Adverse Childhood Experiences, Anger, and Violence Among Justice-Involved Youth: A Network Analysis

Abstract

Background:

Adverse childhood experiences (ACEs) are a complex construct that can be viewed as a network of interconnected adversities, with some interactions being substantially stronger than others. However, relatively little is known about the complex interactions between adversities in justice-involved youth (JIY). Moreover, even less is known about the pathways through which early adverse experiences contribute to violent offending.

Objective:

The present study aimed to identify the most influential adverse event(s) within a network structure of adversity. It also investigated one of the pathways through which ACEs contributed to violent offending.

Participants and Setting:

We used archival data consisting of 4,591 JIY in a large urban juvenile probation department in Texas, United States.

Methods:

We employed network analysis to investigate the network structure of ten types of adversity and how the complex interactions of adversity contributed to violent offending through the pathway of anger.

Results:

The network analysis results revealed that physical abuse and physical neglect were the most influential adversities within the ACEs network. Anger was positively and directly linked to all adversities except parental separation/divorce, controlling for other adversities in the network. Only emotional abuse and anger directly predicted the higher likelihood of violent offending within the one-year follow-up, accounting for the effects of other adversities in the model.

Conclusion:

Given the complexity of existing evidence-based interventions, combined with the limited resources in the juvenile justice system, tailoring these interventions specifically related to physical abuse and physical neglect could enhance feasibility without significantly compromising effectiveness for JIY. The deleterious impact of anger on violent offending highlights the need for interventions that address not only the broader system of environmental risks but also person-level criminogenic needs.

Adverse childhood experiences (ACEs) are potentially traumatic events that occur in childhood (0-17 years), encompassing child-caregiver maltreatment such as emotional abuse, physical abuse, and neglect, as well as household dysfunction, such as growing up in a household with mental health or substance use issues (Centers for Disease Control and Prevention, 2021). ACEs have negative impacts on almost all life outcomes (Kim & Royle, 2024). Despite their ubiquitous nature, ACEs manifest unevenly across different populations. Justice-involved youth (JIY) in the U.S. are exposed to ACEs at significantly higher rates than their peers in the general population (Belisle et al., 2024). Accumulating evidence indicates that ACEs are associated with a heightened risk of delinquency, including violent offending, among JIY (Astridge et al., 2023; Craig & Zettler, 2021; Fox et al., 2015; Malvaso et al., 2018; Yohros, 2023).

Research on the ACEs-delinquency link often employs a cumulative risk score approach, which sums up multiple dummy-coded adversities. Findings from this approach suggest that the greater the number of adverse experiences a JIY has experienced, the higher their likelihood of engaging in delinquency. However, this approach may oversimplify the complexity of ACEs. A growing number of studies utilizing the network approach reveal that not all adversities interact directly, with some interactions being substantially stronger than others (Breuer et al., 2020; de Vries et al., 2022; Pollmann et al., 2023). Consequently, certain adversities may have a direct effect on delinquency, while others influence delinquency indirectly. Despite these insights, relatively little is known about the complex interactions between adversities for JIY. Furthermore, even less is known about the pathways through which early adverse experiences influence violent offending.

In this study, we attempted to gain insights into the complex nature of ACEs and one of the potential pathways by which ACEs contribute to violent offending in a large sample of JIY. Specifically, we used the network approach to identify the most influential adversities within a network structure of ten different types. Targeting these central environmental risk factors could more efficiently disrupt the network, potentially leading to an overall reduction in the recurrence of ACEs, particularly child maltreatment. Additionally, including anger and violent offending in the network analysis allows for examining how various types of adversity contribute to violent offending through the pathway of anger.

The Complex Nature of ACEs

In delinquency literature, ACEs encompass ten types of negative exposure during childhood, including emotional abuse, physical abuse, sexual abuse, emotional neglect, physical neglect, violent treatment towards mother, household substance abuse, household mental illness, parental separation or divorce, and having an incarcerated household member (Baglivio et al., 2014). A cumulative risk score is calculated by summing the ten dummy-coded adversities, representing the total number of adverse experiences a JIY has experienced. The dose-response association between the number of ACEs and the risk of offending highlights the cumulative impact of different types of adversity in shaping delinquent behaviors, but it has significant limitations.

Deriving an overall ACE score by summing up the ten types of adversity assumes that all adverse experiences are equally important for delinquency, which is unlikely (Lacey & Minnis, 2020). An early study investigating the impact of individual adversities on the likelihood of becoming a serious, violent, and chronic juvenile offender (SVC) in a sample of 22,575 JIY showed that abuse (OR = 1.26–1.58), neglect (OR = 1.07–1.2), and family dysfunction (OR = 1.09–2.19) were associated with a higher risk, while sexual abuse (OR = 0.84) was associated with a lower risk of becoming an SVC (Fox et al., 2015). The findings of this study demonstrate varying impacts of individual adversities on violent offending, both in terms of magnitude and direction. Relatedly, the cumulative ACE score overlooks the specific patterning of individual adversities (Lacey & Minnis, 2020). A JIY experiencing parental separation and household mental illness would receive an ACE score of two, as another JIY experiencing physical and emotional abuse. This implies that both have the same risk of engaging in delinquency, which is unlikely. If a child maltreatment intervention were delivered to both JIY based on their ACE scores, its effectiveness on reducing the incidence of child maltreatment and the risk of reoffending would be concerning. Thus, the cumulative risk approach provides limited insights into the unique effects of individual adversities, which are crucial for designing targeted intervention programs that address the particular needs of JIY.

An alternative approach is to emphasize the unique effects of each type of adversity on developmental outcomes, including violent behaviors, referred to as the single adversity approach (Lacey & Minnis, 2020) or the specificity approach (Kim & Royle, 2024). However, this approach fails to take the co-occurrence of adversities into account. There is consistent and compelling evidence that adversities often co-occur in the general population (Dong et al., 2004) and in JIY (Baglivio & Epps, 2015). Considering the complex nature of ACEs, a network perspective has been suggested to circumvent the limitations of the cumulative risk approach and the single adversity approach, providing a deeper understanding of the interactions among various types of adversity (de Vries et al., 2022).

Understanding ACEs from a Network Perspective

The network approach to ACEs entails conceptualizing ACEs as a set of components and the relationships among those components. In network models, nodes represent components (variables), and edges represent relationships (conditional relationships/associations). The nodes represent distinct types of adversity (e.g., physical abuse, emotional neglect), connected by edges representing statistical relationships that can be estimated from data (e.g., partial correlation coefficients conditional on other adversities). ACEs can be characterized in terms of the interactions between discrete types of adversity in a network, reflecting the high level of co-occurrence of different childhood adversities (Baglivio & Epps, 2015; Dong et al., 2004).

Building upon the network perspective, Borsboom (2017) proposed the network theory of psychopathology. This theory posits that symptoms of psychopathology are causally interconnected, such that the activation of one symptom increases the likelihood of other connected symptoms emerging. Although this theory was developed to understand psychopathology, its major assumptions are also relevant for uncovering the complex nature of ACEs. Similar to psychopathology, ACEs can be contextualized as active environmental experiences that can affect one another and thereby contribute to the persistent activation of the overall network. The complex interactions between individual adversities involve both the quantity and magnitude of associations; not all adversities directly interact, and some interactions are significantly stronger than others. For example, a recent study examining the network structure of 14 adverse experiences for a sample consisting of 1,029 Dutch adolescents showed that emotional abuse and physical abuse exhibited the strongest association than any other associations included in the network, and parental illness and parental addiction were only indirectly linked to emotional and physical abuse through financial difficulties and family conflicts (de Vries et al., 2022).

Indirect interactions between adversities lead to varying levels of importance within the network, with some adversities playing a more central role than others. To that end, one major application of network analysis is to identify highly central nods (Costantini et al., 2015). High-centrality nodes have strong connections to numerous other nodes, serving as hubs that link other unconnected nodes. In contrast, low-centrality nodes are situated on the network’s periphery, with fewer and weaker connections to other nodes. When a highly central adversity is activated, it will influence other adversities to become activated as well, maintaining the ACEs network. A study applying network analysis to a sample of 384 psychiatric inpatients to evaluate the network structure of 10 types of adversity found that emotional neglect, physical neglect, and violence against the mother were the most influential adversities (Breuer et al., 2020). Addressing these adversities is likely to destabilize the entire network, suggesting that these forms of neglect and violence could be targets for developing effective and efficient interventions.

Although the network approach to ACEs shows promise in identifying effective and efficient intervention targets, there is limited evidence on how the commonly studied adversities interact with each other for JIY. Furthermore, even less is known about how the complex interactions among adversities contribute to delinquency, including violent offending, and the mechanisms underlying the ACEs-delinquency link.

ACEs, Anger, and Violence

There are a variety of mechanisms underlying the ACEs-delinquency link. Among them, anger is one of the most frequently discussed mechanisms across different theoretical rationales (Agnew, 1992; Cicchetti & Toth, 2005; Ford et al., 2012), yet it remains less tested in JIY. Various forms of adversity share a common characteristic: they represent a deviation from the expected environment that demand substantial psychological, social, and neurological adaptation from a typical child (McLaughlin, 2016). In general strain theory (GST; Agnew, 1992), such deviations—where individuals are not treated as they would like—are referred to as strains or stressors. Because noxious stimuli such as abuse and neglect may be perceived as unjust or as threats to a child’s goals, values, needs, activities, and/or identities, such strains increase the likelihood of negative emotions, especially anger and frustration. These negative emotions generate pressure to take corrective action, with crime being one possible response.

Though the strain-anger-delinquency link has received a significant amount of empirical support, tests of GST have focused overwhelmingly on the general population (Brezina, 1998; Hay & Evans, 2006; Maschi et al., 2008; Sigfusdottir et al., 2004). Kimonis et al. (2010) found that anger mediated the association between violence exposure and violence perpetration, but the sample is constrained to incarcerated boys. Thus, it remains unclear whether and how anger connects ACEs and delinquency for JIY in general. Additionally, there is evidence that anger is an intervening mechanism between adversities and violent offending rather than with non-violent offending (Aseltine et al., 2000). Indeed, anger is conceived as a casual factor of aggression such as violent offending (Anderson & Bushman, 2002). Thus, the present study focused on anger and violent offending.

The Current Study

In the present study, we applied network analysis to a large archival data from the Harris County Juvenile Probation Department (HCJPD), Houston, Texas, to identify the central adversities within a network structure of ten different types commonly studied in criminology literature. Additionally, we aimed to investigate the extent to which the ten types of adversity contributed to violent offending directly and indirectly through anger, providing insights into a potential pathway through which ACEs contribute to violent offending.

Method

Sample

HCJPD maintains a centralized database containing records of all alleged offenses involving youth under its jurisdiction. Currently, it holds information on 91,894 JIY who had at least one contact with HCJPD between January 1, 2006, and June 30, 2023, within Harris County. Since September 2017, HCJPD has implemented the Positive Achievement Change Tool (PACT), which encompasses measures of ACEs (ten types of childhood adversity) and anger. As we used only the youth who completed the full PACT, the sample size was narrowed down to 6,059 JIY (see Assessments for details). Further refinement was made by limiting the sample to JIY under 17 years old at the time of completing a stay or commitment at a residential facility, allowing for a one-year follow-up period for violent offending, resulting in a final sample size of 4,591.

To characterize the analytical sample, we investigated the disparities between JIY included and excluded from the analysis after September 2017. In the analytical sample, there was a greater proportion of male JIY (81.2%) compared to those excluded (68.7%), χ²(1) = 240.95, p < .001. Furthermore, racial and ethnic distinctions were noted between the included and excluded samples, χ²(4) = 142.34, p < .001. Specifically, a higher percentage of Black and Hispanic JIY were present in the included sample compared to the excluded sample (46.9% vs. 41.3% and 44.1% vs. 42.4%, respectively), while the proportion of White JIY was lower in the included sample (8.1% vs. 15.1%). Both the included and excluded samples had comparable proportions of Asian individuals (0.7% vs. 0.9%) and JIY with incomplete race/ethnicity data (0.2% vs. 0.3%). The proportion of committing violent offending was higher in the included sample (54.9%) than the excluded sample (39.3%), χ²(1) = 311.53 p < .001.

In general, the analytical sample differed from the excluded sample, aligning with the fact that the PACT was administered to JIY after adjudication. This implies that male youth of color with a higher risk of committing violent acts were more likely to be detained and thus included in the analytical sample. Ethical and administrative approvals were obtained from the University of Houston Committee for the Protection of Human Subjects and the jurisdiction officials. The data are available on request from the authors.

Assessments

ACEs

The PACT comprises a pre-screen and a full assessment. Measures pertaining to the ten types of ACEs and anger are included in the full assessment. The present study utilized the first ever full PACT assessment for each JIY to extract ACEs. This allows appropriate temporal order for testing the impact of ACEs and anger on violent offending. Table 1 presents the ten types of childhood adversity and their corresponding environmental experiences derived from the full PACT. Consistent with existing research focusing on ACEs among JIY (Baglivio et al., 2014; Baglivio & Epps, 2015; Fox et al., 2015), we dichotomized each form of adversity to signify the absence (0) and presence (1) of adversity. For example, if a youth has experienced verbal intimidation, yelling, heated arguments, and/or threats of physical abuse, it indicates the presence of emotional abuse; otherwise, it suggests the absence of emotional abuse. Figure S1 in the Supplementary Material shows the frequency distribution for each type of adversity before dichotomizing.

Table 1. Measures of Childhood Adverse Experience.

Adversity	Environmental experience
Emotional abuse	Verbal intimidation, yelling, heated arguments Threats of physical abuse
Emotional neglect	Inconsistently willing to support youth Little/no willingness to support youth Hostile, berating, and/or belittling of youth
Physical abuse	Victim of violence or physical abuse at home Victim of violence or physical abuse in a foster or group home Victimized by family member Victimized by someone outside the family Attached with a weapon
Physical neglect	Victim of neglect
Sexual abuse	Sexually abused/raped by family member Sexually abused/raped by someone outside the family
Witnessing family violence	Has witnessed violence at home Has witnessed violence in a foster/group home Has witnessed violence in the community Family member killed as a result of violence
Parental substance abuse	Parental alcohol problem history Parental drug problem history
Parental mental illness	Parental mental health problem history Parental drug problem history
Parental separation/divorce	The youth is not living with biological mother and biological father
Incarcerated parents	Mother/female caretaker had a history of jail/imprisonment Father/male caretaker had a history of jail/imprisonment

Anger

History of anger/irritability distinguishes between having no history of and/or occasional feelings of anger/irritability (= 0) and those with a history of consistent feelings and/or aggressive reactions to feelings of anger/irritability (= 1).

Violent Offending

Violent offending included all crimes against persons, including homicide, assault, violent sexual offenses, and other delinquency against persons. Youths younger than 17 years old were tracked for one year from the date they completed a stay or commitment at a residential facility, according to the court disposition. Adult records and offenses taking place outside the state of Texas were not tracked by the jurisdiction and thus were not included in the study.

Analytical Plan

We utilized an Ising model (van Borkulo et al., 2014) to estimate the network structure, which characterizes relationships among binary variables through pairwise log-linear associations, akin to partial correlation coefficients within a Gaussian Graphical model (Epskamp, Waldorp, et al., 2018). To mitigate potential spurious associations, we employed the eLasso penalty guided by the Extended Bayesian Information Criterion (EBIC; Ravikumar et al., 2010). We adopted the default parameter values provided in the package, with the EBIC hypertuning parameter established at 0.25. All analyses were conducted using R-version 4.3.3 on February 29, 2024. For network estimation, we used the estimateNetwork function in the bootnet package (Epskamp, Borsboom, et al., 2018) using the IsingFit package (van Borkulo et al., 2014; version 0.4). Networks were visualized using the qgraph package (Epskamp et al., 2012; version 1.9.8). Due to the relatively small number of nodes, no edges were hidden by setting the cut command of the qgraph function to 0. The network nodes were displaced according to the Fruchterman-Reingold algorithm (Fruchterman & Reingold, 1991), whereby nodes that are stronger and/or more connected are positioned in the center of the network, whereas less strong and/or less connected nodes are positioned in the periphery.

To further quantify the relative importance of a node in the network structure, we examined strength, betweenness, and closeness as centrality measures (Costantini et al., 2015). A node’s strength represents the sum of all absolute edge weights connected directly to a node. A node’s closeness denotes the average shortest path between a specific node and all other nodes in the network. A node’s betweenness quantifies the number with which a given node resides on the shortest path between any two other nodes. Together, these indices quantify the position of a node within the network.

To assess the accuracy of the edge weight estimates, we used nonparametric bootstrapping based on 1,000 bootstrap samples implemented in the bootnet package (Epskamp, Borsboom, et al., 2018; version 1.6). To assess the stability of the edge strength estimates, we used a case-dropping subtest bootstrap based on 1,000 bootstrap samples embedded in the bootnet package (Epskamp, Borsboom, et al., 2018; version 1.6). In addition, the correlation stability coefficient (CS) was also reported. As a rule of thumb, a value of 0.25 indicates that the network is stable, but a value above 0.50 is desirable (Epskamp, Borsboom, et al., 2018).

Results

Descriptive Statistics

Descriptive statistics for the ten types of adversity are presented in Table 2. The top three most prevalent adversities were parental separation/divorce (84.9%), witnessing violence (56.9%), and parental incarceration (39.3%). The least commonly reported adversities were parental mental illness (3.7%), parental substance abuse (8.5%), and sexual abuse (9.1%). The vast majority of JIY (95.1%) have experienced at least one type of adversity. Regarding anger, approximately one-third of the sample had a history of anger/irritability (33.9%). Regarding delinquency, 611 (13.3%) JIY were rearrested for violent offending during the one-year follow-up period. Among JIY who experienced at least one type of adversity, 13.6% were arrested for violent offending.

Table 2. Frequency for Adverse Experiences.

Studied variables	n (%)
Emotional abuse (EMA)	867 (18.9%)
Emotional neglect (EMN)	759 (16.5%)
Physical abuse (PHA)	1,035 (22.5%)
Physical neglect (PHN)	662 (14.4%)
Sexual abuse (SEX)	417 (9.1%)
Witnessing violence (WIV)	2,614 (56.9%)
Parental substance abuse (SUB)	389 (8.5%)
Parental mental illness (MEN)	168 (3.7%)
Parental separation/divorce (DIV)	3,897 (84.9%)
Parental incarceration (INC)	1,804 (39.3%)

The ACE Network Structure

Figure 1 depicts the undirected network structure, where the nodes represent the ten types of adversity. The edges represent partial associations (i.e., controlling for all other variables in the network) between these nodes. Edge weight estimates are provided in Table S1 in the Supplementary Material. The network was relatively dense (average weight = .382), with 34 nonzero edges over 45 possible edges. This suggests that not all adversities directly interact. The thickness of each edge represents the magnitude of association between any two nodes. All edges exhibited the same color, indicating that the ten types of adversity were positively related. Figure S2 in the Supplementary Material shows confidence intervals of edges estimated through non-parametric bootstrapping. In general, the confidence intervals were very narrow, indicating precise results.

Figure 1. Undirected Network Model for Childhood Adversity.

Note. Each indicator is represented as a node (circle) in the network and edges between indicators represent conditional dependencies. The thickness of each edge indicates the strength of the association between nodes. EMA = emotional abuse; EMN = emotional neglect; PHA = physical abuse; PHN = physical neglect; SEX = sexual abuse; WIV = witnessing violence; SUB = parental substance abuse; MEN = parental mental illness; DIV = parental separation/divorce; and INC = parental incarceration.

Figure S3 shows the stability of the centrality indices after observing only portions of data. The stability of the three metrics dropped gradually, whereas the stability of node strength was better. The CS-coefficient [CS(cor = 0.7)] was 0.439, 0.361, and 0.672 for betweenness, closeness, and strength, respectively, exceeding the lower bound of the recommended value (Epskamp, Borsboom, et al., 2018). Hence, the three-centrality metrics are stable and trustworthy.

Figure 2 shows the centrality metrics for strength, closeness, and betweenness. Physical abuse had the highest strength, followed by physical neglect. Physical neglect had the highest betweenness and closeness. As shown in Figure 1, physical abuse exhibited the highest strength in the network, because it was interconnected with all types of adversity and showed particularly strong connections with physical neglect, sexual abuse, and witnessing violence. Change in physical abuse would thus strongly affect the remaining nine types of adversity. Physical neglect exhibited the highest betweenness and closeness because it was closely connected to all nine other adverse experiences. Therefore, any changes in physical neglect would affect large parts of the network. In contrast, the four adversities associated with household dysfunction exhibited relatively lower centrality metrics and were located peripherally in the network.

Figure 2. Centrality Indices for the Network Structure Involving Childhood Adversity Alone.

Note. Centrality indices are shown as raw scores. The y-axis represents the nodes while the x-axis represents the raw scores of centrality metrics. EMA = emotional abuse; EMN = emotional neglect; PHA = physical abuse; PHN = physical neglect; SEX = sexual abuse; WIV = witnessing violence; SUB = parental substance abuse; MEN = parental mental illness; DIV = parental separation/divorce; and INC = parental incarceration.

Relationships with Anger and Violent Offending

We integrated anger and violent offending into the ACE network to investigate how discrete adversity is conditionally linked to them. Edge weight estimates are provided in Table S2 in the Supplementary Material. Figure S4 in the Supplementary Material shows confidence intervals of edges estimated through non-parametric bootstrapping. In general, the confidence intervals were very narrow, indicating precise results. Comparing Figure 1 with Figure 3, the overall network structure of the ten types of adversity remained largely unchanged, suggesting stability of the network structure.

Figure 3. Undirected Network Model for Childhood Adversity, Anger, and Violent Offending.

Note. Each indicator is represented as a node (circle) in the network and edges between indicators represent conditional dependencies. The thickness of each edge indicates the strength of the association between nodes. EMA = emotional abuse; EMN = emotional neglect; PHA = physical abuse; PHN = physical neglect; SEX = sexual abuse; WIV = witnessing violence; SUB = parental substance abuse; MEN = parental mental illness; DIV = parental separation/divorce; INC = parental incarceration; ANG = history of anger/irritability; and VIO = violent offending.

As shown in Figure 3, anger was directly connected to all adversities except parental separation/divorce. Anger exhibited stronger associations with emotional abuse, physical abuse, and physical neglect, than with parental substance abuse, parental mental illness, parental separation/divorce, and parental incarceration. Only emotional abuse and anger positively predicted violent offending one year later, conditional on other variables in the network model. This suggests that the other nine types of adversity contribute to violent offending indirectly through emotional abuse and anger.

Discussion

The present study aimed to identify the most influential adverse experience(s) within a network structure of ten different types of adversity and investigate one of the potential pathways through which ACEs contributed to violent offending among JIY. The results of the network analysis revealed that physical abuse and physical neglect were the most influential adversities within the ACEs network. Additionally, emotional abuse and anger emerged as key factors in contributing to the increased risk of violent reoffending during the one-year follow-up period among JIY.

Adversities Are Not Equally Important

The single adversity approach and the cumulative risk approach are the two primary approaches to conceptualizing ACEs (Kim & Royle, 2024; Lacey & Minnis, 2020). The single adversity approach highlights the unique effects of each adversity on developmental outcomes but overlooks the co-occurrence of ACEs. In contrast, the cumulative risk approach considers the co-occurrence of ACEs but unrealistically assumes that all adversities contribute equally to the overall cumulative risk and developmental outcomes. The findings of the present study advance the understanding of the complexity of ACEs by showing that not all adversities directly interact, and some adversities play a more central role in the network of ACEs than others.

A quarter of the edges in the ACE network were zero, indicating that 11 pairs of adversities were no longer related controlling for other adversities in the network. Our findings offer new insights into the complexity of ACEs. In the seminal work, Dong et al. (2004) showed that all ten categories of ACEs were associated with each other in the general population. Similar findings have also been found among JIY (Baglivio & Epps, 2015). This body of research suggests that experiencing one adversity increases the likelihood of exposure to any other type. However, our findings suggest that this conclusion may be misleading. For instance, we found that emotional neglect was not directly related to parental mental illness or parental separation/divorce conditional on other adversities. A more accurate summary, therefore, is that experiencing one adversity increases the likelihood of exposure to certain other types. This rephrasing could facilitate covariate selection for researchers using the single adversity approach to conceptualize ACEs. For instance, it is not necessary to control for parental mental illness and parental separation when the associations between emotional neglect and developmental outcomes are of interest.

Despite substantial variations in the studied populations and the adversities included in network structures, existing research has consistently shown that not all adversities directly interact with one another, and some interactions are significantly stronger than others (Breuer et al., 2020; de Vries et al., 2022; Pollmann et al., 2023). Our findings align with the existing results. Collectively, these findings suggest that some adversities play a more central role in maintaining the ACE network. Our findings showed that physical abuse exhibited stronger associations with physical neglect, witnessing family violence, and sexual abuse than with other adversities. Physical neglect was located centrally in the network and were highly interconnected with all types of adversity. These connection patterns are further supported by centrality measures: physical abuse held the highest strength, while physical neglect showed the highest betweenness and closeness. Physical neglect connects to more adversities, while physical abuse has stronger connections within the network, suggesting they may exert greater influence on the overall network activation. As a result, they are particularly important intervention targets.

Some caveats should be considered when interpreting the central roles that physical abuse and physical neglect play in the network. All edges in the network structure are undirected, meaning the network does not indicate the direction of relationships. Nonetheless, substantive theories suggest that the five types of child maltreatment (physical abuse, physical neglect, emotional abuse, emotional neglect, and sexual abuse) and witnessing family violence are likely to interconnect, whereas the four types of household dysfunction (parental substance use, parental mental illness, parental separation/divorce, and parental incarceration) are likely to serve as antecedents to child maltreatment (Hughes et al., 2017). Thus, the high centrality of physical abuse and physical neglect means that emotional abuse, emotional neglect, sexual abuse, and witnessing family violence are more likely to arise when physical abuse and physical neglect are activated. Additionally, household dysfunction is likely to impact developmental outcomes through physical abuse and physical neglect. Our findings carry important implications.

Given the high prevalence rates and developmental consequences of experiencing childhood adversities, a growing number of interventions focused on reducing child maltreatment have been developed and evaluated. Evidence-based interventions demonstrated to effectively reduce maltreatment include the Parent–Child Interaction Therapy (PCIT; Chaffin et al., 2004), the Triple P-Positive Parenting Program (Triple-P; Sanders et al., 2014), and Multisystemic Therapy for Child Abuse and Neglect (MST-CAN; Swenson et al., 2010). The effectiveness of PCIT and Triple-P is supported by meta-analytical evidence (Sanders et al., 2014; Zhang et al., 2024), while MST-CAN is supported by findings from randomized controlled trials (Swenson et al., 2010).

Despite their significant potential, none of these interventions have been tested for JIY. Therefore, it is crucial to systematically evaluate and tailor these interventions to meet the needs of JIY. When adapting these interventions, it is crucial to recognize that their effectiveness comes at the cost of simplicity. The parenting-focused interventions—PCIT, Triple-P, and MST-CAN—primarily aim to improve overall parental knowledge, skills, and competency. As a result, these interventions commonly consist of multiple modules or components, each of which requires several sessions, and involve the participation of both parents and their children. Given the comprehensiveness and complexity of these programs, combined with the resource constraints in the juvenile justice system, implementing any of these interventions for JIY without adaptation is nearly impossible. An optimal intervention for JIY should balance effectiveness with efficiency.

Our findings suggest that interventions focusing on physical abuse and physical neglect may have the potential to reduce the incidence of emotional abuse, emotional neglect, sexual abuse, and witnessing family violence, as well as mitigate the adverse effects of household dysfunction on developmental outcomes. Therefore, rather than addressing all aspects of maltreatment, retaining and tailoring components of these interventions specifically related to physical abuse and physical neglect could enhance feasibility without significantly compromising effectiveness for JIY.

Anger As a Potential Pathway Linking ACEs with Violent Offending

The overarching goal of disrupting the ACE network is to lessen the negative impacts of adversity on developmental outcomes, ultimately promoting child well-being. Therefore, a better understanding of the relationships between ACEs and developmental outcomes is equally important as identifying efficient ways to disrupt the network. Our findings showed that anger was directly connected to all adversities except parental separation/divorce controlling for other variables in the network. The findings of the current study align with GST (Agnew, 1992), which predicts that when individuals are not treated as what they would like, negative emotions, particularly anger and frustration, would arise. For instance, experiences of abuse and neglect may threaten a child’s goals, values, and needs, leading to feelings of anger and frustration. In fact, research on developmental psychopathology reveals that exposure to chronic and/or intense strains/stressors at developmental vulnerable times such as early childhood or adolescence compromise the brain and autonomic nervous system, that are likely to increase anger and impulsivity while reduce the youth’s ability to inhibit these reactions (Ford et al., 2012). Thus, it is unsurprising that anger is a common consequence of various types of adversity.

Only emotional abuse and anger were found to positively predict the likelihood of violent offending within the one-year follow-up, even after accounting for the other adverse experiences in the model. Our findings align with studies examining the strain-anger-violence pathway in both the general population (Aseltine et al., 2000; Maschi et al., 2008) and in male JIY (Kimonis et al., 2010). The findings of this study confirm the generalizability of GST to a broader population, and more importantly, offer value practical implications.

The network structure involving ten types of adversity, anger, and violent offending indicates that the remaining nine adversities indirectly influence violent offending through emotional abuse and anger. These findings highlight the importance of targeting both environmental risk factors (e.g., ACEs) and individual risk factors (e.g., anger) in interventions. As upstream risk factors, ACEs create conditions that increase the likelihood of downstream consequences such as emotional, behavioral, and cognitive dysregulation, all of which are associated with a heightened risk of violent offending. This suggests that ACEs contribute to violent offending both directly and indirectly through individual dysregulation. Addressing the recurrence of maltreatment can mitigate its direct impact, but it is equally important to address the resulting downstream issues to fully reduce the risk of violent offending. Therefore, in addition to interventions focused on improving overall parenting skills, interventions that target individual regulation should also be considered.

Trauma-Focused Cognitive Behavioral Therapy (TF-CBT) has been shown to effectively support children and adolescents who have experienced various forms of abuse and trauma—including physical abuse, emotional abuse, sexual abuse, and exposure to intimate partner violence—in processing traumatic memories, dealing with maladaptive thought patterns and behaviors, and developing coping and interpersonal skills (Wang et al., 2022). Moreover, meta-analysis found that CBT was associated with reduced likelihood of recidivism (Pappas & Dent, 2023). Notably, CBT interventions that included anger control were more effective than those without anger control components (Landenberger & Lipsey, 2005). Thus, an integrated combination of parenting intervention with TF-CBT would ideally reduce the risk of violent offending among JIY while also improving their overall well-being.

Limitations and Future Directions

Some limitations of this study are important to keep in mind. The sample is not representative despite its relatively large size (N = 4,591). Similar to most existing studies on ACEs in JIY, archival data from risk assessment instruments like PACT served as the primary source for extracting measures of ACEs. However, because these measures are only included in comprehensive assessments designed for JIY, who typically commit more severe delinquency, our sample is not representative of the broader JIY population. However, the unrepresentative sample does not necessarily weaken our findings, as the implications of the current study are drawn by considering it in conjunction with findings from previous research. Future research can test the generalizability of our findings by investigating gender and racial/ethnic differences in the network structure.

Moreover, although utilizing dichotomized measures of ACEs is the common practice in studying ACEs, it fails to take the intensity and chronicity of exposure into account (Anda et al., 2020). In fact, screening and assessing ACEs is not routinely conducted in the juvenile justice system. Most studies examining ACEs have extracted data on adverse experiences from violence risk assessments, such as the PACT. Given the harmful impacts of ACEs, there is a growing call to consider incorporating ACEs screening into juvenile justice systems (Folk et al., 2021). Berman et al. (2022) provided a list of measures allowing for the capture of the frequency and severity of adversities.

Lastly, while the range of ACEs examined in the current study aligns with that of most studies on ACEs for JIY, this scope may be either overly broad or excessively narrow. In fact, the ten types of ACEs have been widely examined since its adoption from Felitti et al. (1998) by Baglivio et al. (2014). However, there was no rationale for why these adversities were examined (Lacey & Minnis, 2020). The number of ACEs included in the network model may have a significant influence on the network structure. For instance, including additional types of adversity (nodes) may lead to the removal of certain edges from the network. Conversely, removing some adversities from the ten types of adversity may result in the emergence of new edges. Therefore, the selection of ACEs for future research should be guided by reasonable rationale.

Conclusion

Both the single adversity approach and the cumulative risk approach provide valuable insights into the dose-response relationship between adversity and various negative outcomes, as well as the ability to identify potential risks for these outcomes. Nevertheless, ACEs are a complex construct. Viewing ACEs as a network of risks that influence each other offers new insights into how to design effective and efficient interventions to prevent the recurrence of adverse events. We showed that physical abuse and physical neglect were the most influential adversities within the network of ACEs. Given the complexity of existing evidence-based interventions, combined with the limited resources in the juvenile justice system, our findings suggest that tailoring these interventions specifically related to physical abuse and physical neglect could enhance feasibility without significantly compromising effectiveness for JIY. Moreover, anger played a key role in the deleterious impacts of ACEs on violent offending among JIY, highlights the need for interventions that address not only the broader system of environmental risks but also person-level criminogenic needs.