Displacement in Somalia: Exploring the Direct and Indirect Relationships Between Social Networks and Livelihoods Outcomes

2026-06-15

2026-12-31

1. The size and density of women's social networks
2. Access to childcare - both actual access over the previous month, as well as hypothetical access (willingness to ask others for help; perception that they would receive assistance with childcare if requested)
3. Reduction in stress and improvement of psychosocial wellbeing
4. Increased labor force participation

This study evaluates a low-cost, low-intensity intervention -- facilitated "Mothers' Groups" -- designed to increase social network size and social capital among internally displaced mothers in Somalia. The study is structured as a randomized controlled trial conducted across major IDP-receiving areas in Hargeisa and Kismayo.

Eligible participants are mothers aged 18–50 who have been displaced within the previous three years due to conflict, drought, or related environmental and economic challenges, and who have at least one child between 10 months and 3 years of age. Recruitment occurs through two channels: outreach at health clinics providing antenatal or postnatal care, and direct household recruitment in IDP areas using a random-walk methodology with a short screening survey.

The treatment consists of 13 semi-monthly group sessions over approximately 6.5 months. Groups are targeted at 11 members and are geographically clustered to minimize travel. Sessions last roughly two hours and include informal group discussion, informational sessions on household management and childrearing, and mother-child interactive play. Mothers attend with their children, as part of the program's goal of modeling interactive, play-based parenting.

The study uses an oversubscription design. All recruited mothers complete a baseline survey before learning their assignment, which serves both to collect pre-treatment data and to function as a costly signal of interest, encouraging less motivated recruits to drop out before randomization. This protects internal validity by concentrating attrition in the pre-randomization phase. Post-randomization, participants who drop out are not replaced, with one exception: if a group falls below eight active members within the first four sessions, additional mothers are added to preserve the social dynamics essential to the intervention. These replacements are excluded from the analysis. Randomization is conducted at the individual level within geographic clusters, using block randomization on mother's age, child's age, and recency of displacement.

The study examines treatment effects along three causal pathways. First, a direct pathway through expanded social networks, hypothesized to improve information-sharing about livelihood opportunities and job referrals. Second, an indirect pathway through childcare access -- as network size and trust grow, mothers gain access to reliable informal childcare, enabling labor force participation. Third, an indirect pathway through mental health -- social support is expected to reduce stress and depression, improving participants' capacity to pursue employment. The study measures both intermediate outcomes (network size and strength, childcare access, mental health) and final outcomes (employment, self-employment, and hours worked).

Data will be collected through baseline and endline surveys supplemented by higher-frequency surveys conducted during group sessions, capturing granular labor market data such as weekly hours worked. The pilot sample includes 176 treatment and 176 control participants. The primary estimands are the intent-to-treat effect and the complier average causal effect, estimated using difference-in-differences models. A detailed pre-analysis plan will be developed following the baseline survey.

Not available

Assignment of women to treatment occurs at the individual level, within geographic areas. We will define a “recruitment cluster” after mapping all recruits; for the sake of illustration, we can imagine a recruitment cluster to consist of 35 recruits who live within 1 mile (1.6 kilometers) of one another, though the size of this area may need to be adjusted depending on recruitment density. The baseline survey will be conducted with all recruited mothers in the cluster.

We employ an oversubscription design, in which 1) being recruited and 2) participating in the baseline survey constitutes the application typically employed in such designs. In other words, participating in the baseline when selection into treatment is explicitly uncertain, signals a strong interest in participation that we believe will reduce attrition amongst both the treatment and control groups. Control group participants will also be informed that they will have first priority for inclusion in future rounds of implementation, and will be provided – at the time of the endline survey – with small tokens of appreciation from the program, specifically a set of 4-5 Somali-language children’s books that feature in the mother’s groups. This should reduce attrition from the control group over time.

Following the baseline, we will block-randomize into treatment and control on the basis of mother’s age (binned to the extent necessary to allow matching within a small per-cluster sample, likely 5-year bins), age of the target child, and recency of displacement. We opt for block-randomization, as opposed to pairwise randomization due to concerns about attrition within our relatively small sample. We propose randomization on mother’s age and recency of displacement because both are predictors of at least one of our outcomes, social capital: women at the upper end of the childbearing age range have stronger social networks by some measures and are more likely to know of issues in their community that require social capital ; IDPs who arrived more recently tend to have narrower social networks within their new locations. The age of the target child is relevant for blocking in part because we are planning a parallel study of child cognitive outcomes with same cohort of women.

We do not propose blocking directly on the outcomes of interest for three reasons. First, we have multiple outcomes of interest and their correlation structure is unknown. Even leaving aside the aforementioned parallel study, the degree of correlation between stress, social network size, and engagement in the labor force is unknown. Second, several of our outcome measures (e.g., social network size, stress) are likely to be continuous, or ordinal variables with multiple levels, which makes blocking in small samples more difficult. Third, we expect a high degree of noise in our outcome measures and we plan to measure most outcomes in multiple ways; blocking on a single of those measures risks reducing precision for measures that may prove more valuable or informative at the time of analysis.

Individual (mothers of young children)

Yes

32 groups of approximately 11 mothers each. Note that treatment assignment is randomized at the individual level, but treatment involves the formation a group and is therefore clustered in terms of application

352

16.