The umbrella initiative for girl’s education: a randomized field experiment in Bangladesh

The intervention involves providing umbrellas to 8th and 9th grade female students from underprivileged backgrounds. Adolescent girls face movement challenges due to social, religious, and geographic factors. Providing umbrellas to the girls would serve multiple purposes: protecting them from rain and heat, and enhancing their sense of security when outdoors. We will provide umbrella to the girls in February 2025, ahead of the summer heat (March-May) and the monsoon season (June-September). As the students primarily commute on foot, they will use the umbrellas while traveling to and from school to protect themselves from direct sunlight and rain. The foldable umbrellas are easy to carry and will be in suitable colors for girls.

There is no study on the use of umbrellas in Bangladesh. Informal discussions with several schoolteachers indicate that many students attend school without umbrellas during the rainy season and summer, primarily due to a lack of umbrellas in their households. The provision of umbrellas is intended to help girls from underprivileged families attend school even in adverse weather conditions. Additionally, the umbrellas will increase the mobility of adolescent girls for other outdoor activities, especially in rural and semi-urban areas where females tend to cover themselves from males in outdoors.

2025-02-01

2025-11-30

1. School attendance: Measured using daily attendance records from schools.
2. Dropout rates: Measured through school records at the beginning of the next academic year.
3. Academic performance: Evaluated using standardized literacy and numeracy test scores before and after the intervention.
4. Seasonal sickness: Collected through surveys by asking questions on weather-induced sickness incidence during the year.

In consultation with the Directorate of Secondary and Higher Education, 500 secondary schools will be selected from a list of nearly 2000. The selected schools will be randomly assigned to either the treatment group or the control group. Randomization will be conducted at the school level to avoid contamination between the groups.Within each selected school, around 12 girls from grades 8 and 9 will be chosen to participate in the study. Selection criteria would be no more than one workable umbrella in the household.

Sample size: The study will involve approximately 6,000 students from 500 secondary schools. This sample size is based on power calculations targeting an effect size of 0.10 to 0.15 with a minimum of 85% power.
Stratification: Schools will be stratified into three categories: urban, semi-urban, and rural to ensure that the sample is representative of different geographical areas and socioeconomic contexts.
Randomization: In consultation with the DSHE, 500 secondary schools will be selected from a list of nearly 2000. The selected schools will be randomly assigned to either the treatment group or the control group. Randomization will be conducted at the school level to avoid contamination between the groups. Within each selected school, around 12 girls from grades 8 and 9 will be chosen to participate in the study. Selection criteria would be no more than one workable umbrella in the household.
•Treatment group: Schools in the treatment group will receive the intervention, which involves distributing foldable umbrellas to the selected girls.

•Control group: Schools in the control group will not receive any intervention during the study period. After the study period the control students will get a similar umbrella (optional).
Umbrella distribution: In association of the local DSHE offices, GDRI will manage the distribution of umbrellas. They will leverage their expertise in community engagement to ensure smooth and efficient delivery of the intervention. The umbrellas will be procured locally by the PI. The umbrellas will be durable and capable of providing protection from both heavy rains and extreme heat.
Timing: The umbrella distribution will take place at the beginning of the academic year (February 2025). This timing ensures that the umbrellas can be used throughout the rainy season and the summer months.
Baseline data collection: Before the intervention, baseline data will be collected in January 2025. This data will include:
•School attendance records: Daily attendance records for the previous year from each students will be collected to establish baseline attendance rates.

•Academic performance: Standardized literacy and numeracy tests will be administered to assess the baseline academic performance of the students.

•Health surveys: Surveys will be conducted to gather information on the incidence of weather-induced illnesses (e.g., colds, heatstroke) among the students.

•Socioeconomic information: Data on household characteristics, including the number of functional umbrellas, socioeconomic status, and parental education levels, will be collected.
End-line data collection: End-line data will be collected at the beginning of the next academic year (February 2026) to measure the impact of the intervention. The same measures used in the baseline data collection will be employed to ensure consistency. Additionally, school admission records in the next year will be reviewed to measure dropout rates.

Randomization will be done by a computer software.

School will be the unit of randomization

Yes

500 schools

6000 female students

250 schools in the treatment group; 250 schools in the control group

Utilizing the Optimal Design Software, we conduct a power calculation to determine the optimal sample size needed for this study. Since we lack pre-intervention data on variables of the students, we are unable to estimate the expected effect size of the intervention precisely. As such, we are conservative in our estimates and hypothesize a low to medium effect size (0.15 - 0.2) in our power calculations, assuming to a modest improvement in student outcomes due to the interventions. Such an effect size is commonly observed in low-cost educational research (Kraft, 2020). Given that the students belong to similar socio-economic backgrounds and come from the same locality, we expect a moderate intra-cluster correlation coefficient (0.25), suggesting sizeable similarity among students within each school. This similarity implies a larger sample size to achieve the desired statistical power. Given the sample size of 6,000 from 500 clusters and expected values for relevant parameters, our study will achieve more than 85% power in detecting an effect of 0.15 to 0.2.