Abstract
Child abuse is considered to be an essential factor in the development of aggressive behavior. The intensity of the positive relations between child abuse and aggressive behavior differed considerably among researches despite the fact that abundant studies have observed this relation. According to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) approach, a three-level meta-analysis was employed to obtain reliable estimates for the sizes of effects and investigate some potential moderators of the relation between child abuse and aggressive behavior. The present study obtained 51 studies (30,566 participants; 680 effect sizes) through performing the detailed literature search. It was found that child abuse was positively associated with aggressive behavior in the current study. In addition, the present meta-analysis observed significant moderating effects for type of child abuse, culture, measurement of child abuse, and publication year in the association between child abuse and aggressive behavior. This study suggests that child abuse is a predictor for the development of aggressive behavior in humans. Moreover, child abuse is an important aspect for consideration in efforts toward strengthening of interventions targeting individuals’ aggressive behavior.
Aggression, a notable externalizing problem in humans, has been focused upon by numerous fields in current society universally (Megías et al., 2018). It has been found to cause harm to individuals and others, and leads to serious consequences to families and society (Salimi et al., 2019; Vuoksimaa et al., 2021). Mounting studies have shown that child abuse is associated with aggressive behavior (e.g., Moore et al., 2020; Zhu et al., 2020). Although risk factors for child abuse varies depending on socioeconomic background, sampling methods used, and type of abuse, there is consensus among certain risk factors having a major role to play, such as social isolation, parental educational background, parental history of childhood maltreatment victimization, and family and social environment (Deb & Ray, 2022). Social learning theory and developmental-ecological model could provide a framework for understanding child abuse (Espinosa et al., 2017; Schelbe & Geiger, 2017). The former theory emphasizes a transmission across generations, whereas the latter focuses on the relationship between child abuse and the risk factors in individual, familial, community, and environmental domains (Espinosa et al., 2017; Schelbe & Geiger, 2017).
It is important to note that there are some inconsistencies in sample characteristics, publication characteristics, and assessment and study design characteristics of existing studies examining the association between child abuse and aggressive behavior (e.g., Estévez et al., 2016; Han et al., 2018). Moreover, there is a lack of a systematic review statistically summarizing the association. Given that it is important to synthesize empirical evidence to better understand the relationship between the two variables, the study employed a three-level meta-analysis to quantitatively summarize the findings of preliminary empirical researches. Our meta-analysis contributes to clarifying possible explanations for the abovementioned inconsistent findings, and it could be of particular importance for understanding the role of child abuse in aggressive behavior and strengthening interventions that target individuals’ aggressive behavior.
Conceptualization of Child Abuse and Aggressive Behavior
Despite that various definitions of child abuse are there as shown in Table 1 (A1–A3), there is consensus that it can be described as any form of physical and emotional ill treatment, sexual abuse, neglect, and exploitation, resulting in actual or potential harm to children’s dignity, health, or development (A1) (Subedi et al., 2019). Some researchers have suggested mounting negative effects of child abuse on their psychological development (Ashraf et al., 2020; Longobardi et al., 2022). For instance, it has been found that child abuse can interfere with the development of children’s emotion regulation and interpersonal relationships, undermine their self-esteem, and induce feelings of shame and inadequacy (Longobardi et al., 2022). As depicted in Table 1, child abuse falls into four categories: psychological abuse (B1), physical abuse (C1), sexual abuse (D1), and neglect behavior (E1) (Azzopardi et al., 2019; Gülırmak & Orak, 2021; Sarkar et al., 2020; Scoglio et al., 2021; Wu et al., 2022). There are some inconsistent findings in the global prevalence of the different forms of child abuse. For instance, some scholars have found higher global prevalence in psychological abuse (44.0%) than physical abuse (14.0%), whereas others observed a reversed result (psychological abuse: 27.9%; physical abuse: 45.4%) (Chandraratne et al., 2018; Leppäkoski et al., 2021). One can speculate that these inconsistencies may be due to differences in the definitions, sources of information, and sampling methods used.
Definitions of Child Abuse and Aggression.
Note. The asterisk symbols (*) indicate the definitions that are adopted in our study.
Aggression can be regarded as a forceful behavior, attitude, or action which is conveyed physically, verbally, or symbolically (see F1 of Table 1) (Benaroya-Milshtein et al., 2020). Previous studies have demonstrated that aggression is highly stable, leading to maintaining a person’s harmful intention over time (e.g., Lansford, 2018). Aggression can be classified in multiple ways: (a) the purpose of aggression (instrumental vs. hostile aggression); (b) the cause of aggression (proactive vs. reactive aggression); (c) the form of aggression (verbal vs. physical vs. relational aggression) (X. Li, 2021). Although violence is one kind of aggressive behavior, it is obviously different from this externalizing problem behavior as it is aimed at some extreme harm, such as death (Anderson et al., 2002).
Association Between Child Abuse and Aggressive Behavior
Plenty of researches have investigated the association between child abuse and aggressive behavior (Fagan, 2020; Handley et al., 2021). More specifically, individuals who have experienced child abuse are more likely to exhibit aggressive behavior. According to social learning theory, parents are regarded as a crucial reference for their children (Yoon et al., 2017). If individuals have been abused in their early childhood and people around them often show anger, scolding, or even beating, they will form a coherent and lasting cognitive system of aggressive behavior in a relatively longer period of time, resulting in that they are more likely to be aggressive (Calvete & Orue, 2012).
Despite that some empirical researches have confirmed that child abuse is positively correlated with aggressive behavior (e.g., Moore et al., 2020; Zhu et al., 2020), others have failed to observe such relationship (Norton-Baker et al., 2019; Yoon et al., 2017). Therefore, a meta-analysis and systems review is required to clarify this uncertainty. It is necessary to integrate multiple and diverse literature on the association between child abuse and aggressive behavior. The reviews of Assink et al. (2018) and Mohammadi et al. (2014) focused on individuals who encountered child abuse in early childhood, while individuals’ aggressive behaviors were examined in the studies of Hershcovis et al. (2007) and Zumkley (1994). These reviews imply that past reviews often examined child abuse and aggressive behaviors but not both (Assink et al., 2018; Hershcovis et al., 2007; Van der Put et al., 2018). However, both child abuse and aggressive behavior have been included in our meta-analysis.
Impact of Moderator Variables
Despite that gender differences in the relationship between child abuse and aggressive behavior have been investigated in the relevant literatures, there are some inconsistent findings (M. C. Chung & Chen, 2020; Jung et al., 2017). Specifically, some studies have shown the gender differences in the relationship between these variables, whereas others failed to find these gender differences (M. C. Chung & Chen, 2020; King et al., 2019; Norton-Baker et al., 2019). Therefore, we employed a moderating analysis to test whether gender played a vital role in the association between child abuse and aggressive behavior.
Participants’ age might be also a possible moderator variable. Researches have shown that individuals’ age plays a role in the association between child abuse and aggressive behavior (Godinet et al., 2014; Tucker et al., 2013). According to viewpoints of developmental pathology, the effect of individuals’ negative experiences on their behaviors may vary over time (Holmes, 2013). Previous empirical studies, however, have failed to examine that the age influences the association between child abuse and aggressive behavior (He & Xiang, 2021; Mangold & King, 2021). Thus, the current meta-analysis aimed to make it clear whether participants’ age is a significant moderating variable.
Some ethnographies have suggested that the level of children’s aggressive behavior varies in different cultural contexts (Korbin, 2003). Compared with parenting in Eastern cultures, that in Western cultures was less affected by collectivist values, leading to ignore family hierarchy, parental authority, and high academic achievement of children (Huang & Gove, 2015). In this regard, parents in Western society may be less likely to abuse their children, resulting in that they are at lower risk of aggression (Ma et al., 2022). Hence, culture may influence the strength of the association between child abuse and aggressive behavior.
As mentioned, there are four forms of child abuse: psychological abuse, physical abuse, sexual abuse, and neglect behavior (Ahad et al., 2021). Existing studies have suggested that type of child abuse might be one moderating factor between child abuse and aggressive behavior (Debowska et al., 2018; Vandenberg & Marsh, 2009). Compared with sexual abuse, psychological abuse was significantly correlated with a higher level of aggressive behavior (Yoon et al., 2017). Furthermore, individuals who encountered physical abuse are at a lower risk of externalizing problems than those people that suffered psychological abuse (Allen, 2011; Moore et al., 2020). Finally, in contrast to people with experiences of neglect behavior, those with experiences of psychological abuse are more likely to be aggressive (Han et al., 2018; Xie et al., 2020).
Present Study
There is considerable uncertainty in the association between child abuse and aggressive behavior (Han et al., 2018). Therefore, the first aim of the present three-level meta-analysis was to investigate the overall relationship between child abuse and aggressive behavior (Research Question 1). More specifically, our meta-analysis quantitatively summarized preliminary findings to help people understand the true effect for the relationship between these two variables. The second goal of our study was to examine the moderating variables that impacted the relationship between child abuse and aggressive behavior (Research Question 2). Hence, moderator analyses were conducted to investigate potential variables that might modulate the relationship between the two variables. Based on the main theoretical framework and previous research results, it may be assumed that sample, publication, assessment and research design characteristics might affect the strength of the association between child abuse and aggressive behavior (e.g., Debowska et al., 2018; Holmes, 2013; Vandenberg & Marsh, 2009).
Methods
The present meta-analysis used PRISMA checklist as guidelines (Moher et al., 2015; Ran et al., 2022). Furthermore, in order to facilitate transparency and prevent unconscious duplication of effort, we preregistered the protocol of the meta-analysis at the PROSPERO (the International Prospective Register for Systematic Reviews; registration number: CRD42022309440; URL: https://www.crd.york.ac.uk/prospero/) (Lakens et al., 2016).
Data Sources and Study Selection
In January 2022, relevant studies were searched through the following databases, including the PsycINFO, Web of Science, ScienceDirect, Web Online Library, Google Scholar, and CNKI. Two authors (GR and QZ) independently performed the present literature search. The database search strategy developed by the authors was in consultation with two peer researchers.
To select keywords of search terms, we looked up all terms about our topic in the dictionaries. Then, we extracted the keywords of high frequency in the multiple databases and the keywords of low frequency being removed. Subsequently, the remaining keywords were evaluated independently by two authors (GR & QZ) based on their relevance to our topic. All discrepancies were resolved via discussion between these two raters. Finally, the selected keywords were evaluated by a peer expert. The search string included the following two elements: (a) child abuse and (b) aggressive behavior. For the search element of child abuse, as displayed in Table 2, the following key words were used “child abuse” OR “children abuse” OR “child maltreatment” OR “children maltreatment” OR “child mistreatment” OR “children mistreatment” OR “kid abuse” OR “kids abuse” OR “kid maltreatment” OR “kids maltreatment”. For the element of aggressive behavior, the following key words were collected “aggressive behavior” OR “aggression” OR “externalizing problems” OR “attack behavior” OR “assailment”. Relevant studies, which contained at least one key word in the title, abstract, and key words of the databases, constituted the primary studies. In addition, some supplementary searches (e.g., manual search of related journals, forward searching, and backward searching) were conducted to search as much literature as possible.
Key Words of Two Search Elements.
Primary studies were eligible for this meta-analysis if they (a) measured child abuse and aggressive behavior with objective instruments; (b) reported at least one correlation coefficient between child abuse (four forms of child abuse) and aggressive behavior; (c) showed statistics that could be transformed into correlation coefficients; (d) were cross-sectional or longitudinal studies; (e) were written in Chinese or English. Besides, exclusion criteria were as follows: (a) participants experienced abuse in adulthood; (b) participants showed intimate partner violence; (c) studies included clinical samples; (d) one number of a sample was less than 30; (e) studies were not empirical research articles, such as review articles. The flow chart of systematic search for studies was demonstrated in Figure 1.

The PRISMA flow chart used to identify studies for detailed analysis of child abuse and aggressive behavior.
Coding the Studies
Following the established guidelines for meta-analysis research (Lipsey & Wilson, 2001), a reliable coding scheme was created for various descriptors and characteristics in the selected study. Descriptors of each study consisted of the basic information of primary studies, including title of the article, the name of the first author, effect size, and sample size. At the same time, study characteristics which might moderate the relationship between child abuse and aggressive behavior were grouped into three forms of moderator variables: sample characteristics, publication characteristics, and assessment and research design characteristics (Ran et al., 2022). Since the coding work was done by the two authors independently, it was necessary to calculate the interrater reliability. For the interrater agreement, intraclass correlation coefficient (ICC) was used to calculate the continuous variables and Cohen’s Kappa (k) was used to calculate categorical variables (Ran et al., 2021, 2022). For the sample characteristics, one interrater reliability coefficient was calculated for gender (ICC = 0.983), age (k = 0.980), culture (k = 1.000), and person’s status (k = 1.000). For the publication characteristics, an interrater reliability coefficient was acquired for publication year (ICC = 0.994) and publication status (k = 1.000). For the assessment and research design characteristics, one interrater reliability coefficient was obtained for study design (k = 1.000), type of child abuse (k = 0.979), measurement for child abuse (k = 1.000), and measurement for aggressive behavior (k = 0.992). In sum, the results showed a good interrater reliability, manifesting a fine agreement in the characteristics of studies between the two authors. The coding manual was available upon request (Milner et al., 2022).
Sample Characteristics
Four kinds of sample characteristics were included in our meta-analysis. Firstly, gender was coded on the basis of the percentage of men. Secondly, participants’ age was coded as three categories: children (0–11 years), adolescents (12–18 years) and adults (over 18 years). It was noted that in the studies, grouping according to the participants’ educational status was coded as follows: (a) the college participants were coded as adults; (b) the junior and senior high school participants were coded as adolescents; (c) the primary school participants were coded as children. Thirdly, culture was coded as either Eastern or Western, according to the culture in participants’ country. Finally, a person’s status was coded based on whether the participant was a criminal or not.
Publication Characteristics
Publication characteristics consisted of publication year and publication status. Publication year was coded as a continuous variable. In addition, we coded publication status as either a published or unpublished publication status. The published status was published on one academic journal, the unpublished status included master-doctoral dissertation and conference abstracts. We also made attempts to consult authors of some articles for grey literatures.
Assessment and Research Design Characteristics
The current study coded two categories of study design: cross-sectional study and longitudinal study. Moreover, the variable of child abuse was created as a five-category variable (Psychological abuse vs. Sexual abuse vs. Physical abuse vs. Neglect behavior vs. Other). In the empirical literature, child abuse was measured by the Childhood Trauma Questionnaire (CTQ), Child Psychological Abuse and Neglect Scale (CPANS), Childhood Psychological Maltreatment Scale (CPMS), and Comprehensive Child Maltreatment Scale (CCMS) (Deng et al., 2007; Higgins & McCabe, 2001; Pan et al., 2010; Zhao et al., 2005). Moreover, there were some measurements of aggressive behavior: the Aggression Questionnaire (AQ), Child Behavior Checklist (CBCL), Reactive-Proactive Aggression Questionnaire (RPQ), and Displaced Aggression Questionnaire (DAQ) (Achenbach & Rescorla, 2001; Buss & Perry, 1992; Denson et al., 2006; Raine et al., 2006). Thus, we created four measurement categories of child abuse (CTQ vs. CPANS vs. CPMS vs. Other) and aggressive behavior (AQ vs. CBCL vs. RPQ vs. Other).
Supplement Information
The effect sizes in this study generated from independent samples. When a study contained multiple independent samples, each one was coded independently. Additionally, we coded separately the correlation coefficients in a boy and girl sample. Finally, as a result of the data of the meta-analysis containing data from multiple sources, coding errors were inevitable. To address these coding errors, each preliminary study was coded independently by two authors (GR and QZ).
Data Analysis
Given that correlation coefficients (rs) were widely used and easily interpreted, these coefficients were selected as the effect size index in our meta-analysis. Each correlation coefficient (r) was first transformed to Fisher’s z, and then converted back to r after conducting meta-analysis (Ran et al., 2022). As mentioned above, we extracted multiple effect sizes in the same sample. But one of the key requirements in a standard univariate meta-analysis was that the effect sizes in the dataset were independent of each other, indicating that one effect size could be included in each primary study (Assink & Wibbelink, 2016; Lipsey & Wilson, 2001). The results of univariate meta-analysis would be exaggerated when multiple effect sizes were extracted in the same study (Assink et al., 2015). The multiple-level meta-analysis, however, could adopt multiple effect sizes that extracted form one study because the independence was considered (Konstantopoulos, 2011). Therefore, the current study employed a three-level meta-analysis to examine the overall relationship between child abuse and aggressive behavior and conduct moderator analyses.
The model of three-level meta-analysis comprised three forms of sources for variance: Level 1 estimated sampling variance in effect sizes; Level 2 estimated variance between effect sizes that extracted from the same sample; Level 3 estimated the variance between studies (Gao et al., 2019; Yan et al., 2020). Compared with the univariate meta-analysis, the three-level meta-analysis was qualified for realizing maximum statistical power (X. Li et al., 2021). Statistical analyses of the present meta-analysis were carried out in R version 3.5.3, in accordance with the model guidelines of multiple-level random-effect that was formulated by the tutorial with the metafor package (Assink & Wibbelink, 2016).
Publication bias was attributed to the fact that the included studies were not well representative of the field (Franco et al., 2014). As many eligible articles were included as possible to minimize the impact of this bias on the results of our meta-analysis. The funnel plot for the studies and Egger’s regression test were employed to assess the publication bias (X. Li et al., 2021). For the funnel plot of the studies, a symmetrical funnel shape might indicate that the publication bias could be ignored (Li et al., 2019). For the Egger’s regression test, publication bias could be regarded as non-existent, if the results of linear regression were non-significant (p > .05) (Duval & Tweedie, 2000). The trim-and-fill technique was employed to adjust asymmetrical funnel plot when the publication bias was present observed in the meta-analysis (Duval & Tweedie, 2000).
Results
Study Characteristics
As shown in Table 3, a total of 680 effective sizes from 51 studies were acquired in this study, with 30,566 participants. The sample size of studies ranged from 97 to 1,951, and the participants’ average age was 21.13 years. The maximal number of effect size in a sample was 48, and the minimal one was 1. There were two forms of moderator variables: continuous and categorical moderator variable. For the continuous variables, the numbers of effect sizes for gender were 636 and for publication year was 680. Moreover, for the categorical variables, the number of effect sizes were as follows: age (children: 61; adolescents: 277; adults: 342), culture (Eastern: 542; Western: 138), status (criminal: 101; non-criminal: 579), publication status (published: 425; unpublished: 255), study design (cross-sectional: 668; longitudinal: 12); type of child abuse (psychological abuse: 269; sexual abuse: 61; physical abuse: 72; neglect behavior: 226; other: 26), measurement of child abuse (CTQ: 270; CPANS: 180; CPMS: 126; other: 104), and measurement of aggressive behavior (AQ: 489; CBCL: 19; RPQ: 30; other: 142).
Characteristics of the 51 Studies Included in the Meta-analysis.
Note. K = number of effect sizes; N = number of participants; Gender = percentage of males; A = adults; Ad = adolescents; C = children; E = eastern; W = western; Non = non-criminal; Cri = criminal; CA = child abuse; AG = aggressive behavior; CTQ = childhood trauma questionnaire; CPANS = child psychological abuse and neglect scale; CPMS = childhood psychological maltreatment scale; AQ = aggression questionnaire; CBCL = child behavior checklist; RPQ = reactive-proactive aggression questionnaire.
Publication Bias
An examination for the symmetry of the funnel plot could address the issue of the publication bias (Duval & Tweedie, 2000). As depicted in Figure 2, the funnel plot was symmetrically distributed, indicating the absence of publication bias in our study. We, however, still cannot judge whether the funnel plot was completely symmetrical by subjective judgment. Therefore, to further investigate the publication bias, we performed the Egger’s test. The p-value in Egger’s test was not significant (t = 1.182, p = .238), which also implied that the publication bias did not exist in the present meta-analysis. Given that both the shape of funnel plot and Egger’s test showed that the results of current meta-analysis were relatively reliable. Hence, the trim-and-fill technique was not required to further assess bias issues in our meta-analysis (Ran et al., 2021; Yan et al., 2020).

Funnel plot of the association between child abuse and aggressive behavior.
Overall Effect Sizes
The overall relationship between child abuse and aggressive behavior was calculated based on a random effect model. The results of analyses revealed that there was a significant positive correlation between child abuse and aggressive behavior (r = .236, p < .001). In addition, logarithmic likelihood ratio test showed a significant heterogeneity in both variance within a study (Level 2) and variance between studies (Level 3) (p < .001). For the total variance, 9.20% was attributed to Level 1, 28.67% and 62.13% were attributed to Level 2 and Level 3, respectively. This distribution indicated that investigation in the moderating effects of characteristic variables on the overall relation was fruitful (Ran et al., 2022). Therefore, we performed the analyses of moderation effect to investigate whether the magnitude of the relationship between child abuse and aggressive behavior was influenced by these characteristic variables.
Moderator Analyses
The results of moderating analysis of the relationship between child abuse and aggressive behavior are shown in Table 4. The moderating effect of culture was significant (F (1, 678) = 8.886, p = .003). This relationship was lower for Western individuals (r = .182) than Eastern ones (r = .270). In addition, publication year moderated the overall relation (F (1, 678) = 17.529, p < .001). The relationship between the two variables was stronger in more recent studies. Besides, the study found that type of child abuse was a significant moderator (F (4, 649) = 29.783, p < .001). Specifically, the association between child abuse and aggressive behavior was stronger when children experienced psychological abuse (r = .265) than those that encountered sexual abuse (r = .183). What’s more, it was found that there was a significant moderator for measurement of child abuse (F (3, 676) = 6.455, p < .001). A stronger relationship between child abuse and aggressive behavior was found when child abuse was measured using CPMS (r = .330) than using CTQ (r = .218). No significant moderator effects were found for participants’ gender (F (1, 634) = 0.010, p = .919); age (F (2, 677) = 0.225, p = .799); person’s status (F (1, 678) = 0.085, p = .771); publication status (F (1, 678) = 0.121, p = .728); study design (F (1, 678) = 0.373, p = .541); and measurement of aggressive behavior (F (3, 676) = 2.333, p = .073).
Results of Categorical and Continuous Moderators for the Association between Child Abuse and Aggressive Behavior.
Note. #ES = number of effect sizes; mean z = mean effect size (Fisher’s z); 95% CI = 95% confidence interval; β1 = estimated regression coefficient; r = mean effect size expressed as a Pearson’s correlation; df = degrees of freedom; Levels 2 variance = variance between effect sizes extracted from the same study; Levels 3 variance = variance between studies; for child abuse scale: CTQ = childhood trauma questionnaire; CPANS = child psychological abuse and neglect scale; CPMS = childhood psychological maltreatment scale; For aggressive behavior scale: AQ = aggression questionnaire; CBCL = child behavior checklist; RPQ = reactive-proactive aggression questionnaire. RC = reference category.
Omnibus test of all regression coefficients in the model.
p-Value of the omnibus test.
p < .05. **p < .01. ***p < .001.
Discussion
Aggressive behavior has been regarded as one severe mental health issue among adolescents in both Eastern and Western cultures (Jia et al., 2016). Child abuse was particularly harmful because it increased risks of learning difficulties, psychological problems, physical injuries, neurologic sequelae, and death (Ashraf et al., 2020). Mounting studies have shown that child abuse was associated with aggressive behavior (e.g., Moore et al., 2020; Zhu et al., 2020). However, there are some inconsistencies in existing studies examining the association between these two variables (e.g., Estévez et al., 2016; Han et al., 2018). Besides, a systematic review for the association between child abuse and aggressive behavior was not available. Thus, we employed the three-level meta-analysis to quantitatively summarize the findings of preliminary empirical researches to synthesize empirical evidence to better understand the relationship between the two variables.
Fifty-one studies were identified with a total of 680 effect sizes and 30,566 participants in our meta-analysis. The results of current meta-analysis showed that there was a significant positive correlation between child abuse and aggressive behavior, indicating that individuals encountering more child abuse might exhibit a high level of aggressive behavior. Furthermore, the strength of the association was affected by type of child abuse, culture, measurement of child abuse, and publication year, implying that these moderator variables played vital roles in the relationship between child abuse and aggressive behavior. Our meta-analysis could be of particular importance for understanding the role of child abuse in aggressive behavior and strengthening interventions that target individuals’ aggressive behavior.
Overall Association Between Child Abuse and Aggressive Behavior
The results of this meta-analysis verified a significant positive correlation between child abuse and aggressive behavior, suggesting that individuals showed a high risk of aggressive behavior when they experienced abuse in early childhood. Based on Bowlby’s attachment theory (Byrne et al., 2005), people who had experiences of child abuse were more likely to develop insecure attachments, which undermine their social skills, and ultimately lead to aggressive behavior (G. Yu et al., 2020). Besides, according to the theory of trauma-induced offense cycle (Zhu et al., 2020), it was thought that child abuse promoted individuals’ stable pattern of affective cognitive responses characterized by feelings of danger, anger, helplessness, and hostility attribution bias, resulting in their high level of aggressive behavior in situations that triggered trauma-related symptoms. In sum, both theories suggested that child abuse was positively related with aggressive behavior.
Explaining Heterogeneity With Moderators
It was observed that the association between child abuse and aggressive behavior was moderated by type of child abuse. Specifically, the relationship between child abuse and aggressive behavior was stronger when individuals experienced psychological abuse as compared to those who suffered sexual abuse. This difference in the strength of the relationship was indicated in two previous empirical studies (Ji, 2013; Yoon et al., 2017). This might be due to the fact that one individual experiencing psychological abuse in early childhood showed a high risk of externalizing problems (e.g., aggressive behavior) than a person encountering sexual abuse (Hailes et al., 2019; Hodgdon et al., 2018). This was consistent with social control theory and attachment theory, which suggested that psychological, but not sexual abuse might interfere with the development of prosocial parental bonds and secure attachments, resulting in a person’s aggressive behavior toward others (Auslander et al., 2016). Besides, biosocial theory hypothesized that psychological abuse combined with biological tendencies could lead to aggression or other problematic behaviors because it was a strong predictor of dysfunctional emotions (Allen, 2011).
In addition, this study found that culture had a significant effect on the association between child abuse and aggressive behavior, that was, the correlation between these two variables was lower in Western cultures. This finding provided support for the perspective of cross-cultural differences in child psychological development (Arslan, 2016; Bergmüller, 2013). From an individualist-collectivist dimension viewpoint, characteristics such as individual thinking, importance of autonomy, self-expression, and personal choice were highly valued in individualist countries (Nunes, 2008). In this regard, compared with parents in collectivist societies, the ones in individualist countries showed a lower level of harsh discipline and physical punishment (low controlling), and often characterized by high warmth, resulting in the outcome that child abuse was generally lower in its global prevalence (Arslan, 2016; W. Zhang et al., 2017). This suggested that, in individualist societies, there were less negative effects of child abuse on individuals’ psychological and behavioral problems, such as aggressive behavior (Arslan, 2016).
The current study observed that the associations between child abuse and aggressive behavior were moderated by measurement of child abuse. In more detail, the association between these variables was stronger when the measuring instrument of child abuse was CPMS than when it was CTQ. This might be due to the fact that there were differences in the measurements (Prevoo et al., 2017). For instance, CPMS was a scale developed by Pan et al. (2010) to measure psychological abuse, with a total of 24 items in 5 dimensions. CTQ, however, was a scale for measuring four forms of child abuse compiled by Bernstein and Fink (1998), which had a total of 28 items in 5 subscales. It was noted that, compared with CTQ, CPMS measured fewer types of child abuse, but more accurately measured psychological abuse.
The current results suggested that publication year was also a significant moderating factor. The relationship between child abuse and aggressive behavior was stronger in studies published later than those published earlier. In recent years, people have become more anxious (Salari et al., 2020), leading to an outcome that individuals experiencing child abuse showed more aggressive behaviors (J. E. Chung et al., 2019). In addition, more and more children have been bullied in recent years (Hinduja & Patchin, 2019). Such trauma could trigger the higher association between child abuse and aggressive behavior based on the theory of trauma-induced offense cycle, which was described carefully in the above text (Zhu et al., 2020).
This meta-analysis failed to show a significant moderating effect of gender. This implied that gender differences for aggressive behavior might not be explained by differential effects of child abuse. Despite that some studies have shown the gender differences in the relationship between child abuse and aggressive behavior, social learning theory suggested that girls learned abusive and aggressive parenting behaviors in ways that appeared more similar to boys (M. C. Chung & Chen, 2020; Jung et al., 2017; Norton-Baker et al., 2019). Moreover, participants’ age didn’t moderate the overall relation, suggesting that association between child abuse and aggressive behavior may be stable across ages. This was consistent with the viewpoint that there may not be a significant age difference in the frequency of aggressive behavior (Makinde, 2018).
Limitations and Future Directions
It should be pointed out that there were several limitations in the current research and some research directions in the future. One limitation of the meta-analysis was that the empirical studies included in this study failed to report children’s ages at the time of their experiences in encountering child abuse. Therefore, we strongly recommend that future empirical studies examining the association between child abuse and aggressive behavior focus on the age at which abuse began, which would enable researchers to further explain the association between these variables.
In addition, considered from a dynamic perspective (L. Sun et al., 2017), another limitation of this study was that the development process of the association between child abuse and aggressive behavior was not examined, since our meta-analysis mainly included the cross-sectional studies. The technique of longitudinal study should be used in future studies to understand the trajectory of child abuse affecting aggressive behavior over time.
Finally, only children were the informants when researchers measured the child abuse and aggressive behavior in some empirical studies (Y. Li, 2018; T. Yu et al., 2021), which might cause the problem of common method variance (CMV) (Chang et al., 2010; Gao et al., 2017). To address the CMV, the informants of these two variables should be parents, teachers, and peers in future empirical studies.
Conclusion
The present study conducted the three-level meta-analysis to quantitatively synthesize the overall relationship between child abuse and aggressive behavior. The results demonstrated a significant positive correlation between child abuse and aggressive behavior, suggesting that child abuse was an important predictor of the development of aggressive behavior in humans. Subsequently, our moderator analyses indicated some significant moderators: type of child abuse, culture, measurement of child abuse, and publication year. The differences for the strength of the association between child abuse and aggressive behavior might be explained by these four moderators. Our study was of particular importance for understanding the role of child abuse in aggressive behavior and strengthening interventions that target individuals’ aggressive behavior.
Summary of Critical Findings
A total of 51 studies with 30,566 participants were included.
Child abuse was significantly correlated with aggressive behavior.
The moderating effect of culture was significant.
The publication year moderated the overall relation.
The type of child abuse was a significant moderator.
There was a significant moderator for measurement of child abuse.
Summary of Implications for Practice, Research, and Policy
Child abuse may be an important aspect for consideration in the targeting of interventions aimed at preventing or reducing aggressive behavior.
Interventions focusing on child abuse need to take into account the influence of different cultural backgrounds.
Interventions targeting child abuse are equally important for boys and girls.
Interventions targeting individuals who have experienced psychological abuse may be more effective.
Future studies could adopt a multi-stage longitudinal design with adequate follow-up evaluation.
Future research on child abuse and aggression can be conducted in a variety of reporting forms, such as parent reports, teacher reports, and peer reports.
Supplemental Material
sj-docx-1-tva-10.1177_15248380221129596 – Supplemental material for The Association Between Child Abuse and Aggressive Behavior: A Three-Level Meta-Analysis
Supplemental material, sj-docx-1-tva-10.1177_15248380221129596 for The Association Between Child Abuse and Aggressive Behavior: A Three-Level Meta-Analysis by Guangming Ran, Qiongzhi Zhang, Qi Zhang, Jun Li and Jing Chen in Trauma, Violence, & Abuse
Footnotes
Declarations of Conflicts of Interest
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the National Natural Science Foundation of China (No.31900760)
Supplemental Material
Supplemental material for this article is available online.
Author Biographies
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
