Findings from this research uncovered patterns in the relationship between maltreatment and later delinquent and criminal behaviors from adolescence into young adulthood and how these patterns vary by sociodemographics. Specifically, we explored how maltreatment frequency affected the starting point and trajectory over time in predicted offense frequency from ages 12 to 30 and how this pattern varied by sex, race/ethnicity, and sexual orientation.
To answer our first research question, we found that those youth who had experienced maltreatment were more likely to engage in violent offending behavior, a finding backed up by previous research [32,33,34]. We also found that youth who experienced maltreatment were more likely to see a faster increase in the amount of non-violent offending they engaged in. While these are two different components that describe increased offending frequency, they align with our hypothesis that increased maltreatment experiences would be associated with both increased violent and non-violent offending behavior.
This paper also extends our understanding of the relationship between maltreatment and offending later in development. More frequently than exploring the relationship between childhood maltreatment and offending behaviors into adulthood (which has been explored minimally), papers explore the linkage between early childhood experiences of abuse, neglect, or trauma with long-term health outcomes or self-sufficiency [35,36,37]. Prior papers with delinquency outcomes have mostly focused their studies on adolescence or early adulthood (often age 21) [21, 32,33,34, 38]. The analyses here examined whether the decline in criminal behavior that we see in the administrative data extends through the 20’s following maltreatment. Prior papers also tend to have outcomes at specific ages for the whole dataset (e.g., 15–19), and we build on this by having data from respondents at different ages to show the shape of trends from age 12 to 30 and allow for nonlinearity such that we can see that predicted offending frequency peaks at around age 16.
To answer our second question, we explored differences by sex, race/ethnicity and sexual orientation. We did not find differences by race/ethnicity or sexual orientation. Our findings indicate that the link between maltreatment and later offending varies significantly by sex. Specifically, results showed differences in nonviolent offending between males and females, such that, among those who experienced maltreatment, the predicted nonviolent offense frequency was significantly higher for males compared to females. This was contrary to our hypothesis where we expected that even while males commit more offending behavior that the association with maltreatment would be stronger for females. Recent explorations of a similar question have found that the associations can vary across gender by type of maltreatment [39]. These findings have implications for the dialogue surrounding male-perpetrated offending because given recent research into trauma and externalizing behavior [40, 41], understanding males’ experiences of maltreatment could help motivate the provision of needed therapeutic treatment or positive relationships that could reduce negative behaviors [42, 43]. These findings may also shed light on the notions around gender and risky or offending behavior. The stronger relationship between maltreatment and nonviolent offense frequency for males indicates that the higher rate of offending among boys may not only be due to their higher proclivity for risk behavior but also due to an externalizing response to maltreatment. This finding is consistent with previous literature that demonstrates externalizing responses (e.g., delinquency) are more common for males, compared to the internalizing responses (e.g., depressive symptoms) that are more common for females [44, 45].
One important finding in this paper is that there are no differences seen for the relationship between maltreatment and either violent or nonviolent offending by either race/ethnicity or sexual orientation. Previous research with administrative samples has found a linkage by race [12] while other prospective studies also found no linkages between maltreatment and violent behavior by race [46]. We see this as positive in many ways. For instance, the lack of difference indicates that there is not one particular race or sexual orientation where maltreatment is associated with more subsequent offending, violent or nonviolent. More specifically, all youth – regardless of race/ethnicity or sexual orientation – negatively respond to maltreatment. These findings are not necessarily surprising given that it is likely that humans have universal biological and adaptive responses to maltreatment during childhood including how it affects their brains, emotions, and cognitive processes [47,48,49,50]. Rather, they should prompt us to think more broadly about trauma and children’s behavior within the specific context in which they live, allowing us to respond more appropriately to their needs given their specific environmental exposures.
We also hypothesized that LGBQ youth may struggle with their mental health and exhibit more externalizing behaviors [27, 51, 52]. We did not see this in our results. (Note that while we did find small differences in nonviolent offending behavior by sexual orientation that these differences were found overall and were not based on different past experiences of maltreatment. Specifically, youth who identified as heterosexual or homosexual did not report different patterns of offending behavior following experiences of maltreatment than their straight peers. Therefore, while their behavior may be externalizing following other struggles, there do not appear to be differences in externalizing behavior following maltreatment by sexual orientation.) This may indicate either that non-heterosexual youth are doing better overall than we hypothesized and are more similar to their heterosexual peers, or that their struggles are more likely to be exhibited with internalizing symptoms rather than externalizing symptoms [27, 53, 54].
Finally, we hypothesized that we may see differences across race due to different stressors and violence exposure. Despite finding no variation in delinquent or criminal behavior following experiences of maltreatment for adolescents and adults across race, there is substantial evidence for differential treatment after criminal or delinquent behavior occurs. Past studies find that both Black and Latino students are significantly more likely to receive a suspension in comparison to their white counterparts, a discrepancy that appears as early as preschool [55, 56]. This trend continues through adolescence when Black and Latino individuals are more likely to have both contact with police as well as experience arrest and engagement in the juvenile justice system [57, 58]. This is particularly true for boys. Our findings, coupled with past literature, reinforce the need to reexamine areas where inequalities in the trajectory from maltreatment to juvenile delinquency and offending persist so that we can create a more equitable juvenile and adult justice system.
There are several limitations to the analyses. Specifically, while the most recent round of Add Health data is brand new (2016–2018) [59] (we do not use this most recent wave), the respondents are now in their late 30s and early 40s, meaning that the experiences of maltreatment that we are analyzing happened some time ago. Fortunately, reports of childhood abuse and neglect have been declining in the last two decades [60]. This could mean that the relationships we see here may differ in a sample of youth who experienced maltreatment today; however, we also have seen delinquency decrease significantly over the same time period, bolstering the argument that these experiences and behaviors may be intertwined [43].
Additionally, exploring the linkages between specific types and frequencies of maltreatment with specific offending behaviors may be an important next step which we did not do here. Watts and Iratzoqui do look at this by gender in their new paper [39], which explored moderation by gender in how different types of abuse or neglect are associated with different types of delinquency. More research along this strain of questioning could shed light on whether certain types of maltreatment have a stronger relationship with certain types of offending and deserve more attention.
In addition to these challenges, the Cronbach’s alphas for the offending frequency measures were as low as 0.5 at one of the waves, which indicate low internal consistency reliability of our outcome measures, particularly for non-violent offending behavior at Wave I. Previous analyses of offending behavior using these data have constructed similar measures, so we used these measures to remain congruent with the broader field [28]. It makes sense the different behaviors measured by the non-violent offending scale would have lower internal consistency reliability than the violent offending scale as the behaviors in the former cover a wide range of behaviors (e.g., trespassing, theft, and injection drug use). By comparison, the behaviors measured in the violent offending scale seem more conceptually congruent as they all involve violent behaviors. Finally, while we mentioned above that there are pros and cons to self-report data, some research indicates that self-reported retrospective data is more likely to overestimate associations with self-reported outcomes. As our outcomes are self-reported, this is something to consider [61].
There are also strengths to these analyses. First, we are also to stratify by race/ethnicity and sexual orientation because of the sample size, and our data cover nearly 20 years of age. Second, the lack of variation from random effects in the intercept and slope indicate the sample results are well represented by the predicted plots. In other words, if we allowed the predicted lines to diverge to represent groups on one spectrum or the other of the association, the lines would be very close together. Building from these strengths in future research is essential as knowing particularly what experiences are urgently problematic is something that many parents, educators, healthcare providers, judges, and juvenile justice practitioners desperately want to know so that future delinquent behaviors can be prevented. Third, the Add Health study asked respondents how many times a respondent experienced maltreatment, rather than a simple “yes” or “no.” Recent evidence indicates the frequency of maltreatment may matter more than the type of maltreatment, as types of maltreatment tend to co-occur [56, 62].
Additionally, while we discussed the weaknesses above of self-report data, it is important to note here that there are also strengths. Specifically, the rates of both maltreatment and offending behavior are higher in Add Health than in government reports. We likely capture experiences here that were not reported. This may indicate that Add Health was successful at giving adolescent’s a sense of confidence and confidentiality in the survey and allowing them to feel safe self-reporting delinquent or criminal behaviors for which they did not get caught. It also may mean that a young person may have shared an experience they felt happened but under further investigation did not justify government reporting. More importantly though, both child welfare investigations and policing are patterned by socioeconomic status and race [12, 57, 58, 63]. This is important because in this study we can capture youth who did not end up in the welfare or justice systems – who are overwhelmingly youth of color [64] – and therefore can create estimates for the associations for a broader range of youth. This strikes us as particularly important given that race is found to be a significant moderator in other administrative data studies [12] but not in some other prospective studies [46] suggesting that more exploration into the potential for bias here is important. We hope that the results here can be compared to studies of administrative data to better inform the field of potential strengths and biases to using both methods of data collection.
Finally, by using linear mixed effects models, we decreased the models’ vulnerability to endogeneity. There are many potential factors that may be shared predictors of both maltreatment and delinquency, and our data source did not allow us to control for all of them. Other studies have used evaluations or natural experiments to find exogenous patterns but linear mixed effects models, by examining an individual’s change over time, controls for those unobserved factors that are time invariant. This robust method allowed us to look at how these associations change when the frequency and types of maltreatment increased, as well as test for differences by sex, race/ethnicity, and sexual orientation.