This study analyzed how aggression in early adolescence changes over time and examined the role of key variables influencing this change from a longitudinal perspective. The data used in the analysis were panel data from the Korean Children and Youth ...
This study analyzed how aggression in early adolescence changes over time and examined the role of key variables influencing this change from a longitudinal perspective. The data used in the analysis were panel data from the Korean Children and Youth Panel Survey (KCYPS2018), covering six years (2018-2023) from middle school 7th grade to high school 12th grade, with a total of 1,975 adolescent respondents.
First, to identify key factors influencing adolescent aggression, we analyzed SHAP summary plots using the machine learning technique XGBoost. This was used to rank the importance of factors influencing aggression. The results revealed that positive parenting style ranked highest (Mean|SHAP|=0.175) in the SHAP-based variable importance ranking. Therefore, this study analyzed the longitudinal interaction between adolescent aggression and parenting styles, including negative parenting styles (Mean|SHAP|=0.046), which ranked third in importance.
To elucidate the longitudinal relationship between positive and negative parenting styles and adolescent aggression, this study applied the Latent Growth Curve Model (LGCM), Latent Class Growth Analysis (LCGA), and Autoregressive Cross-Lagged Modeling (ARCL) in a stepwise manner. The LGCM analysis revealed a nonlinear pattern of change in adolescent aggression, with a gradual decline over time from the first year of middle school to the third year of high school. Both the initial level variance and the rate of change variance were significant, demonstrating heterogeneous developmental trajectories, with different starting points, rates of change, and directions of aggression across adolescents.
To specifically distinguish these individual differences, a Latent Class Growth Analysis (LCGA) was conducted, resulting in four latent classes identified based on the trajectories of aggression. The low-stable type (23.0%) showed the lowest level of aggression and experienced the highest positive parenting style (average 3.30) and the lowest negative parenting style (average 1.74). The gradual-decrease type (35.3%) showed a pattern of initially high aggression gradually decreasing over time. The high-maintained type (15.1%) showed persistently high aggression and experienced the lowest positive parenting (average 3.02) and the highest negative parenting (average 2.20). The persistently increasing type (26.5%) started with initial low aggression (1.80) and continuously increased, accompanied by a pattern of decreasing positive parenting (3.18→3.03) and increasing negative parenting (1.91→2.03). In particular, the high proportion of the persistently increasing group suggests a unique characteristic related to Korea's competitive educational environment.
To explore the temporal causal structure between positive parenting styles and adolescent aggression, we constructed an autoregressive cross-lagged model (ARCL) that included the covariance between the error terms of the two variables within the same time point. The analysis revealed significant cross-time autoregressive effects for both aggression and positive parenting styles. Positive parenting styles exhibited significant autoregressive effects across all time points, confirming the stability of positive parenting styles for adolescents over time. For aggression, high levels of significant autoregressive coefficients were observed across all paths from Time 1 to Time 6, confirming the relatively stable nature of aggression over time.
An analysis of cross-time interaction revealed a time-asymmetric pattern, with the cross-lagged effects exhibiting different directions and significances across time points. This suggests that parent-child interactions are not simply unidirectional causal relationships, but rather dynamic and reciprocal processes that evolve over time and context. Examining the cross-lagged effect of positive parenting, we found that positive parenting significantly reduced aggression during the transition from Wave 1 to Wave 2 (from 1st to 2nd grade in middle school) (B=-0.077, p<.05). However, it paradoxically increased aggression during the transitions from Wave 3 to Wave 4 (from 3rd grade in middle school to 1st grade in high school) and from Wave 4 to Wave 5 (from 1st to 2nd grade in high school) (B=0.100, B=0.090, respectively, p<.01). This suggests that positive parenting may be perceived as controlling due to the increased need for autonomy in mid-to-late adolescence.
The relationship between negative parenting styles and adolescent aggression was also analyzed using the ARCL model. Negative parenting styles also exhibited significant autoregressive effects at all time points, confirming stability over time. The cross-lagged effect, similar to positive parenting, exhibited a time-asymmetric pattern. During the transition from Wave 1 to Wave 2, negative parenting significantly increased aggression (B=0.108, p<.001). However, during the transition from Wave 3 to Wave 4 and Wave 4 to Wave 5, a paradoxical pattern of decreased aggression was observed (B=-0.245, p<.001; B=-0.110, p<.05). This may reflect a certain degree of normative acceptance of strict parenting in Korea's academic-centric culture, or a process of renegotiation of autonomy between parents and adolescents.
Meanwhile, unlike the cross-lagged effect, which showed inconsistent direction and significance across time points, the concurrent correlation between aggression and parenting style measured at the same time point showed a consistent pattern. A significant negative correlation was found between positive parenting styles and aggression at all time points (r=-0.277~-0.359), and a significant positive correlation was found between negative parenting styles and aggression at all time points (r=0.435~0.455). This means that while positive parenting acts as a protective factor and negative parenting acts as a risk factor cross-sectionally, the longitudinal causal relationship changes dynamically depending on developmental stage and cultural context. In particular, correlation analysis showed that negative parenting (r=0.435) was 1.57 times stronger than positive parenting (r=-0.277), but machine learning analysis showed that positive parenting had a higher predictive importance, suggesting the need for differentiated strategies for preventive intervention and crisis intervention.
In summary, this study used machine learning to identify factors influencing adolescent aggression. The resulting positive and negative parenting styles were then integrated into a longitudinal model, along with aggression trajectories, to conduct a multi-layered analysis of the interactions between positive and negative parenting styles and adolescent aggression. This analysis uncovered the differential impact of parenting styles across aggression trajectories, providing practical implications for early identification and intervention strategies for adolescents at risk for aggression. This study empirically confirmed that positive and negative parenting styles operate as independent and differential mechanisms, and that their effects dynamically change across developmental stages. Furthermore, it revealed that correlations and longitudinal causality can exhibit different patterns, emphasizing the importance of considering longitudinal causality, not mere correlation, when developing intervention strategies. Furthermore, this study proposed a research framework that views adolescent aggression not simply as a temporary problem but as a long-term, structural developmental process. This study provides empirical evidence supporting the need for integrated family-level interventions and tailored policies for each group.