Correlates of intimate partner violence among pregnant and parenting adolescents: a cross-sectional household survey in Blantyre District, Malawi

Study design

Data analyzed in this study were drawn from a cross-sectional survey of pregnant and parenting girls in rural and urban Blantyre District in southern Malawi. The larger study aimed to understand the lived experiences of PPAs aged 10–19 years. As of 2018, there were nearly 860,000 and 500,000 inhabitants in urban and rural Blantyre, respectively [32]. The rate of childbearing among adolescents aged 15–19 in Malawi in 2016 was 29% [33]. The 2015–16 Malawi Demographic and Health Survey showed that 19% of women in Malawi aged 25–49 had their first sexual intercourse before age 15 and 64% before age 18. The percentage of girls in Blantyre aged 15–19 who had already begun childbearing was 32.1%. Those in this age group who were pregnant with their first child was 4.4%; while 16.5% had already had a live birth [33].

Sample size and selection of participants

A total sample of 669 PPAs of the 679 identified completed the survey; 10 refused to participate in the study. The sample size is sufficient to generate 80% statistical power for all the variables in the study. We based the sample estimation on the following parameters: in 2015, 29% of adolescent girls in Malawi had begun childbearing [33]; the proportion of adolescent fertility in the base population is 0.136 in Malawi; we considered a design effect of 1.5, a relative margin of error (RME) of 0.0325; average household size of five members in Malawi, and 5% possible incomplete responses.

We used a two-stage cluster random sampling to select study participants. In the first stage, we randomly selected 66 enumeration areas (EAs) from the Primary Sampling Frame developed by the Malawi National Statistical Office. Malawi is demarcated into small census clusters called EAs and stratified by urban and rural areas. In the second stage, we conducted a household listing in the selected clusters to identify all households with PPAs. We undertook the household listing to create an updated list of households for all selected EAs so that the sampled households represented the total population. All structures in randomly selected 66 EAs were listed, and households were identified. The listing exercise involved enumerating all household members and recording information on age, sex, and relationship to the household head.

Participants were included in the study if they were aged 10–19, ever pregnant (irrespective of the outcome of the pregnancy), currently pregnant, or had ever had a child regardless of their marital or relationship status, and provided consent to participate in the study. All PPAs identified in the households were eligible for the study. Interviews were conducted by well-trained and experienced research assistants using SurveyCTO installed on Android-powered tablets. The data collection took place between March and May 2021.

Ethical consideration

The University of Malawi Research Ethics Committee (UNIMAREC) approved the study, and we observed all guidelines for conducting research with human participants. Research assistants were trained in research ethics before fieldwork. All participants provided voluntary informed consent after our team availed sufficient information about the study. Permission from parents and guardians was obtained for unmarried minors, while participants assented to participate. We anonymized all the data to protect participants' privacy and confidentiality.

Variables and measurementsDependent variable

The dependent variable was IPV. It was measured using 15 previously validated questions on sexual, physical, and emotional violence used in Demographic and Health Surveys. All PPAs reported whether their intimate partners had said something to humiliate, threaten to hurt or harm, or insult or make them feel bad. Eight questions assessed physical violence, focusing on whether intimate partners pushed, slapped, arm-twisted, hair-pulled, punched, beat up, choked, threatened with a knife, or attacked the respondent with a weapon. Four questions assessed sexual violence, encompassing grabbing or fondling, attempting to have sex against respondent's will, and physically forcing the respondent to have sex or perform sexual acts against their will. These questions combined, and any experience of one was considered IPV (yes coded as 1 and no coded as 0). The alpha coefficient for physical, emotional, and sexual violence were 0.83, 0.72, and 0.86, respectively, indicating high internal consistency among the items.

Independent variables

We included individual, household, and community variables based on the socio-ecological framework on violence against women. The socio-ecological model argues that factors that predispose women to violence operate on multiple levels, including individual, household, and community. Therefore, we considered age, marital status, employment, education, transactional sex, and endorsement of wife-beating at the individual level. Age was coded as 13–16, 17, 18, and 19. Marital status was categorized as single, married, and separated/divorced. Employment status was defined as working for pay or not, while education was categorized as no education, primary and secondary education. However, in the analysis, we combined the no education and primary categories because only one participant had no formal education of the 669 respondents.

We used nine questions to measure transactional sex. These questions touched on engaging in sex for money, food, shelter, school fees, phone/airtime, clothes/shoes/beauty products, sanitary pads, protection, and rent. All the questions demonstrated high internal consistency with a Cronbach alpha coefficient of 0.78. Engagement in sex for any items listed was considered transactional sex. Acceptance of wife-beating was measured using five questions specifying conditions when it is justifiable to beat a wife and probing if respondents endorse wife-beating under those situations. These questions demonstrated high internal consistency with an alpha score of 0.77. The scores were grouped into three, with zero indicating not endorsing wife-beating. A score of one indicated somewhat supports wife-beating, and a score of 2–5 indicated endorsement of wife-beating.

We considered four household-related factors: living with both parents, whether parents are alive or dead, parental support, and partner support (material and emotional provisioning). First, we asked if participants lived with their fathers and mothers and coded responses as living with one, both parents, or not. Similarly, we asked if their fathers and mothers were alive and coded responses as both parents alive, one parent dead, and both dead. Finally, participants rated the support they received from their parents, including material and financial support, as good, fair, poor, or no support. We also asked if participants belonged to a social group they met with regularly.

Three community-level factors were considered: place of residence, neighborhood safety nets, and safety. The place of residence was grouped as rural and urban. Neighborhood safety net was defined as having relationship resources to draw from in the community like friends, adult mentors, and other parents they could turn to if they had serious problems. Seven questions were used to measure safety nets, and all demonstrated high internal consistency (alpha coefficient 0.65). Higher scores indicate more community safety net. Lastly, neighborhood safety was measured using seven questions bordering on the feeling of safety walking around the community during the day and night, feeling scared of being raped, and being touched indecently, robbed, and teased in the past six months in the neighborhood. Higher scores indicate higher community safety.

Statistical analysis

Analysis was performed using Stata 15. We ran descriptive statistics, including means, frequencies, and percentages, for all variables of interest. To answer the study objectives, we fitted multilevel mixed-effect logistics regression models. Given that the factors associated with IPV operate at multiple levels, we used a multilevel logistic regression analysis to estimate covariance at the individual/household and cluster/community level. Multilevel modeling adjusted the estimated standard errors, allowing for the clustering of observations within communities. This means respondents were nested within households and households nested within communities to account for cluster-level effects [34]. Model 1 was a null model with no covariates. We used this model to ascertain if the odds of experiencing IPV vary across randomly selected enumeration areas. Statistically significant intercept shows evidence that IPV exposure varies by EAs. In Model 2, we included individual-level factors such as age, marital status, employment, education, transactional sex, and endorsement of wife-beating.

Models 3 and 4 were used to examine the independent effects of household and community-level factors. Model 5 was a parsimonious model fitted to explore the main correlates of IPV among the PPAs. We used the ‘melogit’ command to fit the models. The log-likelihood ratio (LLR) and Akaike’s information criterion (AIC) tests were used to compare models with the highest log-likelihood and the lowest AIC indicating the best-fit model (see Table 3). Random effects were expressed in terms of community level variance, while the intra-class correlation coefficient (ICC) was used to examine clustering and the extent to which community/contextual factors explain the unexplained variance of the empty model [34]. All models were fitted at a 95% confidence level. P-values less than 0.05 were considered statistically significant.

留言 (0)

沒有登入
gif