Stressors are pervasive in contemporary society and contribute to increased vulnerability to stress-related disorders. Strengthening resilience is crucial for reducing health risks. Among protective factors, self-compassion has been proposed as a promising resource for promoting resilience. This argument is supported by the tripartite model of affect regulation, which posits that self-compassion activates the soothing system and down-regulates the threat system. This affect regulation process is considered a central mechanism underlying resilience within the affect regulation framework of psychological resilience. However, empirical findings on the relationship between self-compassion and resilience remain scattered and inconsistent, with some studies reporting robust associations, while others showing weak or nonsignificant results. The inconsistency raises questions regarding the overall strength of this association and the boundary conditions. To address this gap, the present meta-analysis aimed to quantitatively synthesize both correlational and causal evidence on the relationship between self-compassion and resilience, and to examine potential moderators related to population characteristics, contextual factors, operationalization of resilience, self-compassion types and components, and study design, guided by the systemic model of resilience.
A systematic literature search was conducted across six databases: CNKI (China National Knowledge Infrastructure), Web of Science, PsycINFO, MEDLINE, Scopus, and CINAHL. Two reviewers independently screened the studies and extracted data. A three-level meta-analysis was performed in R 4.2.2 to estimate pooled effect sizes and examine moderators. Study quality was assessed using the JBI critical appraisal tools. Publication bias was evaluated using Egger-MLMA regression and the fail-safe number, and sensitivity analyses were conducted with multiple diagnostics, including studentized deleted residuals, Cook’s distance, and DFBETAS. Results indicated a low risk of publication bias and robust findings.
A total of 124 studies, comprising 128 independent samples and 296 effect sizes from 44, 775 participants, were included in the meta-analyses. Self-compassion was positively correlated with resilience (r = 0.46, 95% CI [0.42, 0.49]), and self-compassion interventions significantly enhanced resilience (Hedges’ g = 0.49, 95% CI [0.32, 0.65]). Moderator analyses revealed that the relationship between self-compassion and resilience was influenced by population characteristics (health status), contextual factors (adversity types, specific adverse events, and cultural backgrounds), resilience operationalization (orientation, measurement and domains of resilient outcomes), and intervention formats. Specifically, stronger associations were observed in cultural contexts more distant from Chinese culture, under social adversity (particularly stigma and interpersonal violence), and when resilience was conceptualized as a trait and assessed as psychological or overall resilience. Different types and components of self-compassion exhibited differential associations with resilience: uncompassionate self-responding was more strongly associated with resilience than compassionate self-responding, whereas common humanity showed the weakest association among the six components. In intervention samples, effects were more pronounced among individuals at higher health risk and when the intervention was delivered in blended online-offline formats. Age, gender, and study design did not moderate their relationship.
The findings support the proposed affect-regulation model of self-compassion in promoting resilience (Affect-Regulation Model, ARM), advancing theoretical understanding of the relationship between self-compassion and resilience. Practically, the results provide evidence for targeting self-compassion in resilience-enhancing interventions, highlight the need for targeted and context-sensitive approaches, particularly for vulnerable populations and under social adversity, and underscore the importance of culturally adapted interventions. Future research should give greater consideration to different operationalizations of resilience, its multidimensional nature, and more fine-grained moderators.

Stress is a pervasive experience that affects individuals across multiple domains of daily life. A survey in the Lancet estimated that more than 500 million individuals worldwide experience stress-related mental health problems annually. Consequently, public concern has grown substantially, making the identification strategies to enhance mental health and resilience in the context of stress an urgent priority. Dose-response theory posits a linear relationship where stress negatively impacts mental health—a perspective widely held by both the public and academia. However, over the past 15 years, research in positive psychology has suggested that moderate stress may foster resilience and adaptive psychological outcomes. According to stress inoculation theory, stress and mental health follow an inverted U-shaped trajectory. Thus, no consensus exists between the linear and curvilinear models.
A systematic search was conducted across Web of Science, ProQuest, EBSCO, and PubMed, as well as the Chinese databases CNKI and WanFang Data to identify empirically studies reporting quadratic effects of stress on mental health. Data were analyzed in R using the metafor package. A three-level meta-analysis synthesized linear (r) and semi-partial (sr) effect sizes and tested for moderators. Publication bias was assessed using funnel plots, Egger-MLMA regression, and the trim-and-fill method. A total of 23 studies (yielding 59 linear and 59 quadratic effect sizes) comprising 31, 422 participants were included to compare the two hypotheses.
Results indicated a moderate negative correlation for linear stress terms and a small negative partial correlation for quadratic terms. Overall, the linear association between stress and mental health remains dominant. However, given statistical power constraints, these findings offer some support for the curvilinear hypothesis, suggesting a potential curvilinear relationship between stress and mental health. Meta-regression analyses further revealed that stress measurement methods significantly moderated the curvilinear association, while population characteristics and types of mental health outcomes moderated the linear relationship. Specifically, stress was negatively associated with mental health in healthy populations, but this effect was non-significant in trauma-exposed groups. When outcomes reflected negative mental health, stress consistently predicted poorer outcomes, but showed no significant linear effect on positive mental health outcomes. By contrast, no statistically significant moderation was detected in association between the quadratic component of stress and mental health. Nevertheless, substantial heterogeneity was observed, indicating that future research should pay greater attention to nonlinear patterns linking stress and mental health and the conditions under which such patterns emerge. In particular, more attention should be given to the potential for individuals to fulfill their potential and experience positive growth under moderate levels of stress.
These findings indicate a small but significant negative curvilinear relationship between stress and mental health, suggesting that moderate levels of stress may promote more favorable psychological outcomes. The variability of the linear relationship across populations underscores the importance of considering sample characteristics and outcome measures in future research. A more precise understanding of these mechanisms and boundary conditions is critical to clarifying the complex role of stress on mental health.

Childhood environmental unpredictability, defined as the spatial-temporal variation of adverse experiences, has been identified as an essential feature of adversity that has important developmental implications. Prior work has linked environmental unpredictability to mental health problems and dysregulation of the hypothalamic- pituitary-adrenal (HPA) axis. However, most studies have treated unpredictability as a static construct, overlooking variability in environmental experiences over time. Although the distinct effects of the two dimensions of childhood adversity (i.e., threat and deprivation) on adolescent outcomes are well-established, the unique characteristics of unpredictability within these dimensions remains poorly understood. Furthermore, resilience factors that may buffer the negative effects of environmental unpredictability have received little attention. To address these gaps, the present study investigated how unpredictability related to threat (i.e., abuse) and deprivation (i.e., neglect) predicts adolescents’ diurnal cortisol rhythms and depressive symptoms, and whether a positive school climate moderates these associations.
The sample consisted of 211 adolescents (T1: M_age = 10.82 years, SD = 0.85; 34.6% girls) who participated in a four-wave longitudinal survey spanning two years. Linear mixed models (LMMs) were estimated to capture trajectories of adversity exposure, and the residuals (the differences between predicted and observed values) at each time point were used to compute the root mean squared error (RMSE) as an indicator of unpredictability in abuse and neglect. Linear regression analyses were then conducted to test both main and interaction effects. For significant interaction effects, the Johnson-Neyman technique was applied to plot the simple slopes across the full range of the moderator.
Results indicated that unpredictability in abuse significantly predicted flatter diurnal cortisol slopes but not depressive symptoms. A significant interaction with average abuse levels was observed, such that unpredictability in abuse was associated with fewer depressive symptoms when average abuse exposure was high. Unpredictability in neglect showed no significant associations with either diurnal cortisol rhythms or depressive symptoms. School safety and order moderated the link between abuse-related unpredictability and diurnal cortisol slopes, such that unpredictability in abuse was associated with flatter slopes only when perceived school safety and order were low. Similarly, school support and acceptance moderated the association between neglect-related unpredictability and diurnal cortisol slopes, with effects evident only when school support and acceptance were low.
Findings revealed distinct effects of unpredictability within threat (i.e., abuse) and deprivation (i.e., neglect). Specifically, abuse-related (but not neglect-related) unpredictability predicted flatter diurnal cortisol rhythms, above and beyond the effects of average abuse levels. Moreover, the impact of abuse-related unpredictability varied by the average levels of abuse exposure, highlighting the importance of considering broader contextual features including environmental severity, in understanding developmental consequences. A positive school climate has emerged as a resilience factor, with significant implications for both educational practice and policy. Enhancing school safety and order may be particularly effective in buffering the negative consequences of abuse-related unpredictability for adolescent health, whereas strengthening teacher-student support and acceptance may better mitigate the negative effects of neglect-related unpredictability.

Childhood adversity is one of the most important environmental risk factors that leads to stress cardiovascular dysregulation, which is innervated by the autonomic nervous system. However, few studies have examined the interaction effects of environmental factors and genetic polymorphisms on cardiovascular reactivity to stress among adolescents. Adopting a multilocus genetic profile score (MGPS) approach, the present study aimed to explore the associations between childhood adversity and adolescent cardiovascular reactivity to stress, and test the moderating role of the autonomic nervous system genetic polymorphisms.
Two independent samples (sample 1 included 256 adolescents, M_age =12.95 ± 0.79 years old; Sample 2 included 358 adolescents, M_age =12.98 ± 0.74 years old) were recruited from two middle schools in northwest China. Cardiovascular reactivity was induced by a speech task and a mental arithmetic task in sample 1 and sample 2, respectively.
Results showed that childhood adversity was negatively associated with cardiovascular reactivity to stress, and these associations were moderated by MGPS. Specifically, compared to adolescents with lower MGPS, those with higher MGPS exhibited lower (or blunt) cardiovascular reactivity to stress when exposed to high childhood adversity, and exhibited greater cardiovascular reactivity to stress when exposed to low childhood adversity.
Findings suggest that childhood adversity and autonomic nervous system genetic polymorphisms interactively impact stress cardiovascular function among adolescents, and have implications for interventions targeted at improving stress cardiovascular function in adolescents with childhood adversity.

Cyberbullying is a prevalent form of online aggression that threatens mental health, especially among adolescents and young adults. Compared with traditional bullying, it is more persistent, less visible, and harder to control, often leading to anxiety, depression, stress, social withdrawal, and reduced well-being. Psychological resilience is an important protective resource that helps victims coping with adversity. Among the psychological factors related to resilience, self-esteem and self-compassion are especially relevant because they shape how individuals evaluate themselves and respond to suffering. Existing interventions for cyberbullying victims often lack personalization, emotional interaction, and scalability. Against this background, the present research examined whether large language model (LLM)-facilitated dialogue interventions could enhance resilience in cyberbullying victims, and whether an intervention targeting self-esteem and self-compassion would be more effective than general psychoeducation.
The research consisted of two independent studies. Study 1 used a three-arm randomized controlled design with 59 participants who reported previous cyberbullying experiences. They were randomly assigned to a self-esteem/self-compassion dialogue condition, a psychoeducational dialogue condition, or a psychoeducational reading control condition. Participants completed baseline measures, received a 10- to 15-minute intervention, and then completed posttest assessments. Primary outcomes were state resilience and trait resilience, and secondary outcomes included anxiety, depression, stress, and well-being. Study 2 recruited a new sample of 105 participants and compared the two dialogue conditions only. In addition to the outcomes used in Study 1, Study 2 assessed state self-esteem and self-compassion and applied network intervention analysis to examine the mechanisms underlying the two interventions. Both studies used a WeChat mini-program connected to ChatGPT-4o, and all procedures were completed with informed consent and ethical approval.
In Study 1, repeated-measures analyses showed that both dialogue-based LLM interventions improved resilience-related outcomes more clearly than the reading control. For state resilience, the time × group interaction was significant, indicating that change over time differed across conditions. After baseline state resilience was controlled, posttest state resilience in the self-esteem/self-compassion group was significantly higher than that in the reading group. Both dialogue groups showed significant within-group increases in state resilience, but the self-esteem/self-compassion group showed the largest improvement. For trait resilience, only the time effect was significant, suggesting a general pre-post increase without a clear differential group effect. Regarding secondary outcomes, anxiety and depression decreased over time across groups. Stress and well-being showed significant time × group interactions: stress significantly decreased only in the self-esteem/self-compassion group, whereas well-being significantly increased in both dialogue groups but not in the reading group. In Study 2, the overall intervention pattern was broadly consistent with Study 1, supporting the stability of the main findings. Network intervention analysis further showed that, compared with the psychoeducational dialogue, the self-esteem/self-compassion dialogue was uniquely linked to lower isolation, higher mindfulness, higher social self-esteem, and a direct increase in state resilience. These nodes formed a pathway through which changes in self-esteem and self-compassion jointly contributed to resilience enhancement.
Taken together, the findings showed that brief LLM-facilitated dialogue interventions can improve the psychological functioning of cyberbullying victims, and that a targeted intervention focusing on self-esteem and self-compassion has greater specificity in both outcomes and mechanisms. The results suggest that resilience, particularly state resilience, can be strengthened in short-term digital intervention contexts. They also indicate that self-esteem and self-compassion contribute to resilience through interrelated processes involving social self-evaluation, mindfulness, and reduced isolation. In practical terms, LLM-based interventions may provide scalable and responsive support for individuals affected by cyberbullying.

Binocular rivalry is a phenomenon wherein the two eyes receive conflicting visual images, resulting in alternating perceptual dominance rather than image fusion. It offers a unique insight into the neural basis of visual consciousness, as the visual input remains constant despite fluctuations in subjective awareness. A fundamental question in binocular rivalry research concerns identifying the visual pathway along which the competition occurs. According to early theories, rivalry primarily involves monocular neurons in the primary visual cortex (V1) or the lateral geniculate nucleus via interocular inhibition mechanisms. However, electrophysiological studies in monkeys showed stronger correlations between neural activity in higher visual areas, e.g., the inferotemporal cortex, and perceptual states, indicating competition at object-level representations. Since rivalry occurs across multiple hierarchical levels through reciprocal interactions, the hybrid model sought to reconcile these findings. Nevertheless, a critical question remains unaddressed: How might different visual conditions alter the sites at which rivalry competition is most prevalent? Although neuroimaging studies hint at this possibility, there is currently no systematic causal evidence, and this prevents us from understanding whether the brain flexibly recruits different neural resources based on stimulus characteristics.
This study employed transcranial direct current stimulation (tDCS) to systematically examine the causal roles of four candidate brain regions, namely, the occipital cortex, the right parietal cortex, the left parietal cortex, and the prefrontal cortex, across two stimulus types representing different representational levels. Eight independent experiments were conducted based on a 2 × 4 design: two stimulus types (orthogonal gratings as local-feature stimuli vs. face/house images as object-level stimuli) × four brain regions. Each experiment used a within-subjects design with three tDCS conditions, namely, anodal stimulation, cathodal stimulation, and sham stimulation. A total of 160 participants were recruited across experiments (18~22 per experiment; mean age = 20.8 years; all right-handed with normal or corrected-to-normal vision). We used high-definition tDCS with a 4 × 1 electrode montage—comprising one central electrode and four return electrodes—positioned over target regions based on the international 10-10 system. Stimulation intensity (maximum 1.0 mA) was applied for 20 min during each session. Participants completed binocular rivalry tasks using a mirror stereoscope while continuously reporting their perceptual states via key presses. Two dependent variables were measured: eye dominance strength (reflecting competitive bias and relative gain) and perceptual alternation rate (reflecting destabilization and switching frequency).
The results revealed a robust stimulus-dependent pattern of regional involvement. For grating stimuli (Experiments 1~4), occipital tDCS significantly changed eye dominance strength, F(2, 38) = 7.64, p = 0.002, η²_p = 0.28, 1 - β > 0.99. Right parietal tDCS also changed eye dominance, F(2, 38) = 5.36, p = 0.009, η²_p = 0.22, 1 - β = 0.99, and influenced the alternation rate, F(2, 38) = 3.71, p = 0.034, η²_p = 0.16, 1 - β = 0.99. Neither left parietal nor prefrontal tDCS affected grating rivalry. In marked contrast, for face/house stimuli (Experiments 5~8), left parietal tDCS significantly changed eye dominance, F(2, 38) = 3.97, p = 0.027, η²_p = 0.17, 1 - β = 0.99. Prefrontal tDCS also altered eye dominance, F(2, 38) = 4.06, p = 0.025, η²_p = 0.18, 1 - β = 0.99. Notably, neither occipital nor right parietal stimulation affected face/house rivalry. This double dissociation indicates that simple grating rivalry primarily engages the occipital cortex (early interocular competition) and the right parietal cortex (spatial attention modulation), whereas object-level rivalry recruits the prefrontal cortex (cognitive control) and the left parietal cortex (object feature integration).
This study provides causal evidence for a stimulus-dependent, flexible configuration of rivalry sites. Instead of employing a fixed neural network, the brain flexibly allocates hierarchical neural resources based on the stimulus representational level. These findings advance the hybrid model from a descriptive multi-site framework to a predictive theory that specifies how stimulus properties determine the dominant competitive loci, thereby offering new insights into the flexible neural architecture underlying visual consciousness.

Spatial visualization is a core component of spatial cognition. Compared with traditional cognitive training, novel interventions that integrate physical activity with cognitive processing offer new directions for enhancing special visualization, particularly for individuals with poor spatial ability. However, there is currently a lack of behavioral and neural evidence empirically assessing the effectiveness of these interventions.
The aim of this study was to investigate the facilitative effects and neuroplasticity mechanisms of virtual reality (VR)-based motor-cognitive dual-task training on the spatial visualization ability of individuals with poor spatial ability. A two-factor mixed experimental design was adopted with four groups (VR motor-cognitive dual-task, VR motor, VR cognitive, control) and two time points (pre- and post-intervention). Using the Santa Barbara Sense of Direction Scale (SBSOD), 120 college students with poor spatial ability (SBSOD < 3) were screened and randomly assigned to one of the four groups (n = 30 each). All intervention groups underwent an 8-week training program. Spatial visualization ability was measured using a mental rotation task, while functional near-infrared spectroscopy (fNIRS) was used to record changes in prefrontal-parietal network activation before and after the intervention. A random forest machine learning model was then applied to identify the discriminative features of the neural patterns.
The key findings were as follows: (1) At the behavioral level, the VR dual-task group significantly outperformed the other three groups on post-test measurements of spatial visualization. No significant difference was observed between the two single-task training groups, but both performed significantly better than the control group. (2) At the neural level, the VR dual-task group exhibited the highest neural efficiency, with significantly lower activation levels in the left frontopolar area (L-FPA) and right primary motor cortex (R-M1) after the intervention compared with the other three groups. Those in the dual-task group also showed the greatest reduction in activation from pre- to post-test. In contrast, although the right frontopolar area showed significantly lower activation in all three intervention groups relative to the control group, no significant differences were observed among the three intervention groups. (3) The machine learning model further validated that the neural activity patterns of the VR motor-cognitive dual-task group were highly separable from those of the control group. Changes in R-M1 and L-FPA activity were the most discriminative features.
In conclusion, VR-based motor-cognitive dual-task training can significantly improve spatial visualization ability in individuals with poor spatial ability by synergistically optimizing neural efficiency in the prefrontal and motor cortices. This study provides multimodal evidence for the “body-cognition integration” theory and establishes a methodological foundation for precision intervention paradigms based on neural markers.

Creativity arises from the interplay between knowledge structure and cognitive control. This study aims to systematically investigate the role of semantic network organization in creative thinking across multiple psychological levels. Specifically, it examines how the topological properties of semantic networks relate to both creative ability and creative personality, and its potential neural correlates. By adopting a three-part design, this work seeks to bridge cognitive, trait, and neurobiological perspectives to provide a more comprehensive understanding of how knowledge organization supports creative thinking across different psychological levels.

A well-established cognitive feature of depression involves a working memory bias towards negative stimuli. Due to limited capacity, working memory requires filtering efficiency—a selective mechanism that dynamically coordinates the admission of relevant information and the exclusion of irrelevant input. Whether depression impairs filtering efficiency and subsequently leads to working memory bias remains unclear. According to the subcomponents model of depression, the processing bias towards negative emotional information and mental imagery maintain depressive states. However, researchers have overlooked the role of mental imagery in working memory bias research. Given that mental imagery and working memory share overlapping neural representations in the primary visual cortex, depression likely influences filtering efficiency through its effect on mental imagery.
Depressive tendency, which refers to a subclinical depressive status between clinically diagnosed depression and a healthy state, can predict the onset of depressive disorders two years later. This predictability makes individuals with these tendencies an ideal target for early detection and intervention. Therefore, the present study included three experiments to investigate the moderating role of mental imagery in the relationship between depressive tendency and the filtering efficiency of task-irrelevant emotional stimuli. A filtering change detection task with emotional face stimuli was employed to explore this relationship. In Study 1, participants (n = 108) completed the Center for Epidemiologic Studies Depression Scale (CES-D), the Vividness of Visual Imagery Questionnaire and the filtering change detection task. The experimental design included two emotional filtering conditions (two neutral targets with two positive distractors and two neutral targets with two negative distractors) and two nonfiltering conditions (two neutral targets and four neutral targets). These conditions allowed for exploration of the relationship amongst depression, filtering efficiency across emotional valences and the vividness of mental imagery. Furthermore, the study classified participants into a depressive tendency group (CES-D scores ≥ 16, n = 59) and a control group (CES-D scores < 16, n = 45) to examine the influence of depressive tendency on filtering efficiency. Study 2 (n = 25 in the depressive tendency group, n = 25 in the control group) and Study 3 (n = 24 in the depressive tendency group, n = 25 in the control group) examined the effects of positive and negative mental imagery on filtering efficiency for negative distractors, respectively.
Results showed that depression was negatively associated with mental imagery ability and the filtering efficiency of negative information. Individuals with depressive tendency exhibited lower filtering efficiency for negative distractors than for positive distractors; however, no such difference appeared in the control group. Compared with neutral mental imagery, positive mental imagery increased the filtering efficiency of negative information for individuals with depressive tendency but not for controls. Conversely, negative mental imagery decreased the filtering efficiency of negative information in the control group but did not affect individuals with depressive tendency.
These findings indicate that mental imagery moderates the relationship between depression and the filtering efficiency of task-irrelevant negative information. Specifically, positive mental imagery can guide individuals with depressive tendency to ignore task-irrelevant negative information, thereby improving their filtering efficiency. These findings expand the depression subcomponent model and provide a new perspective for early intervention for individuals at risk of depression.

Posner divides the components of attention into orientation, maintenance, and disengagement. In some groups with mood disorders, the characteristics of their attention to stimuli are different from those of healthy groups. According to Beck's theory of depression, in groups with mood disorders, such as the subclinically depressed group, they are more inclined to process negative stimuli. At present, few studies have scrutinized subclinically depressed college students on the two characteristics of attention maintenance and disengagement. The present study examined the attentional features of verbal emotional information processing in subclinically depressed college students with the help of eye-tracking technology. Firstly, Experiment 1 explored the processing characteristics of subclinically depressed students in the attention maintenance stage and the timing of their attentional bias by means of a free-viewing paradigm. Second, Experiment 2 explored the processing characteristics of this group in the attention disengagement stage through the gap-overlap paradigm.
The BDI-II-C and CES-D questionnaire were used to screen subclinically depressed participants. In Experiment 1. A two-factor mixed experimental design of 2 (group: subclinically depressed group, control group) × 3 (lexical valence: positive, neutral, negative) was used to examine the attentional characteristics of subclinically depressed participants using a free viewing task. The results revealed that (1) The total fixation time for negative words was significantly longer in the subclinically depressed group than in the control group. (2) The total fixation number of negative words was significantly greater than that of neutral words in subclinically depressed group; The total fixation number of positive words was significantly greater than that of neutral words in control group; (3) On the attentional bias scores, the negative attentional bias scores of the subclinically depressed group were significantly higher than those of the control group. The positive attention bias score of the control group was significantly higher than that of the subclinically depressed group.
Experiment 2 used a gap-overlap paradigm to examine the attentional disengagement of subclinically depressed college students. The experiment was a three-factor mixed experimental design of 3 (lexical valence: positive, neutral, negative) × 2 (group: subclinically depressed group, control group) × 2 (task type: gap task, overlap task). The results found that (1) Only in the overlap task, the saccade reaction time to negative words was significantly longer than that to positive words in the subclinically depressed group; the saccade reaction time to positive words was significantly longer than that to negative words in the control group. (2) The negative disengagement cost was significantly higher than the positive disengagement cost in the subclinically depressed group; The negative disengagement cost of control group was significantly lower than that of neutral disengagement cost.
Based on the above findings, the following conclusions were drawn in this study:
(1) In terms of the characteristics of attention maintenance, the attention bias of subclinically depressed college students does not appear in the early stage, but in the late stage, which is manifested as the attention bias to negative words.
(2) In terms of the characteristics of attentional disengagement, subclinically depressed college students have difficulty in attentional disengagement with negative words.
(3) Subclinically depressed college students showed attentional avoidance of positive words.

Learning multiple non-adjacent dependencies is fundamental to processing the hierarchical structures of natural language and constitutes one of the most challenging aspects of language acquisition. Such dependencies are typically acquired automatically and implicitly; however, how to effectively facilitate their learning remains unresolved. A particularly powerful factor in improving learning and generalization is variability in the input examples. However, its role in the implicit learning of multiple non-adjacent dependencies remains unclear, particularly whether such variability enables learners to move beyond value-value mappings to operations over variables, thereby acquiring transferable abstract rules. Accordingly, this study first examines how different levels of example variability influence implicit learning performance (Experiment 1) and then explores whether the effect of variability operates at the level of abstract variable-variable mappings (Experiment 2).
The study employed multiple non-adjacent rules defined over the tones with which Chinese syllables were spoken. Specifically, the tone types (ping or ze) of the first five syllables predicted those of the final five by an inversion relation in strings of length ten. In Experiment 1, three conditions of variability (low, medium, and high) were established by manipulating the number of syllable types used in the strings. The experiment comprised a training phase and a test phase. During the training phase, participants listened to a series of tone-syllable strings that followed the multiple non-adjacent rules and repeated them. In the test phase, participants were told that all strings they heard during the training phase followed a grammatical rule, and were asked to judge whether a set of new strings conformed to the learned rule. In Experiment 2, to further examine the level at which variability facilitates learning, two types of materials were constructed. One included only partial tone mappings (e.g., Tone 1-3 or Tone 2-4), while the other included complete tone mappings between level tones (Tones 1/2) and contour tones (Tones 3/4). Unlike in Experiment 1, during the training phase of Experiment 2, participants were exposed only to strings containing partial mappings. They were then tested in a non-transfer test (strings with the same partial mappings as in training), and either a mixed transfer test (strings with complete mappings; Experiment 2a) or a novel transfer test (strings with novel partial mappings that conflicted with training; Experiment 2b).
The results revealed a U-shaped relationship between example variability and implicit learning performance in Experiment 1: participants showed significant implicit learning effects in both the low- and high-variability conditions, whereas no learning effect was observed in the medium-variability condition. Performance was greater in the low- and high-variability conditions than in the medium-variability condition. In Experiment 2, the participants demonstrated implicit learning in the non-transfer test (Experiment 2a) across all variability conditions. However, their ability to transfer this learning to new materials varied significantly by condition. Participants in the low- and high-variability conditions were able to implicitly extract abstract rules from partial mappings and generalize them to complete mappings (Experiment 2a). Remarkably, this transfer ability was maintained even when participants were confronted with novel partial mappings that directly contradicted the training material (Experiment 2b). In contrast, participants in the medium-variability condition failed to transfer their knowledge to new materials in either transfer tests. This indicates that both low and high variability facilitated the implicit learning of transferable abstract rules (variable-variable mapping level), whereas participants in the medium-variability condition implicitly learned only surface-level value-value mappings.
Taken together, this study provides a comprehensive account of how example variability influences the implicit learning of multiple non-adjacent dependencies. It reveals a nonlinear relationship between example variability and implicit learning and identifies that this effect operates primarily at the level of abstract variable-variable mappings. From the perspective of example variability, the findings further indicate that the implicit acquisition of abstract rules is constrained by specific boundary conditions. The findings not only deepen the understanding of the mechanisms underlying implicit learning, but also provide theoretical implications for optimizing instructional design.

Recent research on adolescent internet gaming disorder (IGD) has increasingly emphasized an analytical perspective integrating dynamic, heterogeneous, and holistic approaches. This study combined growth mixture modeling with network analysis methods; it incorporated 15 risk and protective factors across family, school, community, and individual domains, in order to explore the heterogeneous developmental trajectories of adolescent IGD, the causes of heterogeneous development, and their dynamic variation characteristics.
This study used a cluster sampling method to conduct a one-year longitudinal survey with three measurement waves of 1, 852 adolescents from Shandong Province, China. Participants were recruited from three junior high schools (grades 7-9), two general high schools (grades 10-11), and two vocational high schools (grades 10-11). The participants completed a questionnaire assessing 15 risk and protective factors, and then an IGD scale. Preliminary analyses were performed with SPSS, with missing values imputed via the Expectation-Maximization algorithm. After that, Mplus was used to construct growth mixture models and identify heterogeneous trajectory subgroups of IGD development. Next, network analysis and network comparison analysis were performed using R software to construct relational networks of IGD and the multi-domain factors for every subgroup at different time points, and to compare the similarities and differences in network models and properties across subgroups at every time point. Finally, by integrating these comparative results, we examined the underlying mechanisms of heterogeneous IGD development and its dynamic characteristics in adolescents.
This study identified four distinct developmental trajectories of IGD among adolescents. Based on research objectives and sample representativeness considerations, two trajectories characterized by low and stable IGD scores were then merged into a normative development group, and they were subsequently compared with a more representative addictive development group, which was characterized by high initial IGD levels with significant growth and high sample proportion, from the other two trajectories. Comparison results indicated that in early IGD development, no significant differences were observed between the two groups in overall network connectivity patterns or global strength, nor in important factors, namely, proximal factors and high strength centrality nodes. Deviant peer affiliation, peer victimization, parental psychological control, teacher support, community resources, and self-esteem served as proximal factors for both of the groups. School belongingness and self-esteem were consistently high strength centrality nodes in both groups, and no specific factors unique to the addictive development group were identified. The primary difference between the two groups lay in that the addictive development group scored significantly lower than the normative development group on all of the protective factors, while scoring significantly higher on risk factors, thus suggesting that IGD development may be more associated with the accumulation of multi-domain risks characterized by “high risk-low protection.” In the later stage of IGD development, all of the direct connections between factors and IGD were not visible in the addictive development group, possibly indicating a hysteresis phenomenon in adolescent IGD development, namely, that IGD enters a self-sustaining steady state under the cumulative effect of risks, such that even when external risks are diminished, IGD cannot be readily alleviated. However, future research is required to verify this inference based on the present findings.

College freshmen navigating early adulthood experience a critical phase of social role reconstruction, facing new challenges in interpersonal adaptation and environmental integration. While previous studies have examined the relationship between social media use and interpersonal adaptation among college students, these have predominantly employed cross-sectional or long-term longitudinal designs, making it difficult to capture the daily, real-time dynamics between these variables during freshmen’s first month at college. Furthermore, the mechanisms underlying this relationship, as well as its boundary conditions within intensive time frames, remain underexplored. Therefore, using the first month of attending college as a temporal window and employing the residual dynamic structural equation model (RDSEM), this study examined the fine-grained temporal associations between social media use and interpersonal adaptation. It further investigated the intensive mediating effect of self-concept clarity and the cross-level moderating influence of envy.
A 30-day intensive longitudinal survey was conducted among 306 freshmen during their first month of college (M_age = 18.18 years, SD = 0.63; 71.9% female). Every day at 8:30 p.m., participants received an online questionnaire that they had 30 min to complete. If no response was received within 20 min, the system automatically sent a reminder, and the questionnaire closed 10 min thereafter. On average, participants spent approximately 5 min completing the survey daily.
RDSEM analyses were conducted via Mplus 8.7 and used to examine the fine-grained cross-lagged effects among the variables and a moderated mediation model. The results provided three major insights. First, social media use at time t - 1 significantly and negatively predicted the interpersonal adaptation of freshmen at time t. Second, social media use at time t - 1 significantly and negatively predicted self-concept clarity at time t, which in turn positively predicted the interpersonal adaptation of freshmen at time t + 1. Third, freshmen with high levels of envy were particularly susceptible to the negative effects of social media use at time t - 1. This susceptibility manifested in low interpersonal adaptation at time t + 1 and a marked decline in self-concept clarity at time t.
These findings indicate that the adverse effects of social media on interpersonal adaptation can be mitigated by guiding freshmen to appropriately regulate their social media use during the initial stage of college and to cultivate an integrated and stable self-concept. In addition, freshmen with high levels of envy should be identified as a key group for targeted support, thereby helping reduce their tendency to perceive others’ positive self-presentations on social media as threats. Overall, by investigating the mechanisms underlying the negative effects of social media during the early college adjustment period, this study offers practical insights for promoting smoother adaptation among freshmen in the modern digital environment.

With the growing integration of pets into urban life, pet-related injuries have become more common. These incidents often involve ambiguous causes, leading to public disagreement over responsibility. Psychologically, how people assign blame in ambiguous situations is not well understood. We propose that pet contact influences responsibility attribution, hypothesizing that more pet contact would result in less blame being assigned to the pet party diad (the pet and its owner), especially when the cause of the incident is unclear, and that this effect would occur through reduced perceived pet harmfulness. This reduction in perceived harmfulness was also expected to subsequently decrease support for victim compensation, extending the attributional chain to behavioral intentions.
Four studies were conducted with Chinese adults recruited online and locally. Study 1 (N = 156) was a cross-sectional survey measuring self-reported pet contact and responsibility attribution across two types of pet injury scenarios: clear and ambiguous. Study 2 (N = 189) experimentally manipulated pet contact via an imagination task. Study 3 (N = 210) used video priming. Study 4 (N = 156) employed real interactions in a real pet store versus a real store. Measures included pet contact frequency, perceived harmfulness, responsibility attribution to the pet party, and support for victim compensation. The analyses used ANOVA, linear mixed models, and mediation analysis, controlling for pet ownership, age, gender, and pet liking.
Study 1 showed that more pet contact predicted less responsibility attribution to the pet party diad in ambiguous scenarios (β = -0.45, p = 0.010), but no effect in clear situations. Study 2 found that the pet contact group attributed less responsibility to pet party diad (p < 0.001) and perceived pets as less harmful, with mediation confirmed (95% CI [-0.25, -0.04]). Study 3 replicated this finding and demonstrated a sequential mediation: from pet contact to lower perceived harmfulness, then to reduced responsibility, and finally to lower support for victim compensation (95% CI [-0.44, -0.10]). Study 4, implemented in a real setting, again supported the sequential mediation (95% CI [-0.60, -0.10]). A mini meta-analysis of Studies 2-4 showed a consistent medium effect (Cohen’s d = -0.44, 95% CI [-0.81, -0.07]).
This research shows that pet contact reduces responsibility attributions by lowering perceived harmfulness, and supports the application of attribution theory to human-animal interactions. The findings highlight how personal experience can systematically shape causal inference in ambiguous contexts. Practically, they help explain public disagreements over pet-related disputes and suggest that pet contact history may unconsciously bias judgments of blame and compensation. Future research should examine boundary conditions such as pet type and negative contact experiences.

A reverse mentoring relationship is a developmental workplace relationship in which younger employees provide guidance and support to older employees. In the digital-intelligence era, older employees still possess valuable work experience, contextual judgment, and tacit knowledge. Still, they may face difficulties in adapting to emerging digital tools and platform-based work systems. Existing research has mainly relied on Western frameworks derived from traditional mentoring, leaving the role reversal, power dynamics, and cultural features of reverse mentoring relationships in Chinese organizations insufficiently specified. This study examined the dimensional structure of the reverse mentoring relationship, developed a measurement scale, and tested its practical effects for older protégés and younger mentors.
Three studies were conducted. Study 1 used grounded theory methods based on semi-structured interviews with 15 participants and open-ended questionnaire responses from 13 participants. Through open, axial, selective, and theoretical coding, 890 valid semantic segments were extracted. The coding process showed high inter-coder agreement, with an agreement rate of 87.4% and Cohen’s kappa = 0.83. Study 2 developed and validated the reverse mentoring relationship scale using three independent samples: 200 participants for exploratory factor analysis, 492 participants for confirmatory factor analysis and measurement invariance testing, and 249 participants for discriminant validity testing. Study 3 adopted a three-wave matched design, including 240 older protégés and their supervisors and 292 younger mentors, to examine the scale’s practical applicability.
Study 1 identified three dimensions of the reverse mentoring relationship in Chinese organizations: technical support, mental inspiration, and competence recognition. Based on these dimensions, a 15-item scale was developed. In Study 2, exploratory factor analysis extracted three factors, explaining 61.69% of the total variance. Confirmatory factor analysis supported the three-factor model. The second-order three-factor model also showed acceptable fit, indicating that the reverse mentoring relationship is a multidimensional construct. Measurement invariance tests across gender, age, education, position level, traditional mentoring role, relationship type, and organizational ownership supported the scale's cross-group stability. Discriminant validity analysis showed that the new scale was related to, but distinct from, Chen (2014)’s reverse mentoring relationship measure. In Study 3, the reverse mentoring relationship positively predicted older protégés’ task performance and self-efficacy; self-efficacy further predicted task performance. The reverse mentoring relationship also negatively predicted younger mentors’ turnover intention, and positively predicted perceived insider status; perceived insider status negatively predicted turnover intention.
By integrating Chinese cultural characteristics with the context of digital-intelligence transformation, this study clarified the conceptual connotation and structural dimensions of the reverse mentoring relationship. It developed a measurement tool applicable to Chinese organizational contexts in the digital-intelligence era. It not only provides a contextualized refinement and localized extension of existing Western frameworks but also offers a key methodological foundation for future scholars to examine the effects and mechanisms of the reverse mentoring relationship. In addition, it offers new insights and directions for research on talent development mechanisms and management practices in Chinese organizations.