Callous-Unemotional Traits and Risk of Gun Carrying and Use During Crime

The sample included 1,215 male youths, from three regions of the United States, who were arrested for the first time. Participants were eligible if they spoke English, were arrested for an eligible offense of moderate severity, and were between 13 and 17 years of age (see Table 1 for characteristics of the sample). Institutional review boards at all institutions approved the study procedures. Parental informed consent and youth assent were obtained at the time of assessment until the participant turned 18, at which point consent was received from the participant. Participants were informed that participation was voluntary, that participation would not influence their relationship with the justice system, that they were able to withdraw from the study at any time without penalty, and that their data were protected by a privacy certificate from being subpoenaed for use in court. Youths completed the baseline assessment within 6 weeks of the disposition date for their first arrest. They were then reassessed every 6 months for 36 months and then again at 48 months (seven follow-up assessments). Additional sample characteristics and study procedures have been published previously (6).

Given the mild to moderate severity of the crimes committed by youths recruited into the study, very few were ever placed into detention. Specifically, 77.2% (N=884) of the sample were never incarcerated at any point in the study, 10.9% (N=133) were incarcerated for between 1 and 6 months (although not necessarily consecutively), 7.5% were incarcerated for between 7 and 12 months, and 8.7% were incarcerated for between 13 and 37 months. When participants who were incarcerated for more than 6 months across the entire 48-month follow-up period were excluded from the analysis, the results were consistent with those reported for the full sample. Thus, results of the full sample are reported here.

To assess demographic control variables, participants’ self-reported age and race/ethnicity were collected. Race/ethnicity was dichotomized such that endorsement of the ethnicity or race was coded as 1, and no endorsement was coded as 0 (i.e., 1=black, 0=not black, 1=Latino, 0=not Latino). Intelligence was assessed at baseline (mean score=88.43, SD=11.59) using the matrix reasoning and vocabulary subtests of the Wechsler Abbreviated Scale of Intelligence (24). Self-reported lifetime offending prior to a first official arrest was assessed at baseline using the Self-Report Offending Scale (23) by asking adolescents if they ever in their life engaged in a variety of illegal behaviors (Cronbach’s alpha=0.76). Items include a range of antisocial behaviors of various types and severity, including “purposely set fire to a house, building, car, or vacant lot,” “been in a fight,” “forced someone to have sex with you,” “stolen something from a store,” and “gone joyriding.” The measure also included three questions related to gun use prior to a first arrest, including “carried a gun,” “shot at someone, where you pulled the trigger,” and “shot someone, where the bullet hit the victim.” Peer delinquency was assessed at baseline using the Peer Delinquency Scale (22), which asks youths to state how many of their friends have engaged in illegal behaviors (Cronbach’s alpha=0.93). To reduce redundancy, we calculated the total peer delinquency score using 11 items, removing two items related to peer gun carrying and ownership. Exposure to violence was measured at baseline using the Exposure to Violence Scale (25), which asks participants if they were victimized by five different types of violence or witnessed someone else victimized by violence. Impulse control was assessed at baseline using the eight-item self-report impulse control subscale of the Weinberger Adjustment Inventory (26) (Cronbach’s alpha=0.73). Parental supervision was measured at baseline using the Parental Monitoring Inventory (27), which assesses how much the caregiver tried to know and actually knows about domains of the adolescent’s life and how often the caregiver required various forms of curfew (Cronbach’s alpha=0.78). Neighborhood disorder around the adolescent’s home at baseline was measured using 21 items to assess the physical disorder and social disorder of the neighborhood (Cronbach’s alpha=0.94) (28).

Multiple imputation was conducted in SPSS, version 25 (IBM, Armonk, N.Y.), using regression-based imputation, with 20 imputations (29). All variables, including covariates, were used in the imputation. The data were missing completely at random (χ²=106.48, df=112, p=0.63). Eight variables had no missing data, and eight variables had some missing data. The percentage of missing data across these variables ranged from 5.3% for parental monitoring to 0.4% for peer delinquency. Among all participants (N=1,215), 11.6% (N=141) had some missing data, which led to 0.77% of all values missing.

To test the main hypotheses, we conducted a series of negative binomial and Firth logistic regression analyses, all of which were two-tailed tests. Negative binomial regressions were utilized when predicting total self-reported gun carrying because this variable was a count with a large number of “0” values, and it followed a skewed, overdispersed distribution, such that the variance of the dependent variable was greater than the mean. Further, as noted above, none of the participants were unable to engage in gun carrying or gun use across the entire follow-up period, and only a few had a reduced opportunity due to short periods of detention. Firth logistic regressions were utilized when predicting self-reported gun use during the commission of a violent crime and arrests for gun-related crimes because of a low base rate of endorsement of these crimes. Firth logistic regression, conducted in R, uses penalized likelihood, which reduces bias in the prediction of low base rate events in maximum likelihood estimation (30). Callous-unemotional traits, peer gun carrying, and peer gun ownership were mean centered and, with their interactions, entered as independent variables.

To assess whether callous-unemotional traits would predict the frequency of gun carrying after a first arrest, we conducted negative binomial regressions. The analysis revealed a significant main effect of callous-unemotional traits (B=0.078, SE=0.004, p<0.001, 95% CI=0.071–0.086). As shown in Table 2, this result remained even after controlling for other known risk factors, including age, IQ, race/ethnicity, self-reported offending prior to first arrest, impulse control, parental monitoring, neighborhood dysfunction, exposure to violence, and peer delinquency. After controlling for covariates, every one-point increase in callous-unemotional traits was associated with a 7.6% increase in the likelihood of carrying a gun.

Next, we conducted a Firth logistic regression to test the prediction that callous-unemotional traits would increase the probability of using a gun during the commission of a serious crime within the 48 months after a first arrest. Again, the analyses revealed a main effect of callous-unemotional traits (B=0.102, SE=0.014, p<0.001, 95% CI=0.075–0.130) that remained significant after controlling for covariates (see Table 2). Specifically, with every one-point increase in callous-unemotional traits, there was a 6.9% increase in the probability of using a gun during a violent crime, after accounting for the covariates.

To examine whether callous-unemotional traits would moderate the relationship between peer gun carrying (and peer gun ownership) and participant gun carrying, we ran two separate negative binomial regressions, first with callous-unemotional traits, peer gun carrying, and their interaction and then with callous-unemotional traits, peer gun ownership, and their interaction. For both analyses, there were conditional effects for callous-unemotional traits and for peer gun carrying and ownership, and a significant interaction. As indicated by the regression coefficients in Table 3, the form of this interaction was the same for both variables. Thus, we plotted only the interaction with peer gun carrying in Figure 1. Specifically, we calculated and plotted the marginal mean differences in peer gun carrying across levels of callous-unemotional traits in the prediction of total gun carrying across the 48-month follow-up period. As predicted, peer gun carrying led to increased adolescent gun carrying only at low to average levels of callous-unemotional traits. That is, at scores below, but not above, 42 on callous-unemotional traits, or two standard deviations above the mean, the association between peer gun carrying was significantly related to adolescent gun carrying.

Firth logistic regression analyses were conducted to test the prediction of use of a gun during a crime following first arrest (Table 3). Again, there were main effects for callous-unemotional traits and both peer gun variables (carrying and ownership), but the interaction was not significant (p<0.06) for callous-unemotional traits and peer gun carrying. Although not significant, the coefficients indicate a form of interaction similar to that reported in Figure 1.