Optimizing Cash Transfers with Social Empowerment Interventions to Reduce Child Marriage in Bangladesh

BRAC's Social Empowerment and Legal Protection (SELP) program in Bangladesh has launched a two-year (2023-2025) multi-layered intervention targeting 2,400 villages in 240 sub-districts from 30 districts. We are conducting the study in 450 cluster/villages from 40 upazillas of 5 districts. This program aims to tackle child marriage by empowering adolescent girls (aged 13–17) and their families.

The intervention has two key components:
Shopno Sharothi (SS): Each cluster village has a team of 20-25 girls called "Shopno Sharothi (aka the Dreamer)". This girl-led initiative focuses on:
1. Developing life skills for the girls.
2. Fostering teamwork and problem-solving abilities.
3. Raising awareness about the negative consequences of child marriage and advocating for its prevention.

Conditional Cash Transfers: Families with girls aged 15-17 can receive conditional cash transfers for up to two years (or until the girl turns 18). The amount varies (low or high cash) and depends on the girl remaining unmarried. This financial incentive aims to discourage child marriage by providing economic support to families.
The reason for multiple cash transfer options is that it allows us to see if the size of the cash transfer affects the program's effectiveness. Even within the low cash transfer group, some teens receive the cash while others only get the program, allowing us to study spillover effects within the program group.
This program has used a multi-component approach to address child marriage. By combining social empowerment through life-skills training with conditional cash transfers, the program aims to be more effective than single-pronged interventions.

2023-08-01

2025-06-30

Child marriage and Education

In child marriage, we will measure age at first marriage, the likelihood of being married at the age of 17, at the age of 16 and at the age of 16 and at the age of 15. The likelihood of being married at a specific age will be measured as a dummy variable, with a value of 1 if a girl is married at the age of 17, and 0 otherwise.

In education, we will see enrollment, attendance, and academic performance. Academic performance will be measured by the GPA acquired in the last exam.

Parents' attitude towards child marriage, communication, trust, and mutual understanding of adolescents with parents, parents' perceived risks of elopement, incidence of elopement in the community, parents' perception of return to education.

To measure parents' attitudes towards child marriage and communication, trust, and mutual understanding of adolescents with parents, a set of statements (strongly agree/agree/strongly disagree/disagree/neutral) will be asked, and then an index will be calculated using principal component factor analysis.
To measure communication, trust, and mutual understanding of a girl with her parents, we will ask the girl and her parents whether she agrees/disagrees with statements such as "I can express my thoughts to my parents without hesitation or fear."
For parents' attitude towards child marriage, a set of statements such as "Girls who leave school early have better prospects of getting married well".
To measure parents' perception of the return to education, questions will be such as "Suppose, hypothetically, your daughter is twelve years old and were to complete her primary school, and then stop attending school. Think about the kinds of jobs she might be offered and that she might accept. How much do you think she will earn in a typical month when they are about 30 to 40 years old?"
To measure parents' perceived risks of elopement, we will ask the parents “According to you, how concerned are you that your daughter will love and marry of your choice?” We will ask the same questions to the adolescents to check the validity of the measurement.
To measure the incidence of elopement in the community, we will ask about the actual incidence to the parents, girls, community leaders, and Kazi (who conducts local marriages). We will ask multiple stakeholders to get a rough idea of the actual incidence.

We did the randomization at two stages. In the first stage, we randomly selected 30 upazillas (subdistricts) out of 40 upazillas that are eligible to receive low cash and the rest 10 for high cash. In the 30 upazilas, there were 10 eligible villages identified for the interventions and in the second group of upazilas there were 15 villages. In the second stage, we randomly selected 4-5 cluster villages that have received "SS Only" (T1), 3-4 cluster villages that have received "SS + Low Cash" (T2), and 2 villages a control group (C1). For the other 10 upazilas, we selected 10 cluster villages from each for "SS + High Cash" intervention (T3) and 5 villages as control (C2). The reason for first-stage randomization is because of the programmatic constraints of addressing potential backlash by implementing two different sizes of cash transfers within the same upazila. So, we have a total number of villages in each of the arms as follows:

T1. SS only = 140 villages
T2. SS + High cash = 100
T3. SS+ Low Cash = 100 villages
C1. Control (from upazilas with low cash) = 60 villages
C2. Control (from upazilas with high cash) = 50 villages

This design with multiple arms and randomization helps control for external factors that might influence the outcomes and isolates the true impact of the SELP program on child marriage and gender norms. In each of the 450 cluster villages, on average 25 adolescents are selected (overall N=11250), and the targeted adolescent girls will be surveyed to measure the impact of this program. Our randomization was accomplished based on the BRAC’s listing (The listing was conducted on March 2023) data which consists of a sample of 11,250 girls. However, we randomly selected 6,730 girls for the baseline survey that was conducted from September to October 2023.
A midline survey will be conducted after the completion of the program and an endline survey will be undertaken one year after the completion of the program in all 450 cluster villages.

Not available

Randomization is done in a computer using STATA software.

Cluster randomization. Cluster unit is village level.

Yes

450 clusters (villages/communities)

11250 adolescents

T1. SS only = 140 villages
T2. SS + High cash = 100
T3. SS+ Low Cash = 100 villages
C1. Control (from upazilas with low cash) = 60 villages
C2. Control (from upazilas with high cash) = 50 villages

The purpose of the power calculations was to provide us with an approximation of the sample and effect sizes, even though the number of clusters vary across treatment arms. Taking the percentage of girls who get married before the age of 18 as an outcome indicator, a sample of 2,000 adolescents from 100 cluster villages (20 adolescents per cluster village) is adequate to detect a change of 8.2 percentage points (MDE) relative to the mean of the control group with a statistical power of 90% and an ICC of 0.10. In other words, if 50% of the targeted girls from the control cluster villages get married, we are 90% confident to capture a reduction to 41.8% or lower for any of the interventions.