Unveiling common psychological characteristics of proneness to aggression and general psychopathology in a large community youth cohort

Research has consistently linked aggression with a range of psychiatric conditions beyond externalizing disorders, including depression, bipolar disorder, and schizophrenia spectrum disorders. Youths prone to aggression frequently experience poorer general mental health and higher risks for internalizing symptoms, and greater proneness increases the likelihood of aggressive acts under provocation. Although aggression correlates with the total number of lifetime psychiatric disorders, suggesting it as a marker of overall psychopathology, few studies have examined general psychopathology encompassing multiple symptom dimensions alongside proneness to aggression in nonclinical youth. Prior work implicates shared transdiagnostic factors—impulsivity, cognitive control, emotion regulation, negative affectivity, and neurotic/antagonistic personality—in both general psychopathology and proneness to aggression. Neuroimaging evidence also supports overlapping neural substrates for cognitive control, social functioning, and emotion processing across psychopathology and aggression. Adolescence is a developmental period with high prevalence of both psychopathology and aggression, making it critical to understand shared contributing factors to inform prevention and intervention. The primary aim of this study was to investigate the interplay between general psychopathology, proneness to aggression, and contributing factors in a large community youth cohort. The authors hypothesized robust positive associations between proneness to aggression and general psychopathology, overlapping predictive features across outcomes, and successful cross-prediction between models. They further posited that impulsivity, top-down cognitive control, and neurotic/antagonistic personality traits would emerge as shared features.

• Aggression occurs across severe mental illnesses and is associated with the number of psychiatric diagnoses, suggesting a general marker of psychopathology rather than a symptom of specific disorders.
• Dimensional models show strong associations between internalizing symptoms and proneness to aggression, highlighting transdiagnostic relevance.
• Shared transdiagnostic factors—impulsivity, impaired top-down control, emotion dysregulation, negative affectivity, neurotic/antagonistic personality—are linked to both general psychopathology and proneness to aggression in youth.
• Neuroimaging meta-analyses indicate overlapping disruptions in networks for executive functioning, action inhibition, and higher-level cognition across psychiatric disorders and aggression-related phenotypes.
• Despite suggestive evidence of overlap, no prior study systematically examined proneness to aggression, general psychopathology, and a wide set of contributing factors together in nonclinical youth while accounting for inter-variable dynamics.

Design and cohort: Secondary analysis of the Hong Kong Youth Epidemiological Study (HKYES), a population-based cohort of youths aged 15–24 years recruited via multistage stratified sampling since 2016. Data used in this study were queried on 06/21/2021. Interviews used computer-assisted personal interviewing via Qualtrics. Ethical approval obtained; informed consent/assent collected.

Sample: Of 2544 surveyed youths, 2184 were included after exclusions (male n=899; female n=1285; mean age ~20 years). Exclusions: item-wise missing >25% removed at feature level; string-response items excluded; participants excluded if missing any outcome items, age, sex, or >5% missing across all items (n=360 excluded). Final features: 230 items from 29 scales across sociodemographics, cognitive functioning, lifestyle, well-being, and psychological characteristics; none directly measured psychopathology or aggression.

Outcomes:

• Proneness to aggression: 12-item short form of Buss-Perry Aggression Questionnaire (BPAQ), total score (with subfactors: physical aggression, verbal aggression, anger, hostility).
• General psychopathology: first principal component (PCA) across dimensional symptom measures (depression, mania, hypomania, general anxiety, social anxiety, obsessive-compulsive symptoms, psychotic-like experiences, prodromal psychotic symptoms). PCA-based general factor approach per prior literature.

Preprocessing:

• Scale/center numerical variables; dummy-code nominal variables; remove zero-variance features.
• Impute missing data using k-nearest neighbors (k=20).

Data splitting and modeling:

• Dataset split: 3/4 discovery, 1/4 holdout. Within discovery, 3/4 train and 1/4 test.
• Model: LASSO regression with 10-fold CV; λ in [1e-10, 1]; hyperparameter selected by minimizing RMSE.
• Procedure repeated 100 times with random subsampling. For each iteration, recorded λ, RMSE, MAE, accuracy (correlation between empirical and predicted), R², and feature coefficients. Features deemed significant if 95% CI of coefficients across 100 models did not cross zero; insignificant set to zero.
• Mean β coefficients of significant features across 100 runs applied to holdout for validation.
• Overlapping significant features between the two outcome models identified; cross-predictions performed (model trained for one outcome used to predict the other) to assess shared information.
• Sensitivity analysis: Elastic Net with same workflow.

Network analyses:

• Complete-case dataset (n=1673) used. Overlapping significant features were factor-reduced via factor analysis (minimum residual, oblimin rotation); number of factors determined via parallel analysis. Eighteen factors identified.
• Gaussian Graphical Model (GGM) estimated using unregularized model selection (R agraph::ggmModSelect). Variables transformed toward normality using bestNormalize (choosing among multiple transformations by Pearson P statistic). Network visualized with qgraph; edges represent partial correlations; blue positive, red negative; edge width indicates strength. Expected influence computed as centrality measure.
• Stability assessed via nonparametric bootstrapping for edge strengths and case-drop bootstrapping for centrality (bootnet). Flow diagrams illustrated connections from each outcome to features.

Statistical analyses:

• Pearson correlations between proneness to aggression and general psychopathology computed separately in discovery and holdout; correlations between BPAQ subfactors and individual symptom scales; multiple-comparison control via FDR (q<0.05).

• Sample: 2184 youths (41% male), mean age ~19.9 years; discovery n=1638, holdout n=546; no significant differences between splits in age, sex, education, outcomes.
• General psychopathology factor (PCA first component) explained 34.8% of variance; all symptom scales loaded positively.
• Association between aggression proneness and general psychopathology: discovery r=0.56 (95% CI 0.52–0.59, p<0.001); holdout r=0.60 (95% CI 0.54–0.65, p<0.001).
• Correlations between BPAQ subfactors and symptom scales: most significant after FDR correction (q<0.05), except mania, hypomania, and psychotic-like symptoms showing nonsignificant associations with aggression subfactors.
• LASSO selected features: 141 significant predictors for general psychopathology; 157 for proneness to aggression. Overlap: 102 features.
• Holdout prediction performance (all features): • General psychopathology: r=0.793, R²=0.629, MAE=0.767, RMSE=1.06. • Proneness to aggression: r=0.676, R²=0.457, MAE=0.572, RMSE=0.756.
• Models using only 102 overlapping features performed comparably: • General psychopathology: r=0.785, R²=0.616, MAE=0.801, RMSE=1.11. • Proneness to aggression: r=0.659, R²=0.434, MAE=0.587, RMSE=0.784.
• Cross-predictions indicated substantial shared information: • Aggression model predicting general psychopathology: r=0.717 (p<0.001), R²=0.514, MAE=1.05, RMSE=1.39. • Psychopathology model predicting aggression proneness: r=0.606 (p<0.001), R²=0.367, MAE=0.712, RMSE=0.922.
• Sensitivity analysis: Elastic Net performance comparable overall; LASSO superior in cross-prediction, thus preferred.
• Network (GGM) findings (complete-case n=1673): 18 factors derived; impulsivity (F17) exhibited highest centrality; both outcomes showed direct positive associations with impulsivity (F17) and loneliness-isolation (F7). Sleep disturbances (F13) also linked to both outcomes, but only edges from F7 and F17 to outcomes were stably supported by bootstrapping. Centrality stability coefficient for strength was 0.85, indicating high robustness.

The study demonstrates a moderate, robust positive correlation between general psychopathology and proneness to aggression in a large community youth cohort, supporting the transdiagnostic relevance of aggression proneness beyond externalizing disorders. Machine-learning models evidenced overlapping predictive information, as each model cross-predicted the other outcome with moderate to strong correlations, indicating substantial shared variance. Network analyses highlighted impulsivity and social isolation (loneliness-isolation) as key shared psychological characteristics directly linked to both outcomes, with impulsivity emerging as a central node in the network. These findings suggest that impulsivity-related processes (e.g., attentional and motor impulsiveness) and isolation may underlie the co-occurrence of psychopathology symptoms and aggression proneness in youth. Clinically and for public health, targeting these shared factors may yield more effective strategies to reduce aggression risk and improve mental health, rather than focusing solely on psychiatric diagnoses, given that proneness to aggression may more directly influence aggressive behaviors and could confound associations between mental illness and violence.

This study provides evidence of a robust association between general psychopathology and proneness to aggression in nonclinical youth and identifies overlapping psychological characteristics—particularly impulsivity and isolation—that link the two. Predictive modeling and cross-prediction support substantial shared information between outcomes, and network analyses clarify direct connections to key features. These insights can inform prevention and intervention strategies that target common factors to reduce aggression risk and promote better mental health outcomes. Future work should include broader risk factor capture (e.g., passive sensing), external validation samples, and longitudinal or experimental designs to elucidate causal pathways and assess generalizability, including to clinical populations.

• Feature coverage: Despite 230 features across multiple domains, some risk factors may not be captured due to methodological constraints; future studies could incorporate passive data collection.
• External validity: Lack of an external validation sample; results rely on internal discovery/holdout splits within a unique population.
• Causality: Cross-sectional design precludes causal inference among variables.
• Distributional assumptions: GGM assumes Gaussian distributions; some variables (e.g., aggression proneness) only approximately normal despite transformations.
• Variable selection: LASSO may select among collinear predictors and exclude relevant variables; averaging across 100 repetitions mitigates but does not eliminate this issue.
• Generalizability: Findings based on community youths; applicability to psychiatric or other populations requires further study.
• Holistic psychopathology approach may overlook unique characteristics of individual symptom dimensions, suggesting the need to examine both shared and distinct associations.