Masterscriptie Orthopedagogiek Pedagogische en Onderwijskundige Wetenschappen Universiteit van Amsterdam
E. B. M. Van de Put 10648224 mw. dr. M. Hoeve mw. L.A. de Vies MSc
Abstract
The present meta-analysis investigated the relationship between different types of childhood maltreatment and three outcome measures, 1) overall delinquency, 2) general delinquency, and 3) violent delinquency. Also, the moderating role of gender and maltreatment type was
examined. Twenty-four published manuscripts, representing 129 effect sizes, were subjected to a 3-level random effects model. Small mean effect sizes were found for the link between child maltreatment and the delinquency measures (r = .156 for overall, r = .151 for general, and r = .134 for violent delinquency). These links were not moderated by gender. In addition, the relationship between child maltreatment and overall delinquency was weaker for adolescents with a history of sexual abuse, as compared to children with a history of general child
maltreatment. Results build on existing evidence with regard to the ‘cycle of violence’
hypothesis, concluding that childhood maltreatment is related to juvenile delinquency. Childhood maltreatment could therefore be a target in intervention programs in order to reduce juvenile delinquency. Limitations are also discussed.
Keywords: cycle of violence, child maltreatment, juvenile delinquency, violent,
The moderating role of gender in the relationship between different types of childhood maltreatment and general and violent delinquency in adolescence: A meta-analysis.
Childhood maltreatment is a well-known risk factor for antisocial and criminal behavior. Several studies (e.g., Rutter, Giller, & Hagell, 1998; Widom, 1989) have shown that youth who experience some form of abuse during childhood are at risk for developing a conduct disorder, antisocial personality disorder and for becoming violent offenders. Also, the earlier children experience maltreatment, the more likely the risk of developing such problems (Keiley, Howe, Dodge, & Bates, 2001). However, not every child responses the same way to maltreatment. In one study (Widom, 1989), it was found that 50% of the children have an increased risk for developing delinquent behavior, whereas most maltreated children do not become delinquents. Furthermore, there is evidence that there are gender differences with regard to the risks of maltreatment on juvenile delinquency (e.g., Maxfield & Widom, 1996). The present meta-analysis investigated to what extent children with a history of childhood maltreatment become adolescent delinquents, and whether this relationship is different for various types of
maltreatment and different for males and females.
Cycle of Violence
Studies that examined the relationship between childhood maltreatment and delinquent outcomes often rely on the theoretical framework, referred to as the ‘Cycle of Violence’. The cycle of violence hypothesis is derived from social learning theories and assumes that children who are victims of abuse and neglect are at increased risk of becoming violent offenders later in life (Widom, 1989). The social learning theories explain the underlying mechanisms of how childhood maltreatment is related to juvenile delinquency. For example, Sutherland (1939) asserted that individuals learn delinquent behavior, like other behaviors, through a process of
socialization in intimate groups such as family and peer groups. Within these intimate groups, they learn both criminal attitudes (e.g., neutralizing guilt, denying responsibility) and delinquent actions (e.g., committing crimes, eluding detection by police). Further, Bandura (1977)
contended that individuals tend to model the behavior of people whom they consider as authority figures or influential. Hereby, the likelihood of modeling behavior is increased if the observed behavior is perceived to have a desired outcome. Children who are exposed directly (e.g., experience any form of childhood maltreatment) or indirectly (e.g., witnessing a relative being abused) to abuse in the family of origin may develop norms about the suitability of violence used in specific circumstances. More recently, Akers (2009) extended Sutherland’s work in order to emphasize the importance of positive and negative reinforcement in learning processes. Children who are exposed to family abuse not only learn the rationale and commission of abuse, but if the abusiveness is perceived to solve problems, it also may be more likely that they will be
victimized during adolescence and adulthood themselves. According to this theoretical
framework, victims of family abuse may be predisposed to violence later in life. Over time, when children learn a pro-abusive set of norms, they are at greater risk of becoming offenders during adolescence and adulthood in comparison to children who were not exposed to pro-abusive family norms.
A substantial amount of research testing the cycle of violence hypothesis found evidence for the link between physical, emotional and sexual maltreatment during childhood and adverse outcomes, including psychopathology, substance abuse, self-harm behavior, violent and
delinquent behavior (Gilbert et al., 2009; Hillberg et al., 2011; Kessler et al., 2010; Klonsky & Moyer, 2008; Laughlin et al., 2010). This link has been found in both adolescence and adulthood (e.g., Maxfield & Widom, 1996; Seto & Lalumière, 2010; Widom, 1995, 1996; Widom &
Maxfield, 2001; Wilson, Stover, & Berkowitz, 2009). For example, child abuse, in particular physical and emotional abuse, is related to an increased risk of violent and aggressive behavior in adolescence (Egeland et al., 2002; Lansford et al., 2007; Piquero & Sealock, 2000). Furthermore, not only increases child abuse the likelihood of committing violent offenses, it also seems to increase the likelihood of both participation rates and frequency of committing violent offenses (Yun, Ball, & Lim, 2011). In addition, Smith and Thomberry (1995) found that child
maltreatment is associated with both official and self-reported criminal behaviors.
Although there is evidence supporting the relationship between child maltreatment and juvenile delinquency, some inconsistencies have become apparent in the literature. There are some studies that indeed reported a positive relationship between child maltreatment and delinquent behavior later in life, whereas others did not. For example, two studies found that children of sexual abuse are least likely to be arrested for a violent crime as compared to children who experienced other forms of maltreatment (Siegel & Williams, 2003; Widom & Maxfield, 2001). On the contrary, Gault-Sherman, Silver, and Sigfusdottir (2009) stated that sexually abused children were more likely to be involved in violence. The likelihood of taking part in acts of delinquent behavior maltreated children may have depend upon the type of child maltreatment suffered and on other study and sample characteristics. Therefore, it is crucial to synthesize the empirical findings from previous studies in order to examine to what extent different types of child maltreatment are related to delinquent outcomes during adolescence and whether study characteristics explain different findings between studies.
There have been several reviews and two meta-analyses conducted on the relationship between childhood maltreatment and delinquency (Derzon, 2010; Leschied, Chiodo, Nowicki, & Rodger, 2008; McGrath, Nilsen, & Kerley, 2011), indicating that childhood maltreatment is a
precursor to violent and criminal behaviors in adolescence and adulthood. First, McGrath and colleagues (2011) performed a systematic review of literature about the association between child sexual abuse and adolescent and adult criminal behavior. They asserted that the majority of the studies found a positive relationship between child sexual abuse and (violent) delinquency, sex offenses, and adult criminality. One shortcoming was that the review focused on childhood sexual abuse as a predictor in relation to juvenile and adult delinquency, instead of taking into consideration other types of childhood maltreatment. Also, the distinction between adolescent and adult delinquency was not always evident when study results were evaluated. Another general shortcoming with regard to systematic reviews as opposed to meta-analyses was that results are descriptive in nature. Reviews do not summarize and synthesize quantitative results (i.e., calculating an overall mean effect size). Quantifying the strength of relationships is one of the benefits of a meta-analysis. Second, the meta-analysis of Derzon (2010) revealed that general child maltreatment is associated with aggressive, violent, and criminal behavior. No distinction was made between adolescent and adult outcomes of adverse behavior. Also, child maltreatment was considered as one construct and differential maltreatment forms were not taken into account, although maltreatment was assessed as being abused as a child, being mentally injured,
neglected, physically or sexually abused. Thereby, analyses were conducted using weighted effect sizes. Effect sizes within the same study were averaged to one effect size per study in order to obtain independence within studies. Calculating one mean effect size per study might give distorted results when the effect sizes deviate a significant amount from each other. It also restricts researchers that want to differentiate between multiple forms of one predictor (e.g., childhood maltreatment) or if there are multiple predictors measured in one study. Third, the meta-analysis of Leschied and colleagues (2008) reported a significant positive association
between adverse family outcomes (e.g., family conflict, abuse, neglect) and adult criminality (violent and non-violent offenses). However, adverse family outcomes is a very broad concept and not solely focused on childhood maltreatment. Also, the outcome measure contained adult criminality whereas the current meta-analysis examined juvenile criminality. Finally,
longitudinal study designs were the sole source for the meta-analysis, which has limited the scope.
Gender differences
Results from a study by Maxfield and Widom (1996) showed that there are gender differences with regard to the cycle of violence assumption. Abused and neglected females were at increased risk of arrest for violence during adolescence. On the contrary, the relationship between child maltreatment and violent behavior in adolescence was barely significant for males, relative to the control males. These findings are in contrast with earlier study results, indicating that abused and neglected males were at an increased risk for violence, but females were not (Widom, 1989). The latter seems more plausible, because females’ behavioral and emotional responses to abuse differ from males’ (Sullivan, Farrell, & Kliewer, 2006). Males were more likely to respond in overt ways to child maltreatment (e.g., aggressive towards others, violence, conduct problems, sexual risk behavior), while females tend to respond in covert manners (e.g., withdrawal, depression, self-blame, suicidal ideation and behaviors, eating disordered eating) (McClellan et al., 1997).
Thus, differential responses to child maltreatment indicate that males’ and females’ pathways from maltreatment to juvenile delinquency are likely to be different. However, there are some inconsistencies in the literature about gender differences with regard to the cycle of
violence hypothesis. Therefore, it is important to investigate potential gender differences in the relationship between childhood maltreatment and juvenile delinquency.
The present study
In sum, over the past two decades, a developing literature have addressed the link between child maltreatment and the propensity to become a juvenile or adult offender. Most studies found evidence that support the cycle of violence hypothesis, indicating that child maltreatment is associated with delinquent behavior in adolescence and adulthood. On the contrary, there are also studies who did not find this link. There is yet no meta-analysis
conducted about the relationship between different types of childhood maltreatment (e.g. sexual, physical, emotional abuse) and different forms of delinquency (e.g., general, violent) during adolescence. It is in particular essential to examine this link among juveniles, because of prevention purposes. If there appears to be a relationship between child maltreatment and juvenile delinquency, aggravated problems such as adult criminality might be prevented by developing interventions focused on juvenile delinquents. Further, potential gender differences in the link between child maltreatment and juvenile delinquency were not examined either in
previous meta-analyses, and therefore it remains unknown whether this link is gender-specific. Summarizing these results in a narrative review is impediment because of the inconsistencies in the literature, which stresses the importance of the present meta-analytic study. Therefore, the goal of the present study is to provide a comprehensive meta-analysis identifying to what extent child maltreatment is associated with both general and violent delinquency in adolescence and whether this relationship is moderated by gender and specific types of maltreatment. The following research questions will be addressed. First, is childhood maltreatment associated with delinquency during adolescence? How strong is this child maltreatment-delinquency link?
Second, is the relationship childhood maltreatment-delinquency moderated by gender and by specific types of maltreatment (e.g., sexual, physical, emotional abuse)? If childhood
maltreatment is indeed related to juvenile delinquency and this link appears to be different for males and females, gender-specific interventions aimed at delinquent youth may be developed to meet the different needs of males and males.
Method Sample of Studies
Initially, studies were selected based on five inclusion criteria. First, studies were included when the term delinquency was defined as behavior prohibited by the law. In the context of the cycle of violence, we only included studies concerning violent or general
delinquent behavior. Studies that focus on problem behavior were not selected, because problem behavior is considered as behavior that is not necessarily prohibited by the law. Second, the present meta-analysis only focused on delinquent behavior during adolescence. Thus, studies that focused on delinquent behavior in childhood or adulthood were not included. Third, studies had to focus on child maltreatment, meaning behaviors acting toward the child that are also
prohibited by the law such as sexual, physical and emotional child abuse and child neglect. Fourth, several studies have shown that cultural differences exist regarding parenting practices (e.g., Van IJzendoorn and Kroonenberg, 1988). Hence, with regard to the definition of the concepts child maltreatment, we only selected study samples from Western countries (USA, Canada, Europe, Australia and New Zealand). Fifth, to calculate a bivariate test statistic, manuscripts had to present results from a bivariate analysis of the relationship between child maltreatment and delinquency during adolescence in order to provide enough details for the calculation of the overall effect size.
After the search strategy and screening procedure, 144 studies were considered eligible for the current meta-analysis, which was too much for the purpose of this manuscript (master’s thesis). Therefore, two inclusion and three exclusion criteria were added. The sixth inclusion criteria meant that the maltreatment-delinquency link (either direct or indirect) was explicitly mentioned in the title or abstract. Seventh, we set requirements with regard to sample type. Study subjects had to be recruited from the general population, schools, or the high-risk population. If subjects had been recruited from the juvenile justice system, the study had included a control group. The first exclusion criteria meant that studies that only reported mild delinquency
measures from the Child Behavior Checklist or the Youth Self report (CBCL, YSR; Achenbach, 1991) were excluded. Second, studies with only sexual offense as outcome variable were
excluded, because sex offenses are considered as a specific type of offenses, other than general and violent delinquency. Finally, studies only focusing on neglect (not specified as childhood maltreatment) as predictor and studies addressing the specific link sexual child abuse-sexual offenses were excluded.
Literature Search Strategy
In order to find eligible studies, both a computerized and manual literature search was generated. From October until December 2014, studies were collected by searching for articles, books, book chapters, and reviews through electronic databases including PsycINFO, ERIC, MEDLINE (R), Pubmed and Sociological Abstracts. Search terms such as childhood
maltreatment*, child abuse*, child neglect* were cross-referenced with juvenile or adolescent delinq*, crim*, offend*. An additional search term was to not search for intervention studies and studies about mental health during adolescence. The computerized search covered a wide time span (January 1990 – December 2014), which resulted in 1,010 records and after removing the
duplicates in 772 records. After the computerized search, manual literature searches were conducted, meaning that reference lists of articles and reviews were checked in order to find relevant studies that were not found by the searching strategy through the electronic databases. Figure 1 displays a flowchart from the entire literature search procedure.
File Drawer Problem
Publication bias is the tendency of journals to accept manuscripts that report strong significant associations. This problem is also identified as the file drawer problem (Rosenthal, 1979). Consequently, the publication bias or file drawer problem may have implications on the outcome of the present meta-analysis (Rosenthal 1991). There are several methods to address this problem. The best solution is trying to obtain as many unpublished materials as possible (e.g., Mullen, 1989; Rosenthal, 1991). However, unpublished manuscripts are not verified in the research field and were considered of poorer quality as compared to published studies.
Therefore, unpublished manuscripts (i.e., dissertations, paper presentations) were not included in the current meta-analysis.
In order to address the publication bias, Rothstein (2008) recommended to apply additional methods, besides retrieving as many studies as possible, to address potential publication bias. In the current meta-analysis, funnel plot asymmetry was tested. If no
publication bias exists, the distribution of the study’s effect size should be shaped as a funnel. Asymmetry in the funnel plot reflect the presence of a publication bias, meaning that studies showing positive or negative outcomes were included (Sutton et al., 2000). Funnel plot
symmetry was examined by regressing the effect size divided by its standard error, against the estimate’s precision (the inverse of the standard error), which largely depends on sample size (Egger et al., 1997). If there was symmetry, meaning no publication bias, the regression line ran
through the origin and the intercept did not significantly deviate from zero. In the present meta-analysis, the shape of the funnel plot was symmetric and the intercept did not significantly deviate from zero (t = 1.724, p = .087), meaning that there is no indication of publication bias.
Coding of the Study Outcomes and Characteristics
To code study characteristics, a coding system was developed. First, study results (test statistic and value) and sample size were retrieved. Second, data on demographics were gathered. Gender was coded as follows: the percentage adolescent males (M = 56.3%; range 0.0 –
100.0%), and gender of the study sample (both males and females, only males, only females). Age of the study subjects was coded at the time of the child maltreatment measurement (M = 13.6; range 3.5 – 23.0), and age of the subject during the juvenile delinquency assessment (M = 17.5; range 14.5 – 28.3). Further, the ethnicity of study subjects was coded by percentage ethnicity, defined as the percentage non-Caucasian subjects in the sample (M = 41.5; range 12.0 – 93.0%). Sample type refers to the type of the recruited study subject, and was coded in the following categories: general population/ community samples, recruited through schools, recruited through the juvenile justice system with a control group, high-risk group plus control group. Also, data of the country in which the study was conducted was collected (USA/ Canada = 85% of the studies, Europe = 10%, Australia/ New Zealand = 5%). With regard to the
independent variable, childhood maltreatment was coded as follows: general child maltreatment, sexual abuse, physical abuse, and emotional/ psychological abuse and neglect. The outcome variable delinquency was coded in the categories general delinquency and violent delinquency. Finally, it was analyzed whether indicators of study quality influenced the combined effect size. Therefore, data on quality indicators of empirical studies were collected: sample size (M = 854; range 20 – 3,694), study design (cross-sectional, longitudinal), source of maltreatment
(self-report, official records, other, more than one source) and source of delinquency (self-(self-report, official records, other, more than one source) were coded. Another indicator of study quality was the source of maltreatment and juvenile delinquency, in that there was a distinction between official records versus self-report data and one single informant versus multiple informants. Table 1 presents the descriptive statistics of the studies included in the meta-analysis.
Data Coding Reliability
The effect sizes and study characteristics were coded by the first author (principal rater). In order to examine the reliability of the coding process, a selection of 10 articles (42%) was randomly drawn from the total number of included articles (N = 24) and this selection was coded by an independent rater. The 10 articles included 31 effect sizes, which is 24% of the total
number of included effect sizes. Interrater reliability was tested by calculating Kappa coefficients for categorical variables (e.g., study design, sample type, maltreatment type, delinquency type) and intraclass correlations for continuous variables (e.g., sample size, percentage male, age at maltreatment measurement, age at delinquency measurement). Across the categorical variables, the average Kappa coefficient between the principal rater and the independent rater was .926 (ranging from .703 to 1.000). Regarding the continuous variables, the average intraclass correlation coefficient between the principal rater and the independent rater was .893 (ranging from .637 to 1.000). These results indicate that high interrater agreement was obtained for the majority of the variables. Only the continuous variable ‘percentage ethnicity’ and the categorical variable ‘statistical test’ showed an interrater agreement below .800 (.673 and .703 respectively). Cases of disagreement were discussed and resolved until full agreement was obtained.
An effect size for each individual study was calculated. All effect sizes were transformed to Pearson’s r to align the association between child maltreatment and delinquency on a same scale in order to compare them. Before combined effect sizes and moderation analyses could be generated, Pearson’s correlations were transformed in Fisher Z-values to approximate a normal sampling distribution (Lipsey & Wilson, 2001; Mullen, 1989). Outliers were identified but removal was not necessary. Each continuous moderator variable was centered around its mean, and categorical variables were dummy coded.
To generate the meta-analysis, random effect models were utilized (Raudenbush, 2009). Traditional random effect models often perform two-level models, assuming that the observed effect sizes vary across sample variation and between studies. However, studies often report more than one effect size. It seems plausible that effect sizes within one study are more similar as compared to effect sizes between different studies (i.e., effect sizes from one study are based on the same sample type and are often measured with similar instruments). Therefore, in the present meta-analysis a three-level random effects model (Cheung, 2014; Van den Noortgate, López-López, Marín-Martínez, & Sánchez-Meca, 2013) was applied. Variation of effect sizes was taken into account in terms of 1) variance at level 1, referred to as sampling variation of observed effect sizes, 2) variance at level 2, meaning variance between the effect sizes from the same study, and 3) variance at level 3, referred to as variance between studies. Consequently, the three-level model include three residuals, which are assumed to be normally distributed with zero mean:
With an observed effect size in study k, 𝑇𝑇𝑗𝑗𝑘𝑘, is assumed to be equal to an overall population effect size, 𝜃𝜃, plus a random deviation of the mean population effect size in study k from this overall population effect size, 𝜈𝜈𝑘𝑘, plus a random deviation of the j-th population effect in study k from the mean effect in this study, 𝑢𝑢𝑗𝑗𝑘𝑘 , plus a random error deviation of the observed effect size from the population effect, 𝑒𝑒𝑗𝑗𝑘𝑘. In a three-level model, the variances of level 2 and level 3 are estimated. Level 1 variance (sampling variance) is not estimated, because it is considered as known as it can be derived based on for instance sample size. Regarding Fisher’s Z for example, sampling variance is equal to 1/(N-3), with N equal to the sample size (Lipsey & Wilson, 2001). Thus, a three-level approach was chosen, because it allows for variation between effect sizes within one study.
Analyses were conducted as follows: first, the overall effect size of the relationship between childhood maltreatment and juvenile delinquency was estimated. Next, a heterogeneity test of the level 2 and level 3 variances from the intercept-only model was examined, then moderator variables were added (bivariate models), followed by significant moderators added in a multivariate model. Finally, variances of level 2 and 3 were calculated for both the bivariate models and multivariate model. This procedure was performed on three different datasets, each representing a different outcome variable 1) overall delinquency (total dataset, containing both general and violent delinquency), 2) general delinquency, and 3) violent delinquency. In the light of the cycle of violence, we were particularly interested in the influence of childhood
maltreatment on violent delinquency, beyond the effect of child maltreatment on overall delinquency. The analyses were generated in a statistical program, R, using the Restricted Maximum Likelihood Estimation method to estimate the parameters.
Overall Mean Effect Sizes
The overall mean effect sizes are displayed in Table 2. Note that the estimates of the effect sizes mentioned in the table are Fisher-Z transformed.
Overall delinquency. Results revealed a significant overall mean effect size for the
association between childhood maltreatment and overall juvenile delinquency, mean Fisher’s Z = .157, p < .001, k = 129, indicating that adolescents with a history of childhood maltreatment did have higher levels of overall delinquency during adolescence (see Table 2). With regard to the total explained variance, 9.6% is explained by level 1 variance (observed sample variance), 44.7% is explained by level 2 (variance between effect sizes from the same study) and 45.7% of the total variance is explained by level 3 (variance between studies).
General delinquency. The overall mean effect size for the relationship between
childhood maltreatment and general delinquency also appeared to be significant, mean Fisher’s Z = .152, p < .001, k = 79. This suggests that children who were maltreated during childhood had higher levels of general delinquency in adolescence. Regarding the total amount of explained variance, 9.9% is explained by level 1 variance, 45.9% is explained by level 2 and 44.2% of the total variance is explained by level 3.
Violent delinquency. We found a significant overall mean effect size for the link
between childhood maltreatment and violent delinquency during adolescence, mean Fisher’s Z = .135, p < .001, k = 50, indicating that adolescents with a history of childhood maltreatment did have elevated levels of violent delinquency in adolescence. The total amount of variance is for 9.5% explained by level 1, 48.1% by level 2 and for 42.4% explained by level 3.
A hetereogeneity test showed significant heterogeneous variances on level 2 and level 3, meaning that the significant relationship between childhood maltreatment and overall, general
and violent delinquency was not robust and likely affected by study characteristics. Therefore, potential moderators were examined.
Moderating Role of Gender
Overall delinquency. Table 3 displays the results of bivariate analysis of the moderator
variables on the link between childhood maltreatment and overall delinquency. Gender was not found to be a significant moderator in the relationship between child maltreatment and overall delinquency. Neither for the categorical measure of gender, F(22,122) = 0.418, p = .246, nor for the continuous measure of percentage male in study sample, F(1,123) = 0.007, p = .932. This suggests that the relationship between child maltreatment and overall delinquency in adolescence was not different for males and females.
General delinquency. With regard to outcome measure general delinquency, estimate
values of β0, mean Fisher’s Z, β1, omnibus test results and p-values were comparable to the estimate values of outcome measure overall delinquency, they did not deviate much. Therefore, no table is presented for the bivariate analyses of moderating variables (table can be obtained by contacting the first author). Again, gender was not a significant moderator in the association between child maltreatment and general delinquency during adolescence, indicating that this link was not different for males and females.
Violent delinquency. Only the results for significant moderators in the relationship
between childhood maltreatment and violent delinquency in adolescence are displayed in Table 4 (complete table can be obtained by contacting the first author). Gender was not found to be a significant moderator, assuming that the maltreatment-violent delinquency link was not gender-specific and appeared to be equal for males and females.
Overall delinquency. Results showed a trend of the moderator variable childhood
maltreatment type, F(3, 125) = 2.317, p = .079. When comparing the different types of childhood maltreatment, the mean Fisher’s Z of sexual abuse (β0 = .114) was significantly lower (p < .05) than the mean Fisher’s Z of reference category general maltreatment (β0 = .175). This suggests that the association between sexual abuse and overall delinquency during adolescence was significantly weaker than the association between general maltreatment and overall delinquency.
General delinquency and Violent delinquency. The significant relationship mentioned
above for overall delinquency was not apparent for the outcome measures general delinquency and violent delinquency. Childhood maltreatment was no significant moderator, indicating that there was no significant difference in the strength of the relationship between different forms of childhood maltreatment and general delinquency and violent delinquency in adolescence.
Moderating Role of Sample Type
Violent delinquency. Only results for the significant moderator variables are presented
(see Table 4). Results showed that sample type was a significant moderator in the relationship between child maltreatment and violent delinquency, F(3, 46) = 3.028, p = .039. This significant difference was found between study subjects recruited through schools and subjects recruited via the juvenile justice system plus a control group. The mean Fisher’s Z of subjects recruited through schools was .081, while the mean Fisher’s Z of subjects recruited via the juvenile justice system was .233. This means that the link between child maltreatment and violent delinquency was significantly stronger for subjects from the juvenile justice system as compared to subjects recruited through schools. However, this results was only present in bivariate analysis. When adding sample type to a multivariate model (which controls for background variables), no significant result was found for sample type.
Multivariate Analyses
Overall delinquency. Finally, a multivariate analysis was performed in order to examine
the unique contribution of different types of maltreatment on the maltreatment-overall
delinquency link, taken into account background variables. We analyzed one model, including child maltreatment type (this predictor showed a trend in bivariate analysis), gender as expressed in percentage males, and the interaction effects of gender with maltreatment type. Results
revealed that sexual abuse remained significant (p = .010), showing a significant weaker link between sexual abuse-overall delinquency as compared to the link general maltreatment-overall delinquency after taking into account other predictor variables (see Table 5). No significant effect sizes were found for other types of child maltreatment. Analyses of interaction effects showed no significant interaction effect for gender on the relationship between child
maltreatment and overall delinquency. This indicates that for all forms of child maltreatment, the link maltreatment-overall delinquency during adolescence was not significantly different for males compared to females. The multivariate model had a significantly worse fit on the level 2 and level 3 variance as compared to the intercept-only model without explanatory variables (see Table 5), indicating that results are not robust.
Discussion
This meta-analysis summarized and integrated findings from previously conducted studies on the association between childhood maltreatment and juvenile delinquency, and investigated the effect of possible moderating variables, including gender and type of child maltreatment. The cycle of violence assumption proposing that a history of childhood maltreatment (of any type) is related to juvenile delinquency was confirmed by this meta-analytic study. Specifically, the maltreatment-delinquency was found for different types of
juvenile delinquency, that is, overall, general, and violent delinquency. The strength of the link between child maltreatment and overall, general and violent delinquency was small (r = .156, r = .151, r = .134 respectively), according to the criteria of Cohen (1988). Furthermore, the strength of the relationships did not significantly deviate from one another, suggesting that in future research it might not be necessary to distinguish between delinquency type in order to examine the maltreatment-delinquency link. These findings are relatively similar to results from the meta-analysis of Derzon (2010). His study results revealed weighted effect sizes of .213 for the link between childhood maltreatment and criminal behavior, and a weighted effect size of .100 for the relationship between childhood maltreatment and violent behavior. In addition, single study outcomes often report small effect sizes for the child maltreatment-juvenile delinquency link. Thus, results from this meta-analytic study adds to the existing evidence with regard to the cycle of violence assumption.
Investigating the effects of potential moderators revealed that the associations between child maltreatment and overall, general, and violent delinquency were not moderated by gender, suggesting that the likelihood for maltreated children to engage in delinquency during
adolescence was quite similar for males and females. This is partly in contrast with our expectations of sex differences being present, particularly for the outcome on violent delinquency. There are several theories explaining gender differences in delinquency. First, Moffit and colleagues (2001) proposed that male delinquency can be explained in the same way as female delinquency, but that males are more often exposed to risk factors than females. A second theory asserted that the causes and risk factors of delinquency are different for males compared to females. The third assumption stated that males have a lower threshold for becoming juvenile delinquents than females, meaning that females (relative to males) need to
experience more risk factors before becoming delinquent although females and males may be affected by similar risk factors. However, when the threshold is passed by females, it is assumed that the manifestation of delinquency is more severe than in males (Loeber & Keenan, 1994). Several studies found evidence for gender differences in delinquency, in that males were more likely to react in overt manners to child maltreatment, such as vandalism and physical assault (e.g., McClellan et al., 1997). In contrast to males, females tended to react in more covert ways, such as becoming introvert and running away. Consequently, we expected that males (relatively to females) with a history of maltreatment were more likely to get involved in (violent)
delinquency. One possible explanations for the lack in gender differences might be the shift in delinquency involvement among females. Over the years, females became more engaged in offending and violent behavior (e.g., Harrison & Beck, 2006), so the gap between male and female offending became narrower. Another reason for not finding gender differences in delinquency might be that childhood maltreatment might be a shared risk factor, meaning that child maltreatment has similar predictive value for males and females in explaining delinquency (see the second theory mentioned above). An European review was conducted (Wong, Slotboom, & Bijleveld, 2010), presenting an overview of unique and shared individual, familial, school, and peer risk factors associated with male and female delinquency as found in Europe. Overall, many similarities in risk factors were found between females and males. However, clear differences became apparent, in that females seemed to be more influenced by risk factors from the social context, such as negative life events, physical abuse by parents, and internalizing problems. In this meta-analysis, the relationship between physical abuse and juvenile delinquency was not different for males and females, which is an opposite finding compared to the European review. Although there seems to be overlap in risk factors for predicting delinquency, a number of
unique risk factors for both females and males emerged, supporting the second theory about the different causes in explaining gender differences in delinquency. Future research should
investigate whether different forms of maltreatment are shared or unique risk factors and whether overlapping risk factors have the same predictive value for female and male delinquency.
Besides gender differences, we also examined whether the link between maltreatment-delinquency differed for various types of childhood maltreatment. Depending on the type of maltreatment, adolescents were engaged in more or less delinquent behavior. To be more
specific, a negative effect for the relationship between sexual abuse and overall delinquency was found, indicating that children with a history of sexual abuse showed decreased levels of overall juvenile delinquency as compared to children with a history of general child maltreatment. This finding is in line with previous results from two reviews, asserting that victims of sexual abuse were least likely to involve in violent behavior than victims of physical abuse, neglect, and non-victims (McGrath, Nilsen, & Kerley, 2011; Trickett, Negriff, Ji, & Peckins, 2011). An
explanation for this phenomenon is that sexually abused adolescents were more likely to run away than to engage in violent behaviors, compared to adolescents who were physically abused, neglected or non-victim (Siegel & Williams, 2003; Widom & Maxfield, 2001). This stresses the importance of investigating response trajectories of various types of child maltreatment, in particular with regard to sexual abuse. It seems plausible that adolescents with a history of sexual abuse tend to response in more covert manners and that adolescents who were for instance physically abused or neglected tend to respond in more overt ways, such as delinquent and violent behavior.
It is remarkable that we did not find an effect of gender on the relationship between sexual abuse and juvenile delinquency. There are studies revealing that male victims of sexual
abuse were more likely to commit criminal and deviant acts, compared to female victims of sexual abuse (Chandy et al., 1996; Seto & Lalumière, 2010). It is recommended that future studies address this issue.
Interestingly, the link between child maltreatment and violent delinquency was not moderated by type of maltreatment. We expected that physical abuse in particular would be associated with adolescent violence. There are studies that found evidence that physical and emotional abuse are associated with an increased risk of violent behavior during adolescence (Egeland et al., 2002; Lansford et al., 2007; Piquero & Sealock, 2000). Also, in a sample of 1,404 low-income minority children, victims of both physical abuse and neglect were more involved in violent delinquency as compared to non-victims. The two abuse groups were not significantly different, suggesting that each maltreatment type was a significant risk factor for juvenile delinquency (Mersky & Reynolds, 2007). Childhood maltreatment might be overall related to violence, regardless of maltreatment type.
There might also be a biological explanation relating epigenetic mechanisms to the cycle of violence hypothesis. This is a relatively new line of research that studies the effects of
childhood maltreatment on epigenetic changes in the genome. Social scientists found evidence that children exposed to violence are presumed to be at greater risk for juvenile and adult violent behavior (Maxfield & Widom, 1996; Widom, 1989; Wilson, Stover, & Berkowitz, 2009). One recent study (Yang et al., 2013) found differences in methylation values between maltreated and nonmaltreated children, and another study (Essex et al, 2013) revealed that childhood stress is prospectively linked to differences in methylation in adolescence. This indicates that childhood maltreatment affects epigenetic changes on the long term. Furthermore, there is evidence
al., 2014). Also, there are several genes that have been found to be related to antisocial behavior, including MAOA, DAT1,DRD2, 5HTT, COMT (Dick et al, 2011; Raine, 2013; Tielbeek et al, 2012). Based on these findings, it seems plausible that epigenetic changes may affect these particular genes, and in turn mediate the relationship between childhood maltreatment and juvenile delinquency. This is a promising line of research that is recommended to be further explored.
Practical implications
The current meta-analysis has some practical implication with regard to prevention and intervention programs. Based on our results, it might not be necessary to develop gender-specific intervention programs, since the maltreatment-delinquency link appeared to be similar for males and females. According to Welsh and Farrington (2007), the benefits of a prevention program can be maximized when early risk factors are known. For instance, in early childhood, family factors including child maltreatment and coercive parenting play a more significant role than peer influence (this is affecting adolescence). Besides child maltreatment, there are other risk factors related to the development of juvenile delinquency that become apparent in childhood. Therefore, child maltreatment and other childhood factors could be targets in prevention programs for reducing juvenile delinquency.
Limitations
There were several limitations of the present meta-analysis that should be addressed. First, we only included published studies. Although non-published studies are often considered of poorer quality and there was no indication of publication bias, Mullen (1989) and Rosenthal (1991) both recommended to obtain as many manuscripts as possible to obtain the best
childhood maltreatment and juvenile delinquency were included, because of comparability reasons. With multivariate analyses, unique contributions of child maltreatment to juvenile delinquency can be calculated, while simultaneously controlling for other factors. However, Lipsey and Wilson (2001) stated that meta-analyzing multivariate associations is problematic, because the mean effect size of the dependent variable is affected by control variables in the multivariate model. Therefore, effect sizes derived from multivariate models cannot be compared to effect sizes from bivariate models (these effect sizes do not depend on other variables). Next, study subjects were categorized in one type of maltreatment, not taking into consideration that they might experience multiple forms of maltreatment. The intensity of the maltreatment might also have an influence on the strength of the maltreatment-delinquency link. One study found that adolescents who reported three types of maltreatment (physical abuse, sexual abuse, and neglect) showed significantly more delinquent behavior as compared to adolescents who
reported less types of maltreatment (Arata, Langhinrichsen-Rohling, & O’Brien, 2007). Also, we did not take into account the developmental timing and duration of childhood maltreatment, which might partially explain the lack of the moderating role of maltreatment type. For further research, intensity, developmental timing, and duration of childhood maltreatment should be taken into consideration when evaluating the relationship between childhood maltreatment and juvenile delinquency.
Besides these limitations, there are several strengths that we would like to stress. This is the first meta-analysis that investigated the moderating effect of gender on the relationship between different types of maltreatment and different types of juvenile delinquency. Further, the present meta-analytic study used a 3-level random effects model, controlling for an additional variance, known as the variance within the same study.
To conclude, the present meta-analysis confirmed that child maltreatment was related to different types of delinquency during adolescence. This association was not moderated by
gender, indicating that there are no differences in this relationship between males and females. In addition, the relationship between child maltreatment and overall delinquency was weaker for adolescents with a history of sexual abuse (as compared to children with a history of general child maltreatment).
Figure 1. Selection and screening process. This flowchart displays the entire in- and exclusion procedure of collecting articles to be included in the meta-analysis.
Reasons of exclusion, after reading full text: - No bivariate analysis
- Only measurements of mild delinquency - Sample type
Records from each database (after removing duplicates): PsycInfo = 396
ERIC = 153 MEDLINE (R) = 34 PubMed = 1
Sociological Abstracts = 188
Number of records identified by searching through databases:
772
Number of records identified by manual literature search:
24
Number of records screened (based on initial inclusion criteria):
796
Number of manuscripts/ studies assessed for inclusion in meta-analysis
(by reading full text): 48
Final number of manuscripts/ studies included in meta-analysis:
24 Additional inclusion/ exclusion criteria:
- Link child maltreatment – juvenile delinquency explicitly mentioned in abstract;
- Sample type: general population, schools, high-risk population or juvenile justice system with control group; - Studies with only mild delinquency measures (e.g., CBCL, YSR) were excluded.
- Studies with only sex offense as outcome variable were excluded;
- Studies with only neglect as form of maltreatment were excluded.
Number of records included, based on the title and abstract:
144
Excluded: 652
Table 1
Descriptive Statistics of the Studies Included in the Meta-Analysis Study Year Country N n
maltreated group
n non-maltreated group
Sample type Male % Ethn % Delinquency typea Age at delinquency measure (in years) Maltreatment typeb Age at maltreatment measure (in years) Arata et al. 2007 USA/
Canada
536 121 415 High risk group + control group 46.7 64.5 General 15.6 General; Sexual; Physical; Emo + Neglect 15.6
Crooks et al. 2007 USA/ Canada
1,788 - - Schools 47.6 12.0 Violent 14.5 General 14.5
Ellis & Wolfe 2009 USA/ Canada
1,558 - - Schools 48.4 22.0 General 15.0 Physical;
Emo + Neglect 15.0
Fox, Benson, & Spohn 2000 USA/ Canada 1,575 - - Juvenile justice system + control group 49.3 33.0 General - General -
Study Year Country N n maltreated group n non-maltreated group
Sample type Male % Ethn % Delinquency typea Age at delinquency measure (in years) Maltreatment typeb Age at maltreatment measure (in years) Gover 2002 USA/ Canada 3,694 - - Juvenile justice system + control group 100.0 68.0 Violent 16.2 General 16.2
Herrenkohl et al. 2003 USA/ Canada
298 - - High risk group + control group 54.3 16.0 Violent 18.5 General 3.5 Herrera & McCloskey 2003 USA/ Canada
141 43 98 General population 0.0 44.0 General; Violent 14.9 Sexual; Physical 9.1 Krischer & Sevecke
2008 Europe 44 - - Juvenile justice system + control group 52.7 14.7 General 16.9 Sexual; Physical; Emo + Neglect 16.9
Lahlah et al. 2013 Europe 318 - - Juvenile justice system + control group; Schools 100.0 26.9 Violent 16.5 Sexual; Physical; Emo + Neglect 16.5
Lansford et al. 2007 USA/ Canada
574 69 505 Schools 52.0 17.0 General; Violent
Study Year Country N n maltreated group n non-maltreated group
Sample type Male % Ethn % Delinquency typea Age at delinquency measure (in years) Maltreatment typeb Age at maltreatment measure (in years) Lemmon 1999 USA/ Canada
495 130 365 High risk group + control group 100.0 81.6 General 18.0 General; Emo + Neglect 18.0 Maxfield & Widom 1996 USA/ Canada 1,050 605 445 Juvenile justice system + control group 49.8 33.0 General; Violent 18.0 General 11.0 Salzinger, Rosario, & Feldman 2007 USA/ Canada
153 78 75 High risk group + control group 61.0 93.0 Violent 17.0 Physical 10.5 Siegel & Williams 2003 USA/ Canada
411 206 205 High risk group + control group
0.0 83.5 General; Violent
28.3 Sexual 8.3
Smith et al. 2008 USA/ Canada
872 217 655 General population 50.0 82.1 General; Violent 19.5 General 20.0 Smith & Thornberry 1995 USA/ Canada 908 124 784 Schools 74.0 85.0 General; Violent 15.8 General 12.0 Stouthamer-Loeber et al. 2001 USA/ Canada 156 52 104 Schools 100.0 57.5 General; Violent 19.7 General 18.3
Study Year Country N n maltreated group n non-maltreated group
Sample type Male % Ethn % Delinquency typea Age at delinquency measure (in years) Maltreatment typeb Age at maltreatment measure (in years) Swanston et al. 2003 Australia/
N-Z
155 82 73 High risk group + control group 22.5 - General; Violent 18.4 Sexual 9.5 Tyler, Johnson, & Brownridge 2008 USA/ Canada
306 76 230 High risk group + control group 41.9 49.0 General 15.2 Sexual; Physical; Emo + Neglect 12.2 Wekerle & Wolfe 1998 USA/ Canada
359 - - Schools 39.9 20.0 General 15.2 General 15.2
Widom 1996 USA/
Canada
1,160 493 667 High risk group + control group
- - General - General -
Widom & Ames 1994 USA/ Canada 980 313 667 Juvenile justice system + control group 44.4 31.6 General; Violent 18.0 General; Sexual; Physical; Emo + Neglect 11.0
Wolfe et al. 2001 USA/ Canada
710 - - Schools 50.0 13.0 General;
Violent
Study Year Country N n maltreated group n non-maltreated group
Sample type Male % Ethn % Delinquency typea Age at delinquency measure (in years) Maltreatment typeb Age at maltreatment measure (in years) Yun, Ball, &
Lim
2010 USA/ Canada
3,472 - - General population 46.1 33.6 General; Violent - General; Sexual; Physical; Emo + Neglect -
Note. N = mean number of participants; n maltreated group = mean number of participants who were maltreated; n Non-disorder group = mean number of participants who were not maltreated; Male % = mean percentage of male participants; Ethn % = mean percentage of participants with a minority status; Emo + Neglect = emotional/ psychological maltreatment and neglect; N-Z = New-Zealand.
a Delinquency type was based on self-report or official records. b
Table 2
Results for the Mean Effect Sizes of the Relationship between Childhood Maltreatment and Juvenile Delinquency (Overall, General and Violent) Outcome variables # studies # ES Mean
Fisher’s Z
95% CI t-statistic p-value Variance level 2a Variance level 3b Overall delinquency 20 129 .157 .117; .196 7.800 < .001*** .005*** .006*** R2 level 1c = 9.6% R2 level 2a = 44.7% R2 level 3b = 45.7% General delinquency 15 79 .152 .106; .197 6.601 < .001*** .006*** .005*** R2 level 1c = 9.9% R2 level 2a = 45.9% R2 level 3b = 44.2% Violent delinquency 15 50 .135 .084; .185 5.376 < .001*** .006*** .005** R2 level 1c = 9.5 % R2 level 2a = 48.1% R2 level 3b = 42.4%
Note. # studies = number of independent studies; # ES = number of effect sizes; Mean Fisher’s Z = mean effect size in Fisher’s Z; CI = confidence interval. a Variance between the effect sizes from the same study.
b
Variance between studies.
Table 3
Results for all Moderator Variables in the Relationship between Childhood Maltreatment and Overall Delinquency (Bivariate Analyses) Moderator variables
(categorical)
# studies # ES β0, mean Fisher’s Z
(95% CI)
β1 (95% CI) Omnibus test p-value Variance
level 2a Variance level 3b Design F(1,127) = 1.049 .308 .005*** .005*** Cross-sectional (RC) 11 57 .175 (.122; .228) Longitudinal 9 72 .134 (.075; .193) -.041 (-.120; .038) Country F(2,126) = 1.470 .234 .005*** .007*** USA/ Canada (RC) 18 112 .155 (.111; .200) Europe 2 14 .148 (.012; .285) -.007 (-.151; .137) Australia/ New Zealand 1 3 .288 (.132; .444) .133 (-.021; .287)
Gender F(2,122) = 0.418 .246 .006*** .005***
Males only (RC) 6 33 .181 (.120; .242)
Females only 4 15 .196 (.127; .266) .016 (-.057; .088) Both males and females 13 77 .138 (.093; .183) -.043 (-.110; .024)
Sample type F(3,125) = 1.205 .311 .005*** .009***
General population (RC) 3 18 .148 (.034; .263)
Recruited schools 8 43 .119 (.051; .188) -.029 (-.162; .104) Recruited juvenile court 4 42 .199 (.126; .272) .051 (-.085; .187) High-risk group 7 26 .194 (.123; .266) .046 (-.089; .181)
Delinquency type F(1,127) = 0.751 .388 .005*** .006*** General delinquency (RC) 15 79 .163 (.121; .206) Violent delinquency 15 50 .147 (.102; .193) -.016 (-.053; .021) Delinquency source F(1,127) = 0.132 .717 .005*** .006*** Self-report (RC) 16 68 .154 (.109; .198) Official record 9 61 .164 (.109; .219) .010 (-.046; .066) Maltreatment type F(3, 125) = 2.317 .079 .005*** .006*** General maltreatment (RC) 12 70 .175 (.130; .219) Sexual abuse 8 20 .114 (.056; .172) -.061 (-.116; -.006)* Physical abuse 10 22 .131 (.075; .186) -.044 (-.098; .010) Emotional abuse + Neglect 8 17 .178 (.120; .236) .003 (-.050; .057)
Maltreatment source F(2,126) = 1.213 .301 .005*** .007***
Self-report (RC) 10 50 .144 (.085; .204)
Official record 8 65 .192 (.129; .255) .048 (-.030; .126) Other (e.g., parent-report) 4 14 .124 (.031; .218) -.020 (-.129; .090)
Moderator variables (continuous)
# studies # ES β0, mean Fisher’s Z
(95% CI)
β1 (95% CI) Omnibus test p-value Variance
level 2a Variance level 3b Publication year 20 129 .155 (.112; .198) .001 (-.006; .008) F(1,127) = 0.111 .740 .005*** .006*** Sample size 20 129 .157 (.117; .197) -.000 (-.000; .000) F(1,127) = 0.012 .911 .006*** .006*** Percentage male 20 125 .157 (.117; .197) -.000 (-.001; .001) F(1,123) = 0.007 .932 .006*** .006*** Percentage ethnicity 20 122 .153 (.110; .195) .000 (-.001; .002) F(1,120) = 0.235 .629 .006*** .006*** Age at delinquency measure 20 116 .162 (.124; .201) -.009 (-.020; .002) F(1,114) = 2.707 .103 .006*** .004*** Age at maltreatment measure 20 116 .164 (.124; .203) .001 (-.008; .009) F(1,114) = 0.036 .850 .006*** .005*** Note. # studies = number of independent studies; # ES = number of effect sizes; CI = confidence interval; RC = reference category
a Variance between the effect sizes from the same study. b Variance between studies.
Table 4
Results for the Significant Moderator Variables in the Relationship between Childhood Maltreatment and Violent Delinquency (Bivariate Analyses) Moderator variable
(categorical)
# studies # ES β0, mean Fisher’s Z
(95% CI)
β1 (95% CI) Omnibus test p-value Variance
level 2a Variance level 3b Sample type F(3,46) = 3.028 .039* .004*** .015*** General population (RC) 3 8 .124 (-.032; .280) Recruited schools 6 22 .081 (-.019; .182) -.043 (-.228; .143)** Recruited juvenile court 3 16 .233 (.116; .349) .108 (-.086; .303)** High-risk group 4 4 .187 (.034; .341) .063 (-.156; .283)
Note. # studies = number of independent studies; # ES = number of effect sizes; CI = confidence interval; RC = reference category a Variance between the effect sizes from the same study.
b Variance between studies.
Table 5
Results for the Moderator Variables in the Relationship between Childhood Maltreatment and Overall Delinquency (Multivariate Analyses)
Moderator variables β (SE) 95% CI t-statistic p-value # ES Omnibus test Variance level 2a Variance level 3b Model 116 F(7,117) = 2.242* 0.005*** 0.006*** Intercept .175 (.023) .130; .220 7.680 < .001*** Sexual abuse -.092 (.035) -.161; -.023 -2.630 .010** Physical abuse -.046 (.029) -.103; .010 -1.619 .108 Emo + Neglect -.008 (.028) -.063; .048 -0.268 .789 Percentage males -.000 (.000) -.001; .001 -0.438 .662 Interaction sexual abuse *
percentage males
-.001 (.001) -.003; .001 -1.377 .171
Interaction physical abuse * percentage males
.001 (.001) -.002; .003 0.533 .595
Interaction emo + neglect * percentage males
.002 (.001) -.000; .005 1.599 .113
Note. # ES = number of effect sizes; CI = confidence interval. a
Variance between the effect sizes from the same study. b Variance between studies.
References
Achenbach, T. M. (1991). Manual for the child behavior checklist/4-18 and 1991 profile. Burlington: University of Vermont, Department of Psychiatry.
Akers, R. L. (2009). Social learning and social structure: A general theory of crime and deviance. Piscataway, New Jersey: Transaction Publishers.
Arata, C. M., Langhinrichsen-Rohling, J. B. D., & O’Brien, N. (2007). Differential correlates of multi-type maltreatment among urban youth. Child Abuse and Neglect, 31, 393 – 415.
Bandura, A. (1977). Social learning theory. New York: General Learning Press.
Chandy, J. M., Blum, R. W., & Resnick, M. D. (1996). Gender-specific outcomes for sexually abused adolescents. Child Abuse &
Neglect, 20, 1219−1231.
Cheung, M. W. L. (2014). Modeling dependent effect sizes with three-level meta-analyses: A structural equation modeling approach.
Psychological Methods, 19, 211-229.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale: Earlbaum.
Crooks, C. V., Scott, K. L., Wolfe, D. A., Chiodo, D., & Killip, S. (2007). Understanding the link between childhood maltreatment and violent delinquency: What do schools have to add? Child Maltreatment, 12, 269 – 280.
Derzon, James H. (2010). The correspondence of family features with problem, aggressive, criminal, and violent behavior: A meta- analysis. Journal of Experimental Criminology, 6, 263–92.
Dick, D. M., et al. (2011). Genome-wide association study of conduct disorder symptomatology. Molecular Psychiatry, 16, 800-808.
Egeland, B., Yates, T., Appleyard, K., & Van Dulmen, M. (2002). The long-term consequences of maltreatment in the early years: A developmental pathway model to antisocial behavior. Children’s Services: Social Policy, Research, and Practice, 5, 249 – 260.
Egger, M., Smith, G. D., Schneider, M., & Minder, C. (1997). Bias in meta-analysis detected by a simple, graphical test. British
Medical Journal, 315, 629–634.
Ellis, W. E., & Wolfe, D. A. (2009). Understanding the association between maltreatment history and adolescent risk behavior by examining popularity motivations and peer group control. Journal of Youth Adolescence, 38, 1253-1263.
Essex, M. J., et al. (2013). Epigenetic vestiges of early developmental adversity: Childhood stress exposure and DNA methylation in adolescence. Child Development, 84, 58-75.
Fox, G.L., & Benson, M.L. (2000), ‘Families, Crime and Criminal Justice: Charting the Linkages’, in G.L. Fox and M.L. Benson (eds.), Families, Crime and Criminal Justice (Vol. 2, pp. 1–21). New York: Elsevier Science.
Gault-Sherman, M., Silver, E., & Sigfusdottir, I. D. (2009). Gender and the associated impairments of childhood sexual abuse: A national study of Icelandic youth. Social Science & Medicine, 69, 1515−1522.
Gilbert, R., Widom, C. S., Browne, K., Fergusson, D., Webb, E., & Janson, S. (2009). Burden and consequences of child maltreatment in high-income countries. Lancet, 373, 68-81.
Gover, A. R. (2002). The effects of child maltreatment on violent offending among institutionalized youth. Violence and Victims,
Guillemin C., et al. (2014). DNA methylation signature of childhood chronic physical aggression in T cells of both men and women. PLoS ONE, 9, e86822.
Harrison, P. M., & Beck, A. J. (2006). Prison and jail inmates at midyear 2005. National Criminal Justice Reference Service Abstracts
Database. Doi: NCJ213133.
Herrenkohl, T. I., Huang, B., Tajima, E. A., & Whitney, S. D. (2003). Examining the link between child abuse and youth violence: An analysis of mediating mechanisms. Journal of Interpersonal Violence, 18, 1189 – 1208.
Herrera, V. M., & McCloskey, L. A. (2003). Sexual abuse, family violence, and female delinquency: Findings from a longitudinal study. Violence and Victims, 18(3), 319-334.
Hillberg, T., Hamilton-Giachritsis, C., & Dixon, L. (2011). Review of meta-analyses on the association between child sexual abuse and adult mental health difficulties: A systematic approach. Trauma, Violence, & Abuse, 12(1), 38-49.
Keiley, M. K. K., Howe, T. R., Dodge, K. A., Bates, J. E., & Pettit, G. S. (2001). The timing of child physical maltreatment: A cross- domain growth analysis of impact on adolescent externalizing and internalizing problems. Developmental Psychology, 13(4), 891-912.
Kessler, R. C., et al. (2010). Childhood adversities and adult psychopathology in the WHO World Mental Health Serveys. The British
Journal of Psychiatry, 197, 378-386.
Klonsky, E. D., & Moyer, A. (2008). Childhood sexual abuse and non-suicidal self-injury: Meta-analysis. The British Journal of
Krischer, M. K., & Sevecke K. (2008). Early traumatization and psychopathy in female and male juvenile offenders. International
Journal of Law and Psychiatry, 31, 253-262.
Lahlah, E., Lens, K. M. E., Bogaerts, S., & Knaap, L. M. van der (2013). Assessing the intersectionality of ethnicity, socio-economic status, parental connectedness, child abuse, and gender attitudes in juvenile violent delinquency. Child Abuse & Neglect, 37, 1034-1049.
Lansford, J. E., Miller-Johnson, S., Berlin, L. J., Dodge, K. A., Bates, J. E., & Pettit, G. S. (2007). Early physical abuse and later violent delinquency: A prospective longitudinal study. Child Maltreatment, 12(3), 233 – 245.
Lemmon, J. H. (1999). How child maltreatment affects dimensions of juvenile delinquency in a cohort of low-income urban youths.
Justice Quarterly, 16(2), 357-376.
Leschied, A. Chiodo, D., Nowicki, E., & Rodger, S. (2008). Childhood predictors of adult criminality: A meta-analysis drawn from the prospective longitudinal literature. Canadian Journal of Criminology and Criminal Justice, 50, 435–467.
doi:10.3138/cjccj.50.4.435
Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis. Thousand Oaks: Sage.
Loeber R., & Keenan, K. (1994). Interaction between conduct disorder and its comorbid conditions: Effects of age and gender.
Clinical Psychology Review, 14(6), 497-523.
Maxfield, M. G., & Widom, C. S. (1996). The cyle of violence. Revisited 6 years later. Archives of Pediatrics & Adolescent Medicine,
McClellan, J., McCurry, C., Ronnei, M., Adams, J., Storck, M., Eisner, A., & Smith, C. (1997). Relationship between sexual abuse, gender, and sexually inappropriate behaviors in seriously mentally ill youth. Journal of the American Academy of Child &
Adolescent Psychiatry, 36(7). 959-965.
McGrath, S. A., Nilsen, A. A., & Kerley, K. R. (2011). Sexual victimization in childhood and the propensity for juvenile delinquency and adult criminal behavior: A systematic review. Aggression and Violent Behavior, 16, 485-492.
McLaughlin, K. A., Green, J. G., Gruber, M. J., Sampson, N. A., Zaslavsky, A. M., & Kessler, R. C. (2010). Childhood adversities and adult psychiatric disorder in the national comorbidity survey replication II: Associations with persistence of DSM-IV disorders. Archives of General Psychiatry, 67(2), 124-132.
Mersky, J. P., & Reynolds, A. J. (2007). Child maltreatment and violent delinquency: Disentangling main effects and subgroup effects. Child Maltreatment, 12, 246 – 258.
Moffitt, T. E., Caspi, A., Rutter, M., & Silva, P. A. (2001). Sex differences in antisocial behaviour: Conduct disorder, delinquency,
and violence in the Dunedin Longitudinal Study. New York: Cambridge University Press.
Mullen, B. (1989). Advanced BASIC meta-analysis. Hillsdale: Erlbaum.
Piquero, N. L., & Sealock, M. D. (2000). Generalizing general strain theory: An examination of an offending population. Justice
Quarterly, 17, 449–484. doi:10.1080/07418820000094631
Raine, A. (2013) The anatomy of violence: The biological roots of crime. New York: Pantheon.
The handbook of research synthesis and meta-analysis (2nd ed., pp. 295–315). New York, NY: Russell Sage Foundation.
Rosenthal, R. (1979). The “file drawer problem” and tolerance for null results. Psychological Bulletin, 86(3), 638–641. Rosenthal, R. (1991). Meta-analytic procedures for social research (Vol. 86). Newbury Park: Sage.
Rothstein, H. R. (2008). Publication bias as a threat to the validity of meta-analytic results. Journal of Experimental Criminology, 4, 61–81.
Rutter, M., Giller, H., & Hagell, A. (1998). Antisocial behavior by young people: A major new review. New York: Cambridge University Press.
Salzinger, S., Rosario, M., & Feldman, R. S. (2007). Physical child abuse and adolescent violent delinquency: The mediating and moderating roles of personal relationships. Child Maltreatment, 12, 208 – 219.
Seto M. C., & Lalumière, M. L. (2010). What is so special about male adolescents sexual offending? A review and test of explanations through meta-analysis. Psychological Bulletin, 136(4), 526-575.
Siegel, J. A., & Williams, L. M. (2003). The relationship between child sexual abuse and female delinquency and crime: A prospective study. Journal of Research in Crime and Delinquency, 40, 71−94.
Smith, C., & Thornberry, T. P. (1995). The relationship between childhood maltreatment and adolescent involvement in delinquency.
Criminology, 33, 451−481.
Smith, C. A., Ireland, T. O., Thornberry, T. P., & Elwyn L.J. (2008). Childhood maltreatment and antisocial behavior: Comparison of self-reported and substantiated maltreatment. Journal of Orthopsychiatry, 78, 173–86.
