We reviewed 337 study reports, which provided 343 prevalence rates, based on self-report from either adults or children. North American studies were most numerous across each category of abuse, whereas South American studies were least numerous. In approximately two-thirds of studies prevalence rates were available for either or both genders. Where differentiated, studies of girls were more common than for boys across all maltreatment categories. Prevalence rates were most commonly available for sexual abuse, then for physical abuse and least commonly for neglect. Median rates of sexual abuse were higher for girls than boys in the three continents with the highest number of studies (North America, Europe, Asia) and there were big differences between continents in actual rates (for example 20.4% and 14.3% for girls in North America and Europe respectively). Median rates of physical abuse were similar for boys and girls in all continents (for example 24.3% and 21.7% respectively in North America) apart from Europe and Africa where it was higher for boys (for example, 60.2 and 50.8 respectively for Africa, while rates varied considerably between continents for both girls and boys. Few studies of emotional abuse were found for Africa, Australia and South America and rates were much higher for girls than boys in North America and Europe but more similar in Asia (33.2% for boys, 26.9% for girls). Finally, a similar picture of study frequency was found for neglect and rates were much higher in North American girls (40.5%) compared to boys (16.6%) but similar across gender in both Europe and in Asia.
Considerable variation in lifetime prevalence rates of self-reported child maltreatment was found between studies, particularly between worldwide studies (between 0.0 and 100.0%), however, the variation in rates reported in UK based studies was still very large (between 0.7 and 77.8%). It is perhaps important to provide some context to the studies that reported the rather surprising extreme rates of 0.0% and 100.0%. Harkness and Monroe (2002) [
25] found that all the females in their study reported that they had suffered neglect at some point, this was a clinical (depressed) sample, and so that may have had a bearing on the results. Khamis et al. (2000) [
26] found that no males in their study had reported sex abuse, the respondents were boys aged 12–16 who were interviewed by school counsellors, it is possible that they therefore may have been reluctant to disclose a history of sex abuse due to discomfort or embarrassment. In both UK and worldwide studies the greatest difference in prevalence rates reported was for neglect. While some of this variation may reflect actual different experiences that children have, there are methodological differences that exist in the research that are likely to give rise to these variations [
7,
9,
27]. We adopted a broad approach to inclusion for the review resulting in a heterogeneous sample of studies and prevalence rates.
Study participants
The age of the participant at time of reporting may have an effect on prevalence rates. One of the most common methodological approaches for collecting maltreatment data involved the use of retrospective adult self-reports of childhood experiences [
28]. Some researchers have raised concerns about the reliability and validity of retrospective recall in adult respondents, especially about childhood events and about events that are emotionally charged [
29], what is known as recall bias [
12,
30,
31]. Concerns include forgetting an experience that happened many years ago [
32], while length of time since the abuse occurred may impact reliability [
33], and adults maltreated as children may experience memory impairment related to the event [
34]. Characteristics of the abuse may influence recall, including the type of abuse, the kinds of acts committed, or severity or chronicity of abuse [
35]. It may be the case however that maltreatment is much more likely to be under-acknowledged rather than forgotten [
36], and respondents may actively choose not to think about or disclose maltreatment experiences to avoid being reminded of them [
37,
38].
Children are also asked to self-report maltreatment, and studies sometimes included both adults and children, and many of the methodological issues related to retrospective recall by adults can be problematic for children. Some researchers have been reluctant to question children directly about their experiences on account of ethical and procedural complications related to reporting requirements [
39].
Comparison of prevalence rates from studies that collected self-reports from adults with those that involved children is problematic [
11], for example, a study conducted in 2017 may include self-reported maltreatment as far back as the 1930s or 1940s for adults, but only as far back as the 1990s for children, the time lapse may have an effect, as well as social and legal changes in the definition and recognition of child maltreatment [
36]. What individuals consider to be abusive behaviour may change between generations, for example, smacking a child was socially acceptable in the UK as recently as the 1980s [
40], and still may be today. In principle however it may be possible to compare adult and child reports for time periods that coincide.
Gender of the participant may influence reporting, some evidence suggests that men may be less likely to reveal a history of maltreatment [
33,
41]. The results of the current study seem to support this notion, particularly in relation to sexual abuse, however, the number of studies found concerning sexual abuse in men was relative low at 33% (115/345) compared to those concerning sexual abuse in women (56%, 195/345), it may be the case that there are true differences in prevalence rates between males and females [
42]. It has been suggested that definitions of maltreatment do not capture the experiences of males adequately, specifically sexual abuse [
15], or that fear of being labelled as weak or being flagged as homosexual might underestimate prevalence in males [
43].
The population of study participants may affect prevalence rates [
16], studies variously derived their samples from large samples of participants from the general population [
9], clinical or service user samples, convenience samples such as university or college students, school pupils, or self-selecting volunteers. Prevalence estimates tended to be lower for samples drawn at random from general populations and convenience samples than those based on research with volunteers or service user samples [
9,
43], for example Cawson et al. (2000) [
44] found lower prevalence rates in all four types of maltreatment when using a population sample as compared to Fisher et al. (2011) who used individuals presenting to mental health services with psychosis [
45]. University students may also be more aware of the study’s aims and thus more liable to response biases [
16], while Goldman & Padayachi (2000) somewhat controversially suggested that university students may be a psychologically healthier group which may be associated with lower sexual abuse prevalence [
43]. Drawing inferences from clinical samples can be problematic if the clinical setting from which the respondents are sampled is related to child protection intervention; it may be difficult to sort out causal order among the variables [
11]. To demonstrate the impact that such variation can have on prevalence rates our additional figures showed results based on ‘non-clinical’ study samples. This did not always reduce the prevalence rates, although this was the general direction of effect. The study design, sampling framework adopted (for example, the application of staged and sub-group over-sampling) and the eligibility criteria applied could still exert a substantial effect of apparent prevalence rates even in non-clinical samples.
Data collection mode
The measures used to collect data in self-report studies can be broadly divided into those that require the presence of a researcher presenting questions to a participant, and those that are self-administered. Method of data collection can artificially influence participant response, and some studies have shown that face-to-face interviews result in higher reporting rates compared to self-completed questionnaires [
27]. Amodeo et al. (2006) found that the prevalence of sexual abuse in their sample was higher based on a combined questionnaire and interview rather than a questionnaire alone [
46]. Face-to-face methods can also give opportunities for clarification and probe ambiguous responses, and remind participants of expectations for honesty [
47,
48]. Face-to-face interviews have the advantage of allowing for greater rapport, participants may prefer this method [
47], disclosure may be promoted [
48] through understanding and support on the part of the interviewer while others have not reported such a difference [
27]. It may also be the case however that interviewer presence may hamper disclosure if participants are reluctant to reveal sensitive information directly, may also cause participants to be more vulnerable to the effect of social desirability [
11,
12]. Not everyone however, is equally prone to discomfort relating to sensitive questions, even at a young age [
36].
Definitions of child maltreatment
Participants’ ideas of what constitutes maltreatment can vary [
5], and this may affect self-reported prevalence rates. Participants make a personal judgment about whether what took place was abusive if the questions asked are not specific [
36,
49,
50]. Answers provided will therefore be influenced by participants’ subjective perceptions of abuse [
16], which may be influenced by intergenerational changes in attitudes and cross-cultural differences, amongst other things. Previous studies have found that many people do not perceive childhood experiences such as ‘being whipped or beaten to the point of laceration’ as maltreatment, and there is a tendency to believe that discipline experienced as a child was normal [
51,
52]. This however, should not affect responses to descriptive questions [
5]; direct and specific questions tend to be used in validated measures, and are tested for internal consistency and pre-test reliability [
9]. Age-appropriate questions that give behavioural descriptions of events help respondents to think about specific incidents and are preferred over questions that use legal terminology or ask respondents to label themselves as maltreated [
53], and some have found that using broad questions are associated with lower prevalence rates of sexual abuse than more specific questions [
54]. Furthermore, both the context and the number of questions asked can affect number of reports [
27].
Some researchers specified an age range when asking participants about their maltreatment experiences, Bebbington et al. (2011) defined child sexual abuse as occurring before the age of 16 [
36], and some did not. Diaz-Olavarrieta et al. (2001) asked participants if as a ‘child’ they experienced physical or sexual abuse [
55], this may affect reported prevalence rates as one person’s idea of a ‘child’ may vary from another’s. When researchers defined child maltreatment as something that happens before the age of 16, those who were maltreated at ages 17 and 18 are missed. The definition of the perpetrator of the maltreatment may also affect prevalence rates, most studies do not specify details about the perpetrator, however, some focused narrowly on perpetrators as caregivers and family members, for example Annerbäck et al. (2010) [
56]. It should also be noted that studies will under estimate infant and toddler abuse as the reporters may not be recall these events.
Some studies focused on one form of abuse, 34% (114/339) of the studies reviewed in this paper focused on sexual abuse only, with 56% (189/339) including more than one form of maltreatment. Although Bentley et al. (2017) reported that neglect was the most common reason for a child being subject to a protection plan or on the child protection register in the four UK countries [
6], a disproportionate amount of studies have been conducted on the prevalence of sexual and physical abuse. Perhaps this is a reflection of perceived or actual seriousness of the various types of abuse, or possibly the understanding of what emotional abuse is or thresholds for neglect and whether neglect is always physical neglect or emotional neglect. The definitions used to assess the prevalence of abuse and neglect vary greatly between studies, and this may affect prevalence rates [
30]. Radford et al. (2011) asked participants a series of very specific questions about experiences they may have had as a child [
9], whereas Diaz-Olavarrieta et al. (2001) simply asked participants if they had experienced persistent physical/sexual abuse as a child [
55], allowing participants to impose their own definition of abuse. Most studies, such as that by Diaz-Olavarrieta et al. (2001) [
55] do not present their maltreatment definitions in enough detail in published papers [
10].
Pereda et al. (2009) noted differences in definitions of what constitutes sexual abuse, including the age difference between the perpetrator and the victim, the age used to define childhood, and the type of sexual abuse [
27]. Edgardh and Ormstad (2000) [
57] and McCrann et al. (2006) [
58] defined sexual abuse as when the perpetrator was at least five years older than the victim, this is often done to rule out sexual activity among peers [
16]. There are also cultural and legal differences between countries in the age of consent to sexual intercourse which affects definitions [
44]. The acts that constitute sexual abuse are a crucial part of a definition and would almost certainly affect prevalence rates, for example non-contact abuse such as exhibitionism can be more commonplace and may yield higher prevalence rates than contact abuse only [
16].
Definitions of physical abuse may suffer from cultural preconceptions. As previously mentioned smacking is still legal in the UK but outlawed in some parts of Europe [
40]. In spite of this, often too much is made of cultural differences, and there is a general consensus in many cultures about what constitutes maltreatment [
40], cultural differences may therefore only play a small role in differences in reported rates of maltreatment.
Definitions of neglect vary greatly because recognition of neglect can be difficult; children who are victims of neglect experience multiple types of neglect and it is mostly persistent and rarely traceable to a single incident [
59]. Definitions of neglect have been criticised for imposing middle-class values on lower-class families [
60], and that they do not take cultural differences into account [
59]. There has been debate on whether the focus of the definition should be around either caregiver behaviours, or of the experiences of the child, regardless of who is to blame [
11]. Risk and protective factors can change with age and developmental ability; this can affect definitions [
11]. Some researchers have purported that definitions of neglect should consider the frequency, duration, and severity of the neglect, the age of the child, and potential consequences to the child’s development [
59,
61,
62]. Tonmyr et al. (2011) noted that emotional or psychological abuse can also have particularly ambiguous definitions [
63].
Some forms of maltreatment overlap, for example, sexual abuse often also involves physical abuse, and all forms of maltreatment include an element of emotional or psychological abuse, this can complicate definitions [
44].
Some of the reasons for differing prevalence rates described above are expected, for example, it’s unsurprising that there are variations in self-reports of different types of abuse and neglect, these expected reasons are less likely to represent error. Some of the differences in prevalence rates found however are more likely to represent error, for example, whether data collection is self-administered or requires the presence of an interviewer.
Strengths and limitations
We have reviewed the literature and collated data on the lifetime prevalence of self-reported child maltreatment worldwide. PubMed, Ovid SP and grey literature from the NSPCC, UNICEF, The UK Government, and WHO from 2000 to 2017 were searched. These databases were selected as they were thought to likely contain literature on the prevalence of child maltreatment, and indeed yielded a large amount of articles on the subject. The authors recognise however that it is possible that other databases not utilised could have yielded additional papers. Literature that were not in the English language were excluded, this was due to budget restriction on translation work as this review was part of a PhD. All four types of child maltreatment were included in this review, and studies which did not specify the type were also included. Including all types of child maltreatment in the same review has not been done for some time and this is a strength of the current piece of work. For some studies no upper age limit was provided, contacting the authors of these papers was not justifiable given the current resources and so the authors assumed the upper age limit of 100 for those studies. The authors planned to conduct a meta-analysis on the prevalence reported rates however, studies varied considerably in the data they collected, the tools to collect the data, and the populations included. It was therefore not possible to form sufficiently large groups to warrant a meta-analysis.
Although a portion of all titles and abstracts were triple-screened against the inclusion/exclusion criteria by three additional reviewers, just a single reviewer was responsible for reviewing all the other abstracts, however, reviewer agreement was very high and so we believe that the review process was completed systematically.