Using data from a prospective longitudinal study with a large, population-based sample, we identified developmental trajectories of conduct problems from ages 4–17 years, and investigated links between abuse experienced at different times during development and the derived conduct problem trajectories. In contrast to previous research using developmental trajectories of conduct problems that focused on abuse experienced during childhood [14–17], we used measures covering both childhood and adolescence, which enabled us to explore the impact of abuse timing and persistence. We found that abuse exposure was associated with substantially greater odds of being in the early-onset persistent and adolescence-onset conduct problem classes, particularly when it was present across both childhood and adolescence. We did not find stronger associations between child abuse and membership of the early-onset persistent compared to the adolescence-onset class, which is in contrast to some previous findings [14–16]. We also did not replicate previous findings showing an association between abuse exposure and increased odds of being in the childhood-limited class (compared to the low conduct problems class) [16, 17]. Overall, our findings suggest that conduct problems with an onset in adolescence show similar associations with abuse to conduct problems that emerge in childhood and persist, with any differences between these trajectories being quantitative (i.e., common risk factors) rather than qualitative (i.e., distinct risk factors) in nature.
The current study builds on previous research by examining timing of exposure to child abuse in relation to developmental trajectories of conduct problems. Importantly, our findings support the hypothesis that persistent abuse has a more detrimental effect than time-limited abuse [22]. Thus, in line with the cumulative risk hypothesis, abuse exposure in both childhood and adolescence was associated with greater odds of being in the early-onset persistent and adolescence-onset classes, with effect sizes twice the size of those observed for childhood-only abuse. In addition, different patterns were observed for childhood-only versus adolescence-only exposure when using the aggregate measure of abuse (‘any abuse’). Specifically, whereas childhood-only abuse was associated with increased odds of being in the early-onset persistent and adolescence-onset conduct problem classes, adolescence-only abuse was not associated with membership of any of the elevated conduct problem trajectories. The latter observation runs counter to previous research suggesting that adolescence-only abuse has more detrimental effects than childhood-only abuse [23–25]. On the contrary, the current results indicate that abuse occurring in childhood may be more influential than that occurring in adolescence, suggesting there may be a sensitive period in which abuse is particularly likely to lead to persistent conduct problems. Alternatively, it may be that abuse occurring specifically in adolescence, versus in childhood or in both developmental periods, is experienced differently by the individual or arises for different reasons, given that significant conflict in the parent-child relationship is relatively common (and possibly normative) during adolescence. However, these findings for childhood-only versus adolescence-only abuse were not replicated in an exploratory analysis that examined the impacts of physical and psychological abuse separately. More precisely, adolescence-only abuse also emerged as predictive of these trajectories, alongside the positive associations already identified for childhood-only abuse. In line with many studies published in this field, small cell sizes mean that caution is essential in interpreting these findings. They also prevented us from investigating the specific impact of sexual abuse, which has been consistently linked to adolescent conduct problems [12, 13], because of particularly low frequencies in our sample. As shown in the correlation matrix (see Supplementary Fig. 1), physical and psychological abuse were highly correlated, whereas correlations between these forms of abuse and sexual abuse were much weaker. This may indicate shared risk environments in which both physical and psychological abuse occur, which may explain the similar pattern of effects for these abuse subtypes. Although sexual abuse was most commonly reported in adolescence, it might be less likely to result in conduct problems if experienced within this developmental period. Consequently, the inclusion of sexual abuse in our aggregate measure of abuse may have suppressed associations with adolescence-only abuse. Future studies with larger and/or high-risk samples with a higher prevalence of sexual abuse are needed to further investigate the association between sexual abuse and conduct problem trajectories.
In contrast to the effects observed for the early-onset persistent and adolescence-onset classes, we did not find any evidence of associations between abuse and childhood-limited conduct problems, which contradicts some previous findings in this area [16, 17]. These studies, however, also included child neglect, a form of child maltreatment not investigated in the current study, which may have influenced associations. Alternatively, individual risk factors may be particularly pronounced in these individuals [10], and, thus, neurodevelopmental problems may be more relevant in the etiology of this trajectory as compared to environmental risk factors such as child abuse.
The relationship between child abuse and the early-onset persistent and adolescence-onset conduct problem trajectories may be explained with recourse to social information processing theory [44]. Children with aggressive behavior show biases in social information processing (e.g., hostile attributional biases) [45]. These biases have been shown to mediate the relationship between harsh and abusive parenting and conduct problems [46, 47]. Children may internalize their parents’ aggressive and threatening behaviors, and, as a result, rely on these aggressive schemata in future social interactions. Equipped with this limited repertoire of behaviors, children may struggle to generate non-aggressive responses to situations of conflict and may also evaluate physically and verbally aggressive responses more positively than their non-abused peers [44]. Furthermore, there are well-established bidirectional effects in the relationship between harsh and abusive parenting and child conduct problems [48]. Consequently, children showing conduct problems may become ensnared in coercive exchanges with their parents [49]. By contrast, abuse experienced in adolescence might be less likely to be internalized and viewed as a behavior to emulate, which may explain the null findings for adolescence-only abuse when using the aggregate measure of abuse. Alternatively, the non-significant associations for adolescence-only abuse may partly reflect the relative rarity of abuse only occurring in this developmental period.
First, the findings should be interpreted in the context of limitations relating to our measures of abuse in the current study. A highly varied set of experiences could lead to an individual being classified as having experienced child abuse. This problem is inherent in any measure that attempts to capture something as complex as exposure to adversity in a scale score, but is compounded in cohort studies where low prevalence of child abuse necessitates the use of categorical variables. In addition, child abuse was assessed using retrospective self-report at age 22 years, which may have been subject to recall bias. Prospective and retrospective measures of child abuse often show poor agreement, representing two constructs with limited overlap [50]. However, most instances of child abuse are not reported to authorities [51], which compromises the representativeness of officially documented child abuse cases – the main alternative to self-report. Therefore, while retrospective self-report measures have limitations, it is difficult to develop feasible and ethically acceptable alternatives, particularly in large prospective cohort studies. Second, as already noted, despite this study deriving from a large, representative birth cohort, frequencies of some forms of abuse were low across the different conduct problem trajectories. The findings relating to physical and psychological abuse particularly require replication, and we were not able to examine sexual abuse as a separate category due to very small cell counts for some classes. Third, the present study suffered from high attrition rates, possibly resulting in systematic differences between the sample included in the analyses estimating conduct problem trajectories (n = 10648) and the sample for whom retrospective data on child abuse were available (n = 3345). More precisely, those with missing data were more likely to be male and in the early-onset persistent and childhood-limited conduct problem classes. However, we employed IPW to minimize the impact of this bias by allocating sample weights to complete cases [41], and the findings of weighted and unweighted analyses yielded almost identical results. Fourth, relying on parent-reported conduct problems in adolescence may have underestimated behavioral problems, as parents may be unaware of their child’s antisocial behavior in this developmental phase. However, the use of different informants for conduct problems versus abuse experiences minimizes potential for inflation of effects by informant bias. Moreover, considering the age range of our sample from 4–17 years, neither parent- nor self-report would have perfectly captured conduct problems occurring in both childhood and adolescence. Crucially, we were able to validate our conduct problem trajectories using self-reported measures of antisocial behavior during adolescence, which showed higher rates of antisocial behavior in the elevated conduct problems classes. This information, which is not typically available for studies of this type, supports the validity of our derived trajectories. Finally, the temporal overlap between our derived conduct problem trajectories and measures of child abuse precludes causal inferences. Thus, child abuse may be a risk factor for conduct problems, conduct problems may elicit more harsh and abusive parenting, or both may interact with each other.