Abstract:

While school bullying is widely recognized as a pervasive group phenomenon, bystanders’ defending behaviors play a critical role in terminating bullying events and mitigating the long-term psychological harm to victims. However, previous research has predominantly relied on static, linear analytic approaches to identify isolated predictors of defending behaviors. These traditional approaches fail to elucidate the dynamic, intrinsic psychological processes underlying why bystanders make divergent behavioral choices in complex, real-world contexts. To address this critical theoretical and methodological gap, the present study shifts from a static trait perspective to a dynamic situational interaction perspective. It aims to construct a novel Three-Stage Psychological Decision-Making Framework to systematically reveal the shaping mechanisms of bystander defending behaviors and develop evidence-based, systemic governance pathways.
The proposed framework conceptualizes bystander defending not as an automatic prosocial reflex, but as a complex, dynamic decision-making process influenced by multi-level factors.
(1) Situational Cue Processing Stage: Bystanders quickly process environmental cues to determine the necessity of intervention, generating an "initial defending intention."
(2) Psychological Calculation Stage: Bystanders engage in deep cognitive processing, systematically weighing perceived social risks (e.g., peer rejection, retaliation), potential benefits (e.g., moral satisfaction, social status enhancement), and their own defending self-efficacy. This calculative stage modifies, strengthens, or inhibits the initial intention, transforming it into a "stable defending intention."
(3) Behavioral Selection and Implementation Stage: Based on the stable intention, bystanders strategically select specific behavioral outputs—opting for direct defending (e.g., confronting the bully), indirect defending (e.g., comforting the victim, reporting to teachers), or passive bystanding. Crucially, the model posits that individual psychological traits and school environmental factors (such as classroom anti-bullying norms) act as boundary conditions. Rather than merely predicting behavior directly, they alter the relative weights of risk, benefit, and efficacy during the psychological calculation stage.
To empirically validate this framework and develop targeted interventions, this study employs a cutting-edge, multi-method approach across three interconnected sub-studies.
Study 1: Factor Identification and Configuration Analysis. Utilizing a mixed-methods approach, Study 1 integrates machine learning algorithms (Support Vector Machine, Random Forest, XGBoost) and SHAP value analysis with qualitative in-depth interviews to accurately identify the average marginal contributions of individual, interpersonal, and environmental factors. Furthermore, fuzzy-set Qualitative Comparative Analysis (fsQCA) is applied to uncover the asymmetric, configurational pathways through which these multi-level factors jointly shape defending behaviors, breaking away from traditional isolated variable analyses.
Study 2: Experimental and Computational Modeling. Study 2 transitions from establishing variable relationships to precise process explanation. Studies 2a-2d utilize traditional scenario-based experiments, employing highly controlled, text-only situational descriptions to isolate and test the effects of specific cues—particularly the severity of the bullying incident itself—alongside risk-benefit assessments and self-efficacy. Building upon this empirical foundation, Study 2e introduces cognitive computational modeling. By constructing a specific utility function that mathematically integrates perceived social risk, potential benefit, and self-efficacy, and utilizing a multinomial logit model, this study maps subjective utility values to exact behavioral choice probabilities.
Study 3: Intervention Simulation and Field Validation. To bridge theoretical modeling and practical application, Study 3 designs and tests evidence-based intervention pathways. Study 3a employs Agent-Based Modeling (ABM) to simulate the dynamic interactions among heterogeneous student agents within a virtual school environment. By manipulating individual and environmental parameters, the ABM identifies the most potent and risk-controllable intervention strategies, significantly reducing real-world trial-and-error costs. Guided by the simulation results, Study 3b implements a quasi-experimental longitudinal field intervention. Utilizing multi-source data (self-reports, peer nominations, and teacher nominations), it rigorously evaluates the effectiveness of optimized school climate interventions in promoting bystander defending and reducing overall bullying prevalence.
This research fundamentally shifts the paradigm of bystander defending research from factor identification to process modeling. By quantifying the psychological calculation process through advanced modeling and simulating interventions via ABM, it offers robust, evidence-based insights for policymakers and educators. It highlights the necessity of developing comprehensive school bullying governance models that prioritize ecological and structural environmental enhancements over mere individual behavioral correction.

Key words: school bullying, bystander defending behavior, shaping mechanisms, governance pathways