Nairobi sub-study
In Nairobi, all informal settlement sub-locations were identified using population data from the KNBS 2009 national census for Nairobi County. These included: (1) all sub-locations with enumeration areas designated as informal settlements by KNBS in the 2009 National census, as well as (2) all sub-locations with a population density greater than 20,000 people per square kilometre in 2009. The latter were included to ensure coverage of other densely populated areas that were not categorized as informal settlements in 2009. The informal settlement sub-locations were then categorized using KNBS designations of Locations, which are then categorized into Divisions. Areas where significant adolescent girls interventions were already being implemented by the study investigators (Viwandani, Korogocho, and Kariobangi locations), or funded by the same donor (Embakasi division), were excluded.
After excluding high socioeconomic-status areas and areas that did not have a sufficient number of adolescent girls to reach the target (based on 2009 Census data), Kibera was identified as the primary research site. In Kibera Division, seven locations were classified as Urban Slums: Kibera, Lindi, Makina, Silanga, Laini Saba, Soweto/Highrise, Gatwikira, and Olympic. Central division (Huruma sub-location in Huruma location and Mlango Kubwa sublocation in Mathare location) was selected as the external control site due to similarity with Kibera on key characteristics such as parental education, adult employment status and religion [
17].
In Nairobi, a complete household listing, to identify eligible girls between the ages of 11 and 14, was conducted in both Kibera and the external control Huruma/Mathare, using maps obtained from KNBS and with the assistance of local leaders. The listing was conducted using Open Data Kit (ODK) software on Android tablets. Enumerators assigned a unique serial number to each household, and if an adult was available they administered a brief screening questionnaire to ascertain whether there was an eligible girl residing within the household. The preferred respondent was the household head, but if unavailable, the second choice was a spouse of the household head, and the third choice was another consenting adult (age 18 or over) residing in the household. Similar to the Demographic and Health Surveys (DHS), a household was defined as one that shares a kitchen (pot) and has the same household head. Household members were defined as individuals who have lived or intend to live in the household for 6 or more months, including school children regularly in residence during the school year (even if they spend time away during school holidays).
A screener was conducted to identify girls in the target population. It had five questions about household demographics but only one (How many girls age 10–15 live in this household?) was used to determine whether the household would be asked further questions. The wider than necessary age range (10–15) was used to ensure eligible girls were not missed due to misreported ages. All households that answered one or more to this question continued to complete a household roster and a brief household survey. The roster asked about all household members and included questions on age, birth year (for adolescents), education, marital status, parents’ survivorship, and number of living children. The household survey included questions on ownership of household assets, source of drinking water, type of toilet facility, number of rooms for sleeping, and the main material of the floor and the roof of the structure.
A total of 64,946 unique households were listed in Kibera and 18,578 households in Huruma/Mathare. The screener was completed with a consenting adult in two-fifths (40 % Kibera; 43 % Huruma/Mathare) of these households, with less than 1 % refusals. Although not completed in the remaining households, the field protocol was designed to ensure age-eligible girls were not missed. In particular, a research assistant visited each household three times and if no one was available inquired from neighbours as to whether there were children between the ages of 10 and 15 years residing in the household. Households with children in that age range were revisited before the research team proceeded to a different enumeration area (or later) such that the number of uninterviewed households with age-appropriate children is likely to be minimal. In Kibera, the listing resulted in 5134 households with at least one girl between the ages of 10 and 15 years, and 4351 girls between the ages of 11 and 14 years. In Huruma/Mathare, the listing resulted in 1348 girls between the ages of 10 and 15 and 1166 girls between the ages of 11 and 14 years.
In Kibera, a total of 4351 girls were determined to be between the ages of 11 and 14 at the time of the listing (based on reported or calculated age). Of these, 611 girls were ineligible because they were immediately identified as being in boarding school (and thus not resident during school term) or residing outside of Kibera for school or other reasons. They were excluded because the aim was to obtain a sample of eligible girls according to the program criteria (i.e., girls who were resident and therefore would be available to participate directly in the program interventions). Of the remaining 3740 girls, one girl per household was randomly selected for the sample, resulting in a sample of 3296 girls in distinct households with 444 siblings (or other girls in the same household) within the target age range. In Huruma/Mathare, a total of 1166 girls were between the ages of 11 and 14, and 153 girls were residing in boarding school or outside of the study area. Of the remaining 1013 girls, one girl per household was randomly selected for the sample, resulting in a sample of 895 girls in distinct households and 118 siblings (or other girls in the same household).
In Kibera, out of the target sample of one randomly selected girl from 3296 distinct households approximately 21 % were confirmed to be ineligible during baseline data collection and could not be successfully replaced by another eligible girl in their same household, based on corrections to their age, enrollment in boarding school (which was not asked during the listing) or no longer being in residence in the community. Of the remaining 2606 eligible girls, 2402 (92 %) were interviewed. In Huruma/Mathare, of the eligible 895 girls from distinct households approximately 18 % were confirmed to be ineligible during baseline data collection and could not be successfully replaced by another eligible girl in the same household, based on age, residence in the community, and being in boarding school. Of the remaining 730 girls, approximately 91 % or 666 were interviewed. The reasons for nonresponse in both Kibera and Huruma/Mathare included refusals by the parent, spouse, or girl herself, incapacitation, death, or inability to locate the household or the girl. The research sample included some girls who were age 14 at the time of the listing but had turned 15 by the time they were interviewed. see Table
3.
Table 3
Respondents interviewed
Kibera | 3296 | 2606 (79 %) | 2402 (92 %) |
Huruma/Mathare | 895 | 730 (82 %) | 666 (91 %) |
Wajir | 2923 | 2297 (79 %) | 2150 (93 %) |
Randomization of individuals to study arms was conducted after the baseline survey in the form of a public lottery for transparency and to minimize confusion and distrust regarding the selection process. Girls were randomly assigned to study arms during a Kibera External Advisory Committee
7 meeting attended by local stakeholders and leaders. An Excel file with a list of girls’ anonymous ID numbers was projected on the screen, and an Excel formula used to generate a random number for each girl. The list was then sorted in ascending order of the random number and divided into four equally-sized groups based. Four stakeholders volunteered to randomly pick from a bag one of four pieces of paper, each with one of the four study arms written on it, and this arm was assigned to the particular group.
Wajir sub-study
In Wajir, clusters were defined as settlements with one public primary school or the primary school-catchment area in settlements with more than one. This ensured that girls had access to a primary school and that they had access to group meeting locations. A total of 80 clusters were identified in Wajir and stratified by district: Wajir West (20 clusters), Wajir East (28 clusters), and Wajir South (32 clusters). Wajir North was excluded because the implementing partner does not operate in that area and it was desirable for the study to work with a single implementing partner in each site. Within the selected three districts, approximately 20 communities were excluded including very small villages with fewer than 8 age-eligible girls, urban areas, peri-urban areas with more than one primary school, and villages on the Kenya-Somali border or other villages where the implementing partner had limited access due to security reasons.
In Wajir, a paper-based listing procedure was used to identify eligible girls in the field, as it was not necessary to randomly select girls for program participation due to the cluster-randomization design. Enumerators visited each household within selected villages, assigned it a unique serial number, and conducted a brief screening interview to obtain the name of the head of the household as well as the number of boys and, separately, girls aged 10–15 years residing in the household. The same definitions for a household and household member were used as described above, as well as the same protocol for the respondent. For households with one or more girls between the ages of 10 and 15, a cover sheet was completed in which enumerators listed the name, age, and sex of each girl and boy in that age group. After listing the entire village, the team leaders collected the cover sheets and counted the number of sheets with a girl in the target age range of 11–14.
A total of 4152 eligible girls were identified in 80 villages, ranging from 9 girls to 181 girls per village. Of these, 2923 were selected for the baseline survey. Selection was determined by the number of eligible girls (ages 11–14 years) in each cluster. For villages with fewer than 40 households with an eligible girl, all households were selected for the baseline sample, and all eligible girls within those households interviewed. In villages with 40 or more households with an eligible girl, team leaders used pre-determined lists of random numbers to randomly select 40 households, and then to randomly select one girl within each household for the baseline survey sample. If that girl turned out to be ineligible and there was another eligible girl in the household, the latter was interviewed.
Of the 2923 girls selected for an interview approximately 21 % were confirmed to be ineligible during baseline data collection, mainly based on residence in the community at the time of the survey or incorrect ages. The survey occurred a few days after the listing, and at that time, these parents and guardians clarified that they had listed girls who no longer resided in the household, or who had migrated away from that community. Of the eligible 2297 girls, 2150 (93 %) were interviewed. The reasons for nonresponse included refusals, incapacitation, death, and inability to locate the girl. see Table
3.
Randomization of clusters to study arms was conducted after the baseline survey, at the district level. In each district, a public meeting was held with stakeholders and local leaders, as well as one representative from each of the clusters. A list of all clusters in the district was displayed on the wall, and a container prepared with the same number of cards as clusters in that district and the cards equally divided among the four study arms, i.e., ¼ of the cards indicating arm 1, ¼ arm 2, etc. A representative from each village selected one card from the container publicly announced the arm selected, and pasted it on the wall next to the name of the cluster. After all the clusters had selected an arm, each representative signed an affidavit acknowledging acceptance of the public lottery results.