Loading...
ISSN 1671-3710
CN 11-4766/R
主办:中国科学院心理研究所
出版:科学出版社

Archive

    For Selected: Toggle Thumbnails
    Academic Papers of the 27th Annual Meeting of the China Association for Science and Technology
    Artificial theory of mind in large language models: Evidence, conceptualization, and challenges
    DU Chuanchen, ZHENG Yuanxia, GUO Qianqian, LIU Guoxiong
    2025, 33 (12):  2027-2042.  doi: 10.3724/SP.J.1042.2025.2027
    Abstract ( 927 )   HTML ( 113 )  
    PDF (587KB) ( 910 )   Peer Review Comments

    In recent years, the rapid development of artificial intelligence (AI) has continuously reshaped our understanding of its capability boundaries. Evaluating the theory of mind capabilities of large language models (LLMs) has received significant attention within the research community. Recent studies suggest that LLMs can successfully complete tasks traditionally used to assess theory of mind in humans. However, controversial questions remain: Do LLMs possess theory of mind? What are the essential differences between artificial theory of mind and human theory of mind? Therefore, this systematic review synthesizes the performance of LLMs on theory of mind tasks, reveals essential differences in the internal processes between human theory of mind and artificial theory of mind, refines the conceptual definition of artificial theory of mind, and outlines key challenges in this field.

    Specifically, we systematically synthesize research on artificial theory of mind from the objects of evaluation and the characteristics of tasks. Following the developmental sequence of core theory of mind sub-abilities in humans, we evaluate the task performance of GPT-4. Results indicate that GPT-4 can consistently pass false-belief tasks and various higher-order theory of mind tasks, suggesting a simulation of human-like theory of mind performance. Nevertheless, recognizing that behavioral task accuracy may be insufficient to reflect true capability, we specifically examine cases in which model performance is limited. Our analysis identifies critical limiting factors including external factors (e.g., limitations in test items, prompt design and human baselines) and intrinsic limitations of LLMs (e.g., hallucinations, hyperconservatism, commonsense errors, heuristics or spurious correlations, and spurious causal inference). This suggests that performance fluctuations may not stem from a lack of artificial theory of mind. Consequently, by integrating GPT-4's high accuracy with its intrinsic and extrinsic factors limiting its performance, we demonstrate that GPT-4 has developed artificial theory of mind that is similar in performance to human theory of mind.

    The advanced capabilities of LLMs underscore the importance of studying the internal processes of theory of mind. Multiple lines of evidence suggest fundamental differences in the internal processes between human theory of mind and artificial theory of mind. To fully distinguish them, we conduct a comparison of the neural foundations and developmental factors supporting both. We highlight that their neural mechanisms exhibit distinct complexity across multiple dimensions, and the factors influencing theory of mind acquisition and development differ between humans and LLMs. This reveals the root cause of their internal process differences. Based on this, we define artificial theory of mind as: the simulation of human theory of mind performance exhibited by LLMs during text generation, achieved through recognizing and matching statistical patterns in the text after receiving a theory of mind task prompts.

    However, as an emerging field, existing research faces three main challenges: lack of standardized evaluation protocols, unclear mechanisms of artificial theory of mind, and theory of mind alignment issues. Regarding evaluation, issues in experimental design, prompt design, and scoring methods can lead to divergent results. Regarding mechanisms, while internal processes are key to distinguishing between theory of mind and artificial theory of mind, current research has neither fully elucidated these processes nor sufficiently addressed​ how humans attribute mental states to LLMs. Regarding alignment, LLMs simulate human theory of mind performance without achieving genuine theory of mind reasoning. Based on this, we propose corresponding research directions.

    In conclusion, this study reveals that LLMs possess an artificial theory of mind that is similar in performance to human theory of mind but different in internal processes, refines its conceptual definition, and clarifies key field challenges. It identifies critical issues and delineates future research directions, offering a foundation for leveraging artificial theory of mind research to advance our understanding of human theory of mind emergence and internal processes.

    Figures and Tables | References | Related Articles | Metrics
    Cognitive decision neural networks based on evidence accumulation framework
    CHEN Siyu, PAN Wanke, HU Chuan-Peng
    2025, 33 (12):  2043-2053.  doi: 10.3724/SP.J.1042.2025.2043
    Abstract ( 446 )   HTML ( 55 )  
    PDF (1274KB) ( 428 )   Peer Review Comments

    Reaction time (RT) is a window into understanding human decision-making processes. The Evidence Accumulation Model (EAM) is a dominant computational framework for modeling RT. However, EAMs, such as the Drift Diffusion Model (DDM), offer statistical descriptions of decision outcomes without detailed algorithms for stimulus encoding or neural mechanisms, thereby omitting the algorithmic and hardware levels in David Marr’s three-level framework (computation, algorithm, and hardware). We suggest that these limitations can be addressed by combining Artificial Neural Networks (ANNs) and evidence accumulation models to simulate the entire decision-making process—from stimulus encoding to decision output. These new models, termed Cognitive Decision Neural Networks, enable in silico modeling of human decision-making, providing a novel approach to understanding cognitive processes.

    Cognitive Decision Neural Networks generally consist of three key modules: stimulus encoding, evidence accumulation & decision-making, and reaction time output. The stimulus encoding module encodes sensory data into decision evidence, capturing task-relevant information. The evidence accumulation & decision-making module integrates task-relevant evidence, sets decision rules to make a choice and generates corresponding decision time, capturing both encoding and decision processes. The output module aligns model’s decision time with human reaction time, ultimately producing the final behavioral output.

    Recently, a few models based on ANNs have combined with the evidence accumulation model to simulate speedy decision-making processes in several tasks. The RTNet employs Bayesian neural networks to model uncertainty and reaction time dynamics for handwriting recognition task. RTify, which integrates a recurrent neural network to simulate evidence accumulation under temporally evolving stimulus, fits the human perceptual decision-making RT data well. Both RTNet and RTify use deep neural networks to handle complex visual stimuli; however, spiking neural networks can also be used. For example, SN-DM (spiking neural decision-making model) implements biologically plausible neural coding through spiking neuron populations to investigate neural mechanisms underlying decision processes. All these examples suggest that ANNs provide a new tool for modeling the stimulus encoding, decision-making process, and the neural dynamics underlying the decision-making process.

    While Cognitive Decision Neural Networks show promise in lightweight, interpretable modeling, their generalizability remains limited, particularly in complex tasks such as social or value-based decisions. Future research may explore advanced ANNs’ architectures (e.g., tiny RNNs) or hybrid evidence accumulation mechanisms to enhance flexibility. Integrating multimodal data (e.g., neural, eye-tracking) and embodied AI (e.g., robotic arms) may also improve the performance and help address issues such as the non-decision time. In sum, the Cognitive Decision Neural Network framework outlines new frontiers for modeling human decision-making and offers new insight for understanding human cognition.

    Figures and Tables | References | Related Articles | Metrics
    Conceptual Framework
    How to turn tourists into long-term visitors? A process-based study on tourist ritual perception and its functioning mechanism
    LU Junyang, DENG Aimin, WEI Junfeng, LONG Qianying
    2025, 33 (12):  2054-2068.  doi: 10.3724/SP.J.1042.2025.2054
    Abstract ( 503 )   HTML ( 62 )  
    PDF (734KB) ( 3195 )   Peer Review Comments

    Amidst China’s national strategy for deep cultural-tourism integration, tourist rituals confront a critical paradox: despite their recognized dual function in cultural revitalization and visitor engagement, they consistently fail to convert transient visitation ("volume") into sustained destination loyalty ("retention"). This persistent dilemma originates from three fundamental limitations in extant scholarship: (1) a predominant static analytical perspective that neglects the phased, non-linear evolution of tourists’ ritual perception; (2) fragmented examinations of ritual impacts isolated at individual, place, or group levels, obscuring cross-level transmission mechanisms and synergistic effects; and (3) insufficient theoretical attention to key boundary conditions governing ritual efficacy across heterogeneous contexts. To address these interconnected gaps, this study pioneers an integrated "Design-Perception-Behavior" framework comprising four theoretically interlocked investigations.

    Study 1 systematically develops the first Tourist Ritual Perception Scale (TRPS) grounded in a dynamic process perspective. Through rigorous tripartite data synthesis (ritual designers × destination managers × tourists) and innovative application of retrospective event diaries, TRPS captures perception evolution across three sequential, qualitatively distinct phases: 1) Initial Contact Phase: Centering on environmental immersion triggered by spatial layout, atmospheric cues, and opening rituals that facilitate psychological detachment from daily routines; 2) Deep Interaction Phase: Emphasizing symbolic decoding of cultural metaphors, procedural engagement with ritual scripts, and emotional synchronization with collective rhythms; 3) Meaning Integration Phase: Focusing on cognitive-emotional synthesis, cultural sense-making, and post-experience reflection that transforms ephemeral encounters into enduring meaning.

    This measurement instrument fundamentally bridges the "supply-demand misalignment" by establishing ritual design characteristics—operationalized as contextual (setting/scenography), symbolic (iconography/narratives), and procedural (choreography/rhythm) dimensions—as core antecedent stimuli, while theorizing tourists’ cultural capital as a critical moderating filter shaping perception formation.

    Study 2 (individual level), anchored in embodied cognition theory, theorizes dual parallel mediation pathways through which ritual perception enhances Intention to Extend Stay: 1) Situational involvement (heightened attentional focus and deep emotional absorption in the ritual present); 2) Meaning construction (active symbolic interpretation, cultural reframing, and personal relevance attribution). It further examines participation mode (participatory enactment vs. observational contemplation) as a pivotal moderator that differentially shapes the intensity and valence of these pathways, with participatory modes predicted to amplify embodied effects through kinesthetic engagement.

    Study 3 (place level) integrates authenticity theory to model ritual perception’s influence on retention through two complementary causal chains: 1) Constructive authenticity pathway: Ritual as staged cultural representation → Cognitive appraisal of symbolic fidelity → Place identity internalization → Behavioral commitment; 2) Existential authenticity pathway: Ritual as liminal space for self-actualization → Affective experience of autonomous being → Place identity internalization → Behavioral commitment. The study’s core theoretical proposition examines cultural distance’s nonlinear moderation effect (conceptualized as an inverted U-shape) on the ritual perception-authenticity linkage, positing optimal effects at moderate cultural differences where novelty stimulates engagement without overwhelming cognitive schemas.

    Study 4 (group level), grounded in interaction ritual chain theory, conceptualizes the transformation of ritual perception into retention through: Ritual-perceived emotional energy → Emotional solidarity (emergent collective identity, affective bonds, and shared moral obligations) → Enhanced retention intention. It further theorizes critical contingency roles of ritual type (distinguishing periodic sacred ceremonies from quotidian performances) and identity congruence (tourists’ psychological/cultural alignment with ritual meanings), which may strengthen or weaken the emotional transmission process.

    Collectively, these studies constitute a Multilevel Process Model that advances three transformative theoretical contributions: 1) Temporality Integration: TRPS resolves the "process black box" by enabling granular tracking of perception evolution across ritual stages, overcoming static approaches’ inability to explain intra-ritual variance and delayed effects. 2) Cross-Level Synergy Framework: The model elucidates how micro-level embodied experiences (individual), meso-level authenticity negotiations (place), and macro-level emotional solidarity (group) interact dialectically—with effects potentially amplifying or constraining each other—to co-determine retention outcomes. 3) Contingency Systemization: It synthesizes five key moderators—cultural capital, participation mode, cultural distance, ritual typology, identity congruence—into a unified boundary condition framework that explains divergent ritual efficacy across contexts, providing crucial theoretical scaffolding for context-sensitive design.

    Practically, this research generates actionable pathways for destination governance: 1) Phase-Specific Design Optimization: Targeted interventions aligned with perceptual stages (e.g., enhancing symbolic legibility during Deep Interaction); 2) Resource Prioritization Matrix: Evidence-based allocation toward high-leverage design dimensions (symbolic systems > procedural elements); 3) Cultural Calibration Protocol: Strategic management of cultural distance through tiered interpretation systems; 4) Participatory Engineering Toolkit: Modular tactics to maximize engagement through mode-selective facilitation.

    By bridging design science, cognitive psychology, and sociological ritual theory, this study establishes the first integrated framework to transform ritual experiences from ephemeral encounters into sustained retention drivers. The TRPS instrument and multilevel model offer foundational tools for advancing scholarly and practical frontiers in cultural-tourism integration globally.

    Figures and Tables | References | Related Articles | Metrics
    Meta-Analysis
    Longitudinal changes in students’ learning engagement in China’s mainland (2006~2024)
    ZHANG Zijian, CHEN Jiwen, PENG Shun, WU Jiahui, WANG Siqian
    2025, 33 (12):  2069-2082.  doi: 10.3724/SP.J.1042.2025.2069
    Abstract ( 635 )   HTML ( 37 )  
    PDF (1436KB) ( 11725 )   Peer Review Comments

    Grounded in a sociocultural perspective on learning engagement, combining human capital theory, this study investigated the longitudinal development of student learning engagement in China’s mainland. By incorporating two complementary sub-studies, the research systematically analyzed how three broad categories of societal factors—economic (GDP, Gini coefficient, urban unemployment rate), educational (government spending on education), and internet (internet penetration rate)—influence student engagement levels over time.

    Sub-study 1 employed a cross-temporal meta-analysis of 406 empirical studies, encompassing a cumulative sample of 393,117 participants. The results revealed a significant and sustained upward trend in Chinese students’ learning engagement over the past 19 years (β=0.46, 95% CI [0.21, 0.71], p<0.001). Although engagement levels temporarily dipped in 2020—likely due to the unprecedented disruptions caused by the COVID-19 pandemic—this decline was short-lived, with engagement rebounding and continuing its upward trajectory. Effect size analysis supported this trend, showing medium to large increases in overall engagement (Cohen’s d=0.45), and its core dimensions: vigor (d=0.62), dedication (d=0.57), and absorption (d=0.55). Regression analyses indicated that GDP growth, increasing education funding, and greater internet access were significant positive predictors of learning engagement. Conversely, income inequality (as measured by the Gini coefficient) and urban unemployment rates were not statistically significant predictors in this context.

    Sub-study 2 drawed on longitudinal data from the China Family Panel Studies (CFPS), comprising 14,623 participants. Using multilevel linear regression models, the sub-study 2 validated the meta-analytic findings, confirming a steady increase in student learning engagement over time (β=0.023, 95% CI [0.022, 0.024], p<0.001), with a noticeable inflection point around 2012. While the influence of urban unemployment appeared inconsistent, the remaining societal variables—GDP, education investment, internet penetration rate, and income inequality—showed stable, statistically significant associations with engagement.

    Together, these two sub-studies offered robust, triangulated evidence for a long-term increase in student learning engagement across China’s mainland. By employing different methodologies and data sources to enhance both the internal and external validity of the findings, they jointly highlighted the pivotal roles of economic development, educational investment, and internet connectivity in shaping students’ academic motivation and behavior.

    The study also introduced a novel theoretical contribution: the proposal of “belief-benefit resonance” mechanism. This concept suggested that during periods of rapid economic growth, prevailing cultural values—such as the belief that “knowledge changes destiny”—reinforce the material benefits of education, thereby motivating and sustaining higher levels of student engagement. However, in times of intensified social tensions or inequality, this synergy may break down, potentially leading to disengagement or motivational decline among learners.

    To further achieve a comprehensive understanding of the mechanisms driving shifts in learning engagement, future research could prioritize the following directions. First, this study only selected one kind of measurement tool from diverse tools of engagement for meta-analysis, which might have missed some researches. Future research might incorporate data from other measurement tools to validate the stability and generalizability of the findings. Second, it is imperative to systematically integrate sociocultural and psychological constructs, such as shifting societal values, perceived educational equity, and collective emotional dynamics, alongside conventional macro-level indicators, including GDP, educational expenditure, and internet penetration. This broader analytical lens is essential for capturing the complex, multilevel interactions between contextual forces and student engagement. Finally, researchers are encouraged to adopt advanced methodological strategies, such as Multiverse Analysis (MA) and Specification Curve Analysis (SCA), to rigorously identify the key determinants of academic engagement and to evaluate the stability and robustness of their predictive pathways across alternative model specifications.

    In sum, this study provided a nuanced and comprehensive account of how macro-level societal transformations—including economic growth, educational reforms, and technological diffusion—shape the psychological processes underlying student learning engagement in contemporary China. The findings not only advanced educational psychology theory but also offered timely, evidence-based guidance for educators and policymakers seeking to enhance learning outcomes in rapidly changing social contexts.

    Figures and Tables | References | Related Articles | Metrics
    Emotional speech recognition in children with autism spectrum disorder: A three-level meta-analysis of prosodic, semantic, and integrative deficits
    CHEN Lijun, JIN Yuexin, ZENG Hanhan, JIANG Xiaoliu
    2025, 33 (12):  2083-2104.  doi: 10.3724/SP.J.1042.2025.2083
    Abstract ( 317 )   HTML ( 33 )  
    PDF (2717KB) ( 379 )   Peer Review Comments

    Vocal emotion recognition is a foundational component of effective social communication. However, children with Autism Spectrum Disorder (ASD) frequently experience difficulties in this domain. While prior research has demonstrated that individuals with ASD show impairments in processing emotional information conveyed through speech, there is little consensus on the precise source of this difficulty. Is the core deficit primarily located in prosodic processing, semantic comprehension, or in the integrative mechanisms required to reconcile multiple emotional cues? To address this critical question, the current study conducted a three-level meta-analysis encompassing 47 independent studies, yielding 93 effect sizes and a total sample of 3,142 participants. This meta-analytic framework allows for the modeling of nested data structures and provides a statistically rigorous estimation of overall deficits and their moderators across varied task designs, cultural contexts, and developmental stages.

    The results revealed a significant overall deficit in the vocal emotion recognition performance of children with ASD compared to their typically developing peers (Hedges’ g=-0.71), reflecting a moderate to large effect size. Importantly, the magnitude of the deficit was systematically related to task type: the largest impairments occurred in integrative tasks requiring the simultaneous processing of semantic and prosodic cues (g=-0.90), followed by tasks emphasizing prosodic features alone (g=-0.61), and then by semantic-only tasks (g=-0.49). This pattern offers robust support for the Weak Central Coherence (WCC) theory, which posits that individuals with ASD exhibit a domain-general impairment in integrating multiple channels of information. Even tasks that appear unimodal on the surface—such as recognizing emotional tone or understanding emotional vocabulary—often entail the coordination of subtle subcomponents (e.g., combining pitch, intensity, and duration for prosody, or word meaning, syntax, and context for semantics), which can disproportionately tax integrative processing in ASD.

    Further, a series of moderator analyses revealed key contextual and methodological factors shaping the extent of impairment. Cultural context was a significant moderator of performance differences (p=0.023). Children with ASD from high-context cultures (e.g., many East Asian cultures, where communication often relies on implicit tone and context) exhibited more pronounced deficits in emotional speech recognition (g=-1.03) than those from low-context cultures (e.g., North America or Western Europe, g=-0.60). This finding underscores the importance of considering the sociocultural environment in understanding emotional development in ASD and designing culturally sensitive interventions. Material type also emerged as a powerful moderator (p<0.001). When semantic and prosodic cues were congruent, group differences between ASD and typically developing children were relatively small. However, when cues were conflicting or ambiguous, ASD children's performance deteriorated substantially. This suggests that the integrative challenge—especially in resolving conflicting information—may constitute the most critical barrier to successful emotion recognition in ASD. Contrary to the developmental delay hypothesis, age did not significantly moderate the observed deficits, suggesting that impairments in vocal emotion recognition are not simply a reflection of delayed maturation, but may instead represent a stable and persistent neurocognitive feature of ASD, thereby calling for intervention strategies that move beyond age-compensation frameworks.

    In addition, we observed that the influence of task type interacted significantly with cultural context (p=0.035) and with emotion type (p=0.019). Specifically, children with ASD from high-context cultures demonstrated the greatest difficulties in vocal emotion recognition during integrated tasks, showing the largest performance gap compared to typically developing children when identifying mixed and complex emotions in such tasks. Further exploratory analyses revealed nuanced interactions between task type and diagnostic subtype (p=0.060). For instance, children with Asperger’s syndrome exhibited relatively better performance on semantic tasks, likely reflecting their preserved linguistic capabilities. However, this advantage did not extend to prosodic or integrative conditions, where performance was similarly impaired as in other subtypes. This pattern further supports the idea that superficial language proficiency in ASD may mask deeper deficits in emotional inference, particularly in contexts requiring multimodal integration.

    This meta-analysis provided a nuanced understanding of vocal emotion recognition deficits in children with ASD. The findings demonstrated that the impairment is robust across contexts, highlighting that the greatest difficulty lies in integrating multiple communicative cues rather than in processing prosody or semantics alone. By identifying integrative processing as a critical weakness—and noting the unique patterns associated with cultural context and ASD subtypes—the study provides a foundation for developing targeted, evidence-based strategies to enhance social communicative functioning in children with ASD.

    Figures and Tables | References | Related Articles | Metrics
    The relationship between family cohesion and well-being: A systematic review and meta-analysis
    GU Yipeng, LIU Jinyuan, ZHANG Yali, LI Weihe
    2025, 33 (12):  2105-2120.  doi: 10.3724/SP.J.1042.2025.2105
    Abstract ( 719 )   HTML ( 65 )  
    PDF (801KB) ( 1186 )   Peer Review Comments

    Well-being is a crucial indicator of individual psychological well-being and represents a beautiful aspiration and eternal pursuit for humanity. Numerous studies have explored the relationship between family cohesion and well-being, yet they hold divergent perspectives and yield inconsistent findings. Attachment theory and basic psychological needs theory posited a positive correlation between family cohesion and well-being, whereas the circumplex model suggested that the linear relationship was not significant but rather exhibited an inverted U-shaped pattern. Empirical findings reported correlation coefficients ranging from 0 to 0.73. Therefore, it is necessary to employ meta-analysis to synthesize prior research outcomes and analyze the reasons for these inconsistencies. In Chinese databases (CNKI Journal Database and Master’s/PhD Thesis Database), keywords such as “家庭亲密度” “家庭凝聚力” “家庭功能” or “家庭环境” were combined with “幸福” for searches. In English databases (Web of Science Core Collection, ProQuest Dissertations and Theses, Medline, EBSCO-ERIC, Elsevier SD, PsycINFO, PsycArticles), keywords “family cohesion,” “family environment,” “family function,” “family intimacy” were paired with “happiness” or “well-being” to retrieve literature containing these terms in their abstracts. A total of 42 studies were ultimately included (comprising 45 effect sizes, 31,427 participants, spanning 15 countries).

    In the publication bias test, the funnel plot showed that effect sizes were concentrated in the upper part of the graph and evenly distributed on both sides of the total effect. The result of Egger's linear regression was not significant (intercept=-1.09, 95% CI [-4.63, 2.44], p=0.54). The p-curve test indicated that the curve presented a significant right skewness (Binomial test: p<0.001; Continuous test: z=-37.99, p<0.001), and 43 out of 44 p-values were below 0.025. These results suggest that there is no publication bias. The main effect test indicated that the overall correlation between family cohesion and well-being was 0.35 (95% CI [0.30, 0.40]). This finding supports the perspectives of Basic Psychological Needs Theory and Attachment Theory but contradicts the Circumplex Model, demonstrating a linear relationship between family cohesion and well-being. Families with higher levels of intimacy provide individuals with stable emotional support and a sense of belonging, thereby fulfilling their psychological needs and enhancing well-being. Moderator analyses indicated: (a) Significant moderating effect of well-being measurement tools. The highest effect size was found with ASWBS, while the lowest was with GWB. (b) Significant moderating effect of sampling year (b=0.01, 95% CI [0.004, 0.019]). Effect sizes increased with later sampling years. (c) Significant moderating effect of individualism index (b=-0.01, 95% CI [-0.011, -0.002]). Effect sizes decreased as individualism index increased. (d) Non-significant moderating effects for gender, age, family cohesion measurement tools, and research design.

    This study preliminarily resolves debates among basic psychological needs theory, attachment theory, and the circumplex model, revealing a strong positive correlation between family cohesion and well-being. This indicates that family cohesion is a significant factor in well-being. The results suggest that enhancing well-being requires fostering emotional bonds and deep communication among family members, creating a supportive and understanding family environment, and cultivating a family ethos of respect and inclusivity. Additionally, the measurement of well-being influences its relationship with family cohesion. Therefore, selecting appropriate assessment tools is essential to accurately reflect its associations with other phenomena. Finally, sampling year and individualism index can impact the relationship between family cohesion and well-being. This underscores the need to consider the effects of social changes on family structures and the role of cultural values when seeking to enhance well-being. Mental health professionals and policymakers should design adaptable family support programs that integrate contemporary and cultural contexts. These initiatives should assist family members in establishing strong emotional bonds while respecting individual autonomy, thereby building more resilient family support systems and cultivating greater well-being.

    Figures and Tables | References | Related Articles | Metrics
    The effects of third-party intervention on prosocial behavior: A three-level meta-analysis
    SHEN Yinqi, CAI Yi, WU Jixia
    2025, 33 (12):  2121-2137.  doi: 10.3724/SP.J.1042.2025.2121
    Abstract ( 334 )   HTML ( 32 )  
    PDF (2421KB) ( 337 )   Peer Review Comments

    Third-party intervention refers to the proactive action of an uninvolved observer who, upon witnessing a behavioral event that violates, conforms to, or exceeds social normative expectations, imposes punishment, delivers rewards, or provides compensation to the actor involved. As a crucial mechanism for maintaining social order, third-party intervention has attracted extensive scholarly attention for its potential to foster prosocial behavior. Nevertheless, empirical findings remain inconsistent, suggesting that its effectiveness may depend on complex boundary conditions.

    To systematically examine the strength of third-party intervention’s prosocial effect and its influencing factors, the present study employed a three-level meta-analytic approach, synthesizing 130 effect sizes from 40 empirical studies with a total sample of 10,289 participants. Results of the main effect analysis revealed that third-party intervention has a moderately strong positive impact on prosocial behavior. Moderator analyses further indicated that greater intensity and higher probability of intervention were associated with stronger prosocial effects. This pattern can be explained by several mechanisms: higher intensity and probability increase the anticipated costs of norm violations, enhance the potential benefits of engaging in prosocial actions, and transmit clearer normative signals, thereby fostering prosocial behavior. All forms of third-party intervention—punishment, reward, compensation, and their combinations—significantly promoted prosocial behavior, though the effect of rewards was relatively weaker. This asymmetry may be attributed to the principle of loss aversion, or the higher informational value of negative feedback relative to positive cues. Demographic moderators, including age and gender, were non-significant, suggesting that the prosocial impact of third-party intervention is stable across populations. Likewise, there was no difference between social versus monetary forms of intervention, indicating that both material incentives and social approval serve as fundamental motivation for prosocial behavior. Furthermore, whether the third-party agent was human or computer did not significantly alter prosocial effects. This may be because participants either lacked a strong psychological distinction regarding the agent’s identity, or because they focused more on the behavioral consequences themselves than on the intentions of the interveners. The presence or absence of intervention costs also did not significantly moderate the effect, which may reflect discrepancies between participants’ subjective perceptions and the coding schemes adopted by researchers. Future research should further investigate whether individuals’ prosocial behavior is differentially affected when they can clearly perceive variations in the level of intervention costs. Finally, methodological moderators, including measurement paradigms and control group setting, did not significantly influence outcomes, underscoring the robustness of the overall effect across experimental conditions.

    Theoretical integration of these findings provides support for deterrence theory, focus theory of normative conduct, and indirect reciprocity theory. Specifically, deterrence theory explains the effect of punishment by emphasizing the increased costs associated with norm violations; focus theory highlights the capacity of intervention to draw attention to and strengthen normative awareness; and indirect reciprocity theory stresses the indirect reputational or cooperative benefits generated by intervention behaviors. Rather than being mutually exclusive, these perspectives are complementary, offering a multidimensional framework for understanding how third-party intervention fosters prosociality. Beyond theoretical insights, the findings also carry practical significance. Comparisons of control group conditions revealed that active third-party intervention consistently yielded stronger prosocial outcomes than either the absence of a third party or mere passive observation, underscoring its distinctive role in reinforcing social norms and sustaining cooperation.

    In conclusion, this meta-analysis provides comprehensive evidence regarding the effectiveness and moderators of third-party intervention in promoting prosocial behavior. By clarifying the robustness of its effects, delineating key boundary conditions, and situating findings within complementary theoretical perspectives, the study advances our understanding of how external normative strategies can be leveraged to cultivate prosociality and collective well-being.

    Figures and Tables | References | Related Articles | Metrics
    Research Method
    The design of virtual reality tests for spatial navigation ability
    CHEN Yan, TIAN Xuetao, LUO Fang
    2025, 33 (12):  2138-2155.  doi: 10.3724/SP.J.1042.2025.2138
    Abstract ( 213 )   HTML ( 19 )  
    PDF (658KB) ( 233 )   Peer Review Comments

    Spatial navigation is a critical cognitive ability that ensures the effective functioning of individuals in their daily work and life. Early spatial navigation tests primarily relied on self-report questionnaires, which lacked objectivity and were prone to social desirability bias. Meanwhile, assessing spatial navigation in real-world environments requires large-scale settings, which are time-consuming, labor-intensive, and difficult to replicate. Such assessments are also susceptible to interference from noise, weather, and traffic, making them impractical for broader application. With the continuous advancement of virtual reality (VR) technology, its suitability for spatial navigation research has become increasingly evident. Compared to real-world testing, VR-based assessments are not constrained by physical space or time, offering greater efficiency, safety, and controllability. VR also allows for environmental design and manipulation, provides natural interaction and immediate feedback, and ensures both standardization and ecological validity. This review examines a series of issues related to the integration of virtual reality (VR) technology with the assessment of spatial navigation ability from a psychometric standpoint.

    Firstly, it is essential to determine whether research findings obtained through VR can be generalized to the real world. Currently, this remains a complex and open question, particularly in the field of spatial navigation, where opinions are divided. Some supportive evidence suggests that spatial navigation performance in VR largely mirrors that in the real world, while other studies have identified discrepancies in perception and behavior between VR and real-world environments. At the perceptual level, differences exist in the sensory information available in VR compared to real-world settings, and individuals may exhibit biases in perceiving distance, size, and speed cues in VR. These variations in perceptual cues and processing modes may influence navigation performance. Given current technological limitations, the divergence between VR and reality is inevitable. However, it is worth noting that VR technology is advancing rapidly, and user familiarity with VR devices is steadily increasing, which may mitigate many of the issues identified in earlier studies. Therefore, researchers are advised to specify the hardware and software versions used in their studies and to remain attentive to technological advancements, periodically validating and updating their findings.

    Simulation-based assessment aims to accurately and comprehensively evaluate target abilities by replicating real-world scenarios or tasks. For spatial navigation ability, the fidelity of an assessment tool is reflected in two key aspects: (1) visual presentation, primarily determined by display devices, and (2) interaction methods, particularly locomotion techniques that simulate real-world movement. Previous VR-based spatial navigation studies have employed diverse hardware configurations. While higher fidelity generally enhances immersion, engagement, and performance, test designers must also consider users' familiarity with the devices and their subjective experience. It is crucial to understand how different display devices and locomotion techniques influence performance and whether these effects vary across tasks.

    VR technology significantly expands the range of assessment scenarios and interaction types, offering test designers considerable flexibility. However, it is equally important to ensure the scientific validity of assessment content, procedures, and formats. By synthesizing the design logic of previous spatial navigation tasks, this review concludes that, apart from specialized tasks measuring abilities like perspective-taking, most paradigms follow a two-phase structure: a learning phase followed by a testing phase, with common evaluation tasks targeting three types of spatial knowledge.

    When designing VR-based scenarios and tasks, environmental factors that influence performance must be thoroughly considered. A key advantage of VR lies in its convenience to manipulate environmental variables. Test designers can adjust task difficulty by adding or simplifying environmental elements and modifying task rules. Environmental factors affecting spatial navigation can be broadly categorized into three groups: (1) task-related factors, (2) visual cue-related factors, and (3) spatial layout complexity-related factors.

    As VR technology becomes increasingly integrated into spatial navigation research, more researchers are adopting VR to adapt classical assessment tasks or develop novel, ecologically valid spatial navigation tests. While this approach overcomes some limitations of traditional assessments, it has also raised concerns about the psychometric quality of such tools, which directly impacts the interpretation of research findings. As a novel assessment medium, VR differs substantially from traditional paper-and-pencil tests in interaction methods, data types, variable control, administration scale, user experience, and application scenarios. Given the complexity of VR-based assessments, developers and users should evaluate their quality across multiple dimensions.

    The interaction between humans and VR is an interdisciplinary topic spanning psychology and artificial intelligence. As VR technology evolves at a rapid pace and user familiarity grows, numerous opportunities and challenges emerge. Future VR-based assessments may undergo technology-driven transformations, such as incorporating multimodal data to enable cross-validated, precise evaluations of spatial navigation ability. Researchers in related fields should prioritize the open-source availability of tools and data, allowing assessment instruments to be reused, adapted, and scrutinized by different research teams. Additionally, they should remain attuned to technological advancements, continuously validating and updating their findings accordingly.

    Figures and Tables | References | Related Articles | Metrics
    Regular Articles
    Neurological and computational mechanisms of hallucinations
    YUAN Motong, CAI Yufei, SUN Hongwei, LI Yanyan, WANG Liang
    2025, 33 (12):  2156-2167.  doi: 10.3724/SP.J.1042.2025.2156
    Abstract ( 325 )   HTML ( 31 )  
    PDF (1870KB) ( 346 )   Peer Review Comments

    Hallucinations, defined as perceptual experiences occurring in the absence of external stimuli, remain a central challenge in psychiatry and neuroscience. They are prevalent across multiple disorders, including schizophrenia, Parkinson’s disease, and affective psychoses, and often predict poor prognosis and treatment resistance. While significant progress has been made in identifying brain regions and networks associated with hallucinatory states, the precise neural and computational mechanisms underlying their emergence are still not fully understood. This review synthesizes recent empirical findings and theoretical developments to provide an integrated account of the neurobiological circuits, computational models, and therapeutic interventions related to hallucinations, while highlighting critical challenges and directions for future research.

    At the neural systems level, accumulating evidence implicates a distributed network rather than isolated regions in the genesis of hallucinations. Key structures such as the prefrontal cortex, temporoparietal junction, insula, hippocampus, and parahippocampal gyrus have been consistently linked to hallucinatory phenomena. However, the precise contribution of each region and the nature of their dynamic interactions remain unclear. A major theme emerging from current research is that hallucinations are likely the product of aberrant communication within large-scale cortical-subcortical networks, rather than dysfunction in a single locus. This distributed view calls for multimodal neuroimaging approaches that integrate structural, functional, and temporal information to characterize how interactions among regions give rise to hallucinatory experiences.

    Computational psychiatry has provided valuable tools for conceptualizing hallucinations in terms of predictive coding and Bayesian inference. Within this framework, perception is modeled as the integration of bottom-up sensory evidence with top-down priors. Hallucinations may result from an imbalance in this process, characterized by overweighted priors or diminished sensory precision, leading to misattribution of internally generated signals as external perceptions. Empirical studies support this account, showing that individuals prone to hallucinations exhibit abnormally high reliance on prior beliefs during perceptual decision-making. Neurochemical findings further indicate that striatal dopamine plays a central role in modulating the weighting of prediction errors, with hyperdopaminergic states biasing the system toward prior-driven inferences. The development of hierarchical Gaussian filter models has expanded these insights by allowing simultaneous modeling of priors, learning rates, and belief stability, offering a more fine-grained computational account of hallucinatory states. Importantly, this modeling framework may help distinguish between subtypes of hallucinations, providing a basis for more individualized clinical interventions.

    Non-pharmacological interventions have also become a focus of translational research. Cognitive-behavioral therapy can help patients reinterpret and manage hallucinatory experiences, while electroconvulsive therapy has demonstrated efficacy in some treatment-resistant cases. Among neuromodulatory methods, transcranial magnetic stimulation has shown promise, particularly when applied to the temporoparietal junction. However, clinical outcomes remain inconsistent, reflecting heterogeneity in patient subgroups, stimulation protocols, and neural targets. Recent work highlights the importance of individualized targeting using neuroimaging-guided navigation and computational modeling to optimize stimulation parameters. Such approaches underscore a growing recognition that hallucinations are not homogeneous phenomena and that effective intervention must account for differences in neural circuitry and symptom profiles across patients.

    Looking ahead, several challenges define the next stage of research. The first concerns the need to move beyond region-specific accounts toward a network-based understanding of hallucinations, leveraging multimodal imaging and computational modeling to characterize distributed interactions. Another challenge lies in the lack of standardized paradigms for cross-modal comparison, which hinders efforts to identify shared versus modality-specific mechanisms across auditory, visual, and somatic hallucinations. Progress will require the development of unified experimental frameworks and the incorporation of cross-diagnostic samples to disentangle disease-specific and modality-specific contributions. Finally, establishing causal evidence remains a pressing need. Most current findings are correlational, and future research must employ causal intervention strategies—such as intracranial stimulation combined with behavioral paradigms and computational modeling—to determine the necessity and sufficiency of specific circuits in hallucinatory states.

    In summary, hallucinations emerge from the interplay of distributed neural networks, altered neurochemical modulation, and computational distortions in predictive processing. By integrating empirical neuroscience, computational modeling, and clinical intervention research, this review demonstrates the value of moving toward a more mechanistic and individualized understanding of hallucinatory phenomena. Advances in this direction have the potential not only to refine theoretical models but also to inform the development of targeted therapeutic strategies, ultimately improving outcomes for patients and reducing the societal burden associated with these profoundly disruptive symptoms.

    Figures and Tables | References | Related Articles | Metrics
    Temporal prediction during turn-taking
    SONG Qingyi, JIANG Xiaoming
    2025, 33 (12):  2168-2181.  doi: 10.3724/SP.J.1042.2025.2168
    Abstract ( 193 )   HTML ( 21 )  
    PDF (1254KB) ( 660 )   Peer Review Comments

    Turn-taking involves the rapid alternation of speakers, with gaps between turns averaging ~200 ms. However, producing even a single word requires at least 600 ms, suggesting that speakers must prepare their responses in advance by predicting turn endings. While numerous studies have identified cues that facilitate turn-end prediction, the underlying neural mechanisms remain unclear. During turn-taking, listeners need to predict at what time the current speaker will finish their turn. In this way, turn-end prediction, in its essence, requires listeners’ temporal prediction ability. Thus, we first reviewed the studies about the neural mechanism related to time processing and prediction. Research on neural timing mechanisms distinguishes between millisecond timing, which governs event-based processing, and interval timing, which tracks longer durations (seconds to minutes). These two kinds of temporal process is related to language processing. In a proposed dual-pathway system, temporal processing is achieved in two pathways: the rapid cerebellar transmission is related to the event-based, discrete temporal processing, while the basal ganglia and striato-thalamo-cortical circuit is related to the interval-based, continuous temporal processing. The dual-pathway architecture explores how the brain processes temporal information, providing theoretical support for the neural basis of temporal prediction. However, since the model does not specify which signals enter the timing system, it remains unclear which and how speech cues are utilized to predict turn endings in conversation.

    In the next part, we reviewed the cues that can be utilized by listeners to predict turn-ends. Lexico-syntactic information plays a well-established role, and despite some debate, prosodic features—particularly those at utterance-final positions—are widely recognized as predictive. Additionally, non-linguistic cues such as gaze and nodding contribute to turn-end anticipation. Although previous studies have confirmed that these cues could be utilized by speakers to predict turn-ends, they did not specify the exact role of these cues. To address this gap, we propose a temporal prediction model for turn-taking. In this model, the lexico syntactic information is transmitted linearly to the cortex via ascending pathways, which is then mapped onto meaning. The prediction of lexical-syntactic information then adjusts neural oscillations to anticipate turn endings through cortico-thalamic feedback. Meanwhile, prosodic cues are rapidly processed via the cerebellar pathway, directing cortical attention to the incoming speech signal and setting the dual-pathway system to “predictive mode”. This model integrates different types of cues for turn-end prediction and suggests a possible predictive mechanism. Within this framework, we summarize the limitations of existing research and propose further directions: future studies should 1) examine the role of different cues in dual-pathway architecture in predicting turn-ends, 2) investigate the relationship between individual temporal prediction ability and turn-taking performance, 3) use M/EEG techniques with high-temporal-resolution to study the relative weighting of lexical/syntactic and prosodic cues in turn-end prediction. 4) use free production paradigms to explore multiple cognitive processes involved in turn-taking, including comprehension, content preparation, turn-end prediction, and actual speech production.

    Figures and Tables | References | Related Articles | Metrics
    Regret and its regulation
    JIN Shuai, LIU Sijia, LI Shuang, LIU Zhiyuan, GUO Xiuyan
    2025, 33 (12):  2182-2195.  doi: 10.3724/SP.J.1042.2025.2182
    Abstract ( 740 )   HTML ( 61 )  
    PDF (578KB) ( 1254 )   Peer Review Comments

    As a negative, painful, and self-blaming emotion, regret arises from upward counterfactual thinking—the belief that a better outcome could have been achieved had a different choice been made. Regret is detrimental to mental health and serves as a significant risk factor for the onset of mental health disorders, such as depression. Furthermore, individuals who experience regret often exhibit a tendency toward regret aversion, which can lead to biased decision-making. Given its broad implications, research on regret is crucial for understanding how to promote mental well-being and mitigate decision-making biases. Consequently, this area of research has garnered extensive attention from scholars across various disciplines, including psychology, management, economics, and artificial intelligence.

    Previous studies have employed tasks such as the wheel of fortune task, the devil task, the perceptual decision task, and the electric shock decision task to induce regret. These paradigms typically present participants with multiple options differing in reward or punishment, which elicits counterfactual thinking and subsequent regret. The action effect of regret, the temporal pattern of the experience of regret, and decision justification theory collectively provide a theoretical framework for understanding the factors influencing regret. Expanding on this foundation, the theory of regret regulation proposes strategies for mitigating regret. Furthermore, research has demonstrated that regret involves multiple cognitive and affective processes, including value assessment, reward processing, cognitive control, and emotional expression. Functional magnetic resonance imaging (fMRI) studies have implicated several brain regions in regret processing, such as the orbitofrontal cortex (OFC), ventral striatum, anterior cingulate cortex (ACC), amygdala, and insula, as well as fronto-striatal functional connectivity, suggesting that these areas may constitute a neural circuit for regret. Additionally, electroencephalography (EEG) studies have identified feedback-related negativity (FRN), associated with outcome evaluation, and the P300, linked to emotional salience, as neural correlates of regret processing.

    Regret is highly sensitive to situational factors, including perceived responsibility, social comparison, and adherence to advice. Research has shown that cognitive reappraisal, attentional deployment, anticipation, and neural modulation techniques can effectively mitigate regret. For instance, attentional deployment not only regulates immediate feelings of regret but also, when trained, induces lasting intervention effects that reduce subsequent regret. Prefrontal cortex and alpha oscillation play a key role in this process. Anticipation, meanwhile, enables individuals to psychologically prepare for potential poor decision outcomes, facilitating proactive regret regulation. This mechanism is linked to activation in the dorsomedial prefrontal cortex (dmPFC). Additionally, non-invasive neuromodulation techniques, such as transcranial direct current stimulation (tDCS) and transcranial magnetic stimulation (TMS), applied to regions like the OFC or dorsolateral prefrontal cortex (dlPFC) have demonstrated efficacy in modulating regret. However, empirical findings in this domain remain inconsistent.

    In order to deepen and enrich this research, future studies could explore the following aspects. First, intensive longitudinal designs could be employed to examine the trajectory of regret fluctuations and the intervention effects following the implementation of regulation strategies in daily life, which would provide valuable insights for managing regret and supporting individual mental health in naturalistic settings. Second, research could investigate the characteristics of information transmission (neural oscillation transfer) between different brain regions within the regret-related neural circuits, as well as the dynamic changes in neural oscillation transfer during regret regulation, thereby contributing to a more comprehensive understanding of the neural mechanisms underlying regret and its regulation. Building on this foundation, personalized stimulation frequencies could be applied to key brain regions involved in regret processing by combining electroencephalography with rhythmic transcranial magnetic stimulation, establishing causal evidence linking neural signal modifications to regret intervention outcomes. Further exploration of these critical aspects through empirical investigation may significantly improve decision-making processes across various domains while simultaneously promoting long-term psychological health and well-being.

    References | Related Articles | Metrics
    Suicide exposure: Negative impacts and postvention
    ZHOU Zhongying, WU Caizhi, YUN Yun, XIAO Zhihua, TONG Ting
    2025, 33 (12):  2196-2216.  doi: 10.3724/SP.J.1042.2025.2196
    Abstract ( 487 )   HTML ( 53 )  
    PDF (703KB) ( 589 )   Peer Review Comments

    Suicide exposure refers to exposure to others’ suicidal behaviors. Suicide exposure can cause psychological trauma to suicide bereaved, first responders to suicide incidents, and mental health professionals who previously served the deceased. t may also worsen symptoms in individuals with pre-existing mental disorders or suicide risk, and increase the likelihood of suicide contagion among adolescents and young adults. Thus, it is essential for professionals and researchers to identify vulnerable individuals following exposure, provide appropriate postvention support, and prevent further contagion. Nevertheless, researchers in China’s mainland have given insufficient attention to suicide exposure and postvention. This paper reviewed existing research on the definition and classification of suicide exposure, summarized vulnerable populations and their negative psychological responses along with potential mechanisms, and examined postvention approaches, guidelines, and their effectiveness, limitations.

    In this review, suicide exposure was redefined as “being exposed to suicidal behaviors (including both completed suicide and attempted suicide) of a real or virtual person, while having awareness of basic information such as who, when and where, the method and outcome of suicide”. Meanwhile, suicide affected individual refers to people who experiences short- or long-term psychological distress as a result of suicide exposure. The types of suicide exposure include direct versus indirect exposure, exposure to completed suicide versus attempted suicide, exposure to suicide by close versus non-close others, and real-world versus virtual exposure. Four groups are particularly vulnerable to suicide exposure: suicide bereaved individuals face complex challenges in meaning-making and responsibility attribution, with core experiences of guilt, self-blame, blaming others, and feelings of being blamed; adolescents and young adults are highly susceptible to peer suicidal behaviors, particular attention should be given to multiple suicide exposure in their social networks and social media; individuals with pre-existing mental disorders or suicide risk may respond pathologically to suicide exposure, which may worsen clinical symptoms and prompt a reassessment of suicidal intentions based on others’ suicide outcomes, affecting their suicide decision-making; for those exposed occupationally, direct exposure can cause psychological trauma and negatively impact both their professional competence and overall workforce stability.

    We found that studies on impact mechanisms primarily focused on trauma/grief maintenance among suicide bereaved individuals and suicide contagion mechanisms. Core experiences of suicide bereaved individuals (guilt and blame), along with shame and perceived stigma, showed significant positive correlations with Prolonged Grief Disorder, PTSD, depression, and suicidal ideation. Several theoretical frameworks have been applied to explain suicide contagion: the Contagion model, Symbolic Interactionist Theory (SIT), Social Cognitive Theory (SCT), the Interpersonal Theory of Suicide (IPTS), and the Integrated Motivational-Volitional Model (IMV). Both SIT and SCT emphasized the critical role of “identification”. For instance, individuals with mentalization deficits may fail to differentiate their own psychological states from those of others, resulting in pathological or projective identification with the suicidal individual. This can trigger intense emotional reactions or imitative suicidal behavior. Furthermore, suicide exposure can increase the accessibility of suicidal ideation, acquired capability, and suicidal imagery.

    A substantial body of postvention research focused on individuals bereaved by suicide. Common approaches include peer support groups, CBT-based groups, community-led initiatives, and online resource provision. The most effective mechanism of these interventions lies in facilitating supportive connections for individuals bereaved by suicide, allowing them to experience a sense of understanding and belonging within an empathetic community. Nevertheless, current approaches showed limited sustainability in maintaining long-term therapeutic benefits. For broader exposed populations, some Suicide Postvention Guidelines and Suicide Cluster Response Frameworks have been developed based on crisis intervention models and expert consensus. These guidelines generally recommend adopting the Public Health Model, implementing tiered interventions according to individuals’ level of impact: universal strategies such as social support and psychoeducation for all affected individuals; selective strategies including counseling, peer support, and mutual aid groups for those mildly impacted; and indicated strategies such as psychiatric treatment or psychotherapy referrals for those experiencing significant distress and/or mental disorders. Despite their clinical utility and practical relevance, these guidelines face numerous implementation barriers and lack sufficient evidence regarding their effectiveness.

    Overall, suicide exposure research focused predominantly on suicide bereaved individuals and lacks robust classification systems and risk prediction models for broader affected populations. Previous discussions of suicide contagion mechanisms relied heavily on theoretical inference with limited high-quality empirical evidence. Similarly, postvention efforts primarily targeted suicide bereaved individuals, focusing on interpersonal support and psychoeducational approaches. Professional evidence-based postventions delivered by mental health practitioners remain scarce, and postventions for non-bereaved suicide exposed individuals are largely overlooked. Future research should prioritize non-bereaved suicide exposed individuals and update existing survey instruments and methodologies. Interdisciplinary approaches—such as qualitative studies, longitudinal designs, and big data analytics enhanced by artificial intelligence—are needed to investigate psychological reactions and trajectories following exposure. Additionally, theoretical research on suicide contagion mechanisms requires strengthening, along with developing classification criteria and risk prediction models for vulnerable subgroups. It is also critical to design tailored postvention strategies for different exposure profiles. These advancements will not only enrich postvention research but also improve clinical practices in contagion prevention, ultimately helping the public and relevant stakeholders understand and respond to suicide incidents more effectively.

    Figures and Tables | References | Related Articles | Metrics