Fear of negative evaluation, often abbreviated as FNE, describes the excessive worry and apprehension a person feels about being unfavorably judged by others. Social anxiety, by contrast, is a broader syndrome that includes not only evaluative fear but also maladaptive self-focused attention, anticipatory worry, safety behaviors, avoidance, and disruptions in daily functioning. The DSM-5 explicitly embeds fear of negative evaluation as a core symptom of social anxiety disorder, a condition that affects roughly 4 percent of the global population at some point in their lives according to the World Mental Health Survey. Yet empirical work has shown that social anxiety can also emerge in people with high shame proneness or low social self-efficacy even without pronounced evaluative fear, suggesting the two constructs are related but not identical. Cognitive-behavioral models, including those of Clark and Wells and of Rapee and Heimberg, position fear of negative evaluation as a maintaining bias within social anxiety rather than a mere symptom, making the precise strength of the association a matter of real theoretical weight.

To resolve the question, the research team followed PRISMA 2020 reporting guidelines and prospectively registered their protocol on PROSPERO. They searched major English-language databases including PubMed, ProQuest, and Web of Science, alongside the Chinese-language repositories CNKI and Wanfang Data, capturing studies published between January 2010 and April 2025. Eligible studies had to report Pearson correlations between fear of negative evaluation and social anxiety, or provide sufficient statistics to convert to correlations, in samples of adolescents or adults with at least 20 valid participants. Two independent reviewers screened titles, abstracts, and full texts, extracted data, and assessed methodological quality using the National Heart, Lung, and Blood Institute tool, achieving an inter-rater agreement kappa of 0.876. The final dataset comprised 143 studies with 146 independent samples and 163 effect sizes, spanning participants aged roughly 12 to 47 years, with about 40 percent of the pooled sample male.

The statistical machinery behind the analysis was deliberately conservative. Because the included studies varied widely in participants, instruments, and cultural contexts, the team adopted a random-effects model a priori, assuming that the true effect size differs across studies. Correlation coefficients were Fisher’s z-transformed to stabilize variance, and heterogeneity was evaluated using Cochran’s Q test, the I² statistic, and prediction intervals, which estimate the likely range of true effects in future studies. Sensitivity analyses used the leave-one-out method and externally standardized residuals to detect whether any single study was driving the results. Publication bias was probed with funnel plots, Egger’s regression test, the Fail-safe N method, and trim-and-fill analysis, and the team also included unpublished dissertations to reduce file-drawer effects.

The headline result was a pooled correlation of r = 0.527, with a 95 percent confidence interval of 0.50 to 0.55, indicating a moderate positive association. That figure was remarkably stable: removing any single study shifted the estimate only between 0.521 and 0.527, and after excluding one statistical outlier the range was 0.524 to 0.529. But the analysis also revealed substantial heterogeneity, with an I² of 88.62 percent and a wide prediction interval of 0.29 to 0.76, meaning the true strength of the association likely varies considerably across populations and settings. The authors emphasize that the pooled estimate should be read as a context-averaged association rather than a uniform number that applies everywhere, a point that matters for anyone hoping to translate the finding into clinical screening or theory.

Publication bias, a perennial worry in meta-analysis, appears unlikely to explain the result. The funnel plot was roughly symmetrical, and Egger’s regression test was non-significant. The Fail-safe N was a striking 77,078, meaning tens of thousands of unpublished null-effect studies would be needed to reduce the pooled result to non-significance. Trim-and-fill analysis did impute 31 potentially missing studies, and after adjustment the pooled correlation dipped from 0.527 to 0.483, but it remained statistically significant and substantial. A moderator comparison between published and unpublished studies also found no significant difference, further supporting the robustness of the association.

The most intriguing findings emerged from the moderator analyses, which asked what explains the variability between studies. The type of social anxiety measurement tool significantly moderated the relationship: correlations were strongest when social anxiety was measured with the Social Interaction Anxiety Scale, at r = 0.601, and weakest with the Social Avoidance and Distress scale, at r = 0.414. The authors attribute this to content overlap, since the SIAS emphasizes emotional experience that aligns closely with the emotional-cognitive nature of fear of negative evaluation, whereas the SAD focuses on behavioral avoidance. In contrast, the type of fear of negative evaluation scale used had no significant effect, likely because the four instruments included all descend from the original Watson and Friend scale and share a common theoretical framework. Notably, gender, age, and publication year were all non-significant moderators, suggesting the association is stable across demographic groups and has not shifted over fifteen years of research despite changing social environments.

Culture told a subtler story. The FNE-social anxiety link was stronger in Western, individualistic samples, at r = 0.592, than in Eastern, collectivistic samples, at r = 0.524. The authors suggest this may reflect differing psychological meanings of negative evaluation: in Western societies, being judged negatively tends to threaten personal worth and adequacy, while in Eastern collectivistic cultures it is more closely tied to maintaining face and relational harmony. However, when the team ran a mixed-effects meta-regression controlling for measurement tool type, the cultural effect shrank to marginal significance, explaining part of a model that accounted for just over 20 percent of between-study heterogeneity. This hints that cultural groups may preferentially use instruments that align with their values and norms, partially confounding culture with measurement, and the authors caution that the broad Eastern-Western binary cannot capture within-group diversity, such as the hybrid characteristics of Japanese culture.

The practical implications reach into the clinic. Because fear of negative evaluation emerges as an important cognitive component of social anxiety rather than its equivalent, cognitive restructuring and emotion regulation strategies that target evaluative fear could be relevant considerations in interventions for socially anxious clients. At the same time, the moderate correlation argues against a practice the authors flag in the literature: some studies have treated fear of negative evaluation scale scores as proxies for social anxiety levels, a move criticized for conflating two theoretically distinct constructs. The new data support keeping the constructs separate while acknowledging their close kinship.

Limitations temper the conclusions. Most included studies relied on adolescent and college student samples, limiting generalizability to older adults and other populations, and the predominance of cross-sectional designs means only linear associations, not causal pathways, can be inferred. A considerable share of heterogeneity remained unexplained even after the moderator analyses, and the heavy representation of Chinese-language studies may constrain how far the findings travel. The authors call for longitudinal and cross-cultural research to clarify the developmental and contextual dynamics of the relationship, including whether it shifts in rapidly evolving digital environments and on social media platforms. For now, the analysis delivers a quantitatively grounded verdict on one of psychology’s most studied links: the fear of being judged is tightly woven into social anxiety, but the thread runs differently through different cultures, instruments, and lives.

Subject of Research: The association between fear of negative evaluation and social anxiety examined through systematic review and meta-analysis

Article Title: The relationship between fear of negative evaluation and social anxiety: a systematic review and meta-analysis

Article References: Peng, S., Zhou, M., Yue, L., Hei, X., Li, Z., Kong, F., Zhang, T., Ren, X., & Song, Y. (2026). The relationship between fear of negative evaluation and social anxiety: a systematic review and meta-analysis. BMC Psychology, 14(1), Article 1458. https://doi.org/10.1186/s40359-026-05211-z

Image Credits: AI Generated

DOI: 10.1186/s40359-026-05211-z

Keywords: fear of negative evaluation, social anxiety, meta-analysis, social anxiety disorder, cognitive-behavioral model, cultural differences, measurement tools, publication bias, moderator analysis, PRISMA, psychology, relationship

The scale of the study alone makes it noteworthy. Among the nearly twenty-five thousand participants, 4,350 adolescents—17.8 percent of the sample—met the operational criterion for problematic internet use, a proportion that underscores how common the phenomenon is in secondary school populations. The researchers measured three psychological domains using validated Chinese-language instruments: the Internet Addiction Test for problematic use, the Multidimensional Anxiety Scale for Children for anxiety symptoms, and the Ruminative Response Scale for rumination, the repetitive, self-focused pattern of thinking in which individuals dwell on their distress and its possible causes and consequences. Adolescents with problematic internet use reported significantly higher levels of both anxiety and rumination than their peers without problematic use, confirming the close association between maladaptive online behavior and dysregulated emotional and cognitive processing that has been described in the cognitive-behavioral model of the condition.

What distinguishes this study is its methodological ambition. Rather than relying on a single analytical technique, the team integrated three complementary approaches: network analysis, latent profile analysis, and a computational algorithm called NodeIdentifyR. Network analysis treats psychological symptoms not as isolated scores on a questionnaire but as nodes in an interconnected system, where each node can activate or reinforce its neighbors. In this framework, the edges between nodes represent statistical associations, and the overall architecture of the network—its global strength and its structure—can be compared across groups. Latent profile analysis, by contrast, is a person-centered statistical method that sorts individuals into unobserved subgroups based on their response patterns, allowing researchers to ask whether the population is genuinely heterogeneous rather than assuming it is. NodeIdentifyR then uses network simulations to nominate specific nodes as candidate intervention targets, essentially asking which symptom, if successfully modified, would most efficiently disrupt the maladaptive dynamics of the whole system.

The person-centered analysis delivered the study’s central insight. Latent profile analysis identified two distinct subgroups among the adolescents with problematic internet use, differentiated by the degree to which anxiety and rumination co-occurred: a low anxiety-rumination co-occurrence subgroup and a high co-occurrence subgroup. In other words, some teens with problematic internet use carry a heavy combined burden of anxious and ruminative symptoms, while others show comparatively little of this emotional-cognitive entanglement. The two subgroups differed significantly not only in their overall symptom levels but also in the global strength of their symptom networks and in the structure of those networks—the pattern of connections linking individual anxiety and rumination dimensions to one another.

This structural difference mattered enormously when the researchers ran their network simulations to identify intervention targets. For adolescents in the low co-occurrence subgroup, the simulations pointed to Harm Avoidance—a dimension of anxiety reflecting the tendency to worry about potential future threats and to avoid situations perceived as dangerous—as the most promising candidate target. For the high co-occurrence subgroup, the simulations instead nominated Symptom Rumination, the component of rumination focused on dwelling on one’s distressing feelings and their implications. The implication is striking: the single most efficient point of therapeutic leverage differs depending on which psychological profile an adolescent occupies, and a one-size-fits-all intervention strategy may therefore miss the mark for a substantial fraction of affected teens.

The technical logic behind these simulations is worth unpacking. In network approaches to psychopathology, symptoms are conceptualized as self-reinforcing systems in which activation spreads along the edges of the network. A node with high centrality—many strong connections to other nodes—can, in principle, sustain the entire dysfunctional system, so temporarily activating or perturbing that node in silico and observing the cascade of effects provides a way to estimate where an intervention would produce the greatest downstream benefit. The NodeIdentifyR algorithm formalizes this perturbation logic, and by applying it separately to each subgroup’s network, the researchers could test whether the same node would emerge as the optimal target in both groups. It did not, which is precisely the finding that gives the study its practical significance.

For clinicians and school mental health programs, the results suggest a practical triage strategy. Before selecting an intervention focus, practitioners might first assess whether an adolescent with problematic internet use falls into the high or low anxiety-rumination co-occurrence profile. A teen whose network is dominated by anticipatory worry and threat avoidance may benefit most from interventions that directly target harm avoidance, such as exposure-based cognitive-behavioral techniques that reduce avoidance behavior and catastrophic forecasting. A teen whose symptoms are organized around repetitive negative thinking may respond better to approaches aimed at disengaging from ruminative loops, such as rumination-focused cognitive behavioral therapy or mindfulness-based strategies. The study’s framing within the cognitive-behavioral model supports this logic, since that model explicitly links problematic internet use to maladaptive cognitive processing and dysregulated emotional responses that can be addressed at specific points in the system.

The authors are careful to frame their conclusions appropriately. The findings are model-based, generated from cross-sectional data, and the network simulations identify candidate targets rather than proven ones. Whether intervening on Harm Avoidance in low co-occurrence adolescents or on Symptom Rumination in high co-occurrence adolescents actually reduces problematic internet use over time is an empirical question that can only be answered with longitudinal designs and, ultimately, randomized experimental studies. The researchers state explicitly that these model-based findings require validation in longitudinal and experimental work before they can be translated into clinical guidelines. Cross-sectional networks also cannot establish the direction of causality: anxiety and rumination may drive excessive internet use, excessive use may amplify anxiety and rumination, or both may arise from shared underlying vulnerabilities.

Even with those caveats, the study represents a meaningful advance in how researchers think about problematic internet use in adolescents. By demonstrating that the population is psychologically heterogeneous—that two subgroups with measurably different network architectures exist within the same diagnostic label—it challenges the assumption that a single intervention target will serve everyone. The work also showcases the value of integrating multiple analytical methods, since neither the latent profile analysis alone nor the network analysis alone could have revealed both the existence of the subgroups and the divergence of their optimal intervention points. Funded by the Shandong Excellent Young Scientists Fund Program and approved by the Ethics Committee of Shandong University, with written informed consent obtained from all participants and their legal guardians, the study offers a template for large-scale, methodologically rigorous research on adolescent digital behavior. As screen time continues to climb worldwide, the message for the field is clear: effective intervention may depend less on finding the one right target for problematic internet use and more on matching the target to the individual adolescent’s psychological profile.

Subject of Research: Heterogeneous anxiety-rumination profiles and network-based intervention targets in adolescents with problematic internet use

Article Title: Are candidate intervention targets consistent across adolescents with problematic internet use: an empirical study integrating multiple methods

Article References: Are candidate intervention targets consistent across adolescents with problematic internet use: an empirical study integrating multiple methods. (n.d.). https://doi.org/10.1186/s40359-026-05727-4

Image Credits: AI Generated

DOI: 10.1186/s40359-026-05727-4

Keywords: problematic internet use, adolescents, anxiety, rumination, network analysis, latent profile analysis, intervention targets, cognitive-behavioral model, mental health, internet addiction, NodeIdentifyR, school psychology