2.1. Study Area and Data Source
This secondary analysis was based on 2019–2020 cross-sectional, Liberian demographic and health survey (LDHS) data. The data collection phase was started in October 2019 and completed in February 2020. Liberia is administratively subdivided into 15 counties and 136 districts, with each district further subdivided into clans. For the last census in 2008, each clan was further subdivided geographically into enumeration areas (EA) that comprised of an average 100 households. LDHS used stratified two stage cluster sampling design. In the first stage, using probability proportional to size method, 325 clusters comprising of EAs were selected. In the second stage, households were listed in each cluster, and using equal probability systematic sampling method, 30 households were selected from each cluster. This resulted in the selection of 9745 households for 2019–2020 LDHS, based on 2008 census and subsequent population projections. All 15–49 year old women in these selected households were deemed eligible for the LDHS. However, in adherence to the World Health Organization’s guidelines on ethical conduct of collecting domestic violence data, only one randomly selected woman in a subsample of households was administered a domestic violence module, while ensuring privacy [19
]. This is the common survey methodology used in the DHSs for the implementation of a domestic violence module. The ‘Domestic Violence’ module in Liberia was administered to a subset of all women who were interviewed. In each household, one woman was randomly selected for this module.
The approval for secondary analysis of LDHS was granted by Measure DHS, using an online request form; datafile was downloaded from the Measure website www.measuredhs.com
(accessed on 20 August 2021). LDHS covered all 15 states and cumulatively had 8364 eligible women aged 15–49 years. Out of these, 8065 (96.43%) women were interviewed. For the domestic violence module, 3166 women aged 15–49 years were selected, but 46 (1.45%) women could not be interviewed owing to lack of privacy, inability to find respondents at home even after repeated visits, or due to interruptions during the interview. Hence, domestic violence was administered to 3120 women who were successfully interviewed. Out of these 3120 women, there were 2331 women who were ever-married. Details of survey methodology, sampling design, generation of sampling weights adjustments for non-response, and survey tools were published in the country report available on the Measure website.
2.4. Explanatory Variables
Based on previous studies [6
], 10 variables at the individual, husband/partner, and familial levels were studied for association with respondents having ever experienced intimate partner violence, i.e., women’s age, women’s educational attainment, women’s occupation, wealth index of the household/family, residential status in terms of urban and rural, number of living children, participation in decision making, acceptance of IPV, husband/partner’s use of alcohol, and having witnessed one’s father physically beating up one’s mother. Details on the derivation of each explanatory variable is provided as under:
Age: Respondents were asked about their age, based on their last birthday. LDHS datafile provides age in seven 5-year age groups, starting from 15–19 years and ending with the 45–49 years group. The age group 15–19 years was used as the reference category.
Educational attainment: Respondents were grouped into four categories of either ‘no education’, ‘primary education’, ‘secondary education’, and ‘higher education’, i.e., more than secondary education. The ‘no education’ category was used as the reference category.
Women’s occupation: Several occupational categories were specified in the LDHS. For this analysis, respondents were grouped into three categories of ‘professional, clerical, sales, services’; ‘does not work’; ‘agriculture self-employed, agriculture-employee, household and domestic work, skilled manual, and unskilled manual’. The ‘professional, clerical, sales, services’ category was used as the reference category.
Wealth index of the family: LDHS gave each household scores based on the ownership of consumer goods that included television, bicycle/car, and housing attributes like toilet facilities, drinking water source, and type of flooring materials. Using principal component analysis, wealth quintiles were compiled and assigned to each household and its individual member. Five categories ranging from ‘poorest’, ‘poorer’, ‘middle’, ‘richer’, and ‘richest’ were calculated. The ‘poorest’ group was used as the reference category.
Residential status: Respondent’s place of residence at the time of survey was grouped into two categories of ‘urban’ and ‘rural’ in the LDHS datafile. The ‘urban’ group was used as the reference category.
Number of living children: Responses were coded as continuous variable in the LDHS, ranging from 0 to 15 children. For this analysis, four categories of ‘no/0 children’, ‘1–2 children’, ‘3–4 children’, and ‘5–15 children’ were created. The ‘zero children’ group was used as the reference category.
Participation in decision making: Respondent having the latitude to make decisions, either alone or jointly with their husband/partner, in either of the three areas of healthcare seeking for herself, large household purchases, and visits to relatives was coded as having participated in the decision making. In LDHS, answers were coded into five groups of ‘respondent alone’, ‘respondent and partner/husband together’, ‘partner/husband alone’, ‘someone else’, and ‘other’. Not being able to participate either alone or jointly with husband/partner in any of the three areas was used as the reference category.
Acceptance of intimate partner violence (IPV): Respondent replying affirmatively to believing that IPV is justified in either of the five scenarios of wife goes out without telling her partner/husband, neglects children, argues with partner/husband, refuses sex, and/or burns food, was coded as accepting of IPV. Not accepting all five scenarios was used as the reference category.
Husband/partner’s alcohol use: Respondents were asked if their husband/partner drank alcohol and answers were either affirmative or negative. Negative answer was taken as the reference category.
Witness parental IPV: Respondent replying affirmatively to the question on ever witnessing her father beat her mother. Negative answer was taken as the reference category.
2.5. Statistical Analysis
Analyses was conducted in STATA version 17.1 (StataCorp, College Station, TX, USA) using the survey procedures to incorporate the complex sample design and sampling weight, accounting for the differential probability of selection. Missing data were not imputed; all hypothesis testing was 2-tailed, with statistical significance set at 2-sided p < 0.05.
LDHS data set was downloaded as STATA format data file. As a first step, for outcome and all explanatory variables, unweighted counts, number of records with missing information, and cumulative weighted percentages were calculated. Secondly, simple binary logistic regression models were run to determine the statistical significance of every explanatory variable’s association with the outcome variable of having ever experienced IPV. Odds ratios, statistical significance, and 95% confidence intervals were calculated. Finally, all those explanatory variables that were found to be statistically significantly associated with the IPV were used in the final binary multiple logistic regression model. Adjusted odds ratios, statistical significance, and 95% confidence intervals were calculated for the final model.
The association of all 10 explanatory variables with each sub-type of the outcome variable of IPV was examined next, using simple logistic regression models. Association of emotional, physical, and sexual violence was individually examined with each explanatory variable, in order to study the results stratified by each type of IPV.