Study setting and participants of the study
This cross-sectional study was conducted among high school adolescents in Bahir Dar city administration and in Mecha District, West Gojjam Zone, Amhara Region from November12-28/2016. There are eight high schools in Bahir Dar and six high schools in Mecha District. Two schools providing 9 – 10 grades education were randomly selected from each site and sections were selected randomly from each school. The study population was randomly selected from the four high schools proportionate to the student population size.
The sample size for the study was determined using single population proportion formula considering 50% proportion, 95% confidence interval with 5% margin of error, design effect (2) and 10% non response rate. The final sample size was 844 adolescents.
Study procedure
The study received ethical approval from Ethical Review Committee of Bahir Dar University. We initially developed the survey questionnaire in English and then translated into Amharic to ease understanding. Prior to the study a pretest was conducted among 35 students in a school not selected for the study and the necessary adjustment in language and content was done.
One week before data collection, we communicated principals in the selected schools with formal letters and we selected students who would participate in the study. For selected participants below age 18, we obtained consent from parents/guardians through letters wrote by the principals in the selected schools and verbal assent of individual participants was obtained after being fully informed of the study purpose and procedures. From study participants aged 18 and above, we obtained verbal consent. We ensured confidentiality by removing all personal identities from the questionnaire. At each school, the questionnaires were self administered in a free classroom in the opposite shift without the presence of teachers. The principal investigators and two research assistants informed the participants carefully about the study and were available throughout the administration of the questionnaires to answer questions from individual students.
Variables and measurements
Background variables
Background variables included age, sex, residence, living arrangement with parents/guardians, and substance use. These variables are reported influence adolescents sexual behavior [11, 12, 21].
Living arrangement: respondents were asked with whom they are living during the last 12 months before the survey. The responses alternatives were with both biological parents, with single parents (with either mother or father), and with others (relatives, friends, guardians).
Substance use: We collected information regarding alcohol drinking and chewing khat to measure adolescents’ substance use behavior. Alcohol consumption was measured using the following item: Do you ever drink any form of alcohol during the last 12 months? Response alternatives were never, sometimes (about once a month), and quite regular (every week). Adolescents’ khat use status was also measured by asking the following item: Do you ever chew khat during the last 12 months? The responses were never, sometimes (about once a month), and quite regular (every week).
Network characteristics
Four items - network size, network tie strength, network homogeneity, and network members’ sexual norm (approval or disapproval of sexual practice) were included in the questionnaire to measure respondents’ network characteristics [22].
Network size
As stated above, the ego is the respondent who directly participated in the survey research, while the alters are people (named by the respondent using the name generator question). Network size is the number of alters named by the ego, who have discussed about sexuality (sexual partner, sexual practice, condom use) during the last 12 months preceding the survey.
Higher number of alters named by the ego indicates higher number of network size.
Network tie strength
Using the name interpreter question, respondents were asked to rate how close they felt to each member of the network mentioned using a three point scale (distant = 1, close = 2, and very close = 3). Respondents would rate their relationship with the network member very close when they had closest contact, met more often, discussed secrets, and supported each other as they wish. They would rate the relationship as close when they met occasionally and they felt that they were friends. They would rate the relationship as distant when the contact happened if necessary. Such approach helps to define one’s closest network members [13]. In addition, a visual display of concentric circle that has three circles was used for illustration to help respondents locate their network members in one of the circles. For example, they would locate the very close network member in the inner most circle and distant network members in the outer most circle. The total score of the measure was then calculated, with higher score indicating embedded within strong tie network.
Network homogeneity
Using the name interpreter question, respondents were asked whether the network members mentioned in the name generator question were kin or non-kin (non-kin = 0, kin = 1). The sum of valid responses divided by the number of valid response was used to compute the summary measure of network homogeneity of the respondents. The value ranges from 0 – 1with higher score indicating embedded within homogeneous tie network.
Network members’ sexual norm
Respondents were asked about the opinion/approval of their network members about sex and sexual practice. The response options were dichotomous (0 = not approved, 1 = approved). Then the summary measure of sexual norm score was obtained by dividing the sum of valid responses by the number of valid responses. The value ranges from 0 to 1, with higher score indicating embedded within networks that approve sexual practice.
Sexual behavior
We adapted three items from a previous study to measure sexual behavior of adolescents [8]. We asked whether the respondents had engaged in sexual intercourse during the past 12 months, the number of sexual partners during the past 12 months, and if the respondents consistently used condom during sexual intercourse. Those having more than one partner or not consistently used condom were considered to be in risky sexual behavior. Then, those engaged in risky sexual behavior were coded “1” and the remaining “0”.
Data analyses
All returned questionnaires were checked for completeness and consistency of responses manually. After cleaning, the items were coded and entered, in to SPSS for Windows versions 23 for analyses.
The dependent variable in this analysis was sexual behavior (coded 1 = risky and otherwise 0). Independent variables included both individual-level variables and network-level variables (network size, network tie strength, network homogeneity, and network members’ sexual norm). Thus, multivariate logistic regression models were used to assess the relationship between independent variables and outcome variable. Nested models were employed to show the unique contribution of social network variables to the understanding of outcome variable. The first was a reduced model including only individual-level characteristics. The second model added network variables to the individual model. Since the first model was nested in models2, we used a likelihood ratio statistic (G2) to test whether the addition of network variables significantly improved the fit of the model [23, 24].