Study Design, Participants, and Procedures
This paper draws on data collected under two research projects nested to the Nairobi Urban Health and Demographic Surveillance System (NUHDSS). These are the Transition-To-Adulthood (TTA) project and the Education Research Program (ERP). In 2008, the NUHDSS mid-year population was 59,570 people living in 24,100 households located in Korogocho and Viwandani slum settlements in Nairobi city. The NUHDSS, TTA and ERP have ethical approval from the Kenya Medical Research Institute’s ethical review board. In addition, all research staff, fieldworkers, and data processors are trained on research ethics. For all studies, potential respondents are first briefed on the study objectives and then invited to participate. Respondents are requested to give verbal or signed consent; for respondents aged 12–17, consent is also requested from their parents or guardians.
The Transition-To-Adulthood project
TTA is a component of the 5-year Urbanization, Poverty and Health Dynamics research program conducted by the African Population and Health Research Center. The TTA’s general objective is to identify protective and risk factors in the lives of adolescents growing up in these two informal settlements in Nairobi and to examine how these factors influence their transition to adulthood. Adolescents were randomly selected within the households in the study area using records of residents in the NUHDSS for the year 2007. Allowing for an annual attrition rate of 16% for Korogocho and 24% for Viwandani, and given the planned 3-year follow-up, 2,478 and 3,028 randomly selected young people were targeted for recruitment from Korogocho and Viwandani, respectively. Between October 2007 and June 2008, about 4,058 (75% response rate) adolescents aged 12–21 were interviewed. A structured questionnaire was administered by interviewers and included questions covering reproductive aspirations (e.g., parenthood, marriage); key health and other concerns (e.g., worry about HIV/AIDS, getting a job, marriage, finishing school, employment); living arrangements and nature of interactions with parents, guardians, teachers, and peers; involvement in youth groups (e.g., religious and social groups); and involvement in risky behaviors (e.g., early sexual debut and delinquency). The complete questionnaire was translated from English to Swahili and administered in Swahili, the language most spoken in the study area.
The Education Research Program
This is a longitudinal study designed to compare educational outcomes between two slum settlements (Korogocho and Viwandani) and two non-slum communities (Harambee and Jericho) in Nairobi city. The ERP has been interviewing all children aged 5–21 years since 2005 using five modular interviewer-administered questionnaires that collect information on household characteristics, school characteristics, school enrolment, and children’s behavior.26 Information on adolescent sexual and other risk behaviors is collected as part of the module that assesses children’s schooling status and experiences, as well as informal training and apprenticeships. The behavior section of the module is completed by respondents aged at least 12 years, and the section must be completed with the child as the respondent. By December 2008, a baseline survey and four waves of data collection were completed by the ERP. Wave 4 was collected from December 2007 to August 2008. Details of the sample design and other survey procedures are available elsewhere.26
Description of the Merged Sample
As both the ERP and the TTA are nested to the NUHDSS, it is possible to merge information collected around the same time from the same individuals by the two studies. The merged file would contain detailed information on risk and protective factors from the TTA and details relating to schooling and substance use from the ERP. We merged data from the TTA and ERP Wave 4 collected between October 2007 and August 2008 using the unique identification numbers that are assigned to all residents in the NUHDSS. Overall, 2,028 respondents aged 12–21 years were found in both the ERP and the TTA databases. In order to provide a better comparison with other studies, the adolescents aged 12–19 years were selected and since involvement in sexual relations was used as one of the measures for problem behavior, we excluded adolescents who were or had ever been married. The final sample had 1,722 never married adolescents. To rule out ‘selection bias’, the characteristics of this sample were compared to the larger ERP and TTA primary samples and were generally comparable for several selected characteristics (gender, age, slum location, parental co-residence, education status). The age group 12–19 is wide and covers adolescents at markedly different stages of their maturation. This was evident in the differences in prevalence of problem behaviors by age. Hence, adolescents were grouped into two age cohorts (12–14 and 15–19 years) for this study. Socio-demographic characteristics of the sample and the prevalence of problem behaviors are presented in Table 1.
Socio-demographic variables used in the analysis were: age in years (continuous); sex (male and female); household size; study site (Korogocho and Viwandani); duration of stay in the study area; parental co-residence (staying alone, with both parents, with one of either parent, or with other relative or non-relative); and schooling status (in school versus out of school). Socioeconomic status was assessed using a three-category wealth index (least wealthy, middle, and most wealthy) constructed using household assets and amenities collected through the NUHDSS in 2007. These included asset ownership (e.g., radio, television set, motorcycle, mattress, kerosene lamp, phone, and sewing machine), building materials (floor material, roof, wall material) and availabilities of amenities (water supply, electricity), etc. Principal components analysis was used to construct the socioeconomic index.27
About 53% of the participants were living in Korogocho and about 47% resided in Viwandani. More than half of all the adolescents reported living with both parents (56%), while 28% were living with one of the parents and 16% were living alone or with other people (relatives or non-relatives). As expected, the proportion of adolescents living with no parent was higher in the older age cohort. About 78% of the adolescents indicated that they were currently in school, with the vast majority of the younger cohort (95.6%) versus close to two-thirds of the older ones being in school. Of the older cohort (more or less secondary going ages) who were in school, 54% were in secondary school.
With regard to problem behavior, few adolescents had ever been pregnant or made someone pregnant (2.3%), while about 6.4% reported having had sexual intercourse before reaching 15 years. Older adolescents reported higher levels of involvement in early sexual activity, ever being pregnant or having made someone pregnant, and drinking alcohol than the younger ones (see Table 1).
Measuring Problem Behavior
A composite eight-item Multiple Problem Behavior Index (MPBI) was constructed assessing delinquent behaviors (three items), early sexual experience (one item), illicit drug use (two items), alcohol consumption (one item), and tobacco smoking experience (one item). Although premarital sexual behavior may not be a problem behavior per se, early sexual activity is problematic because of the adverse health and socioeconomic consequences associated with it, and there are societal pressures to preserve young people’s virginity until marriage or as long as possible. Campaigns to promote abstinence and discussion of health consequences of early initiation of sex have especially been highlighted in the widespread HIV prevention programs for adolescents. Therefore, early sexual intercourse was included as a component of the MPBI, with early sexual experience defined as sex before the age of 15 years. This arbitrary cut-off age is a good representation of the median age at first sex for young people growing up in slum settlements.21 Using this cut-off, early sexual experience among adolescents aged less than 15 years was equivalent to “Ever had sex (yes/early sex =1, No/early sex = 0)” while for those above 15 years the measure was “whether sex was before reaching 15 years (early sex = 1) or after 15 (early sex = 0) or never had sex (early sex = 0)”.
Measuring Protective and Risk Factors
The three types of protective factors (models, controls, supports) and the two types of risk factors (models and vulnerability) were constructed as composite measures. Opportunity risk, the third type of risk factor had very low variability and was, therefore, dropped from the analysis. Controls protection was measured as two separate sub-composites, one assessing personal (individual-level) controls, and the other informal social controls or social regulation. For adolescents not in school, items in reference to school-related controls or protection were inapplicable to them and were coded as zero at analysis stage. Alpha reliability was used to assess the internal consistency of items for each composite measure. A composite score for each type of theoretical predictor was constructed using standardized values of the individual items in each scale (see Table 2). All the resulting predictors were standardized (to have a mean equal to zero and standard deviation equal to one). This was necessary in order to enable reasonable interpretations of any possible moderator or interaction effects that might emerge in the analyses. As can be seen in Table 2, the alpha reliabilities of the explanatory measures are all acceptable (Cronbach’s alpha > 0.6), and for four of the six measures reliabilities are good (Cronbach’s alpha > 0.7).
Data were analyzed using STATA version 10.28 Univariate statistics were computed to describe the respondents’ socio-demographic characteristics by age cohort. To assess the linear relationships among the theoretical predictors as well as with the problem behavior outcome measure, correlation coefficients were computed. The outcome and theoretical predictors were assessed for the assumptions of normality. The MPBI measure was skewed to the right and a natural log transformation of (MPBI + 1) was applied to normalize the distribution of this outcome. Hierarchical linear regression methods were then used to assess the applicability of Problem Behavior Theory by modeling the relation of the theoretical predictors, as well as their interactions, to the log-transformed MPBI outcome measure. First, the log-transformed MPBI outcome measure was fitted by including only the socio-demographic variables as predictors. A second model was fitted by including the socio-demographic variables (as controls) and then adding the four protective factor composite measures. A third model was fitted by now adding the two risk factor composite measures to the first model. Finally, a fourth model was fitted by adding all significant interactions between the four protective and the two risk factor composite measures to the third model. These four models were fitted separately for the two age cohorts. The results were then back-transformed to reflect the true relationship between the predictor measures and MPBI that are presented in the tables.