Abstract
Previous research has reported that elevations on both callous-unemotional (CU) traits and anxiety (measured as trait worry) among adolescents are associated with a particularly severe pattern of aggressive behavior. In the current study, we tested whether elevated trait worry would add only to the prediction of less severe and reactive aggression assessed by self-report but not to more severe violence, proactive aggression, and official records of violent arrests. First-time male juvenile offenders (N = 1,216) were assessed at 6-month intervals for 30 months. Contrary to predictions, our analyses found both CU traits and worry independently predicted self-reported violent offenses (regardless of violence severity) and aggression (both proactive and reactive) across 30 months after their first arrest. However, when using arrest records, only CU traits were associated with violent offenses. This suggests that the additive effects of anxiety and worry in predicting risk for later violence may be limited to self-report.
The construct of psychopathy (i.e., narcissism, impulsivity, callousness, poverty of emotions) has a long history of research in adults because of its association with a severe, stable, and difficult to treat pattern of aggressive and antisocial behavior (Blais, Solodukhin, & Forth, 2014). Through much of this history, theories have been proposed suggesting that there are variants of psychopathy that differ in their underlying causal mechanisms. One especially important and influential theory was proposed by Karpman (1941, 1948), who suggested that there are two distinct psychopathy variants differentiated by level of fear, anxiety, and distress. Specifically, he theorized that a “primary psychopathy” variant is characterized by psychopathic traits and low to normal levels of anxiety, which reflects an innate or heritable deficit in the person’s ability to experience emotions. In contrast, “secondary psychopathy” is characterized by psychopathic traits and elevated anxiety and is theorized to reflect a traumatic reaction to serious environmental stressors, such as parental rejection or abuse. A substantial amount of empirical research has supported many of the core assumptions of this theoretical model in adults (Skeem, Poythress, Edens, Lilienfeld, & Cale, 2003).
More recently, callous-unemotional (CU) traits have been used to capture the affective components of psychopathy prior to adulthood, especially the lack of guilt, absence of empathetic concern for others, and the general poverty of emotions that are hallmarks of most conceptualizations of psychopathy (Hare & Neumann, 2005). Like psychopathic traits in adults, CU traits have been shown to capture a more severe and aggressive subgroup of antisocial youth (Frick, Ray, Thornton, & Kahn, 2014), leading these traits to be included in the most recent revision (5th edition) of the Diagnostic and Statistical Manual of Mental Disorders (DSM–5; American Psychiatric Association, 2013) as a way to classify a particularly severe subgroup of individuals with conduct disorder.
Also, similar to research on adults with psychopathy, there appear to be distinct variants of CU traits in children and adolescents who differ on the basis of whether they exhibit low to average levels of anxiety or distress (i.e., primary variant) or high levels of anxiety (i.e., secondary variant) and who show characteristics that support the theorized differences in causal processes between the primary and secondary variants of psychopathy (Kimonis, Skeem, Cauffman, & Dmitrieva, 2011). For example, the secondary variant seems to experience higher levels of trauma, including physical abuse (Kahn et al., 2013; Tatar, Cauffman, Kimonis, & Skeem, 2012), sexual abuse (Kimonis, Fanti, Isoma, & Donoghue, 2013), neighborhood violence (Docherty, Boxer, Huesmann, O’Brien, & Bushman, 2015), and verbal victimization from peers at school (Docherty et al., 2016). They also show greater problems regulating their emotions (Gill & Stickle, 2016; Kahn et al., 2013; Kimonis et al., 2011; Kimonis, Frick, Cauffman, Goldweber, & Skeem, 2012; Salihovic, Kerr, & Stattin, 2014; Sharf, Kimonis, & Howard, 2014; Tatar et al., 2012; Vaughn, Edens, Howard, & Smith, 2009) and a heightened startle response to aversive images (Kimonis, Fanti, Goulter, & Hall, 2017). In contrast, CU traits in the absence of elevated anxiety or worry are more strongly related to deficits in the processing of emotional stimuli, such as showing a failure to orient more quickly to pictures of persons or animals in distress (Kimonis et al., 2012) and reduced amygdala activation to fearful facial expressions (Marsh et al., 2008; White et al., 2012).
Thus, the role of anxiety for moderating the association of CU traits with important background (e.g., abuse and trauma) and dispositional factors (e.g., response to emotional stimuli) has been critical for causal theories of CU traits (Kimonis et al., 2012). However, anxiety also appears to be important for moderating the association of CU traits with several clinically important outcomes. Importantly, research with children and adolescents has consistently indicated that the secondary variant shows higher rates of violent and aggressive behavior than the primary variant (Docherty et al., 2016; Fanti, Demetriou, & Kimonis, 2013; Kahn et al., 2013; Kimonis et al., 2011; Kimonis et al., 2013; Salihovic et al., 2014; Vaughn et al., 2009). For example, in a mixed sample of high school students and detained adolescents, those high on both CU traits and anxiety exhibited the greatest level of aggressive and violent behavior, as measured by multi-informant composite measures of physical aggression (Docherty et al., 2016). Similarly, in a sample of institutionalized adolescents, 92% of the secondary variant committed at least one violent act over a 2-year study period while institutionalized, compared to 69% of adolescents high on CU traits only (Kimonis et al., 2011). In yet another study of incarcerated adolescents, secondary variants self-reported significantly higher rates of violent offending in the 12-month period prior to incarceration than control or primary groups (Vaughn et al., 2009).
Given the consistency of these findings across several samples of adolescents (see Skeem et al., 2003, for similar findings in adult samples), several theories have been proposed to explain this higher level of aggression and violence in persons with elevated CU traits and anxiety. The most common explanation is that the secondary variant of CU traits shows problems regulating their emotions, possibly as a consequence of their histories of abuse and trauma, which leads to aggressive outbursts in response to provocation (Kimonis et al., 2011). Such an explanation would be consistent with research suggesting that one common pathway to externalizing behaviors is through the development of a broad trait of disinhibition that is defined by problems regulating both emotions (e.g., anxiety, anger) and behavior (e.g., impulsivity, aggression; Krueger, Markon, Patrick, Benning, & Kramer, 2007). Thus, the distinction between the variants of CU traits could reflect different pathways to externalizing behaviors more generally, including aggression. However, there are several limitations in the existing research on the variants of CU traits that limit the confidence that can be placed in the conclusion that the secondary variant engages in more violence.
First, much of the past research on the associations among CU traits, anxiety, and aggression have used measures that typically assess the frequency of aggression and have reported that CU traits in the presence of high anxiety is related to more frequent aggressive behaviors (Docherty et al., 2016, Kimonis et al., 2011). The theoretical explanation provided for these findings has largely focused on a specific type of aggression, that is, aggression that is due to high levels of emotional reactivity in response to perceived provocation from others (i.e., reactive aggression; Marsee & Frick, 2007). However, past work has typically not distinguished between the frequency of reactive aggression from the frequency of another type of aggression: proactive aggression. Proactive aggression is not an impulsive reaction to provocation but instead is premeditated and used for instrumental gain (Marsee & Frick, 2007). Furthermore, adolescents with CU traits show elevated levels of both reactive and proactive aggression (Fanti, Frick, & Georgiou, 2009; Frick, Cornell, Barry, Bodin, & Dane, 2003; Kruh, Frick, & Clements, 2005; Lawing, Frick, & Cruise, 2010). The failure of past studies to consider only frequency and not the type of aggression when comparing different variants of CU traits is important because reactive aggression tends to be displayed at a much higher base rate than proactive forms of aggression (Marsee et al., 2014). As a result, it is possible that youth with a combination of high CU traits and high anxiety show the more frequent reactive aggression but not the less frequent proactive forms of aggression compared to those with elevated CU traits without anxiety. In support of this possibility, a study of incarcerated boys reported that those with CU traits and high anxiety self-reported more violent offenses, consistent with past work, but they did not differ from other boys high on CU traits on their level of self-reported proactive aggression (Kimonis et al., 2013).
A second issue when concluding that the secondary variant of CU traits engages in the most violence is the failure to consider differences in the severity of the aggressive behavior. That is, as noted previously, past research has largely focused on the frequency of a child’s aggressive and violent behavior. Thus, frequent but less severe aggression would contribute more to the variance in these measures of aggressive behavior than less frequent but severe aggression that results in greater harm to others. For example, Kimonis et al. (2011) used items that ranged in severity from “seriously threatened another person” to “raped another person.” In another study, the violence subscale included items that ranged from “serious physical fight” to “shot/stabbed someone” (Salihovic et al., 2014). Because of the methodology in past studies, the problems in emotional regulation found in the secondary variant may lead to more frequent aggressive outbursts that are less severe, whereas the lack of anxiety and distress in the primary variant may lead to less frequent but more severe violence (Skeem et al., 2003). Therefore, it is important to distinguish between the frequency of aggression and violence and the severity of aggressive behavior when comparing the two variants.
Third, previous work in adolescent samples has largely focused on self-report of aggression and violence when studying their associations with CU traits and anxiety. In adult samples, the primary variant of psychopathy has sometimes exhibited similar or higher levels of violence when measures do not rely on self-report (Hicks, Vaidyanathan, & Patrick, 2010; Poythress et al., 2010; Swogger & Kosson, 2007; Swogger, Walsh, & Kosson, 2008; Vassileva, Kosson, Abramowitz, & Conrod, 2005). For example, Drislane et al. (2014) reported that although the secondary variant of psychopathy reported significantly more delinquency, aggression, and externalizing problems in a large community sample of young adult men, the primary variant had significantly higher rates of arrests and convictions for violent offenses when using official records. In one of the few studies to test the consistency of findings across methods prior to adulthood, Kahn et al. (2013) reported that, in a sample of 272 clinic-referred adolescents, youth with CU traits and high anxiety self-reported higher levels of impulsivity and externalizing behavior than youth with only elevated CU traits, but there were no differences between these variants on measures of behavior problems as reported by parents. These authors suggested that the primary variant may be less willing to report on the severity of their behavior, and this finding could call into question previous findings of differences between the two variants that relied on self-report.
A final issue that needs to be addressed is that past studies of CU traits, anxiety, and aggression have typically used a methodology that compares two groups of youth with CU traits who differ on their level of anxiety or worry to a non-CU control group. This type of person-centered analysis is justified on the basis of the theoretical contention that the presence of elevated anxiety classifies an etiologically distinct subgroup of youth with elevated CU traits (Kimonis et al., 2012). However, it is also possible that anxiety is largely a marker of the more severe behavioral disturbance displayed by some youth with elevated CU traits. That is, Frick, Lilienfeld, Ellis, Loney, and Silverthorn (1999) provided data to suggest that children and adolescents with elevated levels of CU traits are distressed by their behavior, as indicated by their higher level of anxiety and worry relative to youth without behavior problems. However, they appear less distressed when compared to youth with similar levels of behavior problems. This explanation was used to explain why CU traits are typically uncorrelated with anxiety in zero-order correlations but become negatively correlated when controlling for level of conduct problems (see Frick, 2012, for a summary of this research). This latter explanation thus suggests that anxiety may be a marker of the severity of the child’s behavioral disturbance even within youth with elevated CU traits. To begin to disentangle these potential explanations, it would require determining if any effects of anxiety are specific to those high on CU traits (i.e., designate variants of those with elevated CU traits) or if their effects are independent of each other. Such a test would require using continuous measures of CU traits and anxiety and testing both main and interactive effects of each in predicting aggression and violence.
Current Study
In the current study we attempted to address these four limitations in the existing research on the associations among CU traits, anxiety, and aggression/violence in a large and ethnically diverse sample of adolescents who were arrested for their first offense in three jurisdictions in the United States. Using a justice-involved sample allows for an increased number of youth with elevated CU traits and thus provides an optimal sample for testing potential differences in those with elevated CU traits who differ on their level of anxiety, measured in the current study as generalized worry. Furthermore, such a sample provides a method for tracking level of aggression and violence that does not rely on youths’ self-report (i.e., official records of arrests for violence). In this study, we attempted to replicate previous findings showing that both CU traits and anxiety would be related to more traumatic experiences (i.e., violent victimization) and more self-reported aggression and violence. However, we also tested the novel prediction that anxiety would add only to the prediction of high frequency but less severe violent offenses (i.e., fighting) and to reactive aggression, but would not contribute above CU traits to the prediction of more severe forms of violence (e.g., armed robbery; physical attacks requiring hospitalization) and proactive aggression. Furthermore, we tested the prediction that the added prediction of anxiety would be limited to self-report measures and would not be found when using official records of violent offending over a 30-month period. Finally, to test these predictions, we utilized variable centered analyses (i.e., regression analyses) to test both the main and interactive effects of CU traits and anxiety in predicting aggression and violence to determine if any associations with anxiety would be limited to those high on CU traits.
Method
Participants
The sample consisted of 1,216 male first-time juvenile offenders from the Crossroads Study, an ongoing longitudinal study in Orange County, California (n = 532), Jefferson Parish, Louisiana (n = 151), and Philadelphia, Pennsylvania (n = 533). Participants were eligible for the Crossroads Study if they were English speakers, were arrested for an eligible offense of low to moderate severity, and were between the ages of 13 and 17 at the time of their first arrest. At the start of the study, the mean age of participants was 15.29 (SD = 1.29). The sample was primarily Hispanic (45.9%) and African American (36.9%), with smaller proportions identifying as White (14.8%) or Other (2.4%). The highest level of education either parent obtained was primarily GED or high school (34.1%), less than high school (27.2%), trade school or some college (20.4%), 4-year college degree (13.5%), and graduate-level education (4.8%). Participants’ intelligence was on average lower than that of the general population (M = 88.50, SD = 11.87) as measured using the Matrix Reasoning and Vocabulary subtests of the Wechsler Abbreviated Scale of Intelligence (WASI; Wechsler, 1999).
Procedures
The institutional review boards at all three institutions approved the study procedures. Parental informed consent and youth assent were obtained for all participants at the time of each assessment period (i.e., baseline and follow-up points), until the participant turned 18, at which point consent was subsequently received at each time point with only the participant. Participants and their parents were informed that participation was entirely voluntary, that it would not influence the youth’s relationship with the juvenile justice system, and that they were able to withdraw from the study at any time without penalty. The youth and parents were informed that the research project had obtained a Privacy Certificate from the Department of Justice, which protected their data from being subpoenaed for use in legal proceedings.
The Crossroads Study includes a battery of tests, largely assessing the youths’ academic, occupational, social, and legal outcomes, as well as attitudes (e.g., psychosocial maturity, future orientation) and contextual factors (e.g., treatment, exposure to violence, parenting, neighborhood disorganization, deviant peers) that could influence these outcomes (Fine et al., 2017; Simmons, Steinberg, Frick, & Cauffman, 2018). Youth completed a baseline assessment within 6 weeks of the disposition date for their initial arrest. They were then reassessed every 6 months for 30 months (5 time points). Interviews lasted on average approximately 2 to 3 hr and were administered using a secure computer-based program on a laptop. Participants were able to select their preferred location to complete the interviews, often at the youth’s home, a local restaurant, a public library, the respective research team’s university, or a secure facility if a participant was incarcerated at the time of a follow-up interview. Finally, if participants moved too far to conduct in person interviews, phone interviews were completed (4.7% of all interviews from baseline to 30 months). Participants were compensated $50 for the baseline interview, and the payment increased by $15 for each subsequent interview (i.e., $65 for the second interview, $80 for the third interview).
Retention rates ranged from 96.57% at the 24-month follow-up to 92.58% at the 30-month follow-up with an average retention rate of 94.72% across the five follow-up assessments. In the current study, only those participants with at least three of the five follow-up assessments were used in analyses (n = 1,159; overall retention of 95.3%). There were no missing data on the arrest outcome that relied on a review of official records, given that this information was collected even if the youth did not provide self-report at a given time point. For self-report outcomes, mean substitution (average of available data points) for each individual was used to impute the missing time point for those who were missing one or two follow-up assessments (n = 136, 11.7%). Participants included in the analyses were compared to participants who were not included (n = 57, 4.69%) on baseline demographic variables (i.e., age, race/ethnicity, and IQ) and on key variables of interest measured at baseline (i.e., CU traits, anxiety, all self-report measures of aggression, violence, and victimization). The groups did not differ significantly on any variable (all ps > .05, η2 = .000–.006).
Measures—Baseline predictors
Callous-unemotional traits
CU traits were assessed at baseline using the self-report version of the Inventory of Callous-Unemotional Traits (ICU; Kimonis et al., 2008), a 24-item instrument that utilizes a 4-point Likert-type scale (0 = not at all true, 3 = definitely true) to indicate how accurate each statement describes them. The scale contains equal numbers of items worded in the callous (e.g., “I do not feel remorseful when I do something wrong”) and noncallous (e.g., “I am concerned about the feelings of others”) direction, and the noncallous items are recoded so that higher scores indicate higher levels of CU traits. The total ICU score has been consistently associated with antisocial behavior (Essau, Sasagawa, & Frick, 2006; Fanti et al., 2009; Kimonis et al., 2008; Roose, Bijttebier, Decoene, Claes, & Frick, 2010) and negatively associated with prosocial behavior (Eremsoy, Karanci, & Berument, 2011) in adolescent samples. The internal consistency for the baseline ICU total score in this sample (M = 26.27, SD = 8.03) was acceptable (Cronbach’s α = .76).
Anxiety
The measure of anxiety consisted of six items from the Generalized Anxiety Disorder (GAD) subscale of the Revised Children Anxiety and Depression Scale (RCADS; Chorpita, Yim, Moffitt, Umemoto, & Francis, 2000) that focused on the participants’ level of trait worry collected at the baseline assessment. The items are rated on a 4-point Likert-type scale (0 = never, 3 = always), and the six items used in the current study include “I worry about things,” “I worry that something awful will happen to someone in my family,” “I worry that bad things will happen to me,” “I worry that something bad will happen to me,” “I worry about what is going to happen,” and “I think about death.” This methodology was consistent with several studies that have assessed primary and secondary variants of CU traits in samples of adolescents using scales that assess proneness to worry (Dolan & Rennie, 2007); Gill & Stickle, 2016; Kahn et al., 2013; Salihovic et al., 2014). Furthermore, the GAD subscale of the RCADS has been significantly correlated with other measures of trait anxiety. For example, in a sample of 246 adolescents in the community (44.3% male; M age = 12.20, SD = 2.89) the RCADS GAD scale was correlated (r = .68, p < .01) with a measure of trait anxiety Revised Children’s Manifest Anxiety Scale [RCMAS]; Chorpita et al., 2000). Also, in a sample of 513 adolescents (67.4%; M age = 12.9, SD = 2.7) referred for mental health assessment, the RCADS GAD scale was similarly correlated (r = .65, p < .01) with a measure of trait anxiety (RCMAS; Chorpita, Moffitt, & Gray, 2005). In the current study, the internal consistency for the worry items at baseline (M = 5.28, SD = 3.72) was acceptable (Cronbach’s α = .80).
Measures—Victimization and violence outcomes
Violent victimization
Victimization was measured using the 5-item Self-Report Violent Victimization subscale from the Exposure to Violence (ETV) scale, which asks whether participants were victimized by different types of violence since the last interview (e.g., “Have you been attacked with a weapon, like a knife, box cutter, or bat?”, “Have you been shot?”; Selner-Ohagan, Kindlon, Buka, Raudenbush, & Earls, 1998). Scores on this scale have been associated with increased self-report offending (Selner-Ohagan et al., 1998) and posttraumatic stress symptoms in at-risk adolescents (Muller, Goebel-Fabbri, Diamond, & Dinklage, 2000). Although the ETV also includes items assessing the youth’s witness of violence toward others, these items were not used for the purposes of the current study. Total victimization scores were created by summing the number of violent victimizations endorsed across all time points from baseline to 30 months. The mean number of instances of violent victimization in the sample was 2.24 (SD = 7.30), and the stability of the violence victimization from the 6-month to the 30-month follow-up was significant (r = .24, p < .001).
Aggression
Proactive and reactive aggression was measured using the Peer Conflict Scale (PCS; Marsee et al., 2011). The PCS is a 40-item scale designed to provide extensive coverage of aggression expressed physically (i.e., intentional physical harm to others) and relationally (i.e., intentional harm to others social relationships). Only the physical aggression items were used in the current study, with 10 items assessing reactive aggression and 10 items assessing proactive aggression. Items are rated on a 4-point Likert-type scale from 0 (not at all true) to 3 (definitely true), and as a result, the PCS assesses aggressive traits and not specific aggressive behaviors. Factor analytic support for the ability to the PCS to separate reactive and proactive aggression was reported in a large sample of older children and adolescents (N = 855; age range = 12–18 years; Marsee et al., 2011). Reactive and proactive aggression were also associated with different responses to provocation (e.g., reactive aggression was associated with aggressive responses to low provocation) in a detained sample of adolescent boys (Muñoz, Frick, Kimonis, & Aucoin, 2008). Total, proactive, and reactive aggression scores were created by summing the items across all time points from 6 months to 30 months. The internal consistency estimates for total overt aggression (α = .87–.90), proactive overt aggression (α = .75–.83), and reactive overt aggression (α = .83–.86) were acceptable across the five time points, and the stability of the measures of aggression from the 6-month to the 30-month follow-up was significant for total aggression (r = .44, p < .001), proactive aggression (r = .32, p < .001), and reactive aggression (r = .48, p < .001).
Self-report violent offending
Total violent offending was measured at each follow-up point using nine items of the Self-Report Offending Scale (SRO) that assess crimes against other individuals (Huizinga, Esbensen, & Weiher, 1991). The violence items include rare but serious forms of violence (i.e., “killed someone,” “forced someone to have sex with you,” “shot someone (where the bullet hit the victim),” “shot at someone (where you pulled the trigger),” “taken something from another person by force, using a weapon,” “taken something from another person by force, without a weapon,” “beaten up or physically attacked someone so badly that they probably needed a doctor,” “beaten up, threatened, or physically attacked someone as part of a gang”). It also includes one item that assesses frequent but nonserious fighting (i.e., “been in a fight”). Scores on this scale have been shown to correlate with other measures of aggression and official records of offending across diverse samples (Farrington, Loeber, Stouthamer-Loeber, van Kammen, & Schmidt, 1996; Piquero, Macintosh, & Hickman, 2002; Thornberry & Krohn, 2000). Each item asks participants (yes or no) if they have ever engaged in each crime, and if yes, how many times since the last interview. The SRO variety score is often used to evaluate the number of different crimes (i.e., offense types) the individual endorses over a time period. However, because our study hypotheses are specific to the number of violent acts endorsed overtime, we used the frequency score in the current study (Piquero & Brame, 2008). The total frequency of each violent offense across all five time points from 6 months to 30 months was summed to create a total violent offending composite score (all nine behaviors) and a severe violent offending-only composite score (excluding fighting). The stability of these two scores from the 6-month to the 30-month follow-up was significant for total violence (r = .81, p < .001) and severe violence (r = .93, p < .001). As expected, the fighting item accounted for the majority of the variance in the total violence measure. Specifically, the mean total violence score across the 30-month follow-up was 8.46 (SD = 53.14), whereas the severe violence score, eliminating only the fighting item, was 3.45 (SD = 37.03). 1
Arrests
Data from participants’ official records of both juvenile and adult arrests were obtained within the jurisdictions in which the participant was initially arrested. Approximately 40.7% of the sample were rearrested at any point during the follow period, with 24.1% of the sample being arrested for a violent crime. Crimes were classified as violent on the basis of whether the offense required physical harm to the victim (e.g., assault). The most frequent violent crimes committed across the follow-up period were assault or battery (12.8%), aggravated assault or battery (7.6%), robbery or robbery with serious bodily injury (4.4%), and fighting (1.4%). Because of the small proportion of the sample who were rearrested for a violent crime and because of the assumption that only more severe aggressive acts would lead to an arrest, all violent offenses were used in the creation of this variable and no attempt was made to differentiate more or less severe forms of violent offenses. Because of the low frequency of arrests for violent crimes among our sample, we dichotomized this variable such that any arrest for a violent offense between baseline and 30 months was coded as 1 (n = 272, 23.5%), and participants with no arrests for a violent crime were coded as 0 (n = 887, 76.5%).
Analytic plan
First, zero-order correlations were conducted to test the association between the demographic variables and the main study variables. Second, to test the main hypotheses focusing on the main and interactive influences of CU traits and anxiety, we conducted a series of negative binomial regression analyses. Negative binomial regressions were utilized because most of the dependent variables were count data with a large number of 0 values, and followed a skewed, over dispersed distribution such that the variance of the dependent variable was greater than the mean. CU traits, anxiety, and their interaction were mean centered and entered as independent variables. The one exception was that logistic regression analyses was used to test the prediction of any violent arrest during the follow-up period using official records. All analyses controlled for age, race/ethnicity (i.e., Hispanic and African American), and IQ. African American and Hispanic were coded 1 for endorsing each race/ethnicity, and 0 for all other individuals.
Results
Preliminary analyses
Zero-order correlations among demographic and main study variables are reported in Table 1. First, age and IQ were negatively correlated, and ethnicity (being African-American) was positively correlated with official records of arrests for violent offenses. Thus, the demographic variables were included as controls in the main multivariate analyses. Second, CU traits at first contact with the juvenile justice system were positively associated with participants’ baseline anxiety, violent victimization, and all measures of violence and aggression over the 30-month follow-up period. Third, anxiety at first contact with the juvenile justice system was positively associated with violent victimization, total self-report violence, and all self-report measures of aggression across the follow-up period, but it was not significantly correlated with the self-report of more severe violent offending or official records of violent offending.
Zero-Order Correlations Among Demographic and Main Study Variables
Note: N = 1,159. CU = callous-unemotional traits; SR = self-report. African American and Hispanic are coded 1 for endorsing the race/ethnicity, 0 for all other individuals. Boldface type indicates significant correlations.
p < .05. **p < .01. ***p < .001.
Violent victimization
The first negative binomial regression (controlling for IQ, age, and race/ethnicity) analysis tested if CU traits, anxiety, and their interaction at baseline predicted the number of instances of self-reported violent victimizations that occurred across the follow-up periods. The results of these analyses are presented in Table 2. The analyses revealed significant main effects of both CU traits and anxiety in predicting the frequency of violent victimization but no significant interaction between these two predictors. These findings suggest an additive effect of both CU traits and anxiety on violent victimization such that the influence of CU traits on victimization did not rely on the level of anxiety and the influence of anxiety on predicting victimization did not rely on the level of CU traits.
Negative Binomial Regression Testing the Predictions of Self-Reported Violent Victimization and Self-Reported Violent Offending and Aggression
Note: N = 1,159, df = 7. CU = callous-unemotional traits; ANX = anxiety; CI = confidence interval. Parameters are after controlling for age, IQ, and race/ethnicity.
p < .05. ***p < .001.
Self-report violent offending and aggression
The next series of negative binomial regressions tested the main and interactive effects of both CU traits and anxiety for predicting the various self-report measures assessing frequency of violence and aggression across the follow-up periods. The results of these analyses are also reported in Table 2. These analyses revealed significant main effects for both CU traits and anxiety but no interaction for most of the measures of aggression (e.g., proactive, reactive) and violence (e.g., total violence and more severe violence). These main effects remained significant after controlling for inflated error rates associated with multiple statistical tests using Bonferroni correction with an adjusted alpha of p = .007. Again, this pattern supports the additive effects of CU traits and anxiety for predicting violent offending, given that each predicted offending independent of the other and without a significant interaction. The one exception to this trend was that, in addition to main effects, CU traits and anxiety showed a significant interaction when predicting total violent offending. This interaction is illustrated in Figure 1, showing that the highest rate of self-reported violent offending was found in the adolescents high on both CU traits and anxiety. However, this interaction needs to be interpreted cautiously given that it was the only interaction to be found across all analyses and it did not remain statistically significant after controlling for multiple comparisons.
The interaction between CU traits and anxiety for predicting self-reported violent offending.
Arrests for violent offending
Finally, logistic regression analyses were used to test the main and interactive effects of CU traits and anxiety for predicting the likelihood of being arrested for a violent offense, as measured by official arrest records (Table 3). In contrast to the self-report analyses, only a significant main effect of CU traits (odds ratio = 1.022, p = .013, 95% confidence interval, or CI = [1.005, 1.04]), and not anxiety (odds ratio = 1.001, p = .947, 95% CI = [0.964, 1.040]), was found to increase the likelihood of being arrested for any violent offense. Furthermore, there was no significant interaction between anxiety and CU traits in this analysis (odds ratio = 1.001, p = .534, 95% CI = [0.997, 1.006]).
Logistic Regression Analyses Testing the Prediction of Violent Arrests Using Official Records
Note: CU = callous-unemotional traits. ANX = anxiety. Parameters are after controlling for age, IQ, and race/ethnicity.
Pseudo-R2 = .05. bp = .013. cp = .947. dp = .534.
Discussion
Elevated CU traits has become an important construct for understanding antisocial youth, given their association with aggressive and violent behavior (Frick et al., 2014). Thus, clarifying this association has important implications for systems that serve youth with serious behavior problems, like the mental health and juvenile justice systems. Consistent with a rather substantial body of research, our results support previous research suggesting that youth with high levels of both CU traits and anxiety/distress are at particularly high risk for being both victims of violence and for self-reporting a very high rate of aggression and violence (Docherty et al., 2016; Fanti et al., 2013; Kahn et al., 2013; Kimonis et al., 2011; Kimonis et al., 2013; Vaughn et al., 2009).
However, our results advance this past work by suggesting that this is not limited to mild aggression that is reactive in nature (e.g., in response to perceived provocation). Instead, anxiety and CU traits both showed main effects for predicting greater levels of self-reported serious violent offending (e.g., assault) and both self-reported reactive and proactive (i.e., premeditated for instrumental gain) aggression over a 30-month period. Thus, theories developed to explain the aggression in youth high on both CU traits and anxiety need to consider reasons for both reactive and proactive forms of aggression. For example, histories of abuse and other forms of victimization could lead to hypervigilance toward threatening stimuli, which could lead to reactive aggression to perceived threats (Kimonis et al., 2012). Furthermore, this hypervigilance could operate at the expense of being able to adequately detect other important cues in the environment (i.e., signs of distress in others) and thereby interfere with the development of perspective-taking skills that could result in the child focusing on the instrumental gain of the act while simultaneously ignoring its harm on others (Kahn, Frick, Golmaryami, & Marsee, 2017; Pollak, 2008; Shields & Cicchetti, 1998).
Our results suggest, however, that this interpretation needs to be made in the context of the finding that the higher rate of aggression and violence in those high on both CU traits and anxiety was limited to self-report. That is, when analyses focused on violent rearrests as reported by official records, there was only an association between violence and CU traits, with no added main effect of anxiety and no significant interaction. These findings are important because much of the past work in this area has used self-report measures or composites that included self-report of anxiety or worry (e.g., Docherty et al., 2016; Kimonis et al., 2011). Furthermore, the one study of adolescents that included measures of other informants (along with self-report) and tested differences between informants (self-report and parent-report) also found no differences between youth elevated on CU traits with and without high anxiety when using other report methods (parent-report; Kahn et al., 2013). It is not clear then whether the youth with high anxiety over-report on their behavior problems or whether those high on CU traits without anxiety under-report the severity of their aggression. In support of the latter interpretation, Kahn et al. (2013) had clinicians rate the credibility of the youth and they rated the youth high on CU traits but low on anxiety as less credible informants compared to those elevated on both dimensions. However, it is important to note that in both our results and in the results of Kahn et al. (2013), the findings do not suggest that CU traits in the absence of anxiety were related to more aggression when not relying on self-report. Use of other sources simply eliminated any additive effects of anxiety.
Importantly, past research on the variants of CU traits in adolescent samples have largely focused on comparing groups of youth with elevated CU traits with and without high levels of anxiety to a non-CU control group (Docherty et al., 2016; Fanti et al., 2013; Kahn et al., 2013; Kimonis et al., 2011; Kimonis et al., 2012). Critically, this methodology made it unclear if the association between anxiety and aggression was specific to those high on CU traits. Our variable centered analyses consistently found main effects for CU traits and anxiety but found very little evidence for interactions (see Table 2). That is, anxiety was associated with more aggression and severe violence irrespective of the level of CU traits, and the high level of self-reported violence in those high on both CU traits and anxiety seemed to be due to additive effects of both predictors. These results would be consistent with the possibility that anxiety may simply be a marker of the severity of behavioral disturbance, rather than an indicator of a group of youth with distinct etiologies underlying their CU traits and aggressive behavior (Frick, 2012). It is still possible that the high rate of trauma experienced by those high on CU traits and high in anxiety could be indicative of a unique etiological process underlying the CU traits for this group (Kimonis et al., 2012). However, being a victim of trauma also showed a continuous association with anxiety, irrespective of the level of CU traits, in our sample. Thus, it is also possible that youth who are more aggressive also are more likely to experience violence directed toward them (Fontaine, Hanscombe, Berg, McCrory, & Viding, 2016) and both anxiety and violent victimization may be better considered as a consequence of the youth’s aggressive and violent behavior rather than indicative of a distinct causal process underlying the youth’s CU traits. It is also possible that the presence of anxiety may be a marker of higher levels of trait disinhibition that are indicative of more severe problems regulating emotions and behaviors more generally (Krueger, 1999; Krueger et al., 2007). As a result, future research should consider this possibility and include a measure of disinhibition.
This difficulty in interpreting potential bidirectional effects highlight a critical limitation of the current study. Our measure of arrests for violent reoffending captured an important but low base rate outcome. Thus, to have enough variability in this outcome to test associations with anxiety and CU traits, we had to dichotomously code any arrest for a violent offense across the 30-month follow-up. Because one of our research questions involved comparing results across self-report and official records, the self-report had to be coded to capture aggression in a similar way across the time period. As a result, we could not fully utilize the longitudinal design to test associations with growth in aggression over time or to test cross-lagged relationships that might support bidirectional effects with aggression. 2 However, even with more sophisticated longitudinal analyses, our study could not detect early abuse and exposure to trauma that may have predated the development of CU traits and anxiety, which would have provided a stronger test of the theory that anxiety is indicative of a distinct causal pathway to CU traits.
In addition, participants in the current study were limited to boys and thus the generalizability of our findings to girls is unclear. This is a particularly important limitation, given that CU traits and anxiety may be more highly correlated in girls (Euler et al., 2015; Humayun, Kahn, Frick, & Viding, 2014). Furthermore, our study focused on a sample of first-time juvenile offenders who had committed offenses of only modest severity. This was an important methodology because it likely increased the variability in both CU traits and aggression relative to a community sample, which would have been more restricted at the upper levels of these variables, and relative to a more severe offending population, which would have been more restricted lower levels of these variables. However, this methodology also means that it is not clear if the associations we found would replicate in other samples of adolescents and juveniles who have committed more severe offenses.
Another important limitation of the current study is that it used a measure of anxiety that solely assessed proneness to worry, instead of trait anxiety, which also includes symptoms of physiological arousal (e.g., tension, jitteriness). This is a theoretically important issue, given that proneness to worry is a narrower construct than trait anxiety and the two aspects of anxiety are differentially correlated with depressive symptoms (Krueger, 1999; Mineka, Watson, & Clark, 1998; Vaidyanathan, Patrick, & Iacono, 2011) and several indices of neural activity (Heller et al., 1997; Moser et al., 2013). Importantly, previous research on variants of CU traits have used a mixture of measures to assess anxiety and distress, with some using measures that included both worry and physiological arousal (Docherty et al., 2016; Kimonis et al., 2012; Kimonis et al., 2013; Kimonis et al., 2017; Sharf et al., 2014; Skeem, Johansson, Andershed, Kerr, & Louden, 2007; Tatar et al., 2012), some using measures of more general negative affect that includes indicators of both anxiety and depression (Euler et al., 2015; Kahn et al., 2013) and others using measures that focused only on worry (Gill & Stickle, 2016; Salihovic et al., 2014). There were no systematic differences in findings across these studies, despite these variations in methodology. However, future research should systematically study whether CU traits relations to important outcomes differs depending on how anxiety is measured.
Finally, a strength of this study was our ability to have a clinically important measure of violence that did not rely on self-report (i.e., official arrest records of violent arrests). However, this measure itself is limited in that many acts of violence may not come to the attention of the justice system (Skogan, 1997). In support of this possibility, the base rate of violent arrests was much lower than the base rate for self-reported violent offending in our sample. As a result, future research should test the influences of CU traits and trait anxiety in the prediction of aggression using a number of different methods (e.g., laboratory measures of aggression, report of parents and peers), recognizing that there are limitations in any single method of assessing violence and aggression. Furthermore, our findings that anxiety did not predict violence using an alternative method than self-report is consistent with past research that also found no differences between the variants of CU traits on externalizing behaviors when using an alternative method of assessment (e.g., parent-report; Kahn et al., 2013).
Within the context of these important limitations, our findings clearly support the association between CU traits and aggression that has been reported in multiple past samples (Frick et al., 2014) and which led to the inclusion of CU traits in the diagnostic criteria for Conduct Disorder in the DSM–5 (American Psychiatric Association, 2013). However, they also place some important caveats on the emerging view that those high on CU traits and anxiety are the most aggressive and violent. That is, the findings for the effects of both CU traits and anxiety on later aggression and violence was limited to self-report measures. Our results alone do not allow a clear way to determine if this is due to over reporting of those high on anxiety or the underreporting of those low on anxiety, although some past work supports the latter possibility (Kahn et al., 2013). They do, however, clearly suggest that future studies testing the associations among CU traits, anxiety, and aggression should include both self-report and other methods for measuring aggression to help in clarifying these findings, as well as using analytic methods that utilize variable centered approaches, instead of relying solely on person-centered approaches. Finally, future research should consider designs that provide for better ways of disentangling bidirectional effects, given the possibility that both anxiety and being a victim of violence could be a consequence of the adolescent’s aggressive behavior. Such designs would be crucial for further determining the utility of the distinction between primary and secondary variants of CU traits.
Footnotes
Acknowledgements
We are grateful to the many individuals responsible for the data collection and preparation.
Action Editor
John J. Curtin served as action editor for this article.
Author Contributions
E. L. Robertson, P. J. Frick, and T. D. Wall Myers developed the study concept. Testing and data collection were performed by E. L. Robertson, J. V. Ray, L. C. Thornton, and T. D. Wall Myers. E. L. Robertson and P. J. Frick performed the data analysis and interpretation, and drafted the manuscript. J. V. Ray, L. C. Thornton, T. D. Wall Myers, L. Steinberg, and E. Cauffman provided critical revisions. All the authors approved the final manuscript for submission.
Declaration of Conflicting Interests
The author(s) declared that there were no conflicts of interest with respect to the authorship or the publication of this article.
Funding
The Crossroads Study is supported by grants from the Office of Juvenile Justice and Delinquency Prevention (2010-JF-FX-0612) and the John D. and Catherine T. MacArthur Foundation (09-94942-000 HCD and 10-95802-000 HCD).