Background: The general aggression model has shown that both individual and situational factors can predict aggression. However, past research has tended to discuss these two factors separately, which might lead to inconsistency. This study addresses this gap by examining the importance of each predictor of aggression in a Chinese compulsory drug treatment population and further explores the predictors of aggression in various substance use disorder populations.
Method: Analyses were conducted using a sample of 894 male participants (mean = 38.30, SD = 8.38) in Chinese compulsory drug rehab. A machine learning model named LightGBM was employed to make predictions. We then used a game-theoretic explanatory technique, SHAP, to estimate the effect of predictors.
Results: In the full-sample model, psychological security, parental conflict, and impulsivity were the top 3 predictors. Depression, childhood abuse, and alexithymia positively predicted aggression, whereas psychological security, family cohesion, and gratitude negatively predicted aggression. There were significant differences in the predictive effects of depressants and stimulants. Although the importance of predictors varied between drug-use groups, several individual and situational factors were consistently the most important predictors.
Limitations: All participants in this study were male, and the data were acquired through self-reports from the participants. Domestic and nondomestic aggression are not distinguished. Additionally, our findings cannot support causal conclusions.
Conclusion: This study tested a series of classical theories of the predictors of aggression in China's compulsory drug treatment context and extended the ideas of the GAM to various substance use disorder groups. The findings have important implications for aggression treatment.
Keywords: Aggression; GAM model; Individual and situational factors; Machine learning approach; Substance addiction.
Copyright © 2022 Elsevier B.V. All rights reserved.