Study design
The paper uses data from the 2016 Uganda Demographic and Health Survey (UDHS), accessed with permission from DHS Program [33]. This was a cross-sectional nationally representative survey that employed a two-stage stratified sampling procedure and cluster sampling design based on the sampling frame from the 2014 National Population and Housing Census [1, 34]. A detailed description of the sampling procedure is reported in the 2016 UDHS report [1].
The sample for the domestic violence (DV) module was 4,032 men. From this sample, we extracted a weighted sample of 2,559 ever-married men (formerly married, currently married or cohabiting), for further analysis. We used the domestic violence weighting variable (dm005) included in the UDHS data and the Stata survey (svy) command to weight the data during the analyses to account for the complex survey design [35].
The domestic violence module was based on the shortened and modified version of the Conflict Tactics Scale (CTS) [36]. The survey was carried out based on the World Health Organization’s (WHO) ethical and safety recommendations for research on domestic violence [37]. Specifically, informed consent was sought from all subjects that took part in the UDHS survey [1] and an institution (International Coaching Federation) approved its survey protocol. The respondents in the UDHS survey were briefed about the purpose of the survey before they could be interviewed. Furthermore, they (respondents) were assured of confidentiality, and their anonymity was ensured.
Variables and measures
Outcome variables
The outcome variables were the three different forms of IPV (physical, sexual, and emotional), which could have different predictors. The paper operationalized physical IPV as any physical act that results in abuse by a current or former partner within 12 months before the survey [27]. Sexual IPV referred to being subjected to sexual abuse, sexual harassment, and/or sexual humiliation by a spouse or former partner within 12 months before the survey [38]. Emotional IPV referred to being systematically and repeatedly degraded, humiliated, insulted, verbally oppressed as well as having one’s dignity violated by a spouse or former partner within 12 months before the survey [38].
In addition, an aggregate measure of IPV (any IPV), which combined all the three forms of violence was generated. Any IPV was operationalized as physical, sexual, or psychological harms as well as controlling behaviors aggravated by a current or former partner within 12 months preceding the survey [5, 39]. Similar outcome variables were used in our earlier publication [5]. The dependent variables were based on the following set of questions asked to men in the survey. Men indicated whether their wives/partners had ever or did;
-
a)
Hit, slap, kick or do anything else to hurt them physically?
-
b)
Force them to have intercourse or perform any other sexual acts against their will?
-
c)
Say something to humiliate them in front of others, threaten to hurt them or someone they care about, insult them or make them feel bad.
The response expected was either ‘yes’ or ‘no’; with ‘yes’ to the questions a, b, and c implying experience of physical, sexual, and emotional IPV respectively and ‘no’ implying non-experience of IPV. In addition, a ‘yes’ to any of the three questions a, b, and c implied experience of any IPV and ‘no’ implied non-experience of any IPV.
Explanatory variables
In this paper, independent variables were categorized into five categories. First, socio-demographics including men’s age, place of residence, education, region, and wealth status. History of witnessing parental violence was the second category, it was measured by respondents indicating whether their fathers ever beat their mothers, and it had a binary outcome (0 = No, 1 = Yes).
The third group of variables comprised wife/partners’ behaviors, namely, controlling behaviors, alcohol consumption, and frequency of getting drunk. To measure the partner’s controlling behaviors, male participants were asked, “Does your partner ever or did; a) Prohibit you to meet male friends? b) Limit you contact your family? c) Insist on knowing where you are at all times? d) Is jealous if you talk with other women? and e) Frequently accuses you of being unfaithful?” These were merged into one variable called the “partner’s controlling behaviors.” Any affirmative response (yes) to any of the above questions implied the presence of the partner’s controlling behaviors and no to all the questions implied the non-existence of such behaviors. The partner’s alcohol consumption was measured by responses to the question, “Does your partner drink alcohol?” and it had a binary outcome (0 = No, 1 = Yes). The frequency of a partner being drunk was a follow-up question to those respondents whose partners indicated that the partner drank alcohol.
The fourth category of explanatory variables considered marital factors, which included duration of the relationship, number of wives, total children ever born, and age at first marriage. A detailed description of the categorization of these marital variables is reported in our earlier publication [5]. The term “partner” in this paper includes wives as well as partners in cohabiting unions.
The fifth category was on men’s economic empowerment indicators and included their ownership of property (land and house) and type of earning from a man’s work. Ownership of land, as well as a house, was recoded into two categories: man alone/jointly with the partner as the empowered category and partner alone/others as the other. Type of earnings from a man’s work was recoded into binary outcomes - paid either cash only, in-kind only, and cash & in kind as the empowered category and not paid as the non-empowered category. Similar explanatory variables were used in our earlier publication [5] and also in mainland Tanzania [39].
Statistical analyses
We used frequency distributions to describe the characteristics of respondents. Pearson chi-square tests were used to test initial associations. The level of statistical significance using p-values was set at p<0.05. Selection of variables was based on the ecological model and published studies about IPV among women, and these variables have been found to be strongly associated with IPV among women [5, 27, 28, 40]. Therefore, we aimed to investigate the possible associations using men’s data. After all, there are few studies about IPV among men.
We tested univariate regressions with variables suggested as strongly associated with IPV. Furthermore, we conducted multivariable logistic regressions to examine the association between selected explanatory variables (with p-values less than 0.05) and the experience of emotional, sexual, physical IPV and any form of IPV as an aggregate measure of the three variables. We presented the results in the form of Odds Ratios (OR) and their 95% confidence intervals. All analyses were weighted and performed in Stata version 14.