Experiences of violence among female and male grade eight learners: baseline findings from the Girls Achieve Power (GAP Year) trial across three South African townships

Study design and setting

A baseline cross-sectional analysis was conducted in 3432 grade 8 participants enrolled into the GAP Year cRCT, between April 2017 – September 2018. Participants completed a knowledge, attitudes, perceptions, and behaviour survey at baseline. Twenty-six public high schools were selected across three townships (Soweto and Tembisa in Gauteng Province and Khayelitsha in the Western Cape (WC)). Schools were selected using the following inclusion criteria: mixed-sex public high schools in Tembisa, Soweto and Khayelitsha; in quintiles 1–3¹ which had not been exposed to any asset building interventions in the past six months. A one to one (1:1) random stratification scheme was employed, assigning 26 schools to either intervention or control groups.

Sample size justification

The sample size was calculated based on the study’s primary outcome measures, namely dropout rate and increased reporting of GBV among adolescent girls, computed using cluster-randomized size methodology suggested by Hayes and Bennett¹⁹. The effect size of dropout was factored from other similar studies to account for a large conservative and representative sample size to measure outcomes. Based on other local studies, we hypothesize a reduction in drop-out rate from 17.8% as reported by Branson, Hofmeyr²⁰ to less than 14% (estimated effect size of 20%), with an anticipated attrition rate of 5% per year based on a similar local study²¹. This resulted in a conservative sample size of 2730 adolescent girls and 1850 boys to determine the association between intervention and control school’s dropout rate and GBV variables. This was the upper limit for the sample size for the cRCT. For the cross-sectional study, we only included those who consented into the trial.

Study participants

Grade 8 learners were selected for the trial as this grade marks the end of compulsory education in SA. The grade 8 learner age range is approximately 12–14 years; however due to learners repeating grades and other reasons, the age range is commonly wider with older learners enrolled up to 18 years^20,22. All grade eight learners, at selected schools were eligible to participate in the baseline survey, irrespective of sex, age, or race.

Measures

The baseline interview consisted of an interviewer-administered survey collecting information on demographics, socio-economic status and knowledge and attitudes pertaining to school safety, social support and social networks, sexuality, gender and norms, sexual reproductive health and rights, and care-seeking behaviours. Following completion of this component of the interview, the participant was asked to complete a behavioural audio computer-assisted self-interview (ACASI), allowing learners to hear questions through headphones and respond on a tablet themselves, aiming to reduce social desirability bias. The majority did the interviewer administered survey first, although this was not always the case. The ACASI section, 20–30 minutes in duration, asked sensitive questions regarding the participant’s actual practices and behaviour, covering questions on multiple sexual partners, sexual history, substance abuse (alcohol, drugs) and experiences of different types of violence (physical, sexual, psychological, economic, corporal punishment, neglect, and cyberbullying), as well as the perpetrators, and place and time of violence. In the survey²³, respondents were asked “Please specify the type of violence you have experienced (Select those that apply to you)” and were able to select all those types of violence that they had experienced²³. Each of the types of violence experienced were then captured as separate variables in the dataset, as presented in Table 2.

For the purposes of this analysis the single item variables were then combined as a multiple response variable, for both boys and girls. Therefore, out of 621 participants that stated they had ever experienced violence, there were 744 responses to the types of violence experienced, – as some participants had experienced more than one type of violence.

Data management and analysis

The data for the baseline interview was collected and managed using REDCap (Research Electronic Data Capture) electronic data capture tools hosted at Wits Reproductive Health and HIV Institute²⁴. REDCap is a secure, web-based software platform designed to support data capture for research studies, providing 1) an intuitive interface for validated data capture; 2) audit trails for tracking data manipulation and export procedures; 3) automated export procedures for seamless data downloads to common statistical packages; and 4) procedures for data integration and interoperability with external sources.

The completed surveys were stored on encrypted password-protected tablets and the synced data was stored on Wits RHI secured servers. All data from REDCap and ACASI systems were exported as csv files and then into StataCorp 2017²⁵. Descriptive analyses, together with a chi² square test of independence, were used to describe provincial and gender differences in the socio-demographics and patterns of violence ever experienced. The outcome ‘’ever experienced violence’’ was used to estimate associations with other variables. A multivariate logistic regression analysis examined the independent association between experiences of violence vs sex and province, controlling for potential confounders. The bivariate risk and factors associated with violence were estimated using crude odds ratios, at 95% confidence interval (CI) and significance reported at alpha less than 5%. Significant p-values less than 20% from the one-way Analysis of Variance (ANOVA) were used to estimate the adjusted odds ratio at the multivariate level, while variables more than 20% were considered collinear variables. Furthermore, n-values of missing data are shown in the tables, but missing values were not included in the calculation of the percentage breakdown for each variable. All variables show the percentage breakdown and the corresponding p-value for recorded responses only. Participants’ age was presented in groups (12–14 years and 15–17 years), rather than a continuous variable. It should be noted that the “no” category has been removed for all dichotomous variables [variables with a Yes or No Response] shown in the tables in the Results.

Ethical approval and considerations

The study was granted ethical approval from the University of the Witwatersrand Human Research Ethics Community (#M160940). Provincial research approval was obtained from Western Cape Department of Education and Gauteng Department of Education. Learner’s enrolment and participation was voluntary. Only those with written parental/guardian consent and individual assent were recruited into the study. Those who were 18 years and older, did not require parental consent but provided their own written informed consent, enabling them to participate. Initially in English, the survey was translated and back translated into Xhosa, commonly spoken in some study sites. All data collection was supervised by the research team. Interviewer environments were set up to ensure confidentiality. Where feasible, interviewers were the same sex as the participant. Participants could stop the interview process at any time and were free to refuse to respond to any question(s) they felt uncomfortable answering. Social workers were employed to provide psychosocial support to participants during data collection and study intervention and a social harm form was developed to facilitate prompt referrals, where needed. To ensure meaningful participant and stakeholder engagement, the study was guided by the Good Participatory Practice guidelines¹⁸.

Socio-demographics

Overall, 3432 eligible learners across 26 schools participated in the baseline survey: we included 2383 in the analysis as this was the number who completed both components of the survey. In some cases, due to lack of time for participants to complete both components at the same time, some learners were unable to complete both components. Table 1 shows the socio-demographic characteristics of the study respondents by sex and province: 63.1% (n=1504) were girls, the majority (81.5%, n=1938) were aged 12–14 years, and 96.9% (n=2309) were Black African. Approximately one-fifth (18.5%, n=440) were aged 15–17 years, 76% (n=337) of who have repeated a grade. Overall, 41.4% (n=967) of respondents resided with both parents and 68.8% (n=1 633) of learners had at least one parent/guardian employed. A similar proportion (67.3%, n=1498) lived in households that were recipients of a Government grant. At the time, 18.3% (n=436) of learners had ever had sex, with 14% (n=65) of these reporting their first sexual experience <10 years of age.

A higher percentage of younger respondents aged 12 – 14 years were females (85.5%, n=1 279) and from Gauteng Province (GP) (84.3% n=1075) compared to males (75.1, n=659) and those in the WC (78.3%, n=863)(p<0.001). WC respondents were more racially diverse, with 6.2% identifying as Coloured, compared to 0.4% in Gauteng (p<0.001). More WC respondents reside with both parents, compared to GP respondents (43.7% vs 39.4% respectively, p=0.002). More respondents’ parents are employed in WC (73.3% vs 64.9% respectively, p=0.001) and more receive a Government grant, compared to GP (74.1% vs 61.8% respectively, p=0.001). Those in the WC and males had a higher percentage of those who had ever had sex: 24.5% WC vs 13% GP (p<0.001), and 31.7% males vs 10.4% females (p<0.001). There were statistically significant differences in the age of sexual debut between females and males, with more males having sex under 10 years old (18.0% males vs 6.4% females, p<0.001).

Sex and provincial differences in experiences of violence

In total, 25.9% of all learners had ever experienced violence (n=617) (Table 2). This was higher among boys than girls (34.3% vs 21.1%, p=<0.001) and higher in WC compared to GP (28.3% vs 23.9%, p=0.014). Of these, physical violence was most common (35.7%, n=276), followed by psychological violence (21.8%, n=169), sexual violence (13.1%, n= 101), neglect (10.6%, n=82), cyberbullying (7.9%, n=59), corporal punishment (6.5%, n=50), and economic abuse (4.8%, n=37). Physical violence was more common in boys, although still high in girls (36.0% vs 35.3%). Psychological violence was more common in girls, compared to boys (22.2% vs 21.4%). The WC had more physical violence than GP (41.0% vs. 30.4%), as well as psychological (24.8% vs. 18.9%), but GP had over double the reports of sexual violence (18.4% vs. 7.6%), respectively (p=<0.001). Of those who reported ever experiencing violence (n=621), 15.3% (n=95) indicated having ever experienced more than one form of violence (polyvictimisation).

Overall, school was the most common place (30.7%, n=272) where violence was experienced followed by the park² (17.5%, n=155) or at their friends’ home (12.9%, n=114). There were statistically significant sex differences in the places of violence, boys were more likely to experience violence at school compared to girls (36.0% vs 26.3% p=0.001). Girls were then more likely to experience violence in the park compared to boys (20.2% vs 14.3% p=0.001). There were no statistical differences across provinces in place violence was experienced.

Most violence took place during the day (57%, n=377), followed by evening/at night (32.4% n=213) and 10.1% (n=66) in the morning. The majority of those who had ever experienced violence knew the perpetrator (66.7%, n=414), while the rest (n=207) preferred not to say. Most violence was perpetrated by peers (38.4%, n=168), followed by strangers (21.1%, n=92), parent/guardians (11.9%, n=52) or siblings (8.9%, n=39). After peers, boys were more likely to experience violence from a stranger (25.8% vs. 15.9% respectively, p=0.001). Table 2 shows the baseline experiences of violence by sex and province.

Factors associated with experiencing violence

Table 3 outlines the factors associated with ever experiencing violence, all seven variables have statistical significance at the unadjusted level: sex (OR 1.97 CI 1.62 – 2.35), age group (1.82 CI 1.45 – 2.26), province (OR 0.79 Ci 0.66 – 0.95), ever used substances (OR 2.24 CI 1.82 – 2.75), ever had sex (OR 0.46 CI 0.36 – 0.57), ever repeated a grade (1.52 CI 1.24 – 1.84) and feeling worthless (OR 1.36 CI 1.12 – 1.64).

Boys (p<0.001), those aged 15–17 years (P<0.001), living in WC (p=0.014), those who had ever used substances (p<0.001), ever had sex (p<0.001), those who had ever repeated a grade (p<0.001) and those who felt worthless (p=0.001) were more likely to ever experience violence. Racial group (p=0.707), household structure (p=0.786), parent/guardian employment status (p=0.389), government grant recipient (p=0.854), place of violence (p=0.278) were not found to be statistically associated with ever experiencing violence and therefore they are excluded from Table 3.

When adjusted, the following five variables remained significant: sex, age, ever used substances, ever had sex and self-worth. The adjusted multivariate analysis showed that boys were 1.57 times more likely to have ever experienced violence (aOR 1.57; 95% CI 1.27-1.94; p<0.001) than girls. Those aged 15–17 years were 41% more likely to have ever experienced violence (aOR 1.41; 95% CI1.07-1.84; p=0.013) compared to those 12–14 years. Those who had ever used substances were 92% more likely to have ever experienced violence (aOR 1.92; 95% CI 1.54-2.37; p<0.001) compared to those who had not used substances. Those who sometimes felt worthless were 35% more likely to have ever experienced violence (aOR 1.35; 95% CI 1.10-1.64, 0.003) compared to those who thought their life was worthwhile. Those who had never had sex were 33% less likely to have ever experienced violence (aOR 0.66; 95% CI 0.51-0.83; p=0.001) compared to those who had ever had sex.

Strengths and limitations

When reviewing these findings, the following strengths and limitations should be considered. A strength of this study is that it was conducted in 26 schools in three highly populated diverse townships of South Africa and is therefore generalisable to other South African settings. Another strength is that it provides data on a variety of different types of violence, often excluded in other manuscripts. We also provide updated school violence data, building on previous studies conducted in 2016⁵ and 2012⁶.

Whilst the final sample size was substantially less than originally planned, the results maintain power to draw these conclusions. Data collection relied upon retrospective recollections which may result in an under-reporting of violence⁵⁸. The study was cross-sectional therefore only representing one point in time. There were no questions in the survey to establish recency of violence making it difficult to establish individual incidence of violence or if it was ongoing. The time of day when violence was experienced was not clearly defined and therefore left up to each participant’s interpretation. In addition, there was insufficient power to determine different factors associated with different experiences of violence.

There were many participants who did not complete both components of the survey and were therefore excluded from this analysis. This was linked to the time of data collection: data collection typically took place after school leaving a short period of time for data collection. Some learners did not have sufficient time to complete both components afterschool and were requested to return the following day to complete the second component. In these instances, we experienced challenges in getting the participants back to complete the second component, possibly due to lack of interest or lack of time. To overcome this, we have only included those who completed both components of the survey in the analysis.

These limitations are unlikely to alter our primary findings.

Underlying data

Harvard Dataverse: GAP Year_Violence REDCap and ACASI data, https://doi.org/10.7910/DVN/AHHWNL²³.

This project contains the following underlying data:

Extended data

Harvard Dataverse: GAP Year_Violence REDCap and ACASI data, https://doi.org/10.7910/DVN/AHHWNL²³.

This project contains the following extended data:

Data are available under the terms of the Creative Commons Zero "No rights reserved" data waiver (CC0 1.0 Public domain dedication).