The study was undertaken in 2008 in three districts in the Eastern Cape and KwaZulu-Natal provinces of South Africa. These form a contiguous area, and include rural areas with communally-owned land under traditional leadership, as well as commercial farms, small towns, villages, and a city, inhabited by people of all South African racial groups, several ethnic groups (predominantly Xhosa and Zulu) and socio-economic backgrounds.
The sample used a two stage proportionate stratified design to identify a representative sample of men aged 18–49 years living in the three districts. Using the 2001 census as the primary sampling frame, 222 census enumeration areas (EAs) were selected as the primary sampling unit, stratified by district and with numbers proportionate to district population size. The sample was drawn by Statistics South Africa. In order to avoid problems caused by outdated household listings, the households in each EA were mapped by the survey team and twenty were systematically selected. In each household one eligible man was randomly selected to take part in the interview. Men were eligible for the study if they were aged 18–49 years and had slept there the night before.
Of the 222 selected EAs, two (0.9 %) had no homes, and in five (2.3 %) we could not interview because permission from the local political gatekeepers was declined (1) or we could not access any eligible home after multiple visits at different times of day (4). In all the latter EAs, we established that many households were ineligible due to age or absence of a man. We completed interviews in 215 of 220 eligible EAs (97.7 %). We sampled a total of 4473 visiting points. Of these, 1353 (37.1 %) were found to contain no eligible man, 2298 (51.4 %) contained at least 1 eligible man, and 822 (18.4 %) could not be rostered for eligibility after a minimum of 3 attempts at contact. We completed interviews in 1737 of 2298 (75.6 %) enumerated and eligible households.
Interviews were conducted in isiXhosa or isiZulu or English with data collected using self-completion on APDAs (Audio-enhanced Personal Digital Assistants), thus participants could hear and read each question and its response options. Interviews took 45–60 min to complete. Only one participant was unable to do this and he asked the fieldworker to enter his responses. During interviews the fieldworkers were in the room to help if needed. Interviews were conducted in private and family members were not around. The confidentiality was ultimately assured through self-completion.
Measurement of rape
Rape perpetration was assessed using seven questions developed for the study and validated through cognitive interviewing, none of which actually used the word ‘rape’ [31]. They were modifications of those used previously in the Eastern Cape [10]. A typical item was “How many times have you slept with a woman or girl when she didn’t consent to sex or after you forced her?” The questions additionally asked about having forced a (former) girlfriend or wife into sex, having forced a woman who was not a girlfriend or wife into sex and having sex with a woman who was too drunk to consent. Two questions assessed multiple perpetrator rape perpetration: How many times have you and other men had sex with a woman at the same time when she didn’t consent to sex or you forced her? How many times have you and other men had sex with a woman at the same time when she was too drunk to stop you? Never, once and more than once were the response options. In this paper we consider a man to have perpetrated a multiple perpetrator rape if he had indicated he had done so in responses to either of the two questions, otherwise he was classified as having perpetrated a single perpetrator rape if he answered affirmatively to any of the five rape questions that did not specify multiple perpetrators, or having never raped, if he responded in the negative to all rape questions.
Other variables
The questionnaire included categorical variables measuring age and income. Questions on men’s childhoods included items on whether and how often their father was at home and mother’s level of schooling.
Data on adverse experiences before the age of 18 were collected using a locally modified version of the short form of the Childhood Trauma Questionnaire [32, 33]. We assessed five dimensions of adversity: emotional neglect, emotional abuse, physical neglect/hardship, physical abuse and sexual abuse using a four point response scale (never, sometimes, often and very often) (Cronbach’s alpha 0.79). A typical question was “before I reached 18 one or both of my parents were too drunk to take care of me”. Men were asked if they had ever been raped by a man (“persuaded or forced to have sex when you did not want to”).
Data were collected on two dimensions of psychopathy. Blame externalisation and Machiavellian egocentricity are two core affective and interpersonal deficits of psychopathy [34]. Blame externalisation is a perception of the world as hostile and others being at fault for one’s problems and Machiavellian egocentricity is a measure of narcissism and ruthless attitudes towards others [34]. Thirteen questions on Machiavellian Egocentricity and Blame Externalisation sub-scales of the Psychopathic Personality Inventory- Revised (PPI-I) were included. The Cronbach’s alpha for the scales together were 0.83. A typical item on the Machiavellian Egocentricity sub-scale was “I get mad if I don’t receive special favours I deserve” and on the Blame Externalisation sub-scale was “I have often been betrayed by people I trust”. Each has a 4 level response option (false, mostly false, mostly true, true). We dichotomised the scales and present the proportion scoring in the upper third of the scale versus the lower two-thirds. For blame externalisation 28.4 % were in the upper third and for Machiavellian egocentricity 18.5 % were in the upper third. These were adapted and reproduced by special permission of the Publisher Psychological Assessment Resources, Inc., 16204 North Florida Avenue, Lutz, Florida 33549, from the Psychopathic Personality Inventory- Revised by Scott O. Lilienfield, Ph.D., Copyright 2005 by PAR, Inc. Further reproduction is prohibited without permission of PAR, Inc.
We asked four items to measure empathy, adapted from Abbey et al. [35] (Cronbach’s alpha 0.80). A typical item was “I am often touched by things that I see happen”. These had a five point response scale (doesn’t describe me well – describes me well). Perceptions of life success were assessed with the following question: “If you compare your life circumstances overall now with those of the people you grew up with, would you say you have done much better for yourself, somewhat better, the same, less well, much less well?”
Attitudes towards gender relations were measured using 10 items from the Gender Equitable Men scale [36] (Cronbach’s alpha 0.78). A typical item is “There are times when a woman deserves to be beaten”. A high score denotes more equitable attitudes. Rape myths were measured on a 4-item scale where a higher score denoted more myth belief. A typical item was “in some rape cases women actually want it to happen” (Cronbach’s alpha was 0.76).
The school bullying perpetration score was an 8-item scale used to measure experiences with sexual bullying at school with four level response options (never, sometimes, often and very often) (Cronbach’s alpha 0.76). These questions were developed for the study. A typical item was “My school friends and I were a group and we would put pressure on a girl to date one of us until she agreed”.
We asked 11 items about lifetime experiences of participation in crime. These were modified for the local context from Tremblay et al. [37] who developed them as a measure of delinquency in childhood. Eight of the items related to theft (Cronbach’s alpha 0.81) and a typical item was “how often have you stolen an animal from someone?” The response options were never, once, 2–3 times and more often. Men were also asked about weapons ownership and possession and arrests.
Practices of gender relations were measured through questions about number of sexual partners, and about transactional sex with women, defined as sex that was primarily motivated by a desire for material gain on the part of the woman. This was defined as providing food, cosmetics, clothes, transportation, items for children or family, school fees, somewhere to sleep, handyman work, or cash [38]. Men were asked their age at first sex, and about lifetime perpetration of physical intimate partner violence, using the modified WHO violence against women instrument [39]. Specific acts of violence were asked in five items ranging from slapping to threats with or use of a weapon.
Recent alcohol consumption in the past 12 months was assessed through a question on frequency of having 5 or more drinks per drinking day. Drug use was assessed through a question on how often the man had smoked dagga (cannabis) in the past 12 months. This drug was selected as it grows well locally and is cheap and so is most widely used, and most men who use other drugs use this too. Men were asked if they had ever been in a gang.
Ethical issues
The men were informed about the study, given an information sheet and signed informed consent. As an incentive, they were given R25 (US $3.2) for the interview and those who gave a blood sample for HIV testing were given a further R25 (data not discussed in this paper). Consent for the interview, completion of the interview and the request for blood for HIV testing were performed in sequential stages so that a man who might decline to give blood for HIV could still agree to the interview. It was unlikely that being asked for blood was a major deterrent to the interview as very few of those eligible who were asked for an interview declined, but many men did decline the blood sample. Since the questionnaire asked men to disclose a range of criminal acts and South African law does not privilege research data, interviews were conducted anonymously. No identifying details of the men or their households were kept after the interview and the consent forms could not be linked. Ethics approval was given by the South African Medical Research Council’s Ethics Committee.
Data analysis
The study design provided a self-weighted sample. Data files were collated and analyses were carried out using Stata 12.0. All procedures took into account the two stage structure of the dataset, with stratification by district and the EAs as clusters. The distribution of social and demographic characteristics, childhood experiences, experience of abuse and rape victimisation, attitudes, psychological measures, aspects of gender relations, substance use and engagement with other violent and anti-social behaviour variables by rape perpetration status were summarised as percentages (or means), with 95 % confidence limits calculated using standard methods for estimating confidence intervals from complex multistage sample surveys (Taylor linearization). Pearson’s chi was used to test associations between categorical variables. Random effects ordered regression models were used to assess bi-variable associations between rape categories and continuous variables.
No efforts were made to replace missing data. However, we tested the impact of this on our findings through an analysis where we imputed values of variables that were missing provided there was some information to guide this. For some scale or score-based variables we have imputed a value based on the information from other variables in the scale/score. Where the range of possible values for the missing variables were limited and the value of the missing derived variable could be reliably imputed we did this. In other cases it was imputed based on the most probable values (taking into account available data). Where more than two variables were missing from a scale or score we did not impute but kept the value at missing. This analysis with missing data imputed did not change the overall results and is not presented.
Ordered logistic regression was used to investigate the association between the rape categories and variables describing the men’s characteristics. To account for clustering of men within EAs, we used a random effects ordered regression model. All the variables shown in the preceding tables were candidates for inclusion in the model. Variables were entered and backwards elimination was performed with variables retained at a conservative p < 0.2. The final model was then derived with variables retained at p ≤ 0.05. We tested for interactions between retained variables and found none. We tested the parallel regression assumption of the model, and examined it for each covarate and found it not violated (Brand test, p = 0.141).