REVIEW
The current prevalence of child sexual abuse worldwide:
a systematic review and meta-analysis
J. Barth
•
L. Bermetz
•
E. Heim
•
S. Trelle
•
T. Tonia
Received: 5 March 2012 / Revised: 25 October 2012 / Accepted: 29 October 2012 / Published online: 21 November 2012
Ó Swiss School of Public Health 2012
Abstract
Objectives Systematic reviews on prevalence estimates of
child sexual abuse (CSA) worldwide included studies with
adult participants referring on a period of abuse of about
50 years. Therefore we aimed to describe the current
prevalence of CSA, taking into account geographical
region, type of abuse, level of country development and
research methods.
Methods We included studies published between 2002
and 2009 that reported CSA in children below 18 years.
We performed a random effects meta-analysis and ana-
lyzed moderator variables by meta-regression.
Results Fifty-five studies from 24 countries were inclu-
ded. According to four predefined types of sexual abuse,
prevalence estimates ranged from 8 to 31 % for girls and 3
to 17 % for boys. Nine girls and 3 boys out of 100 are
victims of forced intercourse. Heterogeneity between pri-
mary studies was high in all analyses.
Conclusions Our results based on most recent data con-
firm results from previous reviews with adults. Surveys in
children offer most recent estimates of CSA. Reducing
heterogeneity between studies might be possible by
standardized measures to make data more meaningful in
international comparisons.
Keywords Child sexual abuse International
Epidemiology Prevalence Systematic review
Meta-analysis
Introduction
The devastating long- and short-term consequences of
child sexual abuse (CSA) on the lives of the victims are
reflected in the high public and scientific interest on this
topic (Bolen and Scannapieco 1999; Edgardh 2002; Pereda
et al. 2009a). The need for reliable overall prevalence
estimates of CSA is crucial for health research worldwide,
especially for allocating economic resources in health care
and estimating the burden.
Two recent meta-analyses consistently showed CSA
prevalence of 18–20 % for women and 8 % for men
worldwide (Stoltenborgh et al. 2011; Pereda et al. 2009b).
The narrative review of Andrews et al. (2004) reports that
8.4–67.7 % of females and 3.8–35 % of males have been
sexually abused during childhood. This wide range is only
partly attributed to the geographical region where the study
was conducted. Several studies found a higher prevalence
in Africa than elsewhere but inconsistent findings exist
with regard to other regions (Pereda et al. 2009b; Andrews
et al. 2004; Stoltenborgh et al. 2011; Finkelhor 1994).
Moreover, Stoltenborgh et al. (2011) analyzed the effect of
the level of economic development of a country on CSA
prevalence. They found that for boys, the prevalence was
higher in low-resource countries than in high-resource
countries, whereas no significant effects of the country’s
economic development level emerged for girls.
J. Barth L. Bermetz E. Heim S. Trelle T. Tonia
Institute of Social and Preventive Medicine (ISPM),
University of Bern, Bern, Switzerland
S. Trelle
CTU Bern, University of Bern, Bern, Switzerland
J. Barth (&)
Division of Social and Behavioural Health Research,
Institute of Social and Preventive Medicine (ISPM),
Niesenweg 6, 3012 Bern, Switzerland
e-mail: mail@juergen-barth.de
Int J Public Health (2013) 58:469–483
DOI 10.1007/s00038-012-0426-1
123
Some authors argue that different prevalence estimates
of CSA are the result of differences in methodology of the
primary studies (Bolen and Scannapieco 1999; Edgardh
and Ormstad 2000; Dhaliwal et al. 1996; Gorey and Leslie
1997; Finkelhor 1994). A higher prevalence of CSA is
suggested to be associated with sample type (i.e. college
populations), CSA definition and the number of questions
asked in combined prevalence estimates (Andrews et al.
2004). However, another meta-analysis showed no influ-
ence of the definition of CSA (broad vs. narrow) on the
pooled CSA prevalence (Pereda et al. 2009b). In addition,
Stoltenborgh et al. (2011) found that number of questions
affected pooled prevalence for girls but not for boys.
Whether the prevalence of CSA changes over time is a
matter of controversy: some researchers found a decrease
of CSA from the mid-1990s to 2005 (Gilbert et al. 2009),
whereas others did not find a significant variation over time
(Goldman and Padayachi 2000). If such change in preva-
lence rates over time exists, summarizing prevalence
estimates of different time points might be problematic.
Previous reviews are mainly based on primary studies
which were published in a broad time range (e.g. 1980 until
now) and include both studies with adults and studies with
children. A study with adults from the 1980s assesses CSA
prevalence in the 1950s, whereas a study with children in
2009 assesses more recent prevalence rates. Moreover,
studies with adults may be more prone to potential recol-
lection bias than studies on children (Andrews et al. 2004;
Halperin et al. 1996), which is a further source of bias if
they are mixed in a meta-analysis.
The aim of this study is to summarize the prevalence of
CSA worldwide using the most current data. To achieve
that, we include only papers published after 2002 and
reporting on data collected from 2000 onwards. Further-
more, we only include studies with child and/or adolescent
populations at the time of the study, in order to reduce
recollection bias. The results will be presented stratified for
gender and type of abuse, which was not applied in earlier
meta-analyses. In addition, using meta-regression, we will
examine how methodological aspects (i.e. design of study,
method of data collection, sampling method) and contex-
tual factors (i.e. Human Development Index (HDI), region)
might explain the variation between studies.
Methods
Literature review
A systematic literature search took place in February 2009.
We searched electronic literature databases (Embase,
Medline, PsycInfo and Psyndex) and identified 4,827
potentially eligible studies. The search terms combined
concepts of the population (child or adolescent), the inci-
dent (sexual abuse, assault, molestation) and the study type
(epidemiology, prevalence, proportion). After removal of
duplicates, we were left with 3,295 potentially relevant
studies. In addition, we consulted 75 experts on CSA from
75 different countries, who in turn provided us with names
and contact details of other experts, whom we additionally
consulted. These experts were asked to revise our list of
included studies and to point us towards studies from their
region which we might have missed. This resulted to the
identification of one additional study. Grey literature and
unpublished reports were not included in this study.
Study selection
We included only empirical studies reporting the preva-
lence of CSA for which the data were collected after 2000
and in which the participants were below 18 years old. We
excluded case studies and studies for which the country
was unknown and the sample size was below 1,000. The
latter criterion was applied to exclude studies with low
statistical precision and low reporting quality (N = 178).
After abstract screening, 3,082 studies were excluded,
leaving 213 for full-text screening (see ‘‘Appendix’’ )
applying the same criteria. We were unable to retrieve nine
publications, most of which were dissertations. We man-
aged to translate most articles that were published in
languages other than English, apart from one publication in
Lithuanian. Twelve studies presented data from the same
populations in several publications. These publications are
marked with
1
in the reference list and were analyzed as
one study each. One publication that reported outcomes
separately for community samples and for schools, and one
that reported outcomes separately for two different coun-
tries were considered as reporting two different studies
(Ruangkanchanasetr et al. 2005; Seedat et al. 2004). Two
studies that presented mixed results for child sexual and
physical abuse were excluded. In nine publications, the
data were presented in a way that made it impossible for us
to extract the necessary information. The final list of
included studies consists of 55 studies reported in 65
publications, presenting information about CSA in 24
countries (see
2
in the list of references, and Fig. 3 for a
flowchart of the process of study selection).
Data extraction
We extracted descriptive characteristics (e.g. publication
date, year of data collection, age of the sample, gender). As
outcomes we coded prevalence rates according to the type
1
In references ** included as secondary source
2
In references * included as primary source
470 J. Barth et al.
123
of abuse. Four categories of CSA were used to come to
estimates for very different events: non-contact abuse
(inappropriate sexual solicitation, indecent exposure),
contact abuse (touching/fondling, kissing), forced inter-
course (oral, vaginal, anal, attempted) and mixed sexual
abuse (when different types of abuse had been inquired but
only one prevalence rate was reported or the type of abuse
not specified). Our categorization of CSA differs from
earlier reviews which used more vague distinction (e.g.
narrow vs. broad definition) since we aimed to reduce
variation of outcome between studies.
As contextual moderators, we extracted data on (a) the
region where the study was conducted and (b) the degree of
development in this region according to the HDI ranging
from 0 to 1 with higher values indicating better develop-
ment (Human Developmental Report 2009). We extracted
three methodological moderators: (a) design of study
(primarily cross-sectional or cross-sectional nested in a
longitudinal study); (b) sampling method (random sample
from the general population, school-based or other specific
population); (c) number of items used for the assessment of
CSA; and (d) method of data collection (self-report,
interview by researcher, official registries).
Prevalence rates were stratified according to type of
sexual abuse and gender. Depending on the information
available for each study, we report prevalence on the total
sample or separately on boys and girls (see Table 1). In
cases where CSA is reported separately for boys and girls
but there is no information about the total number of boys
and girls in the sample, we assumed that half of the par-
ticipants were male and half female.
Analysis procedures
Prevalence estimates were computed using the following
logit transformation z ¼ ln
p
1p
with p denoting the pro-
portion of sexually abused in the sample.
We measured prevalence estimates of CSA for each
study by stratifying by gender and type of abuse. Based on
the information gained from previous reviews and meta-
analyses on CSA (Pereda et al. 2009b; Finkelhor 1994;
Andrews et al. 2004), we assumed relatively high between-
study heterogeneity resulting from moderator variables such
as methodological differences between the primary studies.
Therefore, we used random effects models for all summary
statistics because this method explicitly allows for between-
study variability (Higgins and Thompson 2002; Higgins
et al. 2003). When there were less than five studies included
in the pooled analysis, we interpreted the pooled results
using the 95 % confidence interval (CI). Confidence inter-
vals give an idea of where the true value of the prevalence of
CSA lies. When there were five or more studies available in
the category of interest, we report on the prediction interval
(PI) which gives information about the expected prevalence
of a new study in this field. Confidence and prediction
intervals are only reported if the upper value is below 0.50;
meaning in other cases we only give the information of non-
applicability of prediction interval information (n.a.). Het-
erogeneity among studies was examined using the I
2
statistic
(range 0–100 %), which describes the percentage of total
variation across studies that is attributable to heterogeneity
between studies rather than chance (Higgins and Thompson
2002; Higgins et al. 2003). In other words, the variation of
prevalence rates of primary studies is compared with the
expected statistical variation. I
2
values of 25, 50 and 75 %
were considered as low, moderate and high levels of het-
erogeneity, respectively.
As we expected high heterogeneity between studies, we
assessed the impact of methodological moderator variables
(i.e.sampling method) and contextual factors (i.e. HDI) on
the pooled prevalence estimates in meta-regression analy-
ses. Formal tests for interaction using meta-regression were
done to compare stratum-specific prevalence rates. All
analyses were carried out in Stata Release 10 (StataCorp
LP, College Station, TX, 2007). The number of studies was
only sufficient to calculate pooled estimates for forced
intercourse and mixed sexual abuse, but not for contact and
non-contact abuse.
Results
Study characteristics
The majority of the studies were conducted in Asia (16)
and North America (14). Eleven studies came from Europe,
9 from Africa, and five studies were carried out in Central
and South America. No article from Australia or New
Zealand was eligible for our review, mostly due to the use
of adult samples in the studies conducted in these coun-
tries. Table 1 depicts the main characteristics of the studies
and Table 2 gives an overview of methodological aspects
of the studies. The sample sizes ranged from 106 to
127,097, with an average of approximately 7,500. More
precisely, 23 studies had a sample size of up to 1,000
children, 27 between 1,001 and 10,000 children and five
studies with more than 10,000. Seven of them included
only females and eight only males. None of the studies
reported on the prevalence of CSA in populations younger
than 13. In the majority of studies, a cross-sectional design
was used. Most of the samples were recruited in schools
and were evaluated using self-report instruments which
contained 1–15 questions. Fifty-four percent of the studies
was conducted in countries with a high HDI. Thirty studies
reported on mixed sexual abuse or did not define the type
of abuse, and 23 studies reported on forced intercourse.
Child sexual abuse: current situation 471
123
Table 1 Descriptive information of included studies (N = 55)
References Total
(N)
Male
(%)
Female
(%)
Age (mean
or range)
Region HDI* Year of data
collection
Type of prevalence Prevalence rate by gender and
type of abuse
Male Female Total
Aberle et al. (2007) 2,140 40 60 14–18 Croatia High 2005 Lifetime prevalence 6.0
b
3.3
b
Alikasifoglu et al. (2006) 1,871 0 100 16.3 Turkey Medium 2000 Lifetime prevalence 11.3
b
4.9
c
Andersson and Ho-Foster (2008) 127,097 100 0 15 South Africa Medium 2002 Lifetime prevalence 47
c
Aslund et al. (2007) 5,048 50.45 49.54 15–17 Sweden High 2004 Lifetime prevalence 11.9
b
29.0
b
12.5
c
6.8
c
Assis et al. (2004) 1,685 44.3 55.6 11–19 Brazil High 2000 Lifetime prevalence 11.8
d
Audu et al. 2009 316 0 100 14.9 Nigeria Low Unclear Lifetime prevalence 47.7
c
*
Banerjee et al. (2008) 330 14.8 85.1 8–14 India Medium Unclear Lifetime prevalence 3.9
d
*
Banyard and Cross (2008) 2,101 49 51 \18 USA High 2000–2001 Lifetime prevalence 9.4
e
16.8
e
13.2
e
Birdthistle et al. (2008) 863 0 100 16.8 Zimbabwe Low 2004 Lifetime prevalence 8.1
c
Bonino et al. (2006) 804 58 42 14–19 Italy High Unclear Lifetime prevalence 5.9
c
9.0
c
8.0
c
Champion et al. (2004) 106 Unclear Unclear 14–19 Mexico High Unclear Lifetime prevalence 18.0
c
Chen et al. (2003) 239 100 0 17.53 China Medium 2002 Lifetime prevalence 2.1
c
23
d
Chen et al. (2004b) 2,300 49.8 50.2 17.2 China Medium Unclear Lifetime prevalence 8.8
a
12.9
a
10.9
a
2.8
b
6.5
b
4.7
b
1.6
c
2.3
c
2.0
c
10.5
d
16.7
d
13.6
d
Chen (2004) 565 50 50 17 China Medium 2003 Lifetime prevalence 14.3
d
20.0
d
Chen et al. (2006) 351 0 100 17.6 China Medium 2004 Lifetime prevalence 17.4
a
9.1
b
5.4
c
21.9
d
Cheng-Fang et al. (2008) 1,684 49.1 50.9 14.1 Taiwan Medium 2003 Lifetime prevalence 3.0
d
2.0
d
2.5
d
Dassa et al. (2005a, b) 2,400 50 50 \16 Togo Medium 2003 Lifetime prevalence 4.9
d
*
Decker et al. (2007) 5,919 0 100 14–17 USA High 2001, 2003 Lifetime prevalence 14.0
c
Doocy et al. (2007) 263 58 42 1–16 Nepal Medium Unclear Lifetime prevalence 1.0
d
*
Eisenberg et al. (2007) 124,881 48.9 51 11–18 USA High Unclear Lifetime prevalence 2.9
b
7.9
b
Elbedour et al. (2006) 217 Unclear Unclear 16 Israel Medium 2000 Specific age range
prevalence
16.0
a
14.0
b
4.0
c
472 J. Barth et al.
123
Table 1 continued
References Total
(N)
Male
(%)
Female
(%)
Age (mean
or range)
Region HDI* Year of data
collection
Type of prevalence Prevalence rate by gender and
type of abuse
Male Female Total
Fabijanic et al. (2002) 310 46.13 53.87 Unclear Croatia High 2001 Lifetime prevalence 5.0
d
18.0
d
11.0
d
Haavet et al. (2005) 7,329 Unclear Unclear 15–16 Norway High 2000–2001 Specific age range
prevalence
5.8
d
*
Harrison and Narayan (2003) 50,168 49.3 50.7 15 USA High 2001 Lifetime prevalence 4.0
b
10.5
b
Hasnain and Kumar (2006) 150 0 100 Unclear India Medium Unclear Lifetime prevalence 38.0
d
Helweg-Larsen and Boving Larsen (2006) 5,829 49.9 50.06 15–16 Denmark High 2002 Lifetime prevalence 1.1
d
* 4.4
d
*
Kim and Kim (2005) 1,053 52.99 47 12–18 Korea, Republic of High 2001–2002 Lifetime prevalence 0.4
c
5.2
c
Leung et al. (2008) 6,592 49.9 50.1 14.68 China Medium 2005 Specific age range
prevalence
0.6
b
Lien et al. (2007) 7,305 Unclear Unclear 12–17 Norway High 2000, 2001 Specific age range
prevalence
2.0
d
* 6.0
d
* 3.9
d
*
Mitchell et al. (2008) 1,500 48.9 50.13 10–17 USA High 2005 Specific age range
prevalence
13.4
a
Moran et al. (2004) 2,164 54 46 Unclear USA High Unclear Lifetime prevalence 11.7
d
Ndetei et al. (2007) 1,110 56.6 43.3 16.6 Kenya Medium Unclear Lifetime prevalence 16.5
d
Orozco et al. (2008) 3,005 49.3 50.1 12–17 Mexico High 2005 Lifetime prevalence 0.6
c
2.0
c
1.3
c
7.7
d
4.7
d
1.7
d
Polanczyk et al. (2003) 1,193 45.6 54.4 13–20 Brazil High 2000 Lifetime prevalence 2.0
d
2.5
d
2.3
d
Rosenberg et al. (2005) 16,664 49 51 13–18 USA High 2001 Specific age range
prevalence
5.0
c
8.0
c
Ruangkanchanasetr et al. (2005) 426 41 59 15.5 Thailand Medium 2001 Specific age range
prevalence
3.1
c
*
Ruangkanchanasetr et al. (2005) 426 41 59 15.5 Thailand Medium 2001 Specific age range
prevalence
1.4
c
*
Sears et al. (2007) 633 51.18 48.82 14.64 Canada High Unclear Lifetime prevalence 38.0
d
44.0
d
Seedat et al. (2004) 1,140 43.3 41.9 15.9 South Africa Medium 2000 Lifetime prevalence 15.0
d
12.0
d
14.0
d
Seedat et al. (2004) 901 56.7 58.1 15.6 Kenya Medium 2000 Lifetime prevalence 24.0
d
14.0
d
18.0
d
Sesar et al. (2008) 318 19.1 80 15–20 Croatia High 2003 Lifetime prevalence 21.0
d
* 13.0
d
*
Swahn and Bossarte (2007) 13,639 52.31 47.69 \18 USA High 2005 Lifetime prevalence 3.8
c
10.7
c
7.3
c
Taquette et al. (2005) 173 0 100 16.5 Brazil High 2001–2002 Lifetime prevalence 11.6
c
Thurman et al. (2006) 1,172 51.6 48.3 14–18 South Africa Medium 2001 Lifetime prevalence 0.4
c
8.7
c
3.9
c
Turner et al. (2007) 1,000 Unclear Unclear 10–17 USA High 2002–2003 Specific age range
prevalence
13.6
d
Child sexual abuse: current situation 473
123
