Lighting Up Home to Boost Learning and Livelihoods in Refugee-Hosting Communities in Northern Uganda

The intervention involves the distribution of off-grid, solar-powered LED lanterns to households with a P7 student in treatment schools. Each lantern charges during the day from an attached 3.5 W solar panel, provides up to roughly 6 hours of light on high mode (and far longer on lower modes), and includes a USB output capable of charging a mobile phone. The lantern is a durable, household-level capital input that relaxes the binding constraint of darkness after sunset and improves the quality of available light relative to kerosene lamps and flashlights. The study covers four World Vision Area Programmes in the West Nile Sub-region of Uganda. These area programmes span refugee settlements and their surrounding host communities. Forty primary schools operating under these programmes form the study frame.

Of the 40 schools, 22 are randomly assigned to the treatment arm and 18 to the control arm. In treatment schools, the intervention is offered to eligible households with a P7 student, subject to the available number of lanterns per school. On average, 75 lanterns are provided per treatment school (approximately 1650 units). The evaluation survey sample is a subset of this intervention population: approximately 40 P7 students are sampled per school for detailed household and student surveys. Thus, lantern distribution is implemented at the level of eligible P7 households in treatment schools, while impact measurement focuses on the randomly selected index P7 students and their households. The intervention is delivered to families rather than to children directly, because the lantern introduces light at the household level; targeting families avoids unnecessary complications at the school level. The study follows a waitlist design: control schools receive lanterns (approximately 1,350 units) after the endline survey, so that all 40 schools ultimately benefit, with a total of 3,000 lanterns distributed.

2026-07-15

2027-02-25

Nighttime study time of the index P7 student (hours per day, from a time-use recall/diary).
• Homework completion rate of the index P7 student.
• School attendance of the index P7 student (days present / enrolled days; administrative and self-reported).
• Learning performance: standardized term test scores and the national Primary Leaving Examination (PLE) score from administrative records (UNEB), expressed in standard-deviation units.

Primary Outcomes Explained
Nighttime study time is the most proximate behavioral outcome through which the lantern is expected to affect learning. It is measured from a structured 24-hour time-use recall for “yesterday” administered to the index student (and corroborated by the caregiver), recording activities in the evening and early-morning hours when natural light is absent. We construct total nighttime study hours and, separately, total nighttime productive hours to distinguish study from other uses of the newly available time.
Homework completion is measured as the share of assigned homework the student reports completing, complemented by teacher assessment where available. School attendance is constructed from school registers (primary source) and household report, expressed as the fraction of school days attended over the reference period.
Learning performance is the study's ultimate human-capital outcome. We use (i) standardized scores on term examinations administered during the school year and (ii) the national PLE score obtained from administrative records. The project registers all sampled P7 students for pre-mock, mock, and national examinations and collects individual score data. Test scores are standardized to the control-group distribution so that effects are interpretable in standard-deviation units. Because increased study time need not translate mechanically into learning—its effect is mediated by the quality of study time, school quality, parental support, and possible displacement of sleep—study time and learning are analyzed jointly to characterize the full chain from time use to attainment.

Secondary outcomes are organized into pre-specified families:
• Education (additional): oral reading and numeracy fluency, teacher-assessed effort, and independent-study ability.
• Time use: total nighttime productive hours, domestic/chore hours, paid-work hours, and a child-labour indicator, for the index student, adults, and coresident children.
• Livelihoods and income: adult evening work hours, household income, and total household lighting expenditure.
• Health: respiratory symptoms, eye strain/irritation, and burn injuries.
• Energy and environment: kerosene/candle consumption and expenditure, lighting source mix, and attitudes toward solar adoption.
• Psychosocial and intrahousehold: student self-efficacy and exam anxiety, girls' safety perceptions, women's bargaining/decision-making, and intrahousehold control over the lantern.
• Intergenerational and spillovers: time and quality of child-rearing, and benefits accruing to neighbors and coresident/extended-family children through the non-exclusive nature of the lantern.

Secondary Outcomes Explained
These families capture the broader chain of mechanisms in the conceptual framework: expansion of the production frontier (time use, income), allocative re-optimization across day and night, intrahousehold and gender dynamics, and the health and environmental consequences of displacing kerosene. Coresident children are surveyed because lantern benefits are likely to extend to siblings and cousins, which is socially desirable. Outcomes within each family are combined into standardized indices (following Anderson, 2008) to limit the number of hypotheses tested and to improve power, with individual components reported alongside the index.

The study involves a school-level (cluster) randomized controlled trial. Forty primary schools with P7 classes are randomly assigned to a treatment arm (22 schools, lanterns in Phase 1) or a control arm (18 schools, lanterns in Phase 2 after the endline). Randomization is at the school level rather than within school because P7 enrolment per school is small—typically a single P7 class—so splitting students within a school into treatment and control is infeasible and would invite contamination.
Stratification. The 40 schools are stratified by (a) Area Programme (geographic area) and (b) distance from the refugee settlement border, classified into three groups: Group A, schools far from the settlement border (host community, distant); Group B, schools close to the settlement border (host community, proximate, increasingly enrolling refugee students as funding cuts reduce staffing at camp schools); and Group C, schools located within refugee settlements (camp schools, which also enroll Ugandan nationals). Groups A and B are host-community schools and Group C are camp schools; all three are represented in the final sample. Stratification on these dimensions improves balance and supports the pre-specified refugee-vs-host heterogeneity analysis.

Not available

Randomization is carried out using a Stata code with a documented seed for full reproducibility. Within strata defined by Area Programme and distance to host/refugee community schools are ordered and assigned to treatment or control by pseudo-random draw, preserving the overall balanced 22:18 treatment-to-control allocation across the 40 schools. The field team remains blinded to assignment until after baseline data collection is complete (allocation concealment). The final study frame and randomization are confirmed at the end of May 2026, once Term 2 enrollment stabilizes. If fewer than 40 P7 students are enrolled in a school, all listed P7 students are included in the evaluation sample. The student-level sampling is for measurement only and does not affect the school-level treatment assignment. In treatment schools, lantern distribution follows the intervention eligibility rules for P7 households, while in control schools, eligible households receive lanterns after the endline under the waitlist design.

Treatment is assigned at the school/cluster level; all sampled P7 students (and their households and coresident children) within a school share the same treatment status.
The number of clusters/schools will be 40 (22 treatment, 18 control). The planned number of observations is 1,600 index P7 students (40 per school), comprising approximately 800 students in treatment schools and 800 in control schools, plus approximately 6,200 coresident children of the sampled students (assuming about four coresident children per index student), and the associated 1,600 households. The school/teacher survey covers all 40 schools.

Yes

The intervention will be implemented in 40 randomly selected primary school clusters, 22 of them assigned to receive lanterns in Phase 1 and 18 schools to the waitlist control group, which will receive the lanterns after endline.

Of the 40 schools, 22 are randomly assigned to the treatment arm and 18 to the control arm. In treatment schools, the intervention is offered to eligible households with a P7 student, subject to the available number of lanterns per school. On average, 75 lanterns are provided per treatment school (approximately 1650 units). The evaluation survey sample is a subset of this intervention population: approximately 40 P7 students are sampled per school for detailed household and student surveys. Thus, lantern distribution is implemented at the level of eligible P7 households in treatment schools, while impact measurement focuses on the randomly selected index P7 students and their households. The intervention is delivered to families rather than to children directly, because the lantern introduces light at the household level; targeting families avoids unnecessary complications at the school level. The study follows a waitlist design: control schools receive lanterns (approximately 1,350 units) after the endline survey, so that all 40 schools ultimately benefit, with a total of 3,000 lanterns distributed.

Survey sampling involves households with a Primary 7 (P7) enrolled student in one of these school clusters. Forty P7 students per school are randomly sampled from each of the treatment (40*22=880) and control (40*18=720) school clusters yielding a planned sample of approximately 1,600 index P7 students across the 40 schools.

Under plausible assumptions about intracluster correlation, attrition, and the explanatory power of baseline outcomes and covariates, the design is expected to be best powered for primary outcomes measured for the full index-student sample, including nighttime study time, homework completion, attendance, and standardized learning outcomes. Power will be lower for outcomes with high intracluster correlation, outcomes measured in smaller subsamples, and subgroup or heterogeneity analyses. Table 1. Minimum Detectable Effect Sizes for Some Plausible Intracluster Correlation Coefficients. Intracluster correlation (ρ) MDE, no baseline control (SD) MDE, ANCOVA R²=0.50 (SD) 0.05 0.24 0.17 0.10 0.31 0.22 0.15 0.37 0.26 The preferred specification is an ANCOVA model controlling for the baseline value of the outcome where available, randomization strata fixed effects, and pre-specified baseline covariates. For learning outcomes, the baseline learning assessment is expected to improve precision relative to a post-treatment comparison without baseline controls. Under the preferred ANCOVA specification, the design detects effects of approximately 0.17-0.26 standard deviations of learning outcomes at 80 percent across the plausible ICC range of 0.05 – 0.15 (the range for learning outcomes reported in similar studies) (Table 1). Because the number of clusters is modest, inference will account for clustering at the school level and will rely on randomization-inference p-values and wild-cluster bootstrap p-values, alongside conventional cluster-robust standard errors.