Caregiver Mental Health and Early Childhood Development in Conflict-Affected Settings: A National-Scale RCT of Semillas de Apego in Colombia

Young children and their caregivers in conflict-affected and displaced communities remain one of the most underserved groups in global health and development. While the long-term consequences of early adversity are well established, especially for children under five, few interventions have focused on addressing the immediate and interrelated needs of caregivers and young children in humanitarian settings.
Despite growing recognition of the importance of early childhood in emergencies, scalable and evidence-based interventions remain rare. This gap is especially acute in programs that integrate caregiver mental health and child development, even though a large body of research shows that children’s outcomes are deeply shaped by caregivers’ mental health and their capacity to provide nurturing, sensitive care.

Semillas de Apego addresses this gap by treating caregiver mental health not only as an outcome, but as a mechanism to protect early childhood development. The intervention moves beyond psychoeducational models by creating a relational and reflective space led by trained community facilitators. These facilitators—caregivers themselves—draw on shared experiences to foster a relatable, stigma-free environment. This approach directly counters stigma, builds trust, and supports participation in settings where mental health services are limited and where there are also demand-side constraints for accessing these services.

The program is implemented through a 15-week group-based model with approximately 15 caregivers per group. The program follows four sequential goals: (1) promoting awareness of and tools for emotional regulation of the psychological consequences of conflict and forced displacement; (2) understanding children’s emotional needs and the caregiver-child relationship; (3) developing responsive caregiving practices that foster secure attachment; and (4) restoring social support networks and agency. The approach integrates psychosocial support with attention to the social determinants of caregiving and mental health.

Semillas de Apego offers concrete changes to how ECD and mental health services can be delivered in humanitarian contexts. It aligns with WHO and UNICEF calls for integrated child and caregiver support and demonstrates the potential of task-shifting and community-led delivery to close persistent mental health gaps. Its cultural adaptability, training and supervision models, and local ownership also make it transferable to other low-resource and emergency settings.

For this trial, the program will be implemented in 105 neighborhoods in 12 municipalities, with each neighborhood having an implementation group in the second semester of 2025 (treatment group) and an implementation group in the second semester of 2026 (control group).

2025-08-12

2025-11-21

We will evaluate the program using a rich toolkit of psychometric scales and observational assessments. Based on the program’s theory of change, we identified four main constructs as primary outcomes or constructs: (i) primary caregiver’s mental health, (ii) quality of the child-caregiver relationship, (iii) type of child-caregiver interactions, and (iv) child socio-emotional development. Potentially, conditional on identifying a valid measurement tool, we will also assess the impacts on young children's mental health.

Data will be collected at three points: baseline, one month prior to the implementation of the program’s first cohort; and first and second follow-ups 1 and 12 months after baseline, matching the evaluation design of the original RCT.

We will administer all scales at each wave of data collection. However, we will only conduct the observational assessments of the child-caregiver relationships and child socioemotional development at the final one-year follow-up because of the complexities and costs of conducting field work with the caregiver and the child in this scenario. Further, we have administered all of the instruments in previous evaluations, except for the OMCI, allowing us to compare the results from this trial with those of the small-scale RCT. We plan to adapt and pilot the OMCI and conduct further adaptations of the IDELA to enhance their validity.

We acknowledge that, as in previous evaluations, there may be no significant impact on caregiver mental health at the first follow-up, as the program may simultaneously increase emotional awareness—potentially leading to higher reported symptoms—while also reducing the frequency of these symptoms. To assess potential effects on emotional awareness, we will administer the Emotional Awareness Questionnaire (EAQ; Rieffe et al., 2007), as described below under secondary outcomes. We will also track intermediate outcomes such as emotional regulation, parenting stress, and the strength of caregivers’ social networks, which are hypothesized to mediate the effect of the treatment. At the second follow-up, we will consider including the Strengths and Difficulties Questionnaire (SDQ; Goodman, 1997) to measure emotional and behavioral difficulties as well as prosocial behavior, as it offers a more comprehensive and developmentally appropriate assessment. In addition, we will also consider using children’s drawings, analyzed with LASSO machine learning, as a cost-effective and non-invasive observational tool to assess mental health (Baird et al. 2022). We will also consider using emotion-detecting software to analyze facial expressions and/or voice recordings to capture changes in emotional states.

Because we will administer different scales within a single construct, we will follow Kling et al. (2007) and conduct a latent factor analysis to construct a construct-specific index. Our main results will focus on the four indices or family of outcomes, and then we will unpack the results across individual scales and subscales. We will adjust for multiple hypotheses only for the main indices and will consider the analyses of individual scales and subscales as secondary.

To capture both the short-term and medium-term effects of the intervention, we will conduct two rounds of follow-up data collection– two months after the program ends and another eight months later.

Below we describe the specific scales or assessments that fall within each construct.

1. Primary caregiver’s mental health: First, we will measure caregiver mental health using the Symptom Checklist-90-Revised (SCL-90-R), which captures commonly assessed symptoms such as anxiety and depression while also allowing us to measure phobic anxiety, hostility, and interpersonal sensitivity—dimensions particularly relevant in contexts of ongoing conflict and trauma. We will focus on five subscales reflecting symptoms experienced in the past 30 days: (i) anxiety, (ii) depression, (iii) phobic anxiety, (iv) interpersonal sensitivity, and (v) hostility.

2. Quality of the child-caregiver relationship: We will use three scales reported by the caregiver and an assessment tool to measure the changes in the quality of the relationship between the child and the caregiver. First, we will use the parenting stress index (PSI, Abidin (2012)) to measure three major domains of stress: child characteristics, parent characteristics, and situational/demographic life stress. We will analyze a composite measure of this scale and the individual sub-scales for the following domains: (i) parental distress, (ii) parent-child dysfunctional interaction, and (iii) difficult child.
Second, we will use the Child-Parent Relationship Scale (CPRS), which is a self-reported instrument completed by the caregiver that assesses the caregiver’s perceptions of their relationships with their children and captures both positive and negative aspects of the parent-child relationship, including conflict, closeness, and dependency.
Third, we will use the Parent Behavior Checklist (Fox, 1994), where the caregiver is asked about the frequency in which they engaged in different activities with their child over the last couple of weeks. We will use the subscale on nurturing practices, which are linked to positive socioemotional interactions between the parent and child (Carneiro et al., 2024). Finally, we will use the Observation of Mother-Child Interactions (OMCI, Rasheed and Yousafzai (2015)) as an assessment, whereby the enumerators will assess the quality of interactions between the primary caregiver and the child. We will pilot OMCI extensively to ensure that it is culturally valid, feasible in terms of implementation, and well-understood by the enumerators.

3. Style of the child-caregiver interactions: We will use two additional subscales from the Parent Behavior Checklist (Fox, 1994), where the caregiver is asked about the frequency with which they engaged in different activities with their child over the last couple of weeks. These evaluate discipline practices, encompassing both positive and harsh disciplinary approaches. We will conduct a latent factor analysis to construct two scales: (i) negative discipline index, and (ii) positive discipline index (Carneiro et al., 2024).

4. Child socioemotional development: We will assess child socio-emotional development using age-appropriate evaluations that incorporate structured, play-based tasks. We will use the International Development Early Learning Assessment (IDELA), designed for children aged 3 to 6, to assess five developmental areas: motor skills, emergent literacy, emergent numeracy, social-emotional development, and executive functions. We will focus on social-emotional development and executive functions.

Additionally, we will supplement these assessments with the Brief Toddler Socioemotional Assessment (BITSEA, (Briggs-Gowan et al., 2004)), a caregiver-reported scale that evaluates children’s social-emotional development, including behavioral problems, developmental delays, and deficits in social-emotional competence.

Secondary outcomes include caregivers’ emotional awareness, measured using the Emotion Awareness Questionnaire (EAQ, (Rieffe et al. 2007)); parenting self-efficacy, measured by the Parenting Sense of Competence Scale (PSOC, (Ohan, Leung, and Johnston 2000)); perceived social support, assessed with the Parental Social Support Scale (PSSS, (Cutrona and Troutman 1986)), and Mental Health Stigma (using a questionnaire developed by oir team). These outcomes will be used to test mechanisms of impact.

In addition, we will examine secondary outcomes including caregivers’ aspirations and expectations for themselves and their children, their attitudes toward gender norms (using modules from the Demographic and Health Surveys, DHS), and their participation in household decision-making and employment. We will also include questions to measure social desirability score following Dhar et al. (2022) and interact our treatment indicator with this score to test whether our results can be explained by social desirability bias.

While we are not well-powered for heterogeneity analysis, we will conduct some explanatory analyses to test for potential heterogeneity in the following dimensions: (i) type of childcare provider, (ii) local multigenerational poverty index, (iii) number of caregivers in the household, and (iv) caregiver demographic characteristics.

Since all caregivers in the evaluation will be eventually treated, for long-run non-experimental evaluation, we plan to match survey information with administrative records such as school census records and use a synthetic difference-in-differences approach to examine potential long-term effects of the program.

The experimental design will follow a staggered (wait-list) design in which we will randomize program access across two cohorts. An early cohort, which will start in August 2025 and run until November. A later cohort will start in August 2026 and run until November of the same year. The former will be considered our treatment group, while the latter will correspond to our control group and allow us to compare the evolution in key outcomes to identify the program’s impact.

We recruited a total of 3,800 caregivers of young children from community meetings in the 105 neighborhoods where the program has been implemented since 2023. Of them, 2,348 were randomized into the evaluation sample, and further randomized with equal proportions into the treatment and control groups. The remaining 1,452 were randomly assigned to the implementation cohorts of 2025 or 2026, but will not participate in the impact evaluation.

We chose the staggered design / waiut-list design for three reasons. First, it requires a smaller sample size than in the cluster-level randomization. Second, it ensures that all participants will eventually receive the intervention. Third, it also allows building relationships with community leaders, facilitators, and partners established since 2022. Excluding entire communities, neighborhoods, or participants from the program would not be ethical in this context and would risk the relationships that have been built by the program with the different communities and partners in the field.

Not available

Random assignment into the implementation-only or evaluation samples, and into the treatment (early treatment) group and the control (later treatment/wait list) group among the evaluation sample was conducted using the Stata 'randtreat' command, blocking at the level of each neighborhood.

Individual randomization into the implementation-only or evaluation samples, and the treatment and control groups at the neighborhood level. This is, neighborhoods are our strata; within each neighborhood, we randomized into each of these groups.

No

The program and evaluation take place in 105 neighborhoods in 12 different municipalities, but the treatment was not clustered. The program will be implemented in all 105 neighborhoods.

The evaluation sample includes 2,348 caregivers randomly assigned in equal proportions to the treatment (early treatment) group and the control (later treatment/wait list) group. The implementation sample includes 3,800 caregivers, including the 2,348 caregivers randomly assigned into the evaluation sample and 1,452 caregivers who were randomly assigned to the implementation cohorts of 2025 or 2026, but will not participate in the impact evaluation.

1,174 caregivers assigned to the early treatment group; 1,174 caregivers assigned to the late-treatment group.

We estimate that we have 80 percent power to detect effects of 0.10 standard deviations for each of the four primary outcomes of interest. These MDEs correspond to between 50 and 75 percent of the impacts observed in the small-scale trial of the program at the second followup. These estimates account for the variance explained by the baseline prognostic variables we will control for in our econometric analyses, including baseline control variables and the lagged dependent variable, measured also at baseline. We estimated the variances explained by these prognostic variables using data from earlier evaluations of the program. Our power calculations, however, do not include the variance that will be explained by neighborhood fixed effects, which, as discussed before, are our randomization strata.