Study setting
The study was conducted in Malawi, a small landlocked country of about 118,484 km² in Southern Africa, bordered by Tanzania to the north, Mozambique to the South east, and Zambia to the west. The current population of Malawi is about 17,563,749 people [4]. Eighty-five percent of the population depends on agriculture for their livelihood [4]. This research was specifically conducted in rural agricultural areas approximately five to ten kilometers around Mvera mission hospital in Dowa district (Figure 1) between the months of May and September 2018. Dowa district was purposively selected in central Malawi because of media reports that show increased cases of homicide that emanated from domestic violence and marital problems [26].
Study sample
This descriptive cross-sectional study used a multi-stage cluster sampling technique to select representative research participants [27]. Among the eight-outreach clinics under Mvera hospital, six were randomly selected. The selected outreach clinics were Mkhalanjoka, Gogo, Mvera, Kalinyengo, Mphande, and Ching’amba. During the time of the study, Mvera hospital was serving a population of 27,719 people. Out of the total population, there were 4,820 under-five year old child/mother dyads that were clients of postnatal health services in the selected six-outreach clinics. A Raosoft online software program was used to calculate a sample size as proposed by McCrum-Gardner, 2010 [28].The margin of error was set at 5%, with 95% confidence level, and a response distribution of 50%. A systematic sampling strategy was used to select a sample of 538 mothers with under-five children from the total population of 4,820 that were recorded in postnatal registers. A first name of a mother was randomly picked and subsequently picked every 9th child-mother dyad.
Participant recruitment
Selected mothers were contacted through their antenatal clinic when attending their regular monthly health assessment with their youngest under-five child. In a private clinic consultation room, the health worker asked the mother if she would be interested in taking part in this study. If the mother consented, a health worker research assistant administered the questionnaire orally in the consultation room. In this way, the confidentiality of mothers were protected as there was surety that no one present was aware that the mother took part in this study.
Measures
Outcome variables
The primary outcome variable for this study was prevalence of IPV against mothers of children under five years of age perpetrated by the current or most recent partner. We acknowledge that the definition of IPV is a multidimensional concept [29] but in this study the term was operationalized by focusing on IPV against mothers of under five year old children that was perpetrated by the current or recent male sexual partner. The term current or recent partner was defined as a mother’s sexual partner who was the father of the under-five child that the mother was nursing. They could be still in a sexual relationship or separated at the time of the interview. Mother’s exposure to IPV was screened using a WHO violence against women multi-country questionnaire instrument on women’s health and life experiences that had been validated and used in Malawi [12,30].
The questionnaire contains 18 items that make up four sub-scales measuring different forms of IPV: physical, emotional, controlling behavior, and sexual abuse. Maternal exposure to each [31] form of IPV was defined as the mother giving a positive answer to any one of the questions within each subscale. For instance, if a respondent answered “Yes” to any of the six items under controlling behavior, it was counted that such a mother was a victim of controlling violence. The study conformed to the standard IPV exposure screening instrument that was used in prior studies.
Mothers were considered exposed to psychological abuse if reported ever been belittled, insulted, hurt or scared by the current or recent partner. A mother was considered a victim of physical violence if affirmed encountered one of the following situations: was ever punched, slapped, kicked, pushed, choked, or threatened with a weapon by the current or recent partner. Finally, exposure to sexual violence was screened when a mother answered “Yes” to any of the following questions: If the husband or partner ever physically forced the respondent to have sexual intercourse, forced the wife to have sexual intercourse despite declining it, and was forced to perform any sexual activity against her will.
Explanatory variables
Explanatory variables that were identified in previous studies in SSA were considered probable predictors of IPV against women in the current study [13]. These included the mother’s age, education, religion, number and gender of children, whether the woman’s pregnancy was planned, whether the respondent had a confidant, and the partner’s health risk factors. The mother’s age was categorized as 16-24, 25-34, and 35-49. Education level was based on Malawi education standards and categorized as no education, primary education, secondary education, and tertiary education. Maternal religion was categorized as Presbyterian, Catholic, Pentecostal or no religion. Number of children was coded as one, two, three, four, and five or more. The binary questions such as family planning and whether the mother had a confidant were coded yes/no.
Questions about the husband’s health risk taking behavior such as alcohol consumption, smoking and extra marital affairs were examined. The family type was grouped as monogamous and polygamous with focus on polygyny. Monogamous family refers to a type of a family whereby a man has only one wife while polygyny refers to a family arrangement in which a man marries multiple women at one time [32]. Literature has consistently shown that there is a positive relationship between polygyny and IPV in SSA region [33–35].
Household poverty was measured using the international fixed poverty line that uses purchasing power parity conversion factors of US$1.90 per day [36]. Households that were not able to spend US $1.90/day on individual household needs were counted as living below the poverty level, those who spent that amount, or more were counted as above the poverty level. The age difference of the partners were also considered and was categorized as about the same age, husband is 5 years older than wife, husband is 5 years younger than wife, husband is 6-10 years older than wife, and husband is more than 10 years older than wife. Ethnicity of respondents was also included and grouped as Chewa, Tumbuka, Ngoni, and Yawo.
Household level factors such as food security, and time taken for the mother to fetch water were also included. Household food security was measured by a Household Food Insecurity Access Scale (HFIAS). Food secure household was coded = 0, and food insecure household was coded = 1. The four HFIAS (food secure, and mildly food insecure) were coalesced to food secure, and (moderately, and severely food insecure) to food insecure households [37]. Household food access was measured by Household Dietary Diversity Scale (HDDS). Low dietary diversity households were coded = 0. This was screened when the mother affirmed that household members consumed ≤ 4 food groups in the past 24 hours, and as 1 = minimum dietary diversity, when household members had consumed ≥ 5 food groups in the past 24 hours (38). Household source of domestic water was inquired and was coded 1 = borehole (protected water), and 0 = river/wells (unprotected water). Time that the mother took to fetch a pail of water was captured and coded as 0 = ˂ 30 minutes, and 1 = ≥ 30 minutes. The detailed questionnaire has been documented and can be accessed online [39].
Survey administration
The survey was administered using android tablets. The tablets were loaded with the digital version of the survey using ODK Collect. ODK Collect is an open source android application used to administer surveys that can then collect and organize the survey data. This application allows for immediate data validation in the field. The study was administered by nine female Health Surveillance Assistants (HAS) who were trained in using the WHO protocol for conducting studies of IPV [39,40]. The research training and pretesting of the survey questionnaire took five days. Trainers included a medical doctor, a clinical officer, and the first author who was a PhD Social Work candidate, all of whom had expertise in child and maternal health and domestic violence.
Due to the sensitivity of the research topic, the questionnaire was administered in a private consultation room at the outreach clinic during the regular mother-child clinic visit, ensuring that others present were not aware that she was participating in a research study [41,42]. In the private consultation room, the interviewer asked mothers questions and the responses were entered in the tablet by the interviewers. This was done without the presence of other health workers and postnatal clients. This protocol was designed to maximize the privacy and safety of respondents. In four cases, the consultation room was deemed not private due to interruptions. In these cases, the interviewer agreed with the respondent on a neutral venue that was safe for both. The average duration of the interviews was 63 minutes with a minimum of 56 minutes and a maximum of two hours. The interviews were conducted in a local language Chichewa that was juxtaposed with English in ODK.
Research Ethics Review
Ethics approval to conduct this study was obtained from the McGill University Research Ethics Board in Canada (REB File #: 503-0518), and University of Livingstonia research committee in Malawi (UNILIA-REC-4/18). Written permission was also obtained from the authorities at Dowa district commissioner’s office, Dowa district health office and Mvera clinic. Oral permission were obtained from the research participants based on the advice of the research stakeholders in Malawi. The rationale was that a written consent form could be easily seen by partners due to limited privacy in most households that could potentially put the mother at risk of abuse.
Data analysis
A total of 538 systematically selected mothers with under-five children were interviewed. There were no missing data since the questionnaire was designed so that the interviewer could not scroll to the next page in the android tablet until completing the question. Cronbach’s α was used to assess the internal reliability of the items used to determine maternal exposure to each of the four forms of IPV. In line with the WHO questionnaire, controlling behavior had five items, psychological abuse had four items, physical abuse had six items, and sexual abuse had three items (see Table 2). An α level of 0.70 or higher was considered to be satisfactory [43]. The calculated Cronbach’s α for controlling behavior was 0.81, psychological violence was 0.75, physical violence was 0.83, and sexual violence was 0.87.
Descriptive statistics were used to generate frequency tables of socio-demographic factors for mothers, children, and fathers. Univariate logistic regressions were performed to determine significant risk factors of mothers’ exposure to IPV from the selected independent variables. Four separate multivariable logistic regression analyses were performed to explore predictors of controlling behavior, emotional violence, physical violence, and sexual abuse. The variables that were significant in univariate tests were entered in the multivariate logistic regression models using forward method.
Multicollinearity of independent variables was tested and a variance inflation factor (VIF) of 2.314 was obtained, demonstrating that the tested independent variables were not similar and the regression coefficients estimates were reliable [44]. A fixed effects model was used to account for the intracluster homogeneity effect of the 6 study locations in the analysis [45]. The results of each of the multivariate analyses with 95% confidence interval (CI), including both crude (CORs) and adjusted odds ratios (AORs), are reported in Table 3. A p value of less than 0.05 was considered statistically significant in the study. The data were analysed using an IBM Statistical Package of Social Sciences (SPSS) for Windows version 23.0 (IBM Corp., Armonk, NY, USA).