus 90 comparison students at site 1): about 0.6 SD. With two units, the site-level intention-to-treat estimate at site 1 is informative for large effects only; the study is powered for component 1 at every site and for component 2 in the pooled sample. We will not interpret site-specific null results on component 2 as evidence of no effect.