The health, human capital, and economic impacts of school-based mental health intervention among adolescents in Colombia

This project is being conducted in public schools in the urban areas of Medellín and Barranquilla, Colombia's second- and fourth-largest cities, focusing on students in 9th, 10th, and 11th grades. It is a school-based mental health intervention targeting students experiencing mild psychological distress, as identified through the Strengths and Difficulties Questionnaire (SDQ) screening. All students with parental consent and who provide assent complete the screening questionnaire. Based on established SDQ cutoffs and eligibility criteria, approximately 30% of screened students are expected to be eligible for the intervention.
The study employs a two-stage randomization design. In the first stage, classrooms are randomly assigned to either a treatment arm or a pure control arm, stratified at the school level: approximately 85% of classrooms are assigned to the treatment arm and 15% to the pure control arm. Schools with four or fewer classrooms are assigned entirely to the treatment arm. In the second stage, within each treatment classroom, eligible students are individually randomized, with 50% assigned to receive the intervention and 50% assigned to a within-classroom control group. No individual randomization occurs in pure control classrooms; all eligible students in those classrooms serve as the pure control group.
Among those assigned to treatment, students receive counseling sessions delivered by trained non-specialist mental health counselors. The intervention consists of a low-intensity, transdiagnostic, problem-solving therapy adapted from the PRIDE program (Michelson et al., 2020; Malik et al., 2021), delivered through five one-on-one 45-minute sessions following a standardized protocol that all counselors are trained on. During these sessions, students are guided to identify problems, explore potential solutions, and develop concrete implementation strategies.

2026-05-11

2027-06-30

To measure the impacts on adolescents' socioemotional skills, their problem-solving skills, and mental health, we use the following validated instruments:
Strengths and Difficulties Questionnaire (SDQ) Total Difficulties Score (Goodman, 1997)
Self-Reporting Questionnaire (SRQ) Total score (Beusenberg et al., 1994)
Social Problem-Solving Inventory - Revised (SPSI-R) Total Score (D'Zurilla et al. 2002)
In addition to analyzing the impacts on each of these three scales, we will also create a summary index of those scales, following Anderson (2008).

To measure impacts on academic achievement, we will use the following administrative data obtained from schools and/or the secretaries of education:
School attendance (subject to data availability)
Grade repetition (administrative records)
GPA (administrative records)
School dropout and reenrollment status (administrative records)
Saber 11 standardized test score (overall and by subject)
Tertiary education enrollment and persistence (SNIES administrative records)

To measure long-term economic impacts, we will use administrative data on:
Formal employment (PILA administrative records)
Earnings (PILA administrative records)
Mental health diagnoses (RIPS health claims records)
Crime-related outcomes (administrative records, under exploration)

The RCT employs a two-stage randomization design, designed to identify both the direct treatment effect of Clara-Mente and potential spillover effects on untreated peers. This structure reflects a deliberate trade-off following Baird et al. (2018): our primary objective is to estimate the within-classroom treatment effect, which equals the ITT in the absence of spillovers, with maximum precision, while still being able to detect whether there are spillover effects. To balance these goals, we randomize at two levels. In the first stage, classrooms are assigned to one of two conditions: pure controls or treated; in the second stage, eligible students within treated classrooms are individually randomized to treatment or control. The treated share concentrates statistical power on the direct effect, while the pure-control classrooms provide the clean counterfactual needed to identify spillovers on non-treated students who share a classroom with treated peers.

1. Classroom-Level Randomization: Classrooms are randomly assigned to either a treatment arm or a pure control arm, stratified at the school level. Approximately 85% of classrooms are assigned to the treatment arm and 15% to the pure control arm. Schools with four or fewer classrooms are assigned entirely to the treatment arm.

2. Student-Level Randomization within Treatment Classrooms: In the second stage, within each classroom assigned to the treatment arm, eligible students are individually randomized: 50% are assigned to receive the Clara-Mente intervention (treatment group) and 50% are assigned to the within-classroom control group. This randomization is stratified at the classroom level.

Trained mentors are randomly assigned to schools and students using stratification criteria based on school and classroom.

Not available

Randomization is done in the office by one of the PIs or Research Assistants on the computer.

Two-stage randomization:
At the classroom level, with strata defined by the school.
Within each classroom assigned to treatment, eligible students are the unit of randomization. They are individually randomized.

Yes

The first stage of the treatment is clustered at the classroom level, the second stage is not clustered as it is an individual level random assignment. Number of Clusters: 455 classrooms in 64 schools

3,343 out of 11,142 screened

Classroom level (Stage 1):
~391 treatment classrooms (85%, plus schools with ≤4 classrooms assigned entirely to treatment)
~64 pure control classrooms (15%)
Student level (Stage 2):
Within treated classrooms ~1,129 students control and ~1,129 students treated
In pure control classrooms, ~365 pure control students

Analysis Plan