Abstract
The aim of this study was to empirically examine the theoretical model proposed by Grady, Levenson, and Bolder used to explain the relationships between experiencing trauma, attachment styles, and risk factors associated with sexual offending. The specific risk factors tested were emotional, behavioral, and cognitive regulation deficits. The research questions were as follows: (1) What specific childhood traumatic experiences (physical and sexual abuse and/or other domestic trauma events) are associated with anxious-ambivalent and anxious-avoidant attachments? (2) Are anxious-ambivalent and anxious-avoidant attachment styles differentially linked to dysregulation outcomes (emotional, behavioral and cognitive shift, and inhibitions)? (3) Do insecure attachment styles explain why childhood trauma is associated with dysregulation? and (4) Do these early life experiences contribute to sexual offending behavior? The sample included 200 male youth adjudicated for either a sexual or nonsexual crime and living in the community or a residential facility. The average age of the youth was 17.17 years (SD = 1.81 years). Structural equation modeling was used to determine the direct and indirect relationships between abuse and traumatic experiences, anxious-avoidant and ambivalent-anxious attachment styles, and regulation deficits. Authors found a direct relationship between physical abuse and both of the attachment styles, separately. Both anxious-ambivalent and anxious-avoidant attachment styles related similarly to regulation deficits. Separately, anxious-ambivalent and anxious-avoidant attachment styles served as mediators between physical abuse and regulation deficits. Sexual abuse predicted the later commission of sexual crimes with no influence from attachment style. Finally, domestic trauma independently predicted regulation deficits in the model that included anxious-ambivalent attachment but had no effect on deficits in the model that included anxious-avoidant attachment. Implications for treatment include the need to consider using attachment-based interventions and prevention strategies, and a trauma-informed approach when working with justice involved youth. Suggestions for future research are also discussed.
For decades, practitioners and researchers have been seeking the answer to this question—why do people commit sexual offenses? However, the answer is elusive and complex, making it a challenge for practitioners and researchers to develop effective prevention and intervention strategies aimed at reducing rates of sexual violence. Where does one begin to intervene? What factors have the strongest influence over offense cycles? Assuming the answer lies in a combination of several factors, which ones and under what circumstances? Many etiological investigations of sexual offending have been grounded in theoretical assumptions that are used to explain a variety of maladaptive behaviors, both sexual and nonsexual. However, few researchers have empirically tested a proposed theory of sexual offending. The current study tested the theoretical model proposed by Grady, Levenson, and Bolder (2016) linking adverse childhood experiences (ACEs) to the development of insecure attachments that lead to emotional or behavioral regulation problems that, in turn, contribute to sexual offending behaviors. Specifically, this study empirically examined the relationships between physical abuse, sexual abuse, other experiences of domestic trauma, insecure attachment styles, different forms of dysregulation, and sexual offending with incarcerated youth.
Literature Review
In an attempt to answer the question of why people commit sexual offenses, a myriad of theories have been proposed, including biological (Barbaree & Marshall, 1991; Berlin, 1983), cognitive (Egan, Kavanagh, & Blair, 2005; Ward, Fon, Hudson, & McCormack, 1998), neurobiological (Mitchell & Beech, 2011), social learning (Burton, Miller, & Shill, 2002), trauma (Abbiati et al., 2014; Reavis, Looman, Franco, & Rojas, 2013), attachment (Beech & Mitchell, 2005; Miner, Romine, Robinson, Berg, & Knight, 2014), and cultural (Hetherton, 1999). In an attempt to synthesize some of these potentially disparate ideas, authors have created integrated theories of sexual offending that incorporate various perspectives and related research. Marshall and Barbaree (1990) proposed one of the first integrated theories that included learning experiences, sociocultural factors, and biological processes, as well as attachment and childhood experiences, such as trauma. Another theoretical framework was proposed by Ward and Beech (2006), called the Integrated Theory of Sexual Offending (ITSO), included “biological factors (influenced by genetic inheritance and brain development), ecological niche factors, i.e., social, cultural, and personal circumstances, and neuropsychological factors” (p. 50). A major strength of these theories is their attempt to amalgamate different empirical research studies to understand the complexities of sexual offending. However, although various components of these theories were derived from empirical literature, many integrated theories themselves have yet to be empirically tested.
Theoretical Framework
Grady and colleagues (2016) proposed an integrative theory of sexual offending that links childhood trauma to the formation of insecure attachments that, in turn, contribute to risk factors associated with sexual offending. The results from an empirical examination of the first portion of this framework will first be presented; the current paper will test the second portion; and a third paper will test the final portion of the model. Given that there is overlap in the etiological literature relative to developmental antecedents to sexual violence between adult and youth, the theory proposed by Grady et al. (2016) is nonspecific about population age, and the current study tests the theory in a sample of youth. The following section will briefly review salient components of the theory as related to adolescents who have committed sexual harm. For a full review of the theoretical and empirical literature that informs the theory, readers are encouraged to refer to Grady et al. (2016).
Childhood Trauma
The relationship between childhood trauma and subsequent offending behavior has long been examined in the literature; childhood trauma has been linked with high levels of sexual offending (Anda, Butchart, Felitti, & Brown, 2010; Kenny & Wurtele, 2012; Levenson & Socia, 2015; Levenson, Willis, & Prescott, 2015; Wurtele, Simons, & Moreno, 2014) and other nonsexual antisocial behaviors (Cicchetti & Banny, 2014; DeKlyen & Greenberg, 2008; Ford, Chapman, Connor, & Cruise, 2002; Rutter, Kim-Cohen, & Maughan, 2006). Youth who commit sexual crimes have significantly higher rates of childhood trauma than their nonjustice-involved peers (Abram et al., 2004; Dierkhising et al., 2013; Levenson et al., 2017). Indeed, most theoretical models that aim to understand factors that influence criminal behaviors among youth and adults include trauma as a developmental antecedent (Kenny & Wurtele, 2012; Levenson & Socia, 2015; Levenson et al., 2015; Simons, Wurtele, & Durham, 2008; Ward & Beech, 2008; Wurtele et al., 2014).
The traumatic experiences reported include histories of physical and sexual abuse (A. Brown & Burton, 2010), domestic traumas such as exposure to domestic violence (Davis & Leitenberg, 1987; Fehrenbach, Smith, Monastersky, & Deisher, 1986), and family criminality (Wanklyn, Ward, Cormier, Day, & Newman, 2012). Having a parent with a mental health issue (Ryan, 2010), substance use disorder (Graves, Openshaw, Ascione, & Ericksen, 1996; Zakireh, Ronis, & Knight, 2008), or a history incarceration (Levenson et al., 2017; Zakireh et al., 2008) is prevalent among youth who commit sexual crimes. Even when compared with other incarcerated youth, those who committed sexual crimes had higher rates of sexual and physical abuse than those who committed nonsexual crimes (Levenson et al., 2017). Furthermore, youth who commit sexual crimes with sexual victimization histories have higher rates of other forms of victimization experiences, including physical and emotional abuse, relative to nonsexually victimized sexual and nonsexual offenders (Yoder, Dillard, & Leibowitz, 2017).
Childhood trauma influences biological, social, cognitive, health, behavioral, attachment, and mental health outcomes, many of which are associated with criminal behaviors (Grady et al., 2016; Levenson, 2015; Simons et al., 2008; Wurtele et al., 2014). Traumatic experiences during critical developmental periods alter the structure of the brain found to affect the production of stress-related hormones associated with the fight-flight-freeze responses (Anda et al., 2010; Anda et al., 2006; Trickett, Noll, & Putnam, 2011) that lead to challenges with emotional regulation, cognitive processes, and difficulties in social relationships (Anda et al., 2010; Anda et al., 2006). These neurobiological changes might explain why researchers have found a direct link between childhood trauma exposure and later deficits in self-regulation (DePrince, Weinzierl, & Combs, 2009; Samuelson, Krueger, Burnett, & Wilson, 2010) and that some children diagnosed with posttraumatic stress disorder (PTSD) have significant prefrontal lobe damage with no other known antecedents (De Bellis et al., 2002).
Children with trauma histories may struggle with interpersonal relationships (Baer & Martinez, 2006; Morton & Browne, 1998) that may manifest as insecure attachments. Trauma exposure can lead to children keeping secrets, distorting reality, and learning not to trust others (Elliott, Bjelajac, Fallot, Markoff, & Reed, 2005; Teyber & McClure, 2011). Specifically, families of youth sexual offenders, relative to nonsexual offenders, have been found to harbor more secrets, engage in taboo behaviors, and tell more lies (Baker, Tabacoff, Tornusciolo, & Eisenstadt, 2003; Yoder, Hodge, & Ruch, 2019). Compared with youth who commit nonsexual crimes, adolescents who commit sexual crimes have greater difficulties with social isolation, which partially accounts for anxious attachment styles (Miner et al., 2014). In fact, adolescents who commit sexual offenses tend to feel greater isolation and struggle with social and emotional reciprocity, which also is associated with anxious attachment styles (Miner & Munns, 2005; Miner et al., 2010). Early childhood abuse experiences have even been identified as a precursor to insecure attachment (Yoder, Grady, & Dillard, in press). Therefore, research is beginning to reveal how abuse experiences and other traumatic experiences can form the basis for insecure attachments with caregivers and others, leading to a host of additional challenges.
Insecure Attachments
Attachment theory began with Bowlby’s (1969/1982) work with those he referred to as juvenile thieves in 1944. His work with this population propelled his interest in how early relationships influence developmental trajectories (Bowlby, 1944; Coates, 2004). Attachment theory posits that these early experiences with caregivers form internal models of relationship patterns that are repeated in subsequent interpersonal relationships (Benoit, 2004; Bowlby, 1969/1982). These patterns are based on the level of responsiveness that the caregivers provide. Those children who experienced their caregivers as responsive and attuned to their needs form attachment patterns called secure, which are considered the optimal attachment style for emotional, psychological, and interpersonal growth (Benoit, 2004; Shilkret & Shilkret, 2011). Individuals with secure attachments are more likely to develop and maintain healthy relationships, develop a cohesive and stable sense of self, as well as other characteristics, such as empathy (Shilkret & Shilkret, 2011).
Individuals who do not have responsive caregivers may develop insecure attachments, of which there are three subtypes (Hazan & Shaver, 1987; Main & Goldwyn, 1985). Those exhibiting an anxious-avoidant attachment style often have early experiences marked by caregivers who were neglectful and consistently dismissive of their needs (Benoit, 2004; Shilkret & Shilkret, 2011). Consequently, the child learns to rely on himself or herself and avoids engaging with others rather than be disappointed by another. Those exhibiting an anxious-ambivalent/resistant attachment style experience their caregivers as inconsistent and unpredictable (Benoit, 2004; Shilkret & Shilkret, 2011), leading the child to lack confidence in the caregiver’s presence when needed, either physically or emotionally. In an attempt to remain connected to one’s caregiver, an anxiously attached child tends to rely of dramatic or emotional strategies to maintain the attention of his or her caregiver. The third insecure attachment style, called disorganized, is marked by caregiver experiences children experience as frightening and/or traumatic (Shilkret & Shilkret, 2011). Children with disorganized attachments tend to appear dazed, frozen, and/or confused in interactions with others (Benoit, 2004).
Insecure Attachments and Regulation
Over the last several decades, researchers and practitioners have used attachment theory to understand sexual offending behaviors among youth and adults as many of the characteristics of insecure attachments mirror the risk factors that have been associated with sexual offending (Bogaerts, Vanheule, & Declercq, 2005; Hudson & Ward, 2000; Lyn & Burton, 2005; Vondra et al., 2001). Individuals with insecure attachments struggle with mental health issues, such as PTSD, affective disorders, psychotic disorders, substance use disorders, and personality disorders, including borderline, narcissistic, and antisocial (DeKlyen & Greenberg, 2008; Mikulincer & Shaver, 2007; Rosenstein & Horowitz, 1996). In addition, insecure attachments and caregiving practices have been linked to aggression in a variety of age cohorts, including those involved in the criminal justice system (Felizzi, 2015; Harris-McKoy, 2016; Moitra & Mukherjee, 2010; Ogilvie, Newman, Todd, & Peck, 2014; Patterson, DeBaryshe, & Ramsey, 1990; Savage, 2014).
Among youth and adults who sexually offend, insecure attachment issues have been associated with intimacy struggles (Beech & Mitchell, 2005; Grady, Swett, & Shields, 2014; Marshall, 2010), as well as cognitive (Hudson & Ward, 2000; Hudson, Ward, & McCormack, 1999) emotional (Grady & Shields, 2018) and behavioral dysregulation (Lyn & Burton, 2004; McKillop, Smallbone, Wortley, & Andjic, 2012; Miner et al., 2014). Although there have been some attempts to discern how different attachment styles influence different criminogenic risk factors and offense types (i.e., child molestation vs. rape), there does not appear to be consistent patterns among the styles examined (Grady & Shields, 2018; Ward, Hudson, & Marshall, 1996). One research study revealed youth who commit sexual crimes had greater anxious attachments relative to nonsexual offenders (Miner et al., 2014). Other research revealed that youth who commit sexual crimes had greater difficulties with attachment characteristics including trust, communication, and alienation with maternal caregivers relative to nonsexual offenders (Yoder et al., 2019). The same study demonstrated linkages between negative attachment characteristics and more severe offending profiles among youth who sexually offend. Nevertheless, more research is needed to determine how different attachment styles (e.g., anxious-ambivalent or anxious-avoidant) are related to specific offending patterns and risk factors associated with sexual offending. The dearth of research in this area was further noted a limitation in a large meta-analysis on risk factors for youth sexual violence (Seto & Lalumière, 2010).
Integrated Model
The theory proposed by Grady et al. (2016) links the above research on trauma, attachment, and risk factors to describe a potential pathway to offending, as depicted in Figure 1. The authors propose that, for some children, challenges stemming from childhood trauma make it difficult for children to form secure attachments or that these experiences disrupt an existing attachment due to damaged trust with an adult caregiver, especially, if the perpetrator is a trusting adult in the child’s life. Insecure attachments lead to a number of risk factors associated with sexual offending as described above, including distortions in cognition, emotional dysregulation, impulsivity, and behavioral control (R. J. McGrath, Cumming, Burchard, Zeoli, & Ellerby, 2010). Although these risk factors are strongly associated with sexual offending, few studies have distilled certain types of regulation difficulties in adolescents.

Model of ACE to sex offending behaviors.
In a previous study, Yoder and colleagues (in press) examined the first part of this model through the examination of the relationships between maternal parenting styles, attachment styles, and childhood trauma. They found that children with harsh or indifferent mothers are at increased risk of developing insecure attachments and experiencing trauma. However, experiencing trauma did not lead to insecure attachments, as depicted in the Grady et al. (2016) model; no direct effect was present between childhood traumatic experiences and attachment, and trauma did not mediate the relation between caregiver styles and attachment. However, there was a moderating effect of offender type on the maternal caregiving and childhood trauma, where youth who commited sexual crimes with more harsh/indifferent maternal caregivers were more likely to experience trauma relative to youth who commited nonsexual crimes with high or low reports of harsh/indifferent maternal caregivers.
This examination of the model offers some validation for the Grady et al. (2016) model with regard to the relationship between caregiving patterns, the formation of insecure attachments, and the experience of childhood trauma. However, Yoder et al.’s (in press) study did not examine specific types of traumatic experiences, apart from the caregiving role, such as those that are associated with insecure attachments or how those different styles of attachment are related to criminogenic needs, specifically dysregulation issues. Furthermore, Yoder et al.’s study only examined the moderation effects of early life experiences and sexual offending behavior rather than testing the direct/indirect relationship.
Current Study
The overarching goal of the study was to test part of Grady et al.’s (2016) integrated theory of etiology of sexual offending, postulating that child abuse experiences are related to attachment and dysregulation (Samuelson et al., 2010), which contribute to the commission of sexual harm. This study aimed to answer the following questions:
Based on Grady et al.’s (2016) theory, we proposed the following hypotheses:
Method
Data Collection
Data were collected with adjudicated adolescents (N = 200) residing in the community (n = 54) or in juvenile correctional facilities (n = 146) and who committed sexual (n = 70) or nonsexual offenses (n = 130). After obtaining consent or assent, youth voluntarily participated in a cross-sectional study where they were asked to report on a variety of standardized instruments regarding early life experiences, home environments, attachments, and regulation difficulties. For youth residing in correctional facilities, surveys were distributed to groups of youth comprised of approximately 15 youth per group. For youth residing in the community, surveys were distributed to groups of youth or youth completed the survey individually. The surveys were completed in approximately 60 to 70 min, and youth were given either pizza or gift card incentives. No adolescents reported emotional distress, but there was a mental health professional available if youth were to become distressed. Youth were excluded on the basis of significant developmental delay or psychotic disorder, as indicated by a mental health professional. Youth were also excluded if they were more than the age of 20 years and below the age of 13 years.
The sample included adolescents who committed a sexual (n = 70) or nonsexual crime (n = 130). The average age of the youth was 17.17 years (SD = 1.81 years). On average, the youth completed the tenth grade (SD = 1.82 grades). The average number of months youth lived in a residential placement was 22 months (SD = 22.81). The ethnic make-up of the group included Caucasian (n = 80; 40.6%); African American (n = 40; 20.3%); Hispanic/Latino (n = 71; 36%); Asian (n = 4; 2.0%); Native American (n = 16; 8%); Arab American (n = 1; .5%); and other ethnicities (n = 16; 8%). There was a statistically significant difference between offending groups on race (χ2 = 20.274, p < .001), where sexual offenders were more likely to be White. There were no statistically significant differences between offending groups and other demographic variables, including age (t = 1.22, p = .226), grade level (χ2 = 11.03, p = .087) or length of time spent in any residential or detention facility (t = −1.02, p = .320). The majority of the adolescents were raised in a household with a mother as the primary caregiver (n = 152; 76.0%). The remaining adolescents were raised by single fathers (n = 12; 6.1%); fathers and a partner (n = 6; 3.0%); other relatives (n = 4; 2.0%); grandparents (n =17; 8.5%); or foster homes (n = 7; 3.5%). There were no statistically significant differences in family structure between adolescents who engaged in sexually abusive and nonsexually abusive behaviors (χ2 = 8.45, p = .295). This project was granted approvals from a university’s Institutional Review Board, the state Sex Offender Management Board, and the state Department of Youth Services.
Measures
The current study used two structural equation models to answer the research hypotheses. Latent, observed indexes, and observed variables were used. Items from validated instruments were tested to determine the latent factors. The instruments used for analyses include the Childhood Trauma Questionnaire (CTQ; Bernstein et al., 1994); a researcher-composed tool known as the Domestic Traumatic Experiences Survey; the Attachment Style Classification Questionnaire (ASCQ; Finzi-Dottan, 2012), and the Behavior Rating Inventory of Executive Function–Self-Report (BRIEF-SR; Guy, Isquith, & Gioia, 2004). Although these instruments were previously validated on other samples, this study first conducted independent tests of confirmatory factor analysis (CFA) to verify the factor structure of each instrument. Next, the items were combined in two models (one testing anxious-ambivalent attachment and one testing anxious-avoidant attachment) to determine overall model fit. Finally, the structural portion of the models were run to test proposed hypotheses. The confirmatory factor analyses of the independent measures revealed slight differences between the original measures and the factor structure in this sample. Therefore, first the original measures with subsumed items are presented, then confirmatory analyses are presented, followed by the structural model testing.
CTQ
Specific child abuse experiences were measured using the CTQ, a 25-item standardized measure that asked youth to indicate how often (0 = never, 1 = rarely, 2 = sometimes, 3 = often, 4 = very often) they experienced abuse experiences before leaving home (CTQ; Bernstein et al., 1994). For these analyses, two subscales were used: Physical Abuse with five total items (e.g., “I was punished with a belt, board, cord, or some other hard object”; “People in my family hit me so hard it left me with bruises or marks”) and strong internal consistency (α = .900) and Sexual Abuse with five total items (e.g., “I was sexually molested”) and good internal consistency (α = .840). Researchers have suggested physical and sexual abuse are among the most discernable factors associated with sexually abusive behaviors in adolescents (Knight & Sims-Knight, 2004; Seto & Lalumière, 2010). For the entire sample, 92 youth (46%) reported at least one physical abuse experience, with an average score of .62 (SD = 1.02). Furthermore, 49 youth (24.5%) reported at least one sexual abuse experience, with an average score of .31 (SD = .76). The physical and sexual abuse experiences were used as latent variables in the models, where they were represented by observed items after CFA was conducted.
Domestic traumatic experiences
Other childhood traumatic experiences while youth were in the home of origin were measured through a series of twelve questions that have been used in a number of studies of justice involved youth (Burton, Duty, & Leibowitz, 2011; Yoder et al., 2017). These questions ask the youth to describe their family and/or the home in which they were raised. Some example items include having a parent with an alcohol or drug problem; neglect of children; hitting, slapping, punching, or other violence between parents and children; physical abuse of children (not including the youth); sexual abuse of children (not including youth); illegal acts by family members; being very poor; or having children placed outside of the family. The response items were 0 = not at all like my home, 1 = a little like my home, 2 = somewhat like my home, 3 = a lot like my home, 4 = exactly like my home. The internal consistency was strong for the entire scale (α = .846). Among the entire sample, 170 youth (85%) reported at least one childhood traumatic experience in the home, with an average score of .70 (SD = .68). The items under the Domestic Traumatic Experiences scale were cumulated to create an observed index.
Attachment
Attachment styles were measured using the ASCQ (Finzi-Dottan, 2012), a 15-item standardized assessment tool divided into three constructs rooted in Ainsworth, Blehar, Waters, and Wall’s (1978) three current attachment styles. These are Secure (e.g., “I usually believe that others who are close to me will not leave me”), Anxious-Ambivalent (e.g., “I am sometimes afraid that no one really loves me”), and Anxious-Avoidant (e.g., “I find it uncomfortable and get annoyed when someone tried to get too close to me”). Each subscale contains five items with response items including 1 = all wrong, 2 = wrong, 3 = a bit wrong/a bit right, 4 = right, 5 = very right. For the purposes of this article, the Anxious-Ambivalent and Anxious-Avoidant subscales were used and both had good internal consistency (α = .881 and α = .829, respectively). Among the entire sample, 144 youth (72%) endorsed at least one form of anxious attachment, with an average score of 1.10 (SD = 1.01). Similarly, 170 youth (85%) endorsed at least one form of anxious attachment, with an average score of 1.69 (SD = 1.05). The attachment styles were used as latent variables in the models, where they were represented by observed items after CFA was conducted.
Dysregulation
Dysregulation was measured using the BRIEF-SR (Guy et al., 2004). The BRIEF is a standardized 80-item measure on a 3-point scale (0 = never, 1 = sometimes, and 3 = often) to assess youth’s executive functioning deficits over the past 6 months. There are eight factors that comprise two overarching dimensions of executive functioning: Behavior Regulation Index (BRI) and Metacognition Index (MI). The BRI includes four factors: Inhibit (e.g., “I blurt things out”), Shift (e.g., “I get upset by a change in plans”), Emotional Control (e.g., “I get upset easily”), and Monitor (e.g., “I am unaware of my behavior in a group”). The MI includes four factors: Working Memory (e.g., “I have trouble remembering things”), Plan (e.g., “I don’t think ahead about possible problems”), Organizing Materials (e.g., “My backpack is a mess”), and Task Completion (e.g., “I have difficulty finishing a task on my own”). This study used only the subscales within the behavioral regulation index, measuring them as four separate indexes. The Inhibit Scale (α = .889), Behavioral Shift Scale (α = .844), Cognitive Shift Scale (α = .906), and the Emotional Control Scale (α = .877) all demonstrate good-to-strong internal consistency for this sample. Among the entire sample, 175 youth (87.5%) endorsed at least one problem with inhibitions, with an average score of .57 (SD = .48); 148 youth (74%) endorsed at least one problem with behavior shift, with an average score of .52 (SD = .48); 150 youth (75%) endorsed at least one problem with cognitive shift, with an average score of .43 (SD = .42); and 162 youth (81%) endorsed at least one problem with emotional control, with an average score of .47 (SD = .44). The items under the dysregulation subscales were cumulated to create four separate observed indexes.
Offender group
Offender group was measured by asking youth whether they committed a sexual crime that could get them in trouble with the law (0 = no; 1 = yes) or whether they committed a nonsexual crime that could get them in trouble with the law (0 = no; 1 = yes). The items under the offender group scale were combined to create an observed dichotomous variable (0 = nonsexual offender, 1 = sexual offender).
Data Analysis
CFA
Structural equation modeling (SEM) was used with Mplus Version 7.4 (Muthen, Muthen, & Asparouhov, 2015) as the primary statistical analytics package. The Statistical Package for Social Sciences (SPSS; Scientific Software International., 2015) was the statistical software used to organize the variables prior to transferring to a Mplus file. SEM is the most beneficial approach for testing theoretically supported hypotheses to contemporaneously model pathways between latent and observed variables (Bowen & Guo, 2012). First, CFA tested the indicated items within the CTQ, ASCQ, and the BRIEF to substantiate that they represent their higher order latent constructs including physical and sexual abuse, anxious-ambivalent and anxious-avoidant attachment, inhibition, behavioral shift, cognitive shift, and emotional control. Domestic Traumatic Experiences and offender group constructs were included as observed variables in the model; the offender group was only a two-item variable (three is necessary to create a latent variable, Bowen & Guo, 2012, and a composite for the Domestic Traumatic Experiences instrument [twelve items] was created due to the delays in model estimation resulting from the large number of observed variables in the model.
CFA fit statistics were used to determine model fit. The customary fit statistics include the Comparative Fit Index (CFI), Tucker–Lewis Index (TLI), root mean square error of approximation (RMSEA), and Chi-Square (Bowen & Guo, 2012). Predominant indictors of good fit include the Chi-square (χ2; cutoff of p > .05) and the RMSEA (cutoff of < .06; Hu & Bentler, 1995; Iacobucci, 2010). The chi-square statistic is not always robust to sample size, so the RMSEA is often used to make determinations (Hu & Bentler, 1999). The CFI and TLI (cutoffs of >.95) are additional indicators of a good fitting model (T. A. Brown, 2006; Hu & Bentler, 1995).
When testing the individual instruments, there were minor deviations between original validated tools and the current sample; model fit and factor loadings did not meet the criteria during initial instrument testing. Therefore, item deletion modifications made to the factorial structure of the independent instruments in the following manner: Using modification indices and residual correlation matrices, poor fitting items were methodically examined. Modifications or item deletions were considered if (a) it was theoretically justified; (b) a high residual pattern (>.2) was detected in the residual correlation matrix; and (c) the modification indices revealed a poor pattern (Bowen & Guo, 2012; Byrne, Shavelson, & Muthén, 1989). Item deletion decisions were made using a systematic approach, whereby each item was removed one at a time. Each time an item was removed, the previously removed items were inserted back into the model to determine fit. As such, respective items could load positively for all successive solutions (Bowen & Guo, 2012). Some of the items representing the ASCQ were problematic due to issues of low factor loadings and high modification indices. This resulted in four total item deletions (two per scale). Example items eliminated include “I’d like to be really close to some children and always be with them” and “It’s hard for me to trust others completely.” Furthermore, two of the items representing the CTQ were problematic due to issues of high modification indices. This resulted in two total item deletions (one per subscale of physical and sexual abuse). Items eliminated included “Someone threatened to hurt me or tell lies about me unless I did something sexual with them” and “I was punished with a belt, a board, a cord, or some other hard object.” Finally, seven items representing the four behavioral regulation subscales were problematic due to high multicollinearity and modification indices. Therefore, seven total item deletions were made (1-2 per subscale). Example item deletions included “I become tearful easily” and “I blurt things out.”
For the CTQ instrument, there was a good model fit, χ2 = 499.36, p < .001; RMSEA = .043, confidence interval (CI) = [.033, .052]; CFI = .990; TLI = .989; for the ASCQ, there was a good model fit, χ2 = 39.45, p = .024; RMSEA = .058, CI = [.021, .090]; CFI = .989; TLI = .988); and for the BRIEF instrument (only including the regulation subscale), there was good model fit, χ2 = 440.85, p < .001; RMSEA = .051, CI = [.041, .060]; CFI = .975; TLI = .972). After the individual instruments were tested using the CFA fit criteria and met the threshold for good model fit, they were combined together to determine the overall model fit. There were no necessary item deletions when the structural portion of the model was tested. However, the indexes for each regulation factor (inhibit, shift scales, and emotional regulation) were created and applied in the model given there were too many items representing the latent construct, causing significant delays producing output. The model fit statistics for the final two models are presented in the results.
Structural model testing
The hypotheses were tested in Mplus by running a General SEM to analyze the structural portion of the model. The direct relationships regressed anxious-ambivalent and anxious-avoidant attachment on abuse experiences and domestic traumatic experiences. Other direct relationships regressed the four dysregulation endogenous variables on abuse experiences, child traumatic experiences, and anxious-ambivalent and anxious-avoidant attachment styles. The indirect relationships were tested by modeling the attachment styles as mediators. The weighted least square (WLS) estimation procedure was used; the WLS estimator does not make distributional assumptions made by the default maximum likelihood (ML) estimator (Bollen, 1989) and also addresses the nonnormality that characterizes most ordinal scaled data (Flora & Curran, 2004). The presentation of skewness and kurtosis for each composite is presented in Table 1. Furthermore, Mplus uses a pairwise present technique to handle missing ordinal nonnormally distributed data. This technique is similar to Full Information Maximum Likelihood (FIML) in retaining all cases with at least one data point. The range of missingness for each instrument was very low: Domestic trauma had one to four missing cases; CTQ sexual abuse had eight to nine missing cases; CTQ physical abuse had eight to 10 missing cases, the BRIEF regulation index had seven to 11 missing cases, and the ACSQ had 10 to 13 missing cases. In the Mplus models, there were 13 missing data patterns that were accounted for by the FIML technique.
Descriptive Statistics on Indexes of Interest.
Results
CFA
Results from the anxious-ambivalent attachment model (RMSEA = .022, CI = [.000, .046]; CFI = .998; TLI = .996) and the anxious-avoidant attachment model revealed good fit (RMSEA = .033 CI = [.000, .053]; CFI = .994; TLI = .991). Given the complexity presenting results of the combined measurement portion of the model (latent factors with corresponding observed variables, their loadings, and the pathways), only the structural portion of the model is presented in the figures. Information on the final (after all item deletions) observed variables and factor loadings are presented further.
The factor loadings were strong (>.6; Costello & Osborne, 2005; Tabachnick & Fidell, 2001). The factors included anxious-avoidant attachment with subsumed items including “I don’t feel comfortable trying to make friends,” “Sometimes others get too friendly and too close to me,” and “I find it uncomfortable and get annoyed when someone tries to get too close to me.” The factors included anxious-ambivalent attachment with items including “I sometimes feel that others don’t want to be good friends with me as much as I do with them,” “I’m sometimes afraid that no one really loves me,” and “Children sometimes avoid me when I want to get close and be a good friend of theirs.” Other factors included physical abuse experiences with subsumed items including “Someone in my family hit me,” “People in my family hit me so hard that it left bruises or marks,” “I got hit or beaten so badly it was noticed by someone like a teacher or doctor,” and “I believe I was physically abused” and sexual abuse experiences with subsumed items including “Someone tried to touch me in a sexual way or make me touch them,” “Someone in my family tried to make me do or watch sexual things,” “Someone in my family molested me,” and “I believe I was sexually abused.” Finally, dysregulation consisted of the Inhibit Index subscale with 10 subsumed items; Behavioral Shift Index subscale with five subsumed items; Cognitive Shift Index subscale with four subsumed items; and Emotional Control Index subscale with seven subsumed items.
H1: Direct Effects Between Physical, Sexual, and Domestic Traumatic Events and Attachment
Bivariate correlations between all composites are presented in Table 2. The standardized path results for the anxious-ambivalent attachment model indicated there were statistically significant relationships between physical abuse and anxious-ambivalent attachment, ϒ = .313 (.15), p = .035. Physical abuse and sexual abuse were significantly correlated, φ = .652 (.07), p < .001; physical abuse and domestic traumatic experiences were significantly correlated, φ = .530 (.06), p < .001, and sexual abuse and domestic traumatic experiences were significantly correlated, φ = .326 (.08), p < .001. There were minor differences between the two models on these correlations. Domestic traumatic experiences were associated with greater dysregulation with behavioral transitions, ϒ = .139 (.07), p = .046, inhibitions, ϒ = .186 (.07), p = .003, and emotional control, ϒ = .194 (.07), p = .004. There was a significant amount of variance in anxious-ambivalent attachment (.182, p < .001) explained by the different forms of trauma.
Bivariate Correlations Between Indexes of Interest.
p < .05. **p < .01. ***p < .001.
The standardized path results for the anxious-avoidant attachment model indicated there were significant relationships between physical abuse and anxious-avoidant attachment, ϒ = .298 (.13), p = .024. Sexual abuse was associated with a greater likelihood of dysregulation in inhibitions, ϒ = .277 (.11), p = .016. Domestic traumatic experiences were associated with greater dysregulation with emotional control, ϒ = .134 (.06), p = .043. There was a significant amount of variance in anxious-avoidant attachment (.148, p < .001) explained by the differential forms of traumatic events. Overall, the first hypothesis was partially supported such that physical abuse was associated with both forms of attachment, but sexual abuse and domestic traumatic experiences were not significantly associated with attachment.
H2: Direct Effect Between Attachment and Dysregulation
The standardized path results for the anxious-ambivalent attachment model indicated there were significant relationships between anxious-ambivalent attachment and all forms of dysregulation including behavioral shift, β = .556 (.06), p < .001, cognitive shift, β =.500 (.08), p < .001, inhibitions, β = .484 (.06), p < .001, and emotional control, β = .515 (.07), p < .001. There was a significant amount of variance in behavioral shift (.374, p < .001), cognitive shift (.265, p < .001), inhibitions (.437, p < .001), and emotional control (.298, p < .001) explained by anxious-ambivalent attachment. The path results of these relationships appear in Figure 2.

Model 1: Anxious-ambivalent attachment.
The standardized path results for the anxious-avoidant attachment model indicated there were significant relationships between anxious-avoidant attachment and all forms of dysregulation, including behavioral shift, β = .608 (.06), p < .001; cognitive shift, β = .502 (.08), p < .001; inhibitions, β = .607 (.07), p < .001; and emotional control, β = .523 (.08), p < .001. There was a significant amount of variance in behavioral shift (.312, p < .001), cognitive shift (.255, p < .001), inhibitions (.315, p < .001), and emotional control (.282, p < .001) explained by the model. Overall, the second hypothesis was fully supported such that anxious-ambivalent and anxious-avoidant attachment were associated with all forms of dysregulation. The path results appear in Figure 3.

Model 2: Anxious-avoidant attachment.
H3: Attachment Styles as Mediator
For the anxious-ambivalent attachment model, there was an established correlation between physical abuse and inhibition dysregulation (.168, p < .05) and emotional control dysregulation (.154, p < .05) prior to including anxious-ambivalent styles in model (see Table 2). This suggests that anxious-ambivalent attachment fully mediates these relationships. When the indirect effects were tested, there was a fully mediated effect of anxious-ambivalent attachment in the relation between physical abuse and emotional control, β = .161 (.08), p = .048. For the other dysregulation indexes, there was a sequential linear association between physical abuse, attachment, and cognitive and behavioral shift dysregulation.
For the anxious-avoidant attachment model, there was also fully mediated effect of anxious-avoidant style in the relation between physical abuse and some forms of dysregulation. When the indirect effects were tested, there were mediated effects of anxious-avoidant attachment in the relationships between physical abuse and emotional control, β = .158 (.08), p = .035, and physical abuse and inhibitions, β = .183 (.09), p = .041. For the other dysregulation indexes, there was a sequential linear association between physical abuse, attachment, and cognitive and behavioral shift dysregulation. Overall, the third hypothesis was supported such that the relation between physical abuse and indicated forms of dysregulation was fully mediated by attachment.
H4: Dysregulation Associated With Sexual Offending
In the anxious-ambivalent model, sexual abuse was associated with a greater likelihood of committing a sexual crime, ϒ = .486 (.16), p = .002. However, there were no statistically significant direct effects between dysregulation variables and sexual offending. Similarly, in the anxious-avoidant model, sexual abuse was associated with a greater likelihood of committing a sexual crime, ϒ = .386 (.12), p = .001, but there were no statistically significant direct effect between dysregulation variables and sexual offending. Therefore, the third hypothesis was not supported, and the only variable associated with sexual offending was the experience of sexual abuse.
Discussion
The results of this study indicated that there were direct relationships between physical abuse and both attachment patterns, partially supporting our first hypothesis. We predicted that sexual abuse would be associated with anxious-ambivalent attachment and that a combination of abuses would be associated with both styles of attachment measured. However, only physical abuse had a direct relationship to attachment style. It is possible that the type of attachment relationship influenced these findings, as mothers are more likely than fathers to be the perpetrators of physical abuse (Scott & Meredith, 2014); the vast majority of participants in our sample were raised by single mothers. In contrast, those who commit sexual abuse are more likely to be male, nonprimary caregivers (Scott & Meredith, 2014). Therefore, the direct effect observed between physical abuse and attachment styles might be explained by the primary attachment relationship for these youth when characterized by physical abuse.
The second hypothesis tested the relationship between attachment and different forms of dysregulation. This hypothesis was fully supported and is consistent with the previously presented literature that insecure attachments are associated with a host of behavioral and emotional challenges, including difficulties with emotional regulation (Grady & Shields, 2018; Miner et al., 2014). Although this study was one of the first to delineate differential forms of regulation as a function of insecure attachment styles, the findings can be situated within some available extant research. For example, research has revealed that youth who commit sexual crimes had anxious-ambivalent attachments that were partially explained by social isolation (Miner et al., 2014). Other research demonstrates that among a sample of youth sexual offenders, approximately 84% to 94% of the sample reported dysregulation scores (e.g., shift, inhibition, and emotional control) that exceeded clinically significant levels (Burton, Demuynck, & Yoder, 2016). Therefore, the results from our study greatly contribute to research on how attachment styles may explain why these youth have regulation difficulties.
Our third hypothesis examined the mediating effect of attachment style between different trauma experiences and dysregulation. However, this was modified after only finding a direct effect between physical abuse and attachment. We found that both attachment styles fully mediated physical abuse and specified forms of dysregulation; physical abuse was correlated with emotional and inhibited regulation deficits, and when either insecure attachment style was added, those direct relationships were eliminated. This indicates that, when including the impact of attachment, physical abuse alone is not enough to lead to deficits in emotional control and inhibition regulation. This underscores the importance of assessing for attachment patterns in youth who have experienced physical abuse as the strengthening of their relationships with caregivers may reduce their challenges in regulation that might contribute to sexual offending.
Our final model was not supported; dysregulation was not linked to the offending group. However, in both models, having been sexually abused predicted that the adolescent had committed a sexual crime. Although past studies have reported that physical abuse is a greater predictor of sexual offending among adolescents (Widom, 1995), our findings are consistent with more recent reports about the role of sexual abuse in predicting sexual and other crimes, including violent crimes (S. A. McGrath, Nilsen, & Kerley, 2011). Recently, researchers have reported that childhood sexual abuse is the most significant developmental factor in predicting crime type among juveniles (Basto-Pereira, Miranda, Ribeiro, & Maia, 2016; Braga, Gonçalves, Basto-Pereira, & Maia, 2017). There is a strong linkage between early sexual victimization and later sexual offending (Burton & Meezan, 2004). In fact, social learning theory or the internalization and behavioral manifestation of one’s abuse experience has been often used to explain this phenomenon (Burton et al., 2002). This might suggest the presence of unmeasured mediators or moderators not appearing in our model that would better explain sexual abuse’s relationship to sexual crimes beyond regulation deficits and attachment styles. Sexual abuse was also found to have a direct relationship with the inhibition or impulsivity scale within the anxious-avoidant attachment model. This finding was surprising given the significant literature that links sexual abuse with numerous regulation deficit areas (S. A. McGrath et al., 2011). Nevertheless, physical abuse may have been a more powerful predictor in the model.
We found direct relationships between domestic traumatic experiences and nearly all of the regulation areas in the model for anxious-ambivalent attachment only. There was no direct effect between trauma and regulation in the anxious-avoidant attachment model. One possible explanation for this finding is that the traumatic domestic experiences primarily involve witnessing a primary caregiver suffer from mental illness or being the victim of violence at the hands of another. Another possible explanation for this is that the mediating effect of attachment might be a consequence of the youths’ sensitivity to familial environmental factors. McDonald and colleagues (2016) found that the effects of family environmental factors similar to those measured in our study have a strong impact on youths exposed to intimate partner violence, including indicators consistent with regulation deficits. In their study, there were latent differences among youths exposed to intimate partner violence, categorized by low sensitivity, moderate sensitivity, and high sensitivity. For youths in our study with anxious-avoidant attachment (low sensitivity), it is possible that watching their primary caregiver suffer in some way did not elicit high levels of distress that would cause them to be dysregulated. Their avoidance of intimacy might lead them to ignore or attempt to avoid any negative encounters. However, for children who had an anxious-ambivalent attachment (high sensitivity), witnessing traumatic domestic events might increase the tendency to be more emotional and difficult to soothe, leading them to an increase in dysregulation. In other words, for those already anxious in relationships, witnessing the distress of a primary caregiver would only serve to increase their internal experiences of distress.
A second explanation for the difference between how regulation was affected among those with anxious-ambivalent attachment versus those with anxious-avoidant attachment in our models is that these youth primarily have disorganized attachment styles, which by definition means that there is no consistent pattern in behaviors (Shilkret & Shilkret, 2011). In our sample, 72% of youth endorsed having an anxious-ambivalent attachment style, and 85% an anxious-avoidant one. Given the large numbers of the youth that fell into both categories, it is possible that the majority of youth failed to fall into one discrete category of attachment and are actually disorganized in their attachment style. This possibility seems likely given that children who have experienced trauma are often categorized as having a disorganized attachment style (Main & Solomon, 1990; Shilkret & Shilkret, 2011). However, the instrument we used did not measure disorganized attachment, so we were unable to test this style.
Implications
The findings of this study have several important implications to consider for the prevention and treatment of sexual crimes. For practitioners involved in primary prevention, our findings reinforce the importance of supporting parents in building strong attachments and providing them with positive parenting skills to reduce the risks associated with traumatic experiences among children. Given the relationships we found between physical abuse and attachment, it is essential that caregivers of children be equipped with skills to help reduce the likelihood of resorting to physical abuse during times of frustration. Interventions that address parenting skills, such as Parent–Child Interaction Therapy (Eyberg, 1988) have been found to significantly improve parent–child exchanges and interactions across diverse populations and improve child outcomes in a number of domains of functioning (Thomas, Abell, Webb, Avdagic, & Zimmer-Gembeck, 2017). Foster parents, daycare workers, and other adults who are responsible for the care of children also need training and supports to ensure that they use nonviolent means when working with children under their care.
When practitioners are concerned about the attachment relationships specifically, we recommend an intervention focused specifically on strengthening that relationship, such as Child–Parent Relationship Therapy (CPRT) for example (Bratton, Landreth, Kellam, & Blackard, 2006). The focus of CPRT is on building an empathic and attuned relationship between the parent and child to repair attachment disruptions (Bratton et al., 2006; Bratton, Landreth, & Lin, 2010). As relationships repair, children demonstrate reductions in risk factors, including decreases in behavioral difficulties and increases in empathy (Bratton et al., 2010).
The more our society can do to prevent children from experiencing trauma, the more likely we will prevent future crimes as large-scale studies have confirmed the high rates of trauma among juveniles involved in the criminal justice system (Levenson et al., 2017). The findings reported in this study provide additional evidence that trauma has direct and indirect relationships with all risk factors associated with criminal behavior. Therefore, to reduce the likelihood of criminal sexual behavior in adolescents, children need to be protected from various forms of trauma. However, if a child does experience some form of trauma, it is essential that they receive trauma-based services immediately, such as Trauma-Focused Cognitive-Behavioral Therapy (Cohen et al., 2016), to help them address the potential challenges that can emerge from a traumatic experience.
Finally, given the high rate of trauma among youth who enter the juvenile justice system (JJS), it is essential that the JJS incorporate specific trauma treatments into their rehabilitation programs. In addition to discrete models of intervention, JJS programs should also adopt a system-wide trauma-informed care (TIC) approach (SAMHSA, 2013). TIC is a universal approach to addressing trauma by creating policies and practices that are trauma sensitive throughout every aspect of a program (Levenson, 2014). By helping adolescents process their traumas, they will begin decreasing their risk factors and strengthening their protective factors (Cohen et al., 2016), thereby decreasing the likelihood that they will remain involved in criminal activity in the future.
Future Research
It is clear that variables not measured in this study connect sexual abuse to sexual offending. Therefore, additional research should explore how other risk factors, such as sexual arousal or empathy deficits (R. J. McGrath et al., 2010) might fit in to Grady et al.’s (2016) theory. In addition, other research should explore the latter component of Grady et al.’s theory to examine the relationship between particular criminogenic needs and subsequent offending behaviors. Also, given some of the inconsistent patterns within the models, it would be important to include disorganized attachment styles as a potential variable to see if placing children within that category changes the results of the analysis. Finally, to further assess the relationship between child trauma and attachment, future studies should collect more research on the relationships between the victim and the perpetrator, as well as the severity, frequency, and intensity of the abuse itself.
Limitations
Although considering the findings of this study, they should be examined in the context of the limitations. The data were a cross-sectional survey at one time point of youth’s early developmental experiences. Causal conclusions cannot be drawn based on the findings. Although we considered the temporal associations between variables, longitudinal or time series designs are most appropriate for determining causal steps. Furthermore, all of the findings are based on self-report data and retrospective. As such, it is possible that the youth may have either inadvertently or advertently reported inaccurate information. Second, we did not have information about the specifics including perpetrator information, frequency, duration, or timing of the maltreatment experiences. As such, we were unable to ascertain how various factors, such as the relationship to their perpetrator or severity of the abuse, may have influenced the relationship between the variables. Furthermore, we were limited by the data available within the data set. As such, there may be additional variables that would have been important to include beyond regulation challenges that would explain the relationships proposed by Grady et al. (2016). The measures in the study appropriately captured the constructs, however, there may be additional measures of early life abuse, adversity, and attachment and regulation that could be included as viable indicators.
Finally, it will be important for future studies to explore how racial and ethnic differences emerge in different samples, especially, in samples that draw from a more diverse geographic region. In these studies, it will be important to control for race in addition to location, such as rural versus urban to examine how the pathways may differ depending on the racial identity, geographic context, or both.
Conclusion
This study’s aim was to empirically test the theory of sexual offending posited by Grady and colleagues (2016). Specifically, we empirically examined the relationship between different abuse experiences (physical, sexual, and other domestic trauma), attachment style, and dysregulation (behavioral and cognitive flexibility, impulsivity, and emotional control) among youth who were adjudicated for sexual or nonsexual crimes. Our findings indicated that physical abuse had a direct relationship with the youth’s attachment styles, and this attachment style served as a mediator between the abuse and subsequent regulation deficits. Although sexual abuse did not have a direct relationship with attachment or dysregulation, it did predict the commission of a sexual crime. Therefore, there are likely other variables that influence sexual offending beyond attachment and regulation difficulties. Given the impact that trauma has on youth outcomes, it is essential that society do more to prevent all forms of abuse. If and when abuse does occur, it is critical to provide interventions as quickly as possible to minimize the impact of these experiences. Such interventions are especially important for those who are most vulnerable, including those who enter into the juvenile justice system.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
