Setting
This study took place in Kilifi County, Kenya. Kilifi was selected for the study as it ranked low in both education and RH indicators: for example, the transition rate from primary to secondary was 40% in 2010 compared to the national rate of 72% (18); further, around 22% of girls ages 15–19 have begun childbearing, as compared to the national average of 18% (19).
In Kenya, the school year starts in January and consists of three academic terms per year. Primary school is from grade 1 through grade 8, and secondary school is from grade 9 to grade 12. While universal primary education for girls has nearly been achieved, there remains significant variation at the county level, gaps in the transition to secondary school and challenges with pupil absenteeism (19, 20). The Government of Kenya has committed to sanitary pad distribution in schools; however, evaluations have shown that supply chains of sanitary pads to schools were not reliable, and girls were not assured of equitable pad provision (21).
Data and Study Design
This study assessed the impact of the Nia Project via a longitudinal, cluster-randomized controlled trial in 140 public primary schools in three rural sub-counties (Magarini, Kaloleni and Ganze) of Kilifi County, Kenya.
Study schools were randomly assigned to one of the following four study arms:
-
Control group
-
Sanitary pads distribution (pads only)
-
Reproductive health education (RH only)
-
Sanitary pads distribution + reproductive health education (combined)
The sub-counties and schools were selected in collaboration with the Kilifi County Department of Education. All non-boarding schools in the three sub-counties with at least 25 girls enrolled in primary grade 7 were eligible. A total of 215 schools were mapped, and a one kilometer buffer was created around each school. For schools with overlapping boundaries, one school was randomly selected resulting in a list of 173 schools. Enrollment and school type were verified for each school in the first quarter of 2017. Based on this exercise, 33 schools were excluded leaving a sample of 140 schools: 44 in Magarini, 50 in Kaloleni, and 46 in Ganze (see Fig. 1). All eligible schools (n = 140) were included in the study.
In schools with 25 girls in grade 7, all girls were included in the research sample. In schools with a larger number of girls, 25 girls were randomly selected for the research sample and five additional girls were selected as alternates. A total of 3,489 girls were interviewed as part of the baseline survey. All grade 7 girls, including those who were not in the research sample and those who had not yet started menstruating, were eligible to receive interventions in order to streamline program delivery. Grade 7 was selected as it would allow for observation of the transition to secondary school within the study timeframe.
The Nia Project included the following two components:
1) Sanitary pads: girls received, on a monthly basis for the entire duration of the project, one packet of ten disposable sanitary pads of ZanaAfrica’s Nia Teen brand. In addition, girls received two pairs of underwear at the start of the intervention, and an additional pair at the end of each subsequent school term.
2) RH education: a 25-session curriculum, Nia Yetu, was delivered by trained facilitators during girls-only health clubs held during time allocated for extra-curricular activities in schools. The curriculum covered a variety of topics including puberty, menstrual health management, reproductive systems, self-esteem, gender, human rights, power dynamics, sexual violence, assertiveness, decision making, relationships, teen pregnancy, STIs and HIV, peer pressure, drug use and conflict management. Girls also received a health magazine developed by ZanaAfrica, Nia Teen, designed to appeal to adolescent girls and convey core RH messaging through storytelling using aspirational personal stories, a relatable comic-style story, and activities. The magazine was distributed at the start of each school term for a five-term period. Each issue corresponded to the topics covered in the Nia Yetu curriculum that term.
Table 1 shows take-up of the two Nia Project components by study arms: girls in the pads only and combined arms received on average 17.5 out of 20 packets of sanitary pads and girls in the RH only and combined arms participated on average in 21 out of 25 RH sessions. Additional details on the Nia Project, theory of change, and study design have been published elsewhere (22).
Table 1
Nia Project uptake Among Girls Menstruating at Baseline and Interviewed at Endline
|
Arm 1
Control
n = 627
(mean (SD))
|
Arm 2
Pads Only
n = 632
(mean (SD))
|
Arm 3
RH Only
n = 629
(mean (SD))
|
Arm 4
Pads & RH
n = 656
(mean (SD))
|
Mean no. of pads received (target = 20)
|
0
|
17.5 (4.2)
|
0
|
17.5 (4.2)
|
Mean no. of underwear received
(target = 6)
|
|
5.5 (1.3)
|
|
5.6 (1.3)
|
Mean no. of NIA magazines received (target = 5)
|
0
|
0.03 (0.2)
|
4.5 (1.2)
|
4.7 (1.1)
|
Mean no. of safe space sessions attended (target = 25)
|
0
|
0
|
20.7 (5.8)
|
21.2 (5.4)
|
Note: Numbers shown in all columns to show potential for direct contamination in program implementation or through girls moving schools after program assignment |
A baseline survey was conducted between January to April 2017, prior to the start of the intervention. Face-to-face interviews were carried out by a trained research assistant in Swahili and data was entered directly onto a tablet. Interviews were held in a private location to assure confidentiality, most commonly in the girls’ homestead or school (after school hours). School attendance tracking was carried out in two phases. First, an initial enrollment exercise took place in June 2017 where all students who were present in school were registered. Second, this registration list was updated at the start of each data collection term. Daily attendance was taken by community-based data collectors for a period of four weeks (20 consecutive school days) per term, starting in September 2017 through July 2018, for a total of 60 days of observation across three school terms. Girls who were registered during the enrollment period were entered as absent if they were absent on that particular day, or dropped out of school/transferred to another school during the observation period. Attendance data was entered as missing for girls who were not registered during the enrollment exercise. The intervention was completed in October 2018 and endline data was collected in November and December 2018 using the same technique as the baseline survey. All girls from the baseline sample were eligible for interview, regardless of schooling status. Figure 1 shows the sample flow by arm.
Randomization and Masking
The unit of randomization was the school. At the completion of baseline data collection in each sub-county, prior to the start of the intervention, public lotteries were held and schools within that sub-county were randomly assigned to one of the four study arms. Interviewers and respondents were blinded to study arm at baseline but not at endline.
Outcomes
Outcomes of interest related to education are: (i) school attendance, which was measured as the number of days a girl was attending school out of a total of 60 days. Mean school attendance was measured only for girls who remained in the same school from baseline to the end of the survey; and, (ii) school engagement measured with a 0–8 score constructed as the number of responses reflecting higher school engagement to eight agree/disagree survey items (e.g., “You are attentive in class”).
Menstruation management outcomes were binary measures including reporting having enough sanitary pads to comfortably manage menstruation and having leaked blood at school during menstruation. Outcomes of interest related to the RH education intervention include: (i) RH attitudes among girls who had started menstruating with a 0–12 score constructed as the number of responses reflecting a positive attitude to twelve agree/disagree survey items which captured girls’ feelings of shame, pride and comfort vis-à-vis menstruation (e.g., “I feel ashamed of my body when I have my period”); (ii) a pregnancy knowledge score with range 0–4 constructed as the number of correct answers to four pregnancy related items; (iii) whether a girl could spontaneously name a modern method of contraception; (iv) STI knowledge score with range 0–4 constructed as the number of correct answers to four STI related items; and (v) a HIV knowledge score with range 0–11 constructed as the number of correct answers to eleven HIV related items; (vi) gender norms in marriage with a 0–5 score constructed as the number of responses reflecting an equitable gender norm to five agree/disagree survey items (e.g., “If a husband and wife disagree on using family planning, the husband’s opinion should come first”); (vii) heteronormativity in adolescents with a 0–12 score constructed as the number of responses reflecting an equitable adolescent gender norm to twelve agree/disagree survey items (e.g., “Girls should be as independent as boys”); (viii) gendered sexual norms with a 0–5 score constructed as the number of responses reflecting an equitable sexual norm to five agree/disagree survey items (e.g., “Girls should cover up or they will attract unwanted sexual attention”); (ix) acceptability of intimate partner violence (IPV) was measured with an indicator on whether a girl finds IPV acceptable in any of five situations; and (x) general self-efficacy with a 0–10 score constructed as the number of responses reflecting self-efficacy to ten agree/disagree survey items (e.g., “You always manage to solve difficult problems if you try hard enough”). See Additional Table 1 for the list of survey items included in each outcome.
The following covariates were measured at baseline to assess balance across study arms: girls’ age, cognitive score with range 0–16 measured from a subset of Raven’s Coloured Progressive Matrices, math test score with a range 0–37 derived from a test including progressively harder problems, literacy score with a range 0–4 derived from reading sentences, using excerpts from the Uwezo Kenya National Learning Assessment (23), household wealth quintile, parental living status, and sub-county.
Sample Size and Analytical Sample
Based on findings from previous studies conducted in Kenya and Ghana (24, 25), sample size calculations were conducted to detect a minimum difference between study arms of 1.18 mean days of school missed over a 4-week period and a 10 percentage points increase in RH attitudes, assuming power of 0.80, significance level of 0.05, intra-cluster correlation (ICC) of 0.173 and a standard deviation (SD) of 3.57. A sample size of 35 clusters per arm and 20 girls per cluster at endline (25 girls per cluster at baseline, assuming a loss of 20% by endline) was needed.
The analytical sample for this paper focuses on the sample of girls who had started menstruating at baseline and were re-interviewed at endline. Estimates including both menstruating non-menstruating girls at baseline are presented in Additional Tables 3–6.
Statistical Analysis
To assess baseline balance across study arms among girls interviewed at endline, means and 95% confidence intervals (CIs) were estimated for the set of covariates described above as well as for outcome variables measured at baseline. An analysis was also conducted to assess bias due to potential differential attrition by study arms.
An intent-to-treat (ITT) approach was used to estimate the effect of each intervention arm relative to the control group. For outcomes measured both at baseline and endline, difference-in-differences (DID) models with girl-level fixed-effects were estimated to compare the change between baseline and endline for each intervention arm relative to the control group. For outcomes with no comparable baseline data available, ANCOVA models were used to compare endline outcomes for each intervention arm relative to the control group while controlling for the following covariates measured at baseline: girls’ age, cognitive score, math and literacy scores, household wealth quintile, parental living status, and sub-county. All regressions were estimated with robust standard errors accounting for clustering at the school level. Statistical analysis was conducted using Stata 14.1.