Introduction

Social network refers to online platforms that allow users to create, share, and exchange information, encompassing text, images, audio, and video [1]. The use of social network, a term encompassing various activities on these platforms, has been measured from angles such as frequency, duration, intensity, and addictive behavior, all indicative of the extent of social networking usage [2]. As of April 2023, there are 4.8 billion social network users globally, representing 59.9% of the world’s population [3]. The usage of social network is considered a normal behavior and a part of everyday life [4, 5]. Although social network offers convenience in daily life, excessive use can lead to PSNU [6, 7], posing potential threats to mental health, particularly anxiety symptoms (Rasmussen et al., 2020). Empirical research has shown that anxiety symptoms, including generalized anxiety (GA), social anxiety (SA), attachment anxiety (AA), and fear of missing out (FoMO), are closely related to PSNU [8,9,10,11,12]. While some empirical studies have explored the relationship between PSNU and anxiety symptoms, their conclusions are not consistent. Some studies have found a significant positive correlation [13,14,15], while others have found no significant correlation [16,17,18,19]. Furthermore, the degree of correlation varies widely in existing research, with reported r-values ranging from 0.12 to 0.80 [20, 21]. Therefore, a systematic meta-analysis is necessary to clarify the impact of PSNU on individual anxiety symptoms.

Previous research lacks a unified concept of PSNU, primarily due to differing theoretical interpretations by various authors, and the use of varied standards and diagnostic tools. Currently, this phenomenon is referred to by several terms, including compulsive social networking use, problematic social networking use, excessive social networking use, social networking dependency, and social networking addiction [22,23,24,25,26]. These conceptual differences hinder the development of a cohesive and systematic research framework, as it remains unclear whether these definitions and tools capture the same underlying construct [27]. To address this lack of uniformity, this paper will use the term “problematic use” to encompass all the aforementioned nomenclatures (i.e., compulsive, excessive, dependent, and addictive use).

Regarding the relationship between PSNU and anxiety symptoms, two main perspectives exist: the first suggests a positive correlation, while the second proposes a U-shaped relationship. The former perspective, advocating a positive correlation, aligns with the social cognitive theory of mass communication. It posits that PSNU can reinforce certain cognitions, emotions, attitudes, and behaviors [28, 29], potentially elevating individuals’ anxiety levels [30]. Additionally, the cognitive-behavioral model of pathological use, a primary framework for explaining factors related to internet-based addictions, indicates that psychiatric symptoms like depression or anxiety may precede internet addiction, implying that individuals experiencing anxiety may turn to social networking platforms as a coping mechanism [31]. Empirical research also suggests that highly anxious individuals prefer computer-mediated communication due to the control and social liberation it offers and are more likely to have maladaptive emotional regulation, potentially leading to problematic social network service use [32]. Turning to the alternate perspective, it proposes a U-shaped relationship as per the digital Goldilocks hypothesis. In this view, moderate social networking usage is considered beneficial for psychosocial adaptation, providing individuals with opportunities for social connection and support. Conversely, both excessive use and abstinence can negatively impact psychosocial adaptation [33]. In summary, both perspectives offer plausible explanations.

Incorporating findings from previous meta-analyses, we identified seven systematic reviews and two meta-analyses that investigated the association between PSNU and anxiety. The results of these meta-analyses indicated a significant positive correlation between PSNU and anxiety (ranging from 0.33 to 0.38). However, it is evident that these previous meta-analyses had certain limitations. Firstly, they focused only on specific subtypes of anxiety; secondly, they were limited to adolescents and emerging adults in terms of age. In summary, this systematic review aims to ascertain which theoretical perspective more effectively explains the relationship between PSNU and anxiety, addressing the gaps in previous meta-analyses. Additionally, the association between PSNU and anxiety could be moderated by various factors. Drawing from a broad research perspective, any individual study is influenced by researcher-specific designs and associated sample estimates. These may lead to bias compared to the broader population. Considering the selection criteria for moderating variables in empirical studies and meta-analyses [34, 35], the heterogeneity of findings on problematic social network usage and anxiety symptoms could be driven by divergence in sample characteristics (e.g., gender, age, region) and research characteristics (measurement instrument of study variables). Since the 2019 coronavirus pandemic, heightened public anxiety may be attributed to the fear of the virus or heightened real life stress. The increased use of electronic devices, particularly smartphones during the pandemic, also instigates the prevalence of problematic social networking. Thus, our analysis focuses on three moderators: sample characteristics (participants’ gender, age, region), measurement tools (for PSNU and anxiety symptoms) and the time of measurement (before COVID-19 vs. during COVID-19).

Method

The present study was conducted in accordance with the 2020 statement on Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [36]. To facilitate transparency and to avoid unnecessary duplication of research, this study was registered on PROSPERO, and the number is CRD42022350902.

Literature search

Studies on the relationship between the PSNU and anxiety symptoms from 2000 to 2023 were retrieved from seven databases. These databases included China National Knowledge Infrastructure (CNKI), Wanfang Data, Chongqing VIP Information Co. Ltd. (VIP), Web of Science, ScienceDirect, PubMed, and PsycARTICLES. The search strings consisted of (a) anxiety symptoms, (b) social network, and (c) Problematic use. As shown in Table 1, the keywords for anxiety are as follows: anxiety, generalized anxiety, social anxiety, attachment anxiety, fear of missing out, and FoMO. The keywords for social network are as follows: social network, social media, social networking site, Instagram, and Facebook. The keywords for addiction are as follows: addiction, dependence, problem/problematic use, excessive use. The search deadline was March 19, 2023. A total of 2078 studies were initially retrieved and all were identified ultimately.

Table 1 Retrieval information
Full size table

Inclusion and exclusion criteria

Retrieved studies were eligible for the present meta-analysis if they met the following inclusion criteria: (a) the study provided Pearson correlation coefficients used to measure the relationship between PSNU and anxiety symptoms; (b) the study reported the sample size and the measurement instruments for the variables; (c) the study was written in English and Chinese; (d) the study provided sufficient statistics to calculate the effect sizes; (e) effect sizes were extracted from independent samples. If multiple independent samples were investigated in the same study, they were coded separately; if the study was a longitudinal study, they were coded by the first measurement. In addition, studies were excluded if they: (a) examined non-problematic social network use; (b) had an abnormal sample population; (c) the results of the same sample were included in another study and (d) were case reports or review articles. Two evaluators with master’s degrees independently assessed the eligibility of the articles. A third evaluator with a PhD examined the results and resolved dissenting views.

Data extraction and quality assessment

Two evaluators independently coded the selected articles according to the following characteristics: literature information, time of measurement (before the COVID-19 vs. during the COVID-19), sample source (developed country vs. developing country), sample size, proportion of males, mean age, type of anxiety, and measurement instruments for PSNU and anxiety symptoms. The following principles needed to be adhered to in the coding process: (a) effect sizes were extracted from independent samples. If multiple independent samples were investigated in the same study, they were coded separately; if the study was a longitudinal study, it was coded by the first measurement; (b) if multiple studies used the same data, the one with the most complete information was selected; (c) If studies reported t or F values rather than r, the following formula \( r=\sqrt{\frac{{t}^{2}}{{t}^{2}+df}}\); \( r=\sqrt{\frac{F}{F+d{f}_{e}}}\) was used to convert them into r values [37, 38]. Additionally, if some studies only reported the correlation matrix between each dimension of PSNU and anxiety symptoms, the following formula \( {r}_{xy}=\frac{\sum {r}_{xi}{r}_{yj}}{\sqrt{n+n(n-1){r}_{xixj}}\sqrt{m+m(m-1){r}_{yiyj}}}\) was used to synthesize the r values [39], where n or m is the number of dimensions of variable x or variable y, respectively, and\( {r}_{xixj} \)or \( {r}_{yiyj}\) represents the mean of the correlation coefficients between the dimensions of variable x or variable y, respectively.

Literature quality was determined according to the meta-analysis quality evaluation scale developed [40]. The quality of the post-screening studies was assessed by five dimensions: sampling method, efficiency of sample collection, level of publication, and reliability of PSNU and anxiety symptom measurement instruments. The total score of the scale ranged from 0 to 10; higher scores indicated better quality of the literature.

Data analysis

All data were performed using Comprehensive Meta Analysis 3.3 (CMA 3.3). Pearson’s product-moment coefficient r was selected as the effect size index in this meta-analysis. Firstly, \( {\text{F}\text{i}\text{s}\text{h}\text{e}\text{r}}^{{\prime }}\text{s} Z=\frac{1}{2}\times \text{ln}\left(\frac{1+r}{1-r}\right)\) was used to convert the correlation coefficient to Fisher Z. Then the formula \( SE=\sqrt{\frac{1}{n-3}}\) was used to calculate the standard error (SE). Finally, the summary of r was obtained from the formula \( r=\frac{{e}^{2z}-1}{{e}^{2z}+1}\) for a comprehensive measure of the relationship between PSNU and anxiety symptoms [37, 41].

Although the effect sizes estimated by the included studies may be similar, considering the actual differences between studies (e.g., region and gender), the random effects model was a better choice for data analysis for the current meta-analysis. The heterogeneity of the included study effect sizes was measured for significance by Cochran’s Q test and estimated quantitatively by the I2 statistic [42]. If the results indicate there is a significant heterogeneity (the Q test: p-value < 0.05, I2 > 75) and the results of different studies are significantly different from the overall effect size. Conversely, it indicates there are no differences between the studies and the overall effect size. And significant heterogeneity tends to indicate the possible presence of potential moderating variables. Subgroup analysis and meta-regression analysis were used to examine the moderating effect of categorical and continuous variables, respectively.

Funnel plots, fail-safe number (Nfs) and Egger linear regression were utilized to evaluate the publication bias [43,44,45]. The likelihood of publication bias was considered low if the intercept obtained from Egger linear regression was not significant. A larger Nfs indicated a lower risk of publication bias, and if Nfs < 5k + 10 (k representing the original number of studies), publication bias should be a concern [46]. When Egger’s linear regression was significant, the Duval and Tweedie’s trim-and-fill was performed to correct the effect size. If there was no significant change in the effect size, it was assumed that there was no serious publication bias [47].

A significance level of P < 0.05 was deemed applicable in this study.

Table 2 Characteristics of the selected studies
Full size table

Results

Sample characteristics

The PRISMA search process is depicted in Fig. 1. The database search yielded 2078 records. After removing duplicate records and screening the title and abstract, the full text was subject to further evaluation. Ultimately, 172 records fit the inclusion criteria, including 209 independent effect sizes. The present meta-analysis included 68 studies on generalized anxiety, 44 on social anxiety, 22 on attachment anxiety, and 75 on fear of missing out. The characteristics of the selected studies are summarized in Table 2. The majority of the sample group were adults. Quality scores for selected studies ranged from 0 to 10, with only 34 effect sizes below the theoretical mean, indicating high quality for the included studies. The literature included utilized BSMAS as the primary tool to measure PSNU, DASS-21-A to measure GA, IAS to measure SA, ECR to measure AA, and FoMOS to measure FoMO.

Fig. 1
figure 1

Flow chart of the search and selection strategy

Full size image

Overall analysis, homogeneity tests and publication bias

As shown in Table 3, there was significant heterogeneity between PSNU and all four anxiety symptoms (GA: Q = 1623.090, I2 = 95.872%; SA: Q = 1396.828, I2 = 96.922%; AA: Q = 264.899, I2 = 92.072%; FoMO: Q = 1847.110, I2 = 95.994%), so a random effects model was chosen. The results of the random effects model indicate a moderate positive correlation between PSNU and anxiety symptoms (GA: r = 0.350, 95% CI [0.323, 0.378]; SA: r = 0.390, 95% CI [0.347, 0.431]; AA: r = 0.345, 95% CI [0.286, 0.402]; FoMO: r = 0.496, 95% CI [0.461, 0.529]).

Table 3 Overall association between PSNU and anxiety symptoms
Full size table

Figure 2 shows the funnel plot of the relationship between PSNU and anxiety symptoms. No significant symmetry was seen in the funnel plot of the relationship between PSNU and GA and between PSNU and SA. And the Egger’s regression results also indicated that there might be publication bias (t = 3.775, p < 0.001; t = 2.309, p < 0.05). Therefore, it was necessary to use fail-safe number (Nfs) and the trim and fill method for further examination and correction. The Nfs for PSNU and GA as well as PSNU and SA are 4591 and 7568, respectively. Both Nfs were much larger than the standard 5k + 10. After performing the trim and fill method, 14 effect sizes were added to the right side of the funnel plat (Fig. 2.a), the correlation coefficient between PSNU and GA changed to (r = 0.388, 95% CI [0.362, 0.413]); 10 effect sizes were added to the right side of the funnel plat (Fig. 2.b), the correlation coefficient between PSNU and SA changed to (r = 0.437, 95% CI [0.395, 0.478]). The correlation coefficients did not change significantly, indicating that there was no significant publication bias associated with the relationship between PSNU and these two anxiety symptoms (GA and SA).

Fig. 2
figure 2

Funnel plot of the relationship between PSNU and anxiety symptoms. Note: Black dots indicated additional studies after using trim and fill method; (a) = Funnel plot of the PSNU and GA; (b) = Funnel plot of the PSNU and SA; (c) = Funnel plot of the PSNU and AA; (d) = Funnel plot of the PSNU and FoMO

Full size image

Sensitivity analyses

Initially, the findings obtained through the one-study-removed approach indicated that the heterogeneities in the relationship between PSNU and anxiety symptoms were not attributed to any individual study. Nevertheless, it is important to note that sensitivity analysis should be performed based on literature quality [223] since low-quality literature could potentially impact result stability. In the relationship between PSNU and GA, the 10 effect sizes below the theoretical mean scores were excluded from analysis, and the sensitivity analysis results were recalculated (r = 0.402, 95% CI [0.375, 0.428]); In the relationship between PSNU and SA, the 8 effect sizes below the theoretical mean scores were excluded from analysis, and the sensitivity analysis results were recalculated (r = 0.431, 95% CI [0.387, 0.472]); In the relationship between PSNU and AA, the 5 effect sizes below the theoretical mean scores were excluded from analysis, and the sensitivity analysis results were recalculated (r = 0.367, 95% CI [0.298, 0.433]); In the relationship between PSNU and FoMO, the 11 effect sizes below the theoretical mean scores were excluded from analysis, and the sensitivity analysis results were recalculated (r = 0.508, 95% CI [0.470, 0.544]). The revised estimates indicate that meta-analysis results were stable.

Moderator analysis

The impact of moderator variables on the relation between PSNU and GA

The results of subgroup analysis and meta-regression are shown in Table 4, the time of measurement significantly moderated the correlation between PSNU and GA (Qbetween = 19.268, df = 2, p < 0.001). The relation between the two variables was significantly higher during the COVID-19 (r = 0.392, 95% CI [0.357, 0.425]) than before the COVID-19 (r = 0.270, 95% CI [0.227, 0.313]) or measurement time uncertain (r = 0.352, 95% CI [0.285, 0.415]).

The moderating effect of the PSNU measurement was significant (Qbetween = 6.852, df = 1, p = 0.009). The relation was significantly higher when PSNU was measured with the BSMAS (r = 0.373, 95% CI [0.341, 0.404]) compared to others (r = 0.301, 95% CI [0.256, 0.344]).

The moderating effect of the GA measurement was significant (Qbetween = 60.061, df = 5, p < 0.001). Specifically, when GA measured by the GAD (r = 0.398, 95% CI [0.356, 0.438]) and the DASS-21-A (r = 0.433, 95% CI [0.389, 0.475]), a moderate positive correlation was observed. However, the correlation was less significant when measured using the STAI (r = 0.232, 95% CI [0.187, 0.276]).

For the relation between PSNU and GA, the moderating effect of region, gender and age were not significant.

Table 4 Results of the moderating effects of PSNU and GA
Full size table

The impact of moderator variables on the relation between PSNU and SA

The effects of the moderating variables in the relation between PSNU and SA were shown in Table 5. The results revealed a gender-moderated variances between the two variables (b = 0.601, 95% CI [ 0.041, 1.161], Qmodel (1, k = 41) = 4.705, p = 0.036).

For the relation between PSNU and SA, the moderating effects of time of measurement, region, measurement of PSNU and SA, and age were not significant.

The impact of moderator variables on the relation between PSNU and AA

The effects of the moderating variables in the relation between PSNU and AA were shown in Table 6, region significantly moderated the correlation between PSNU and AA (Qbetween = 6.410, df = 2, p = 0.041). The correlation between the two variables was significantly higher in developing country (r = 0.378, 95% CI [0.304, 0.448]) than in developed country (r = 0.242, 95% CI [0.162, 0.319]).

The moderating effect of the PSNU measurement was significant (Qbetween = 6.852, df = 1, p = 0.009). Specifically, when AA was measured by the GPIUS-2 (r = 0.484, 95% CI [0.200, 0.692]) and the PMSMUAQ (r = 0.443, 95% CI [0.381, 0.501]), a moderate positive correlation was observed. However, the correlation was less significant when measured using the BSMAS (r = 0.248, 95% CI [0.161, 0.331]) and others (r = 0.313, 95% CI [0.250, 0.372]).

The moderating effect of the AA measurement was significant (Qbetween = 17.283, df = 2, p < 0.001). The correlation was significantly higher when measured using the ECR (r = 0.386, 95% CI [0.338, 0.432]) compared to the RQ (r = 0.200, 95% CI [0.123, 0.275]).

For the relation between PSNU and AA, the moderating effects of time of measurement, region, gender, and age were not significant.

The impact of moderator variables on the relation between PSNU and FoMO

The effects of the moderating variables in the relation between PSNU and FoMO were shown in Table 7, the moderating effect of the PSNU measurement was significant (Qbetween = 8.170, df = 2, p = 0.017). Among the sub-dimensions, the others was excluded because there was only one sample. Specifically, when measured using the FoMOS-MSME (r = 0.630, 95% CI [0.513, 0.725]), a moderate positive correlation was observed. However, the correlation was less significant when measured using the FoMOS (r = 0.472, 95% CI [0.432, 0.509]) and the T-S FoMOS (r = 0.557, 95% CI [0.463, 0.639]).

For the relationship between PSNU and FoMO, the moderating effects of time of measurement, region, measurement of PSNU, gender and age were not significant.

Table 5 Results of the moderating effects of PSNU and SA
Full size table
Table 6 Results of the moderating effects of PSNU and AA
Full size table
Table 7 Results of the moderating effects of PSNU and FoMO
Full size table

Discussion

Through systematic review and meta-analysis, this study established a positive correlation between PSNU and anxiety symptoms (i.e., generalized anxiety, social anxiety, attachment anxiety, and fear of missing out), confirming a linear relationship and partially supporting the Social Cognitive Theory of Mass Communication [28] and the Cognitive Behavioral Model of Pathological Use [31]. Specifically, a significant positive correlation between PSNU and GA was observed, implying that GA sufferers might resort to social network for validation or as an escape from reality, potentially alleviating their anxiety. Similarly, the meta-analysis demonstrated a strong positive correlation between PSNU and SA, suggesting a preference for computer-mediated communication among those with high social anxiety due to perceived control and liberation offered by social network. This preference is often accompanied by maladaptive emotional regulation, predisposing them to problematic use. In AA, a robust positive correlation was found with PSNU, indicating a higher propensity for such use among individuals with attachment anxiety. Notably, the study identified the strongest correlation in the context of FoMO. FoMO’s significant association with PSNU is multifaceted, stemming from the real-time nature of social networks that engenders a continuous concern about missing crucial updates or events. This drives frequent engagement with social network, thereby establishing a direct link to problematic usage patterns. Additionally, social network’s feedback loops amplify this effect, intensifying FoMO. The culture of social comparison on these platforms further exacerbates FoMO, as users frequently compare their lives with others’ selectively curated portrayals, enhancing both their social networking usage frequency and the pursuit for social validation. Furthermore, the integral role of social network in modern life broadens FoMO’s scope, encompassing anxieties about staying informed and connected.

The notable correlation between FoMO and PSNU can be comprehensively understood through various perspectives. FoMO is inherently linked to the real-time nature of social networks, which cultivates an ongoing concern about missing significant updates or events in one’s social circle [221]. This anxiety prompts frequent engagement with social network, leading to patterns of problematic use. Moreover, the feedback loops in social network algorithms, designed to enhance user engagement, further intensify this fear [224]. Additionally, social comparison, a common phenomenon on these platforms, exacerbates FoMO as users continuously compare their lives with the idealized representations of others, amplifying feelings of missing out on key social experiences [225]. This behavior not only increases social networking usage but also is closely linked to the quest for social validation and identity construction on these platforms. The extensive role of social network in modern life further amplifies FoMO, as these platforms are crucial for information exchange and maintaining social ties. FoMO thus encompasses more than social concerns, extending to anxieties about staying informed with trends and dynamics within social networks [226]. The multifaceted nature of FoMO in relation to social network underscores its pronounced correlation with problematic social networking usage. In essence, the combination of social network’s intrinsic characteristics, psychological drivers of user behavior, the culture of social comparison, and the pervasiveness of social network in everyday life collectively make FoMO the most pronouncedly correlated anxiety type with PSNU.

Additionally, we conducted subgroup analyses on the timing of measurement (before COVID-19 vs. during COVID-19), measurement tools (for PSNU and anxiety symptoms), sample characteristics (participants’ region), and performed a meta-regression analysis on gender and age in the context of PSNU and anxiety symptoms. It was found that the timing of measurement, tools used for assessing PSNU and anxiety, region, and gender had a moderating effect, whereas age did not show a significant moderating impact.

Firstly, the relationship between PSNU and anxiety symptoms was significantly higher during the COVID-19 period than before, especially between PSNU and GA. However, the moderating effect of measurement timing was not significant in the relationship between PSNU and other types of anxiety. This could be attributed to the increased uncertainty and stress during the pandemic, leading to heightened levels of general anxiety [227]. The overuse of social network for information seeking and anxiety alleviation might have paradoxically exacerbated anxiety symptoms, particularly among individuals with broad future-related worries [228]. While the COVID-19 pandemic altered the relationship between PSNU and GA, its impact on other types of anxiety (such as SA and AA) may not have been significant, likely due to these anxiety types being more influenced by other factors like social skills and attachment styles, which were minimally impacted by the epidemic.

Secondly, the observed variance in the relationship between PSNU and AA across different economic contexts, notably between developing and developed countries, underscores the multifaceted influence of socio-economic, cultural, and technological factors on this dynamic. The amplified connection in developing countries may be attributed to greater socio-economic challenges, distinct cultural norms regarding social support and interaction, rising social network penetration, especially among younger demographics, and technological disparities influencing accessibility and user experience [229, 230]. Moreover, the role of social network as a coping mechanism for emotional distress, potentially fostering insecure attachment patterns, is more pronounced in these settings [231]. These findings highlight the necessity of considering contextual variations in assessing the psychological impacts of social network, advocating for a nuanced understanding of how socio-economic and cultural backgrounds mediate the relationship between PSNU and mental health outcomes [232]. Additionally, the relationship between PSNU and other types of anxiety (such as GA and SA) presents uniform characteristics across different economic contexts.

Thirdly, the significant moderating effects of measurement tools in the context of PSNU and its correlation with various forms of anxiety, including GA, and AA, are crucial in interpreting the research findings. Specifically, the study reveals that the Bergen Social Media Addiction Scale (BSMAS) demonstrates a stronger correlation between PSNU and GA, compared to other tools. Similarly, for AA, the Griffiths’ Problematic Internet Use Scale 2 (GPIUS2) and the Problematic Media Social Media Use Assessment Questionnaire (PMSMUAQ) show a more pronounced correlation with AA than the BSMAS or other instruments, but for SA and FoMO, the PSNU instrument doesn’t significantly moderate the correlation. The PSNU measurement tool typically contains an emotional change dimension. SA and FoMO, due to their specific conditional stimuli triggers and correlation with social networks [233, 234], are likely to yield more consistent scores in this dimension, while GA and AA may be less reliable due to their lesser sensitivity to specific conditional stimuli. Consequently, the adjustment effects of PSNU measurements vary across anxiety symptoms. Regarding the measurement tools for anxiety, different scales exhibit varying degrees of sensitivity in detecting the relationship with PSNU. The Generalized Anxiety Disorder Scale (GAD) and the Depression Anxiety Stress Scales 21 (DASS-21) are more effective in illustrating a strong relationship between GA and PSNU than the State-Trait Anxiety Inventory (STAI). In the case of AA, the Experiences in Close Relationships-21 (ECR-21) provides a more substantial correlation than the Relationship Questionnaire (RQ). Furthermore, for FoMO, the Fear of Missing Out Scale - Multi-Social Media Environment (FoMOS-MSME) is more indicative of a strong relationship with PSNU compared to the standard FoMOS or the T-S FoMOS. These findings underscore the importance of the selection of appropriate measurement tools in research. Different tools, due to their unique design, focus, and sensitivity, can reveal varying degrees of correlation between PSNU and anxiety disorders. This highlights the need for careful consideration of tool characteristics and their potential impact on research outcomes. It also cautions against drawing direct comparisons between studies without acknowledging the possible variances introduced by the use of different measurement instruments.

Fourthly, the significant moderating role of gender in the relationship between PSNU and SA, particularly pronounced in samples with a higher proportion of females. Women tend to engage more actively and emotionally with social network, potentially leading to an increased dependency on these platforms when confronting social anxiety [235]. This intensified use might amplify the association between PSNU and SA. Societal and cultural pressures, especially those related to appearance and social status, are known to disproportionately affect women, possibly exacerbating their experience of social anxiety and prompting a greater reliance on social network for validation and support [236]. Furthermore, women’s propensity to seek emotional support and express themselves on social network platforms [237] could strengthen this link, particularly in the context of managing social anxiety. Consequently, the observed gender differences in the relationship between PSNU and SA underscore the importance of considering gender-specific dynamics and cultural influences in psychological research related to social network use. In addition, gender consistency was observed in the association between PSNU and other types of anxiety, indicating no significant gender disparities.

Fifthly, the absence of a significant moderating effect of age on the relationship between PSNU and various forms of anxiety suggests a pervasive influence of social network across different age groups. This finding indicates that the impact of PSNU on anxiety is relatively consistent, irrespective of age, highlighting the universal nature of social network’s psychological implications [238]. Furthermore, this uniformity suggests that other factors, such as individual psychological traits or socio-cultural influences, might play a more crucial role in the development of anxiety related to social networking usage than age [239]. The non-significant role of age also points towards a potential generational overlap in social networking usage patterns and their psychological effects, challenging the notion that younger individuals are uniquely susceptible to the adverse effects of social network on mental health [240]. Therefore, this insight necessitates a broader perspective in understanding the dynamics of social network and mental health, one that transcends age-based assumptions.

Limitations

There are some limitations in this research. First, most of the studies were cross-sectional surveys, resulting in difficulties in inferring causality of variables, longitudinal study data will be needed to evaluate causal interactions in the future. Second, considerable heterogeneity was found in the estimated results, although heterogeneity can be partially explained by differences in study design (e.g., Time of measurement, region, gender, and measurement tools), but this can introduce some uncertainty in the aggregation and generalization of the estimated results. Third, most studies were based on Asian samples, which limits the generality of the results. Fourth, to minimize potential sources of heterogeneity, some less frequently used measurement tools were not included in the classification of measurement tools, which may have some impact on the results of heterogeneity interpretation. Finally, since most of the included studies used self-reported scales, it is possible to get results that deviate from the actual situation to some extent.

Conclusion

This meta-analysis aims to quantifies the correlations between PSNU and four specific types of anxiety symptoms (i.e., generalized anxiety, social anxiety, attachment anxiety, and fear of missing out). The results revealed a significant moderate positive association between PSNU and each of these anxiety symptoms. Furthermore, Subgroup analysis and meta-regression analysis indicated that gender, region, time of measurement, and instrument of measurement significantly influenced the relationship between PSNU and specific anxiety symptoms. Specifically, the measurement time and GA measurement tools significantly influenced the relationship between PSNU and GA. Gender significantly influenced the relationship between PSNU and SA. Region, PSNU measurement tools, and AA measurement tools all significantly influenced the relationship between PSNU and AA. The FoMO measurement tool significantly influenced the relationship between PSNU and FoMO. Regarding these findings, prevention interventions for PSNU and anxiety symptoms are important.