Social Science & Medicine 55 (2002) 1603–1617

Risk factors for domestic violence: findings from a South

African cross-sectional study
Rachel Jewkesa,*, Jonathan Levinb, Loveday Penn-Kekanaa
a
Gender and Health Group, Medical Research Council, Private Bag X385, Pretoria 0001, South Africa
b
Bio-statistics Research Unit, Medical Research Council, South Africa

Abstract

In 1998 a cross-sectional study of violence against women was undertaken in three provinces of South Africa. The
objectives were to measure the prevalence of physical, sexual and emotional abuse of women, to identify risk factors
and associated health problems and health service use. A multi-stage sampling design was used with clusters sampled
with probability proportional to number of households and households were randomly selected from within clusters.
One randomly selected woman aged 18–49 years was interviewed in each selected home. Interviews were held with a
total 1306 women, the response rate was 90.3% of eligible women. For the risk factor analysis, multiple logistic
regression models were fitted from a large pool of candidate explanatory variables, while allowing for sampling design
and interviewer effects. The lifetime prevalence of experiencing physical violence from a current or ex-husband or
boyfriend was 24.6%, and 9.5% had been assaulted in the previous year. Domestic violence was significantly positively
associated with violence in her childhood, her having no further education, liberal ideas on women’s roles, drinking
alcohol, having another partner in the year, having a confidant(e), his boy child preference, conflict over his drinking,
either partner financially supporting the home, frequent conflict generally, and living outside the Northern Province.
No significant associations were found with partners’ ages, employment, migrant status, financial disparity,
cohabitation, household possessions, urbanisation, marital status, crowding, communication, his having other
partners, his education, her attitudes towards violence or her perceptions of cultural norms on women’s role. The
findings suggest that domestic violence is most strongly related to the status of women in a society and to the normative
use of violence in conflict situations or as part of the exercise of power. We conclude by discussing implications for
developing theory on causal factors in domestic violence. r 2002 Elsevier Science Ltd. All rights reserved.

Keywords: Domestic violence; Gender inequality; South Africa

Introduction by observations that its occurrence varies considerably

Intimate partner violence, or domestic violence, is
increasingly being recognised as a public health problem
and associated with injuries and a wide range of other
mental and physical health problems (Campbell et al.,
forthcoming; Heise, Ellsberg, & Gottemoeller, 1999). Its
causes have preoccupied social scientists for several
decades. Hypotheses that domestic violence might be
biologically determined were significantly undermined
*Corresponding author. Tel.: +27-12-339-8525; fax: +27-
12-339-8582.
E-mail address: rjewkes@mrc.ac.za (R. Jewkes).
between as well as within societies, and in some it has
been reported to be exceptionally rare or even absent
(Levinson, 1989; Counts, Brown, & Campbell, 1992).
Understanding of social causation has been significantly
hampered by the narrow geographical base of research
(most work being from North America), a tendency of
academics and activists to pursue single-factor theories
and the reliance on data from small samples or women
who had succeeded in accessing sources of help such as
shelters (Hoffman, Demo, & Edwards, 1994; Heise,
1998). In the last decade there have been several papers
from well-designed studies conducted in a range of
countries which discuss risk factors and processes using

0277-9536/02/$ - see front matter r 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 2 7 7 - 9 5 3 6 ( 0 1 ) 0 0 2 9 4 - 5
1604 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617
survey (e.g. Hoffman et al., 1994; Ellsberg, Pena,
Herreras, Liljestrand, & Winkvist, 1999; Martin, Tsui,
Maitra, & Marinshaw, 1999; INCLEN, 2000) and
ethnographic methods (e.g. Rao, 1997; Wood & Jewkes,
2001; Bourgois, 1996; Harvey & Gow, 1994). These have
enabled hypotheses to be explored more critically and an
understanding of the social plausibility of relationships.
However the body of evidence available to date falls
very far from the standards of epidemiologists for
establishing disease causation (Hill, 1965).
Household research on violence against women is
regarded as difficult. Many women are reluctant to
disclose abuse1 because of perceptions of shame, fear of
blame or reluctance to be ‘disloyal’ to their partner.
Differences in question wording and the number of times
questions asked can result in widely differing prevalences
(Ellsberg, Heise, Pena, Agurto, & Winkvist, 2001). The
research is difficult for field workers. Listening to a daily
litany of abuse can provoke overwhelming feelings of
powerlessness, as well as difficult memories and reflec-
tions on experiences of the field workers’ own intimate
relationships. Without adequate support, and an ability
to support women interviewees, field workers may burn
out or learn to ask questions in a manner which
discourages disclosure (Jewkes, Watts, Abrahams,
Penn-Kekana, & Garcia-Moreno, 2000; Ellsberg et al.,
2001). Interviewers also bring their own views on gender
to the interview setting and this is also believed to
influence levels of disclosure. Careful interviewer train-
ing, selection and support is essential, however inter-
viewer effects may persist. Recent advances in statistical
methodology have enabled adjustment for interviewer
effects (O’Muircheartaigh & Campanelli, 1998); however
these have not previously been used in violence against
women research.
In understanding the origins of intimate partner
violence (physical and sexual), Heise’s (1998) ecological
framework has recently assumed prominence. Based on
review of the mainly, but not exclusively, North
American literature, it premises that abuse results from
an interplay of personal, situational and socio-cultural
factors at different levels in the social environment.
Heise seeks to present the factors which are predictive at
each level of the social ecology. At an individual level,
factors include being abused as a child or witnessing
marital violence in the home, having an absent or
rejecting father; at the level of the family or relationship
factors include use of alcohol, male control of wealth
and decision-making in the family and marital conflict;
at a community level factors include poverty and
unemployment, social isolation of the woman and male
participation in delinquent peer associations; and at a
societal level factors include male ownership of women,
1
In this paper when discussing intimate partner violence the
terms ‘violence’ and ‘abuse’ are used interchangeably.
ideas of masculinity linked to aggression and domi-
nance, rigid gender roles, acceptance of interpersonal
violence and acceptance of physical chastisement. This
model and other theories (e.g. of status inconsistency
(Gelles, 1974)) were drawn on in the development of a
questionnaire for cross-sectional study of violence
against women in South Africa, and subsequently used
to develop a list of candidate variables for the risk factor
analysis. In this paper we present findings from the first
representative study of violence against women in South
Africa on risk factors for experiencing physical violence
from an intimate partner and we conclude by discussing
their implications for a broader theoretical understand-
ing of the problem.
Background
South African society is very violent. Decades of
apartheid State-sponsored violence and reactive com-
munity insurrection, meticulously described in the report
of the Truth and Reconciliation Commission (1998),
have contributed to a situation in which for many people
physical violence is a first line strategy for resolving
conflict and gaining ascendancy (Simpson, 1991). All
forms of interpersonal violence are very common.
Violence is used regularly, for example, in disputes
between neighbours (Department of Health, forthcom-
ing), male and female peers (Wood & Jewkes, 2001),
nurses and patients or their relatives (Jewkes, Abra-
hams, & Mvo, 1998) and fellow workers (Abrahams,
Jewkes, & Laubscher, 1999). Injury is the major cause of
death among youth and, for example, amongst 14–34
year old, 58% of injury deaths are due to homicide
(Peden, 2000).
Gender-based violence is viewed in Government and
civil society as a major problem (Usdin, Christofides,
Malepe, & Maker, 2000). Physical violence is a
prominent feature of sexual relationships from the start
of dating during teenage years (Wood, Maforah, &
Jewkes, 1998; Jewkes, Vundule, Maforah, & Jordaan,
2001). Forms of violence have been described as lying on
a continuum between slapping, ‘persuading’ a woman to
have sex, threatening to beat, hitting with sticks or other
objects, pushing, assaulting with fists, violent rape,
stabbing with a knife or shooting (Wood & Jewkes,
2001). In a country of approximately 40 million people,
as many as five women are estimated to be killed each
week by an intimate partner (Vetten, 1995). Amongst
the youth, violence is very often associated with a
woman rejecting a man’s ‘proposal’ to have a relation-
ship, their actual or suspected infidelity, attempts to end
relationships, resistance to men’s attempts to dictate the
terms of a relationship, and acts which undermine a
boyfriend’s success with other women (Wood & Jewkes,
2001). Physical violence is very often accompanied by a
 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617
range of emotional abuse by men, particularly including
deliberate belittling, attempts to control a girlfriend or
wife’s social interactions and movement, bringing other
girlfriends to the shared home, evicting the woman and
her children, and not providing money for essential
items at home when money is available (Jewkes, Penn-
Kekana, Levin, Ratsaka, & Schrieber, 2001). Sexual
violence is also very common. The rate of reported rape
to the police is the highest amongst all Interpol members
(Interpol, 1996). Two percent of women interviewed for
the study described in this paper disclosed that they had
been raped or an attempt of rape had been made in the
previous year (Jewkes & Abrahams, in press). This
figure is probably conservative as coerced sex in intimate
relationships is extremely common (Jewkes, Vundule
et al., 2001).
To an outsider, one of the most remarkable
features of gender-based violence in South Africa
is that, within certain boundaries of severity, the
society is extremely tolerant of it (CIET-Africa, 2000;
Kim & Motsei, forthcoming; Wood & Jewkes, 2001;
Jewkes & Abrahams, forthcoming). Older people
often attribute the violence of young men in dating
relationships to normal boyish behaviour (Mager, 1999;
Wood & Jewkes, 2001); despite laws against domestic
violence and rape, police are very often reluctant to
pursue cases (CIET-Africa, 1999; Jewkes & Abrahams,
forthcoming); even women often tolerate intimate
partner violence, something which is notable in the
reluctance of teenage girls to leave violent boyfriends
even when they have no children or financial dependence
on the relationship, and in the way in which being
known to be violent seems to carry little stigma and
interfere little in a man’s ability to get another partner
(Wood & Jewkes, 2001). This widespread tolerance
reflects both ideas that the use of violence is often
‘normal’, inevitable and ideas about gender which
legitimate the use of force by men in establishing
hierarchical control over women.
The control of women by men is very prominent in
South African society. Recent ethnographic research
amongst youth in the Eastern Cape suggests that not
only are young men are expected to control girlfriends,
but the ability to do so is a central feature of ideas about
‘successful’ youth masculinity (Wood & Jewkes, 2001).
Such ideas can also been seen reflected in writings on
gender oppression by historians of the nineteenth
century, suggesting that they may have deep cultural
roots. At that time the control of women by men was
then central to the structure and functioning of African
homesteads (Burman, 1990). Furthermore, Guy (1990)
has argued that control and appropriation of the
productive and reproductive capacity of women has
been described as ‘‘the social feature upon which
[precapitalist] society was based’’ (Guy, 1990, author’s
own emphasis).
1605
Improving the status of women in South Africa has
been identified as a political priority for the Government
and indeed since democracy was established in 1994 in
South Africa the legal status of women has improved
greatly. Many discriminatory statutes have been re-
moved, equality is enshrined in the Bill of Rights, and
women are prominent in all levels of Government and
civil society. New legislation has been enacted to give
women reproductive rights, notably the 1996 Choice in
Termination of Pregnancy Act, and to protect women
against domestic violence, in the form of the 1998
Domestic Violence Act. Both of these pieces of
legislation are viewed internationally as amongst the
most empowering for women of any in the world.
Methods
The objectives of the study were to describe the
prevalence of physical, sexual, financial, and emotional
abuse and aspects of the economic and service implica-
tions; and to identify risk factors, associated health
problems and health service use. This paper focuses on
risk factors for physical violence, a more detailed
description of prevalence of all forms of violence by
province can be found elsewhere (Jewkes et al., 2001).
The authors recognise that domestic violence often has
emotional, sexual, financial and physical elements but
for reasons of style in this article ‘domestic violence’ is
glossed as physical violence by a current or an ex-
husband or boyfriend.
The study was cross-sectional, with research under-
taken in three of the country’s nine provinces, the
Eastern Cape, Mpumalanga and the Northern Province.
These were chosen for pragmatic reasons. The sampling
frame was the list of enumeration areas demarcated for
the 1996 Census, these were treated as clusters. Each
province was stratified into urban and rural areas.
Clusters were sampled with probability proportional to
the number of households. In the urban areas 14
households were randomly selected in each cluster and
in rural areas, 28. Within each province the sample was
approximately self-weighting. The sample of enumera-
tion areas was drawn at random by Central Statistics for
the 1998 South African Demographic and Health
Survey (SADHS), a systematic sample of 50% of this
formed the sample in Mpumalanga and the Northern
Province and of 1 in 6 in the Eastern Cape (as there had
been three-fold oversampling in this Province in
SADHS). This study was conducted in parallel to the
SADHS and provided a source of external validation for
violence questions included in that national survey. 2232
households were selected for interviews: 728 in the
Eastern Cape, 748 in Mpumalanga and 756 in the
Northern Province. One randomly selected woman aged
18–49 was chosen from each and no substitution was
1606 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617
used if a visiting point did not contain a dwelling or a
woman of the right age. Three attempts were made to
contact each woman.
The questionnaire collected information on social and
demographic characteristics of the women, attitudes
towards and experiences of abuse, service use and health
questions. A schedule of questions was completed on
characteristics of the woman’s partners in the past year
and the quality of the relationship. The questionnaire
was based partly on one used in Zimbabwe (Musasa,
1996) and further informed by two focus groups. The
questionnaires were translated into nine languages (all
official languages except Setswana) and administered in
the first language of the interviewee. Women were asked
in four separate questions, whether in the last year they
had been threatened with violence, had been slapped,
punched or beaten, had been kicked, bitten, choked or
burnt, and whether they had been threatened with or
injured by a weapon or object. These questions enabled
the group who had experienced violence or threats in the
past year to be defined. Those responding in the negative
were asked if they had ever been beaten by a current or
ex-husband or boyfriend. All women were then asked if
they had ever experienced violence in pregnancy. The
responses to these two questions were used to identify a
further group of women who had experienced violence
before the last year. These were added to the other group
to form an ‘ever abused’ category.
The study followed the principles of the WHO
guidelines for domestic violence research which were
explained verbally to the study team (World Health
Organisation, 1999). Safety and ethical issues have been
described in detail elsewhere (Jewkes, Watts, Abrahams,
Penn-Kekana, & Garcia-Moreno, 2000). Interviews
were conducted in private. Verbal informed consent
was obtained to ensure anonymity and interviewers were
trained to provide immediate emotional support and
referral information. Ethical approval for the study was
given by the Medical Research Council’s Ethics Com-
mittee.
The data was entered onto a database in the
epidemiological package Epi Info and then validated
through a second entry. Data analysis was carried out
using the statistical package Stata (Statacorp, 1997).
Three response variables were considered: whether or
not the woman had ever experienced domestic violence
by an intimate partner (current or ex-husband or
boyfriend); whether or not the woman had experienced
domestic violence by an intimate partner in the year
before the interview; whether or not the woman had
experienced domestic violence or been threatened with
violence by an intimate partner in the previous year. The
prevalences of these three responses were estimated,
together with 95% confidence limits, using the survey
estimation procedures in Stata which take into account
the stratified multistage design used in the sampling.
Model building
The question for each response was: which are the
most important factors determining whether or not a
woman was abused? The approach was to fit multiple
logistic regression models with a large pool of candidate
explanatory variables, while allowing for the sampling
design and interviewer effects (explained below). The
model with all explanatory variables was fitted and then
explanatory variables were removed in a backward
elimination approach, maintaining variables which were
significant at least at the 10% level. The candidate
variables were mostly those which had been found
associated with abuse in the literature, with some
additional ones. Two approaches are possible to take
into account the complex sampling scheme, namely the
aggregated and disaggregated approaches (Skinner,
1989a). In the aggregated approach the survey design
variables are regarded as nuisance variables which need
to be taken into consideration to obtain robust standard
errors; this approach has been used by Conroy, Elmore-
Meegan, Joyce, McGuigan and Barnes (1996). In the
disaggregated approach, the sampling structure is taken
as reflecting the underlying population structure and
thus reflecting parameters of substantive interest.
Estimation for the resulting multilevel model has been
the subject of much work in the last decade, and this
approach has been widely used in recent analyses of
health surveys (Ecob, 1996; Steele, Diamond, & Amin,
1996). The aggregated approach has an advantage of
being robust against both model misspecification and
heteroscedasticity (Skinner, 1989b). The disaggregated
approach, however, can account for confounding effects
of the level two units (e.g. clusters) (Levin, 1999) and can
be used to allow for interviewer effects (O’Muirchear-
taigh & Campanelli, 1998; Hox, 1994).
The aggregated approach was used to identify the
model, once this was done the model was refitted using
the disaggregated approach using either clusters or
interviewers as the second level units. With a binary
response, full maximum likelihood requires numerical
integration of the likelihood function, which is easily
done using the xtlogit command in Stata for a ‘‘two-level
model’’ i.e. considering either clusters or interviewers
separately as the second level units. Joint consideration
of interviewer and enumeration area effects requires the
use of a multilevel model with crossed level-2 effects; full
maximum likelihood is not possible and an approximate
method such as Penalized Quasilikelihood (PQL)
(Breslow & Clayton, 1993) is required. However PQL
is known to result in downward biased estimates of both
parameters and their standard errors (Goldstein &
Rasbash, 1996). Investigation showed that interviewer
effects were much larger than enumeration area effects
and that PQL used for either interviewer effects alone or
enumeration area effects alone did in fact result in a
 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617
considerable downward bias. Thus it was decided to use
the disaggregated approach using interviewers effects
alone, and to compare the results with the aggregated
approach to ensure robustness.
Results
Of the 2 232 households selected for interview, 1447
had eligible women (the others were vacant or had no
eligible woman). Forty women declined to be inter-
viewed or only completed part of the interview and 101
women were uncontactable after three visits. Thus 1306
interviews were completed, a 90.3% response rate. Of
the women interviewed, 1279 reported having ever had a
husband or boyfriend, and were thus ‘at risk’ of
domestic violence, the subsequent analysis is based on
this group. In all, 1164 women had had a partner in the
previous year. The lifetime prevalence of domestic
violence was 24.6% (95% CI 21.5–27.6). The prevalence
of domestic violence in the past year was 9.5% (95% CI
7.5–11.5) and the prevalence of domestic violence or
threats of violence in the past year was 11.6% (95% CI
9.4–13.8). Of the women experiencing violence, 45.9%
reported injury in the past year.
Table 1 shows the social and demographic character-
istics of women in the three abuse categories and
compares them with the total sample of women. Abused
women were less likely to live in the Northern Province,
to be African, to have post-school education and to be
unemployed. There was no difference in mean age.
Abused women were much more likely to drink alcohol,
with the likelihood being the greatest for those abused in
the last year. Those abused in the past year were more
likely to be in domestic work and to have a non-marital
partner. The ever abused group were less likely to be
married. Abused women were much more likely to have
been beaten frequently (at least once a year) as a child,
to report their mother being beaten, and to have spent
their childhood in urban areas. Despite a higher
employment rate, they were less likely to own a TV,
and those ever abused lived in more crowded homes
than the total sample. Women abused in the last year
were more likely to live in homes where they themselves
or their partner was the main provider.
Table 2 shows variables which indicate the women’s
views on gender and relationships. Abused women were
more likely to have someone to talk with about their
relationship, with the greatest likelihood being amongst
those experiencing violence in the past year. They were
more likely to prefer boy children. Recently abused
women perceived their culture to be slightly less tolerant
of female subservience than other women. Those ever
abused were less likely to perceive their culture to expect
a working woman to give her husband her earnings, a
man to share housework and a woman to need her
1607
husband’s permission to work. They were more likely to
perceive it to expect a man to take girlfriends, to punish
his wife and regard beating as signifying love. Those
abused in the past year were more likely to perceive their
culture to expect women to obey their husbands and a
man to take girlfriends. They were less likely to perceive
their culture to expect a woman to give her earnings to
her husband, a woman to get permission to work, a
husband to punish his wife and to perceive beating as a
sign of love.
All the women held personal views which were less
tolerant of women’s subservience than they perceived
their culture to be. Although the most liberal views were
held by those beaten or threatened in the previous year,
the greatest difference between ideas of culture and
personal beliefs was found in the ever abused group.
Abused women were less likely say that they believed
that a woman should obey her husband, give him her
money, get his permission before working, that he
should have the final say in family matters, and that he
can punish her. Those abused in the last year were more
likely to say that they believed that a man should help
with housework. There was no difference in the
proportion (25%) viewing beating as signifying love or
perceiving that a woman can not stop her husband
taking girlfriends (also 25%). The mean personal
subservience score of women declined with education
from 4.13 in the least educated group to 2.11 in the most
educated group, this difference was highly significant
(po0:0001). The same pattern was seen for women
beaten in the past year and for each educational level the
mean personal subservience score was lower than among
women of the same educational level who were not
beaten.
There was little difference in the proportion in all
groups who agreed that it is sometimes acceptable for a
man to hit his wife. Abused women were more likely
than the total sample to perceive violence to be
acceptable in conflict resolution between adults in some
(more than one) situations. All the variables presented in
Tables 1 and 2 were considered as potential explanatory
variables for the logistic regression model of risk of ever
having experienced domestic violence which is presented
in Table 4.
Table 3 shows the social and demographic character-
istics of the women’s male partners of the last year. If
women reported more than one partner the character-
istics described are of the physically abusive partner (if
there was abuse) or one identified as her main partner.
Abused women were more likely to have had more than
one partner in the previous year, to live with the main
partner, to be a sole wife and to know he had other
girlfriends. Their male partners were much less educated
than the total sample, more likely to have had a rural
childhood, the age difference between partners was less
and they were less likely to have an age difference of over
1608 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617

Table 1
Social and demographic characteristics of abused women, compared with all women (N ¼ 1279)

Ever physical Physical violence in Physical violence and All women

violence % (n) past year % (n) threats in past year % (n) % (n)

Province
Eastern Cape 33.8 (106) 37.8 (42) 36.3 (49) 31.0 (396)
Mpumalanga 37.9 (119) 43.2 (48) 43.7 (59) 32.8 (419)
Northern Province 28.3 (89) 18.9 (21) 20.0 (27) 36.3 (464)

Mean age (SD) 31.5 (8.56) 32.5 (8.38) 31.6 (8.37) 31.3 (8.73)

Racial group: Black African 97.1 (305) 95.5 (106) 94.8 (128) 98.0 (1253)

Education
None 14.7 (46) 16.4 (18) 16.5 (22) 15.6 (198)
Primary 16.9 (53) 16.4 (18) 16.5 (22) 15.1 (192)
Lower secondary 50.2 (157) 50.9 (56) 49.6 (66) 46.0 (585)
Higher secondary 15.7 (49) 15.5 (17) 15.0 (20) 16.8 (214)
Any post-school 2.6 (8) 0.9 (1) 2.3 (3) 6.6 (84)

Drinks alcohol 20.7 (65) 29.7 (33) 27.4 (37) 11.3 (144)

Occupation
Unemployed 56.7 (178) 53.2 (59) 54.8 (74) 60.9 (779)
Trader 14.0 (44) 12.6 (14) 11.9 (16) 11.3 (145)
Domestic worker 12.4 (39) 14.4 (16) 12.6 (17) 10.9 (139)
Professional 1.9 (6) 1.8 (2) 3.0 (4) 3.6 (46)
Office worker 2.2 (7) 2.7 (3) 2.2 (3) 1.8 (23)
Shop worker 3.5 (11) 5.4 (6) 4.4 (6) 3.8 (48)
Farm worker 6.7 (21) 7.2 (8) 7.4 (10) 5.9 (75)
Factory worker 2.6 (8) 2.7 (3) 3.7 (5) 1.9 (24)

Marital status
Church ceremony 11.8 (37) 15.3 (17) 14.8 (20) 15.8 (202)
Traditional ceremony only 27.1 (85) 29.7 (33) 31.1 (42) 31.0 (397)
Widow /divorced/separated 2.2 (7) 0 0 2.7 (35)
Non-marital partner 47.1 (148) 48.7 (54) 48.9 (66) 40.9 (523)
Single, no boyfriend 11.8 (37) 6.3 (7) 5.2 (7) 9.5 (122)

Beaten as a child
Once a year or more often 55.6 (173) 57.7 (64) 54.5 (73) 44.3 (563)
Not every year or never 44.4 (138) 42.3 (47) 45.5 (61) 55.7 (707)

Witnessed her mother beaten 32.2 (101) 31.5 (35) 35.6 (48) 20.4 (260)

Urban childhood 25.8 (81) 33.3 (37) 33.3 (45) 20.7 (265)
Rural childhood 74.2 (233) 66.7 (74) 66.7 (90) 79.3 (1014)

Consumer goods ownership

Radio 76.8 (241) 78.4 (87) 78.5 (106) 76.4 (977)
TV 39.8 (125) 41.4 (46) 43.0 (58) 44.6 (570)
Car 16.6 (52) 15.3 (17) 17.0 (23) 16.3 (209)
Fridge 31.2 (98) 31.5 (35) 32.6 (44) 32.9 (421)

Household crowding index

Mean no. of people per room (SD) 1.92 (1.33) 1.83 (1.35) 1.86 (1.27) 1.84 (1.35)

Main source of money for home

Women and/or her partner 55.4 (174) 63.7 (71) 61.5 (83) 57.2 (731)
Another person 44.6 (140) 36.0 (40) 38.5 (52) 42.9 (548)
 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617 1609

Table 2
Views of women on gender and relationships

Ever physically Physically Physical All women

abused % (n) abused in violence and % (n)
past year threats in
% (n) past year
% (n)

Preference for children

Prefers boys 16.6 (52) 23.4 (26) 21.5 (29) 13.2 (169)
Prefers girls or has no preference 83.4 (262) 76.6 (85) 78.5 (106) 86.8 (1110)

Has someone to talk with about her relationship 74.3 (208) 82.4 (89) 83.3 (110) 63.7 (723)

Cultural tolerance of subservience

A woman should obey her husband 74.3 (231) 91.9 (102) 91.9 (124) 74.9 (951)
A working woman should give her money to her husband 53.2 (167) 53.2 (59) 52.6 (71) 58.3 (741)
A man has the final say in family matters 74.3 (231) 74.1 (80) 75.8 (71) 74.9 (951)
A man should share house work 32.3 (101) 36.9 (41) 37.0 (50) 36.0 (458)
A woman needs her partners’ permission to work 84.9 (265) 82.0 (91) 83.7 (113) 88.1 (1122)
A woman can do nothing if her husband takes girlfriends 54.1 (170) 52.3 (58) 51.9 (70) 48.4 (617)
If a wife does something wrong her husband has a right to punish her 61.5 (193) 55.0 (61) 56.3 (76) 58.1 (740)
If a man beats you it shows he loves you 43.9 (137) 39.1 (43) 41.8 (56) 41.7 (531)

Cultural tolerance score (max.8): mean (SD) 5.33 (2.02) 5.08 (1.96) 5.00 (2.15) 5.27 (1.86)

Personal agreement with subservience

A woman should obey her husband 78.0 (245) 75.7 (84) 76.3 (103) 83.6 (1068)
A working woman should give her money to her husband 33.8 (106) 35.1 (39) 34.1 (46) 41.2 (525)
A man has the final say in family matters 47.1 (148) 50.5 (56) 46.7 (63) 52.5 (670)
A man should share house work 60.4 (189) 67.6 (75) 68.2 (92) 60.1 (767)
A woman needs her partners’ permission to work 66.2 (208) 67.6 (75) 65.9 (89) 71.7 (916)
A woman can do nothing if her husband takes girlfriends 25.8 (81) 24.3 (27) 23.0 (31) 26.4 (337)
If a wife does something wrong her husband has a right to punish her 38.9 (122) 36.0 (40) 34.8 (47) 41.1 (525)
If a man beats you it shows he loves you 24.8 (77) 25.2 (28) 24.4 (33) 25.2 (321)

Personal subservience score (max.8): Mean (SD) 3.54 (2.02) 3.47 (2.09) 3.27 (2.20) 3.81 (2.05)

Believes a married woman can refuse sex with her husband 44.3 (139) 41.4 (46) 42.2 (57) 45.9 (586)

Agrees it is sometimes or always acceptable for a man to hit his wife 14.1 (44) 11.7 (13) 10.4 (14) 13.6 (174)

Acceptability of violence in conflict resolution

In more than one situation it is sometimes or always acceptable for 80.6 (253) 75.7 (84) 77.0 (104) 74.9 (958)
an adult to hit another adult

5 years. The men were less likely to be students and more
likely to be unemployed, or if working, migrant workers.
The male partners of women abused in the past year
preferred boy children and were more likely to have
beaten her children during the year. They earned more
money than the women. The relationships were more
likely to be characterised by her being afraid of him, less
likely to be considered ‘good’, and much more likely to
have conflict often or all the time. There was less
communication about family planning and HIV pre-
vention in abusive relationships. The men were much
more likely to drink alcohol and both partners’ drinking
was often a source of conflict. In addition to the
variables described in Tables 1 and 2 these variables
were potential explanatory variables for the logistic
regression models presented in Table 4 for having
experienced in the past year either domestic violence
alone or domestic violence and threats.
Table 4 shows the results from the three multiple
logistic regression models. The factors which were
significantly positively associated with domestic violence
in all three models were her not having post-school
education, drinking, having someone to confide in about
her relationship, and having more liberal ideas about
women’s position. Lifetime experience of domestic
violence was positively associated with having violence
1610 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617

Table 3
Social and demographic characteristics of main male partner and of relationship

Physically abused in Physical violence and All women

past year % (n) threats in past year % (n) %(n)

One partner only in last year 87.4 (97) 88.9 (120) 95.6 (1113)

Couple married or live together 69.4 (77) 68.9 (93) 63.4 (733)

Husband is polygamous 14.2 (15) 12.2 (17) 18.2 (207)

Man has other girlfriends 37.8 (42) 36.3 (49) 22.6 (263)
None 29.7 (33) 31.1 (42) 34.8 (405)
Unsure/may have 32.4 (36) 32.6 (44) 42.6 (496)

Education (if known): None 25.7 (28) 26.7 (36) 20.6 (232)
Primary 35.8 (39) 31.1 (42) 25.5 (287)
Lower secondary 19.3 (21) 18.5 (25) 20.4 (230)
Higher secondary 15.6 (17) 17.8 (24) 24.1 (271)
Any post-school 3.7 (4) 5.9 (8) 9.3 (105)

Woman does not know man’s educational level 1.8 (2) 2.2 (3) 3.4 (39)

Mean age difference (SD) years 4.5 (7.66) 5.04 (8.00) 6.2 (6.35)

Age difference over 5 years 42.5 (45) 43.9 (57) 54.4 (625)

Working 71.2 (79) 71.9 (97) 72.6 (845)

Studying 3.6 (4) 3.7 (5) 6.0 (70)
Unemployed 25.3 (28) 24.4 (33) 21.4 (249)

Away from home with work: Migrant worker 36.9 (41) 38.5 (52) 45.5 (529)

Urban childhood 69.4 (77) 68.9 (93) 79.6 (927)

Rural childhood 27.9 (31) 28.2 (38) 17.8 (207)
Not known 2.7 (3) 3.0 (4) 2.6 (30)

Preference for boy children 44.1 (49) 43.0 (58) 29.1 (339)
Prefers girls or has no preference 55.9 (62) 57.0 (77) 70.9 (825)

He has beaten her children in last year 27.9 (31) 28.9 (39) 15.5 (180)

He gets more money than her 33.3 (37) 33.3 (45) 29.7 (346)

Frequency of conflict
Often or all the time 39.6 (53) 37.8 (51) 11.6 (135)
Sometimes 55.9 (62) 57.8 (78) 51.8 (600)
Never 4.5 (5) 4.4 (6) 36.5 (423)

Woman was ever afraid of this partner 57.7 (64) 54.8 (74) 38.2 (442)

Woman says relationship is good with partner 67.6 (75) 69.6 (94) 85.6 (990)

Have talked about family planning 46.9 (52) 51.1 (69) 53.6 (622)
Have talked about HIV prevention 32.4 (36) 34.8 (47) 39.2 (455)

Drinks alcohol 75.2 (82) 74.1 (100) 53.8 (605)

His drinking has caused conflict 68.1 (62) 49.6 (67) 16.7 (194)

Her drinking has caused conflict 20.7 (23) 18.5 (25) 5.4 (63)
 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617 1611

Table 4
Multiple logistic regression models

Lifetime abuse Violence in past year Violence or threats in past year

Variables Disaggregated MMLE (adjusting Disaggregated MMLE (adjusting Disaggregated MMLE

for interviewers)a for interviewers)a (adjusting for interviewersa)

Odds ratio 95% CI pvalue Odds ratio 95% CI pvalue Odds ratio 95% CI pvalue
b
Province
Mpumalanga 1.28 0.77–2.12 0.347 1.18 0.60–2.29 0.634 1.31 0.74–2.32 0.353
Northern Prov. 0.72 0.42–1.23 0.230 0.45 0.20–1.00 0.050 0.52 0.26–1.04 0.064

Mother beaten 1.93 1.36–2.72 o0.001

Post-school education 0.29 0.13–0.65 0.003 0.11 0.01–0.91 0.040 0.33 0.09–1.16 0.084

Woman drinks 2.55 1.67–3.88 o0.001 2.73 1.52–4.91 0.001 2.64 1.54–4.55 o0.001

Frequently beaten as child 1.64 1.20–2.22 0.002 1.64 1.00–2.68 0.049

No one to confide in 0.53 0.38–0.75 o0.001 0.54 0.29–0.98 0.042 0.47 0.28–0.82 0.007

Personal subservience 0.90 0.83–0.97 0.006 0.88 0.77–1.00 0.052 0.88 0.78–0.98 0.021

He has a boy child preference 2.13 1.25–3.65 0.006 1.78 1.11–2.84 0.016

Money for house from outside 0.57 0.34–0.97 0.038

Conflict in relationship
Some 5.26 2.00–13.84 0.001 5.99 2.50–14.34 o0.001
All the time 16.72 5.88–47.60 o0.001 18.55 7.12–47.85 o0.001

Conflict over his drinking 3.98 2.41–6.59 o0.001 3.18 2.01–5.02 o0.001

Woman has >1 partner 4.53 1.85–11.10 0.001 3.33 1.46–7.56 0.004
Wald w2 ¼ 82:47; p ¼ o0:0001 Wald w2 ¼ 126:28; p ¼ o0:0001 Wald w2 ¼ 134:31; p ¼ o0:0001
a
The clustering effect due to interviewer was much more important than that due to enumeration areas.
b
Province has been included in all models although the differences were not statistically significant in the first.

in the women’s childhood (both towards self and
mother). Violence in the past year was also positively
associated with having been beaten as a child (but not
mother’s abuse) and having the main finance for the
household coming from either of the partners (rather
than a third person). Some other factors were found to
be significantly positively associated with violence or
threats in the past year or violence alone. These were the
man preferring boy children, having high levels of
conflict in the relationship, the woman having more than
one partner in the last year, conflict over the man’s
drinking and not living in the Northern Province.
Discussion
The study shows that domestic violence is experienced
by many women in South Africa, in marital, co-habiting
and dating relationships. The life time and one year
prevalence reported here were in the middle of the range
of those found in other countries (Heise, 1998). The high
proportion of women abused in the past year who were
injured raises the possibility that less severe forms of
violence (e.g. slaps) have been under-reported. This
problem has been reported from other areas with much
general tolerance of violence (Rao, 1997). It is impos-
sible to estimate the impact of this bias on the study’s
findings.
An important theory of domestic violence causation
relates to the inter-generational cycling of violence. This
is visible in Heise’s framework in her individual level
factors and has been said to be more important for
perpetration of violence than for victimisation (Heise,
1998). Our finding that being beaten in childhood and
witnessing mother’s abuse were important for women’s
life-time experiences of violence, confirm those reported
elsewhere (Hotaling & Sugarman, 1986), although this
finding has not been consistent. Ellsberg et al. (1999)
1612 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617
found that witnessing maternal abuse was not associated
with having ever been abused, however she did not have
information on physical punishment in childhood.
Although the questionnaire did not indicate that we
were only interested in severe beating in childhood (as
opposed to slaps), the patterns of responses suggest that
this is what was reported. In this study the prevalence of
knowing that the mother was beaten was similar in all
three abuse groups but it was only independently
associated in the first model (lifetime domestic violence).
A possible explanation is that some women from violent
homes learn in childhood attitudes which place them at
subsequent risk. These may relate to viewing violence as
normative, with tolerance of types of behaviours by men
which may ultimately be associated with domestic
violence. The non-visibility of this risk factor amongst
the one year prevalence group may suggest that women
whose mothers were beaten are more likely to leave after
violence starts or get it to stop through other actions.
The association between physical punishment in child-
hood and adult domestic violence suggests that beating
teaches children the ‘normality’ of using violence in
punishment and conflict situations. It may also impact
on a woman’s self-esteem, reducing her ability to leave
potentially violent relationships.
There are many theories linking domestic violence to
resources. It has been suggested that men beat women
when they lack other resources to call upon to control
them (Goode, 1970), alternatively that the relationship
between poverty and intimate partner violence is
mediated through stress (Gelles, 1974). A large study
of men in Thailand, however, failed to confirm the stress
hypothesis (Hoffman et al., 1994). More recently
Bourgois (1996) has argued that poverty and unemploy-
ment reduce the ability of men to attain certain ideals of
‘successful’ manhood, particularly those based on ideas
of men as providers for the family. A consequence of
this is the emergence of new ideas of masculinity, which
then are attainable, but emphasise misogyny, substance
use and participation in the underground economy.
Violence against women in normalised as men lash out
at women they can no longer patriarchally control or
economically support.
Research in many different settings has found
domestic violence to be associated with poverty (Martin
et al., 1999; Ellsberg et al., 1999; Hoffman et al., 1994;
Allen & Straus, 1980), male unemployment (Stark et al.,
1981) and status differences between partners (Yllo &
Bogard, 1988). In this study a range of socio-economic
indicators were examined including unemployment.
Status and resource differences indicated by employ-
ment, income and age disparity were also examined.
None of these were found to be independently asso-
ciated with domestic violence.
However, in 43% of all households the main source of
money was from someone other than the woman or her
intimate partner. In just over half of these households
(58%) the money came from the woman’s parents, but
in the rest (17.5% of the total sample) it was another
person. Fieldworkers reported that there was commonly
only one breadwinner among several co-habiting adults
or homes dependant on remittances of a non-resident
earner in the extended family. The women interviewed
from these households were younger than those who
supported themselves (mean age 26.3 v. 34.8 years) but
they were none the less adults and mostly had children.
These women were likely to be particularly poor, yet
being financially supported by a third party was a
protective factor. The apparent paradox could be
explained by this situation reducing the scope for
conflict over resources. Indeed, this was found here.
Conflict levels were lower in relationships where women
received their household resources mainly from a third
party (Chi square for trend, p ¼ 0:064), and lower still in
households where the money came from a third party
who was not a parent (p ¼ 0:007). In relationships where
money came from outside, finance was reported
significantly less often to be a source of conflict
(po0:0001). This appears to confirm the finding from
Thailand (Hoffman et al., 1994) that an important part
of the relationship between poverty and domestic
violence is explained through greater conflict over
resources in poor households.
Education of women was dichotomised for the
purpose of model building because a linear relationship
between different levels of educational completion and
risk of domestic violence was not found. Having some
form of post-school education was found to be
protective against having ever been abused, as was
having none although this protection was less substan-
tial. This been observed elsewhere (Straus, Gelles, &
Steinmetz, 1980). In the model for violence and threats
in the last year having post-school education was only
significant at the 10% level, but has been included as the
confidence intervals are wide due to very small numbers
of women with this level of education. Other authors
have found that women who had higher education
(McCall & Shields, 1986; Straus et al., 1980) or were
economically independent were less likely to be abused
(Rao, 1997; Schuler, Hashemi, Riley, & Akhter, 1996)
(although Ellsberg et al. (1999) did not find it). It seems
likely that the mechanism of protection is not just
through economic independence, as many other women
with less education were economically independent, but
also a greater social empowerment (i.e. social networks,
self-confidence, or an ability to utilise sources of
information and resources available in society). One
indicator of this is the lower personal subservience
scores of more educated women. Ethnographic research
suggests that women who are active in community fora
may be relatively protected (Counts et al., 1992;
Levinson, 1989; Schuler et al., 1996).
 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617
Social isolation of the woman has been shown to be a
risk factor for domestic violence, preceding the onset of
abuse (Neilsen, Russell, & Ellington, 1992). It is also a
strategy of abusive partners to increase their control and
it reduces women’s ability to leave abusive relationships.
In countries where violence is not regarded as a ‘private’
matter domestic violence is less common (Counts et al.,
1992). In this study abused women were more likely to
report having someone to confide in about relationship
problems. This may be because seeking a confidante
(either friend or family) followed experiences of vio-
lence, but it is also possible that (particular types of)
close friendships could be a cause of conflict as men
often try to influence who their partners can associate
with (Wood & Jewkes, 2001).
Domestic violence is strongly related to conservative
ideas about the position of women (Sugarman &
Frankel, 1996). In this study boy-child preference was
used an indicator of the male partner’s conservatism,
and this was shown to be positively associated with
abuse in two of the multiple logistic regression models.
The opposite relationship pertains for women. This
confirms the findings of Sugarman & Frankel (1996)
that abused women have more liberal ideas about
gender roles. It is possible that women who are more
prepared to accept patriarchy are those who have so far
experienced it as a less pernicious force (i.e. have never
been beaten). Alternatively Counts et al(1992) have
argued that in societies where women’s status is in
transition, violence is ‘needed’ to enforce male authority.
It may also be used in these circumstances to resolve
crises of male identity which are provoked by challenges
to patriarchal control (Moore, 1994). It is also possible
that holding strongly conservative views on gender
norms is related to greater reluctance to report abuse in
a survey.
Marital conflict is repeatedly found to be associated
with wife assault (Hotaling & Sugarman, 1986; Straus
et al., 1980; Stets, 1990). This was confirmed in this
study. It is not possible from this data to establish
whether high levels of marital conflict precede domestic
violence or follow it, perhaps because women who
experience violence come to love and respect their
partner less and so become more confrontational.
However, given the ‘normality’ of use of violence in
conflict situations in South Africa the finding of a
relationship is entirely as expected.
Conflict over the man’s drinking, was also found to be
positively associated. Male alcohol consumption is
frequently associated with domestic violence (Rao,
1997; Hoffman et al., 1994). The 1998 South Africa
Demographic and Health Survey found that 67% of
men had always or sometimes used alcohol before recent
episodes of domestic violence (Department of Health
forthcoming). In this study conflict over the partner’s
drinking was more important than his drinking per se.
1613
This may be partly because of associations between
conflict and violence, but there may also be an element
of reverse causation, if the conflict over drinking occurs
because the partner hits the woman when he is drunk.
The reasons why men may be drunk when beating
women are complex. Some South African men drink in
order to give women the beating they feel is socially
expected of them (Abrahams et al., 1999; Soul City,
1997). The disinhibiting effects of alcohol may be more
likely to make conflict when inebriated result in violence,
however social anthropological research suggests that
connections between violence and drunkenness are
socially learnt (McDonald, 1994, p.14). A further factor
is that for a woman to criticise a man is often regarded
as a gender role transgression and if this happens when
he is drunk it may be more likely to result in a fight.
Alcohol consumption by the woman was strongly
positively associated with domestic violence, irrespective
of whether it was itself a source of conflict. It is not
possible to determine whether the woman’s drinking
preceded abuse or followed it, but it seems most likely
that both occur. Women’s alcohol consumption can
cause conflict (and may be seen as a gender role
transgression depending on the type of alcohol and
circumstances of consumption) and can make it more
likely that a dispute will have a violent resolution. The
strength of the association of domestic violence with the
woman’s alcohol drinking is probably further influenced
by the fact that drunkenness amongst men and women
who drink is very common (Department of Health,
forthcoming).
The woman having more than one partner in the last
year was independently and positively associated with
abuse. The mechanisms of risk have been described in
detail by Wood and Jewkes (2001). It may be jealousy;
reverse causality if, after violence, a relationship is
terminated and then another started; taking a second
partner in search of love without terminating the first
relationship for fear of further violence; or violence
associated with trying to leave relationships.
This study also found that women in the Northern
Province were less likely to report abuse. This may
reflect greater conservatism in the province, with a
greater reluctance to disclose, but this was not consis-
tently demonstrated on attitude questions in the survey.
One major difference, however, was in cultural meanings
of domestic violence as 58% of respondents said that it
was culturally accepted that beating signifies love and
34% personally agreed with this. These proportions
were very much greater than those found in the other
provinces (Jewkes, Penn-Kekana, Levin, Ratsaka, &
Schrieber, 1999). There are other indications, however,
that the Northern Province is less violent than other
provinces. In 1999 police rape reports were half the
national average and the murder rate was one third the
national average (CIAC, 2000). Substantial differences
1614 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617
in abuse prevalence within reasonably small geographi-
cal areas have been shown in other studies (Martin et al.,
1999) and may provide important opportunities for
further understanding abuse causation and prevention.
Reflecting on what was not found associated with
abuse is of interest. None of the standard social and
demographic indicators (age, race, urban residence,
education, employment, socio-economic status) were
found to be independently associated with domestic
violence. The fact that relationships with abuse were
only slightly more likely to be co-habiting than those
without is an indicator of the substantial problem of
dating violence in South Africa and the extent to which
this is tolerated (see also Wood et al., 1998). Crowding
at home, measured by the mean number of people per
room, was also found not to be associated with abuse, as
has been previously suggested (Heise, Ellsberg, &
Gottemoeller, 1999; INCLEN, 2000). It was anticipated
that there would be an association between abuse and
perceiving the use of physical force by men sometimes to
be acceptable or beating to be a sign of love. It was not
found here although acceptability could be related to
under-reporting. Communication on family planning
and HIV, as indicators of general levels of communica-
tion, were also not independently associated with
domestic violence.
Implications for understanding causation of domestic
violence
The findings of this research in South Africa provide
support for some but not all of the factors described in
Heise’s ecological framework (1998). Three factors in
the model (male childhood beatings, having an absent
father and gang membership) were not tested in this
study but have been shown to be not independently
associated with domestic violence in a large study of
men and domestic violence in Cape Town (Abrahams,
personal communication). As well as a non-association
found with several of the model’s component factors,
the findings of this paper suggest that it has some
inherent conceptual weaknesses. These would not
detract from its value as an advocacy tool and a pointer
for interventions, but reduce its contribution to under-
standing intimate partner violence.
One problem inherent in the ecological framework is
the difficulty in scientifically conceptualising its different
levels, in particular defining and distinguishing between
‘community’ and ‘society’. Both of these notions are
highly contested in social scientific literature (e.g. Kuper,
1992; Cohen, 1985). Even if these categories are used
casually, as is popular discourse, the factors influencing
intimate partner violence are found to operate at several,
or all, of them. The process of allocation to any one level
is essentially arbitrary and, perhaps more importantly,
conceals the impact of one factor on others in the model
as well as interrelationships between factors. For
example, poverty is said to be a community level factor,
but in the discussion above it has been mentioned that it
may impact on an individual level through its impact on
male identity as experienced by an individual man, at a
relationship level through its impact on conflict over
resources, and at a ‘community’ level through its impact
on shared ideas of successful manhood. Indeed, ‘socie-
tal’ poverty, i.e. living in a low income country, increases
the likelihood of lower levels of education, lower
household incomes and so forth which all may impact
on risk of domestic violence.
Whilst this study in South Africa was cross-sectional
and so on its own cannot demonstrate causation, it adds
to the growing body of evidence on intimate partner or
domestic violence. In the data discussed here and the
broader body of literature to which this is added, two
important factors underlying violence against women by
intimate partners are visible. These are the unequal
position of women and the acceptance of certain forms
of inter-personal violence. We would hypothesise that
these are necessary for domestic violence to be found. If
we are correct, however, these factors would explain
why domestic violence is common in a given setting and
why women are the targets of expressions of violence,
but they fail to explain why certain men, and not others,
beat women within a particular society. In under-
standing this it may be useful to turn to post-
structuralist literature on identity. The feminist anthro-
pologist Moore (1994) has argued that experience of
male identity is bound up with experiences of power.
Challenges to the exercise of power by men are perceived
as threats to masculine identity (and vice versa). She
argues that intimate partner violence occurs when male
self-representations and social evaluations are threa-
tened by the behaviour of others (their female partner),
although often the threat is perceived rather than real.
For example, violence may result from a woman being
seen talking with another man, in which case this
response is to a potential of infidelity rather than proof
of an act of infidelity. An inability to maintain the
fantasy of power triggers a crisis of identity and violence
is a means of resolving this crisis because it acts
to reconfirm the nature of powerfulness otherwise
denied. Moore argues that violence should not be
seen as a breakdown of (social) order but as a sign of
struggle for the maintenance of certain fantasies of
identity and power, specifically those based on ideas of
male superiority over women. This theory has been
shown to be very useful in explaining dating violence of
South African youth, whilst acknowledging that in-
timate partner violence may be used as an expression of
anger, resentment and disappointment (Wood & Jewkes,
2001).
Thus we conclude by suggesting that an alternative
model is needed to summarise the current state of
 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617
Fig. 1. Causation of intimate partner violence.
knowledge internationally on causation of intimate
partner violence. Such a model needs to present it as a
constellation or web of associated and mediating factors
and processes which are centrally influenced by ideas
about masculinity and the position of women in a
society and ideas about the use of violence. A possible
model for this is depicted in Fig. 1. This indicates that
ideologies of male superiority legitimate the disciplining
of women, often for transgressions of conservative
female gender roles, and often also the use of force in
this. They construct women as legitimate vehicles for the
reconfirmation of male powerfulness through beating,
since this is a demonstration of male power juxtaposed
against the lesser power of women. They act to
disempower women through denying equal access to
education, employment and political roles. In so doing,
they impact on women’s perceptions of self-worth and
self-efficacy, as well as reducing their social and
economic ability to leave a relationship, return to family
and/or live alone and thus severely curtail their ability to
act against an abuser. It has been argued that violence
against women is much more common where inter-
personal violence is generally common and it has been
shown that childhood experiences of violence in the
home serve to reinforce the normative nature of violence
and thus increase the likelihood of women’s acceptance
of abuse. Where domestic violence is normative,
sanctions against violent men are usually low and the
exercise of violence is often seen as inherently masculine.
Alcohol contributes to violence through enhancing the
likelihood of conflict, reducing inhibitions and providing
1615
social space for punishment. Similarly, it seems likely
that the effects of poverty are mediated through their
impact on conflict over resources, women’s ability to
leave relationships, ideologies of masculinity and on
men’s ability to see themselves as ‘successful men’.
Ultimately strategies for prevention of violence
against women need to focus on removing the central
pillars through engaging with a broad intersectoral
agenda of women’s empowerment in society and
reducing normative use of violence in conflict situations.
However other strategies which address some of the
proximal factors, for example alcohol consumption or
supporting abused women through NGOs in a commu-
nity, may ultimately impact on levels of violence.
Considerable further research is needed on these
processes and the extent of impact of interventions on
one factor. The pathway for some of these may not be
direct, as interventions addressing some of the periph-
eral factors in the model may impact in the short or
medium term on several other parts. The nature of this
impact may be such that it places women at greater risk
of violence. So, for example, measures which somewhat
empower women may have an immediate impact on
increasing the frequency of gender norm transgressions
or reduce personal agreement with a subservient
position for women and these may place her at greater
immediate risk of violence. Only when empowerment
has reached a critical level will its benefits outweigh this.
This should not be a reason not to empower women but
needs to be further understood and taken into account
in programme planning.
1616 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617
Acknowledgements
This study was funded by the Department of Arts,
Culture, Science and Technology National Innovation
Fund and the Medical Research Council. We would like
to thank: Matsie Ratsaka for contributions towards
study design and fieldwork supervision; Charlotte Watts
for her technical advice; to all the staff of the
Masisukumeni Women’s Crisis Centre who assisted us
with development, piloting and interviewing; the North-
ern Province Office for the Status of Women who assisted
with fieldworker recruitment; Zodumo Mvo for super-
vising the Eastern Cape field work; Tebogo Gumede for
help with coding and sampling; Engela Gerber for
administrative support; Alta Hansen and Zenobia
Kiewiets for data entry and all our field workers. Most
of all we would like to thank the women who shared these
intimate details of their lives with us so openly.
References
Abrahams, N., Jewkes, R., & Laubscher, R. (1999). I don’t
believe in democracy in the home. Men’s relationships with
and abuse of women. MRC Technical Report: Cape Town.
Allen, C. M., & Straus, M. S. (1980). Resources, power and
husband-wife violence. In M. A. Straus, & G. T. Hotaling
(Eds.), The social cause of husband-wife violence. Minnea-
polis: University of Minnesota Press.
Bourgois, P. (1996). In search of respect: selling crack in El
Barrio. Cambridge: Cambridge University Press.
Breslow, N. E., & Clayton, D. G. (1993). Approximate
inference in generalised linear mixed models. Journal of
the American Statistical Association, 88, 1–24.
Burman, S. (1990). Fighting a two-pronged attack: the
changing legal status of women in Cape-ruled Basutoland
1872–1884. In C. Walker (Ed.), Women and gender in
Southern Africa to 1945. Cape Town: David Philip Publish-
ers.
Campbell, J., Dienemann, J., Jones, A., Schollenberger, J.,
Geilen, A., O’Campo, P., & Wynne, C. Physical health
consequences of intimate partner violence in a sample of
female HMO enrolees. Annals of Internal Medicine, forth-
coming.
CIAC (2000). Unpublished data. Crime Information Analysis
Centre, South African Police Service, Pretoria.
CIET Africa (1998). Prevention of sexual violence. A social
Audit of the role of the police in the jurisdiction of
Johannesburg’s Southern Metropolitan Local Council.
Johannesburg: CIET Africa.
CIET Africa (2000). Beyond victims and villains. The culture of
sexual violence in south Johannesburg. Johannesburg: CIET
Africa.
Cohen, A. (1985). The symbolic construction of community.
London: Routledge.
Conroy, R. M., Elmore-Meegan, M., Joyce, T., McGuigan, K.
G., & Barnes, J. (1996). Solar disinfection of drinking water
and diarrhoea in Masai children. Lancet, 348, 1685–1697.
Counts, D., Brown, J., & Campbell, J. (1992). Sanctions and
sanctuary. Boulder, Colorado: Westview Press.
Department of Health. South Africa Demographic and Health
Survey 1998. Final Report. Pretoria: Department of Health,
forthcoming.
Ecob, R. (1996). A multilevel modelling approach to examining
the effects of area of residence on health and functioning.
Journal of the Royal Statistical Society A, 159, 61–75.
Ellsberg, M., Heise, L., Pena, R., Agurto, S., & Winkvist, A.
(2001). Researching violence against women, methodologi-
cal considerations from three Nicaraguan Studies. Studies in
Family Planning, 32, 1–16.
Ellsberg, M. C., Pena, R., Herrera, A., Liljestrand, J., &
Winkvist, A. (1999). Wife abuse among women of child-
bearing age in Nicaragua. American Journal of Public
Health, 89, 241–244.
Gelles, R. J. (1974). The violent home. Beverley Hills, CA: Sage.
Goldstein, H., & Rasbash, J. (1996). Improved approximations
for multilevel models with binary responses. Journal of the
Royal Statistical Society A, 159, 505–513.
Goode, W. J. (1970). Force and violence in the family. Journal
of Marriage and the Family, 33, 624–636.
Guy, J. (1990). Gender oppression in Southern Africa’s
precapitalist societies. In C. Walker (Ed.), Women and
gender in Southern Africa to 1945 (p. 40). Cape Town: David
Philip Publishers.
Harvey, P., & Gow, P. (Eds.) (1994). Sex and violence: issues in
representation and experience. London: Routledge.
Heise, L. (1998). Violence against women: an integrated,
ecological framework. Violence Against Women, 4, 262–
40290.
Heise, L., Ellsberg, M., & Gottemoeller, M. (1999). Ending
violence against women. Population Reports Vol. XXVII,
Number 4. Population Information program, Center for
Communication Programs, The John Hopkins University
School of Public Health.
Hill, A. B. (1965). The environment and disease: association or
causation? Proceedings of the Royal Society of Medicine, 58,
295–300.
Hoffman, K. L., Demo, D. H., & Edwards, J. N. (1994).
Physical wife abuse in a non-Western society: an integrated
theoretical approach. Journal of Marriage and the Family,
56, 131–146.
Hotaling, G. T., & Sugarman, D. B. (1986). An analysis of risk
markers in husband to wife violence: the current state of
knowledge. Violence and Victims, 1, 101–124.
Hox, J. J. (1994). Hierarchical regression models for interviewer
and respondent effects. Sociological Methods and Research,
22, 300–318.
International Clinical Epidemiologists Network. (2000). Do-
mestic violence in India. A summary report of a multi-site
household survey. Washington, DC: International Centre
for Research on Women.
International Criminal Police Organisation ICPO-Interpol
(1996) International Crime Statistics.
Jewkes, R., & Abrahams, N. The epidemiology of rape and
sexual coercion in South Africa: an overview. Social Science
and Medicine, in press.
Jewkes, R., Abrahams, N., & Mvo, Z. (1998). Why do nurses
abuse patients? Reflections from South African obstetric
services. Social Science and Medicine, 47, 1781–1795.
 R. Jewkes et al. / Social Science & Medicine 55 (2002) 1603–1617
Jewkes, R., Penn-Kekana, L., Levin, J., Ratsaka, M., &
Schrieber, M. (1999). ‘He must give me money, he mustn’t
beat me’. Violence against women in three South African
Provinces. Medical Research Council Technical Report,
Pretoria.
Jewkes, R., Penn-Kekana, L., Levin, J., Ratsaka, M., &
Schrieber, M. (2001). The prevalence of physical, sexual
and emotional violence against women in three South
African Provinces. South African Medical Journal, 91(5),
421–428.
Jewkes, R., Vundule, C., Maforah, F., & Jordaan, E. (2001).
Relationship dynamics and adolescent pregnancy in South
Africa. Social Science and Medicine, 5, 733–744.
Jewkes, R., Watts, C., Abrahams, A., Penn-Kekana, L., &
Garcia-Moreno, C. (2000). Conducting ethical research on
sensitive topics: lessons from gender-based violence research
in Southern Africa. Reproductive Health Matters, 8, 93–103.
Kim J., & Motsei M. ‘‘Women enjoy punishment.’’ Attitudes
and experiences of gender violence among primary health
care nurses in rural South Africa. Soc Sci Med (in press).
Kuper, A. (1992). Conceptualising society. London: Routledge.
Levin, J. (1999). The use of variance component models in the
analysis of complex surveys. Unpublished Ph.D. Thesis.
Pietermaritzburg: University of Natal.
Levinson, D. (1989). Violence in cross-cultural perspective.
Newbury Park, CA: Sage Publications.
Mager A. K. (1999). Gender and the making of a South African
Bantustan. A social history of the Ciskei, 1945–1959. David
Phillip, Cape Town.
Martin, S. L., Tsui, A. O., Maitra, K., & Marinshaw, R. (1999).
Domestic violence in Northern India. American Journal
Epidemiology, 150, 417–426.
McCall, G. J., & Shields, N. M. (1986). Social and structural
factors in family violence. In M. Lystad (Ed.), Violence in
the home: interdisciplinary perspectives. New York: Brunner/
Mazel.
McDonald, M. (Ed.) (1994). Gender, drink and drugs (p.14).
Oxford: Berg Publishers.
Moore, H. (1994). A passion for difference: essays in anthro-
pology and gender. London: Polity Press.
Musasa, (1996) Questionnaire for survey of violence against
women in the Midlands Province, Zimbabwe. Musasa
Project, Harare.
Neilsen, J., Russell, E., & Ellington, B. (1992). Social isolation
and wife abuse: a research report. In E. C. Viano (Ed.),
Intimate violence: interdisciplinary perspectives. Washington
DC: Hemisphere Publishing Corp.
O’Muircheartaigh, C., & Campanelli, P. (1998). The relative
impact of interviewer effects and sample design on survey
precision. Journal of the Royal Statistical Society A, 161, 63–
77.
Peden, M. (2000). A profile of Injuries in 15–34 year olds in
South Africa 1999. Report produced by the National Injury
Surveillance System, Medical Research Council, Parow.
Rao, V. (1997). Wife-beating in rural south India: a qualitative
and econometric analysis. Social Science and Medicine,
44(8), 1169–1180.
1617
Schuler, S. R., Hashemi, S. M, Riley, A. P., & Akhter, S.
(1996). Credit programmes, patriarchy and men’s violence
against women in rural Bangladesh. Social Science and
Medicine, 43, 1729–1742.
Simpson, G. (1991). Explaining sexual violence: some back-
ground factors in the current socio-political context. Project
for the Study of Violence, Johannesburg.
Skinner, C. J. (1989). Introduction to Part A. In C. J. Skinner,
D. Holt, & T. M. F. Smith (Eds.), Analysis of complex
surveys. Chichester: Wiley.
Skinner, C. J. (1989a). Domain means, regression and multi-
variate analysis. In C. J. Skinner, D. Holt, & T. M. F. Smith
(Eds.), Analysis of complex surveys. Chichester: Wiley.
Soul City (1997) Violence against women. A report. Social
Surveys, Braamfontein, Johannesburg
Stark, E., Flitcraft, A., Zuckerman, B., Grey, A., Robinson, J.,
& Frazier, W. (1981). Wife abuse in the medical setting: an
introduction for medical personnel. Monograph #7. Washing-
ton DC: Office of Domestic Violence.
Statacorp (1997). Stata Statistical Software: Release 6.0.
College Station, TX: Stats Corporation.
Steele, F., Diamond, I., & Amin, S. (1996). Immunisation
uptake in rural Bangladesh: a multilevel analysis. Journal of
the Royal Statistical Society A, 159, 289–78299.
Stets, J. E. (1990). Verbal and physical aggression in marriage.
Journal of Marriage and the Family, 52, 501–514.
Straus, M. A., Gelles, R. J., & Steinmetz, S. K. (1980). Behind
closed doors: violence in the American family. New York:
Anchor Press.
Sugarman, D., & Frankel, S. L. (1996). Patriarchal ideology
and wife-assault: a meta-analytic review. Journal of Family
Violence, 11, 13–40.
Truth and Reconciliation Commission (1998). The TRC
Report. Http://www.polity.org. za/govdocs/commissions/
1998/trc/index.htm
Usdin, S., Christofides, N., Malepe, L., & Maker, A. (2000).
The value of advocacy in promoting social change:
implementing the new domestic violence act in South
Africa. Reproductive Health matters, 8(16), 55–65.
Vetten, L. (1995). ‘‘Man shoots wife’’. A pilot study detailing
intimate femicide in Gauteng, South Africa. People Opposing
Women Abuse, Johannesburg.
Wood, K., Jewkes, R. (2001). ‘Dangerous’ love: reflections on
violence among Xhosa township youth. In R. Morrell (Ed.).
Changing men in Southern Africa. University of Natal Press:
Pietermaritzburg and Zed Books Ltd, London.
Wood, K., Maforah, F., & Jewkes, R. (1998). ‘He forced me to
love him’: putting violence on the adolescent sexual health
agenda. Social Science and Medicine, 47, 233–242.
World Health Organisation (1999). Putting women’s safety first:
ethical and safety recommendations for research on domestic
violence against women. WHO/EIP/GPE/99.2. WHO, Gen-
eva.
Yllo, K., Bogard, M. (Eds.) (1988). Feminist perspectives on wife
abuse. Newbury Park, CA: Sage.