Skip to main content
NCBI home page
JAMA Network logoLink to JAMA Network
. 2024 Jun 24;178(8):814–822. doi: 10.1001/jamapediatrics.2024.2078

Social Media Use and Internalizing Symptoms in Clinical and Community Adolescent Samples

A Systematic Review and Meta-Analysis

Luisa Fassi 1,2,✉, Kirsten Thomas 3, Douglas A Parry 4, Amelia Leyland-Craggs 1, Tamsin J Ford 2, Amy Orben 1
PMCID: PMC11197453  PMID: 38913335

Key Points

Question

What is the proportion of research on the association between social media use and mental health in adolescent clinical populations and does it differ between clinical and community samples?

Findings

In this systematic review and meta-analysis of 143 studies, few focused on clinical populations, and these showed a positive association between social media use and internalizing symptoms. These results mirrored findings from community samples.

Meaning

The paucity of research on clinical populations limits the generalizability of existing research and hinders a comprehensive evaluation of the association between social media and mental health.

This systematic review and meta-analysis evaluates associations between social media use and internalizing symptoms among adolescents.

Abstract

Importance

In response to widespread concerns about social media’s influence on adolescent mental health, most research has studied adolescents from the general population, overlooking clinical groups.

Objective

To synthesize, quantify, and compare evidence on the association between social media use and internalizing symptoms in adolescent clinical and community samples.

Data Sources

Peer-reviewed publications from MEDLINE, Web of Science, PsycInfo, and Scopus (initially reviewed in May 2022 and updated in October 2023) and preprints from Europe PubMed Central (February 2023) published in English between 2007 and 2023.

Study Selection

Two blinded reviewers initially identified 14 211 cross-sectional and longitudinal studies quantifying the association between social media use and internalizing symptoms, excluding experimental studies and randomized clinical trials.

Data Extraction and Synthesis

PRISMA and MOOSE guidelines were followed, pooling data using a random-effects model and robust variance estimation. The quality of evidence was assessed using the Quality of Survey Studies in Psychology Checklist.

Main Outcomes and Measures

Articles were included if they reported at least 1 quantitative measure of social media use (time spent, active vs passive use, activity, content, user perception, and other) and internalizing symptoms (anxiety, depression, or both).

Results

The 143 studies reviewed included 1 094 890 adolescents and 886 effect sizes, 11% of which examined clinical samples. In these samples, a positive and significant meta-correlation was found between social media use and internalizing symptoms, both for time spent (n = 2893; r, 0.08; 95% CI, 0.01 to 0.15; P = .03; I2, 57.83) and user engagement (n = 859; r, 0.12; 95% CI, 0.09 to 0.15; P = .002; I2, 82.67). These associations mirrored those in community samples.

Conclusions and Relevance

The findings in this study highlight a lack of research on clinical populations, a critical gap considering public concerns about the increase in adolescent mental health symptoms at clinical levels. This paucity of evidence not only restricts the generalizability of existing research but also hinders our ability to evaluate and compare the link between social media use and mental health in clinical vs nonclinical populations.

Introduction

Adolescent mental health has declined substantially in recent years. The proportion of UK adolescents (aged 10-24 years)1 with a probable mental health condition increased from 10% to 25% between 2017 and 2022.2,3 Globally, 1 in 5 children and adolescents have a mental health condition, most commonly internalizing disorders (eg, anxiety or depression).4 The impact of such conditions is wide reaching and long lasting, affecting school attendance, interpersonal relationships, employment prospects, physical health, and suicide risk, with suicide now constituting the second-leading cause of death among 15- to 29-year-olds worldwide.4 Many raise concerns that social media, now ubiquitous (97% of young people are daily users),5 is accelerating current mental health declines.6,7

Scientific research investigating social media’s impact on adolescent mental health has failed to provide clarity. There is converging evidence for a small negative cross-sectional association between time spent on social media and well-being.8,9,10,11 However, longitudinal studies and those measuring social media use beyond time spent or mental health beyond general well-being show diverging results.12,13,14

To understand this heterogeneity, researchers have studied whether individual differences (eg, age, sex, or ethnicity) might moderate the relationship between social media use and mental health.15,16,17,18 However, the potential impact of the mental health status of the examined sample has been largely overlooked. Studies routinely recruit adolescents from the general population through schools, universities, or nationally representative surveys.12,13,14 While these samples can include individuals experiencing mental health symptoms at clinical levels, they often fail to distinguish them from those experiencing symptoms at subclinical or nonclinical levels.

Individuals with mental health conditions face unique challenges, such as interpersonal or sleep difficulties and educational disruptions.19 Adolescents with internalizing conditions, in particular, exhibit heightened sensitivity to social comparison and fear of negative evaluation.20,21 They might therefore use or be impacted by social media differently compared to peers. Failure to account for the nature and severity of mental health indicators could therefore restrict our ability to draw accurate inferences about social media’s relationship with mental health.

We addressed the extent and impact of this oversight in 3 steps. First, we completed a preregistered systematic review to quantify the proportion of studies investigating social media use and internalizing symptoms in adolescent clinical samples compared to community or nonclinical samples. Second, we performed a meta-analysis to extract the pooled association between social media use and internalizing symptoms in clinical samples, differentiating between time spent and other measures of social media engagement. Third, we compared the strength and direction of this association across clinical and community samples, testing whether sample type was a moderator.

This work allowed us to gauge whether and how current research in this area of substantial scientific and public interest can be used to make clinically informative recommendations. It also complements preexisting qualitative reviews,22 synthesizing the quantitative effect sizes in clinical populations and comparing these with community samples. Together, these findings can inform academics by identifying gaps for future research; clinicians, by summarizing research studying relevant populations; and policymakers, by guiding evidence-based decision-making for adolescents at risk.

Methods

Search Strategy, Selection, and Extraction

The protocol for this study was preregistered with Prospero (CRD42022321473), following the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline. MEDLINE, Web of Science, PsycInfo, and Scopus were searched (eAppendix 1 in Supplement 1) initially in May 2022 and updated in October 2023; forward/backward citation tracing via Google Scholar and preprint search via Europe PubMed Central in February 2023. We identified 14 211 records (8997 articles and 5214 preprints) and 8438 (5335 articles and 3103 preprints) remained after duplicate removal. Considering the nature of the study design, no ethical review was needed.

Selection criteria (eAppendix 2 in Supplement 1) were peer-reviewed English-language articles and preprints published in or after January 2007; quantitative time- or engagement-based social media use measures, self-reported or logged; quantitative symptom-based or other validated questionnaires of anxiety, depression, or both; and adolescent populations aged 10 to 24 years (if not provided: mean age ±1 SD in age range).

In terms of social media, we categorized types of engagement into 6 preregistered categories to allow meaningful description and pooling of studies (eAppendix 2 in Supplement 1): time spent and frequency, activities (eg, messaging and posting), content (eg, exposure to appearance-related content), user perception (eg, impact of likes on mood), active vs passive use, and other.

We categorized study samples into clinical, community or nonclinical. Clinical samples included adolescents who either scored above the clinical threshold on a validated questionnaire, reporting an active diagnosis, accessed mental health services, or were psychiatrically hospitalized. Community samples included adolescents across the entire distribution of internalizing symptoms without separation into clinical and nonclinical levels, while nonclinical samples excluded adolescents in the clinical range. We restricted our focus to studies that examined internalizing symptoms, excluding other conditions (eg, externalizing or neurodevelopmental) unless they were comorbid with internalizing symptoms.23

Following title and abstract screening by 2 independent reviewers, 231 records (226 articles and 5 preprints) remained and were full-text screened (Figure 1; eAppendix 3 in Supplement 1). Three independent reviewers double coded 10% of studies to harmonize the coding strategy (eAppendix 4 in Supplement 1; reliability: 95%) and then extracted study information, samples, measures, methodologies, and effect sizes. Risk of bias and quality of studies was assessed using an adapted version of the Quality of Survey Studies in Psychology (eAppendix 5 in Supplement 1).24

Figure 1. Flow Diagram.

Figure 1.

Open in a new tab

Reporting of study identification, screening and inclusion for the systematic review. PMC indicates PubMed Central.

Statistical Analysis

We completed all analyses in R version 4.1.2 (R Foundation; full list of packages on OSF).25 We first conducted descriptive analyses of the studies included in the systematic review. Specifically, we calculated the number of studies and effect sizes (with associated percentages), split by sample type, mental health measure, social media measure, social media data collection, study design, and global population (countries were coded based on the International Telecommunications Union classification26).

Next, we conducted meta-analyses to test the pooled association between social media use and internalizing symptoms for clinical and community samples. These meta-analyses were restricted to cross-sectional studies and the initial cross-sectional wave of longitudinal studies (see eAppendix 2 in Supplement 1 for details on this choice). Our confirmatory meta-analyses were restricted to studies measuring time spent on social media, to allow meaningful pooling of effect sizes due to measurement similarity, while in the exploratory meta-analyses, we examined measures of social media engagement.

The association was defined as positive when increased social media use was associated with increased internalizing symptoms. We used an a priori statistical significance level of α = .05 and interpreted effect sizes in line with Cohen (1988; small: r < 0.10, medium: r = 0.30, large: r > 0.50). For studies reporting effect sizes other than a correlation coefficient, we performed transformations where possible (eAppendix 6 in Supplement 1). We transformed all correlations from Fisher z back to Pearson r for reporting.

We used a random-effects model to calculate summary effect sizes due to the high level of heterogeneity. To account for variance inflation emerging from dependent observations for measures collected from the same participants, we used cluster-robust variance estimation based on the sandwich method with adjusted estimators for small samples and the correlated effects weighting scheme using robumeta (r = 0.80 for the within-study effect size correlation).27,28,29 Sensitivity analyses showed that using different r values did not affect the inferences made.25

Given that longitudinal studies have multiple waves per participant, the meta-analysis included only the effect size from the first wave to minimize variance inflation. However, no differences emerged in the strength and direction when including all waves (eAppendix 7 in Supplement 1).

Risk of Bias Assessment and Moderation

To assess potential bias due to small study effects, including publication bias, we visually inspected funnel plot symmetry and performed the Egger regression test.30,31 Further, we used a contour-enhanced funnel plot with superimposed areas of statistical significance (corresponding to P = .10, .05, and .01), interpreting an overrepresentation of effect sizes in the highlighted areas as indicative of publication bias.31 We conducted influence diagnostics (ie, the Cook distance, covariance ratios, and diagonal elements of the hat matrix) using metafor32 to identify outliers and performed leave-one-out sensitivity analyses with such outliers removed (eAppendix 8 in Supplement 1). To examine heterogeneity in effect sizes, we computed I2, interpreting values around 25%, 50%, and 75% to indicate low, moderate, and high heterogeneity, respectively.

We conducted 3 preregistered moderator analyses to investigate factors contributing to heterogeneity, namely, sample type (clinical or community samples; nonclinical samples were excluded due to a lack of power), mental health measure (anxiety, depression, or internalizing symptoms) and COVID-1933,34,35 (before vs during). We classified studies as happening during the COVID-19 pandemic if any data collection was performed after January 2020.4 Lastly, we conducted exploratory moderation analyses for age, sex, and the type of social media measure for the meta-analysis on social media engagement.

Results

Systematic Review: Quantifying the Proportion of Clinical Samples

After duplicate removal, we screened 8438 manuscripts (5335 articles and 3103 preprints), including 143 studies in the systematic review (141 articles and 2 preprints; Figure 1). Included studies had a combined sample size of 1 094 890 adolescents (mean, 7657; SD, 40 026; median, 680; minimum, 41; maximum, 388 275) and reported 886 effect sizes for the association between social media use and internalizing symptoms (eAppendix 9 in Supplement 1).

Studies investigating adolescent clinical samples were rare: 11% of effect sizes, corresponding to 99 effect sizes from 12 studies (Figure 2A; eAppendix 10 in Supplement 1). Most studies examined community samples (88% of effect sizes; 774 effect sizes from 133 studies), with very few focusing on nonclinical samples (1% of effect sizes; 13 effect sizes from 4 studies). The most common mental health measure was depression (67% of effect sizes; 595 effect sizes from 118 studies), while anxiety (26% of effect sizes; 228 effect sizes from 52 studies) and internalizing symptoms (7% of effect sizes; 63 effect sizes from 16 studies) were less frequently assessed (Figure 2B).

Figure 2. Proportion of Included Effect Sizes by Sample Type and Mental Health Measure.

Figure 2.

Open in a new tab

Grid of 10 × 10 (100%) squares representing the percentage of literature in the systematic review by sample type and mental health measure. The presented proportion is calculated based on the total number of effect sizes (N = 886).

Regarding social media measures, 92% of effect sizes were derived from studies using self-reports (816 effect sizes from 138 studies), while 8% used logged measures (70 effect sizes from 8 studies). Nearly half of the effect sizes were extracted from studies measuring time spent (43%; 381 effect sizes from 91 studies). Less common engagement-based measures included user perception (18%; 160 effect sizes from 36 studies), activity (15%; 131 effect sizes from 31 studies), active vs passive use (7%; 65 effect sizes from 14 studies), content (3%; 29 effect sizes from 4 studies), and other metrics (14%; 120 effect sizes from 21 studies). Most studies (66%; 94 studies) were cross-sectional, while 34% (49 studies) were longitudinal (eAppendix 11 in Supplement 1). In line with previous work,16 the most commonly studied populations were from the Global North (82%; 117 studies), compared to the Global South (18%; 26 studies).26

Overall, approximately half of the included studies (55%; 78 studies) were of acceptable quality based on the Quality of Survey Studies in Psychology Checklist. The remaining 45% (65 studies) were classified as being of questionable quality (eAppendix 5 in Supplement 1).

Meta-Analysis: Quantifying Associations in Clinical Samples

Social Media Time Spent

Seven studies of clinical populations (15 effect sizes) measured time spent on social media. The total sample size was 2893 (mean, 413; SD, 585; median, 224; minimum, 49; maximum, 1722). In our confirmatory meta-analysis, we found a positive and significant meta-correlation between time spent on social media and mental health symptoms (r, 0.08; 95% CI, 0.01 to 0.15; P = .03), with moderate heterogeneity (I2, 57.83) (Figure 336,37,38,39,40,41,42). Further, the Egger regression test showed no evidence of small study bias (β, −2.19; SE, 0.46; P = .98) (funnel plots in eAppendix 12 in Supplement 1).

Figure 3. Time Spent on Social Media and Internalizing Symptoms in Clinical Groups.

Figure 3.

Open in a new tab

Forest plot for the individual and pooled effect sizes representing the association between time spent on social media and mental health symptoms. Effect sizes for clinical samples are shown both individually (ie, separate rows with author, year, and sample size) and as a pooled estimate (model with cluster-robust variance estimation [RVE]), while the effect size for community samples is only presented as a pooled estimate at the bottom. Individual Pearson r coefficients are depicted as filled squares with the size indicating the relative weight, based on sample size, of each effect size estimate for clinical studies in the meta-analysis. Increased time spent on social media was associated with decreased symptoms (to the left of zero) or increased symptoms (to the right of zero). The blue diamond and the dashed line represent the overall summary effect size across all clinical studies (r, 0.08; 95% CI, 0.01 to 0.15; P < .001), calculated using RVE to account for dependencies between effect sizes coming from the same study. The orange diamond and dashed line represent the overall summary effect size across all studies run on community samples (r, 0.12; 95% CI, 0.09 to 0.15; P < .001). The error bars and diamond width represent the 95% CIs for the effect sizes. The dotted reference line at r = 0 represents the point of reference for no correlation.

Social Media Engagement

The need to move beyond time spent measures of social media use has been widely acknowledged, as these measures are simplistic and fail to distinguish between types of activities or content that can differentially relate to mental health.43,44 Researchers have therefore advocated for using engagement-based measures, which we examined in an exploratory meta-analysis.

Four studies of clinical populations (19 effect sizes) used engagement-based measures (eAppendix 2 in Supplement 1), specifically social media activities (10 effect sizes; eg, selfie posting) and user perception (9 effect sizes; importance of social media use to daily life), with a total sample size of 859 (mean, 215; SD, 122; median, 233; minimum, 49; maximum, 343). We found a positive and significant meta-correlation between these social media measures and mental health symptoms (r, 0.12; 95% CI, 0.09 to 0.15; P = .002), with high heterogeneity (I2, 82.67) (Figure 436,37,38,40). Further, the Egger regression test showed no evidence of small study bias (β, −0.55; SE, 0.15; P = .93) (funnel plots in eAppendix 12 in Supplement 1).

Figure 4. Social Media Engagement and Internalizing Symptoms in Clinical Groups.

Figure 4.

Open in a new tab

Forest plot for the individual and pooled effect sizes representing the association between social media engagement and mental health symptoms. Effect sizes for clinical samples are shown both individually (ie, separate rows with author, year, and sample size) and as a pooled estimate (model with cluster-robust variance estimation [RVE]), while the effect size for community samples is only presented as a pooled estimate at the bottom.Individual Pearson r coefficients are depicted as filled squares with the size indicating the relative weight, based on sample size, of each effect size estimate for clinical studies in the meta-analysis. Increased social media engagement was associated with decreased symptoms (to the left of zero) or increased symptoms (to the right of zero). The blue diamond and dashed line represent the overall summary effect size across all clinical studies (r, 0.12; 95% CI, 0.09 to 0.15; P < .001), calculated using RVE to account for dependencies between effect sizes coming from the same study. The error bars and diamond width represent the 95% CIs. The orange diamond and dashed line represent the overall summary effect size in community samples (r, 0.14; 95% CI, 0.10 to 0.18; P < .001). The dotted reference line at r = 0 represents the point of reference for no correlation. More information on the type of social media engagement measured in each study is reported in eAppendix 15 in Supplement 1.

aAll reported measures of social media use are for any platform (no study measured activity or user perception in relation to a specific platform).

Meta-Analysis: Comparing Associations Between Clinical and Community Samples

Social Media Time Spent

We also ran a meta-analysis of the 49 studies (and 99 effect sizes) testing community samples (n = 479 215; mean, 9780; SD, 55 482; median, 442; minimum, 41; maximum, 388 275). We found a positive and significant meta-correlation between time spent on social media and internalizing symptoms (r, 0.12; 95% CI, 0.09 to 0.15; P < .001) (Figure 3; eAppendix 13 in Supplement 1). This is similar to the meta-correlation found in clinical samples (r, 0.08; 95% CI, 0.01 to 0.15; P = .03; I2, 57.83) but shows higher levels of heterogeneity (I2, 98.23).

To further test whether sample type influenced the association between time spent on social media and internalizing symptoms, we ran a combined meta-analysis (56 studies with 117 effect sizes; n = 482 273; mean, 8612; SD, 51 926; median, 372; minimum, 41; maximum, 388 275) and tested sample type as a moderator. We found an overall positive meta-correlation across all sample types (r, 0.12, 95% CI, 0.09 to 0.14; P < .001; I2, 98.0) with no evidence of small study bias (β, −0.86; SE, 0.48; P = .96) (funnel plots in eAppendix 12 in Supplement 1).

After excluding nonclinical samples due to a lack of power (3 effect sizes from 3 studies), we tested sample type as a moderator (clinical vs community sample). We found nonsignificant results (β, 0.05; SE, 0.03; t, 1.6; 95% CI, −0.02 to 0.12; P = .145), and heterogeneity remained high (I2, 98.0) (Table). Sample type was therefore not considered a key factor explaining differences in the association between time spent on social media and internalizing symptoms among adolescents.

Table. Moderation Analysesa.
Moderator Level Studies, No. Effect sizes, No. Estimate (SE) t Value 95% CI P value
Time spent on social media and internalizing symptoms
Sample type Clinical [reference] 7 15 NA NA NA NA
Community 49 99 0.05 (0.03) 1.6 −0.02 to 0.12 .15
COVID-19 Before [reference] 44 100 NA NA NA NA
During 9 12 0.04 (0.04) 1.16 −0.04 to 0.12 .27
Mental health measure Internalizing [reference] 4 5 NA NA NA NA
Depression 48 69 −0.07 (0.08) −0.81 −0.30 to 0.17 .47
Anxiety 19 40 −0.07 (0.08) −0.81 −0.27 to 0.14 .45
Social media engagement and internalizing symptoms
Sample type Clinical [reference] 4 19 NA NA NA NA
Community 62 217 0.01 (0.02) 0.72 −0.05 to 0.80 .52
COVID-19 Before [reference] 51 196 NA NA NA NA
During 10 31 −0.06 (0.05) −1.23 −0.15 to 0.04 .24
Mental health measure Internalizing [reference] 3 8 NA NA NA NA
Depression 55 169 0.02 (0.03) 0.74 −0.08 to 0.12 .53
Anxiety 24 59 0.03 (0.04) 0.78 −0.11 to 0.17 .49
Social media measure Other [reference] 11 20 NA NA NA NA
Active vs passive 10 42 0.00 (0.04) 0.02 −0.09 to 0.09 .99
Activity 22 69 0.05 (0.05) 0.90 −0.07 to 0.17 .38
Content 2 25 −0.07 (0.04) - 1.74 −0.34 to 0.19 .27
Perception 30 80 0.10 (0.05) 2.06 −0.00 to 0.20 .06
Open in a new tab

Abbreviation: NA, not applicable.

a

Results of moderation analyses for the meta-correlation of internalizing symptoms with time spent on social media and engagement-based social media use measures.

We also tested whether the mental health measure (anxiety, depression, and internalizing symptoms) and COVID-19 (before vs during) were moderators. Neither the mental health measure (depression vs internalizing: β, −0.07; SE, 0.08; t, −0.81; 95% CI, −0.30 to 0.17; P = .47; anxiety vs internalizing: β, −0.07; SE, 0.08; t, −0.81; 95% CI, −0.27 to 0.14; P = .45) nor COVID-19 (β, 0.04; SE, 0.04; t, 1.16; 95% CI, −0.04 to 0.12; P = .27) explained heterogeneity in the meta-correlation between time spent on social media and mental health (Table). There was also no moderation for age or sex (eAppendix 14 in Supplement 1).

Social Media Engagement

We repeated the same analyses for studies measuring social media engagement. As with the meta-correlation found in clinical samples (r, 0.12; 95% CI, 0.09 to 0.15; P = .002; I2, 82.67), we found a positive and significant association between social media engagement and mental health symptoms (r, 0.14; 95% CI, 0.10 to 0.18; P < .001) (Figure 4) in community samples (217 effect sizes from 62 studies; n = 65,799; mean, 1061; SD, 1607; median, 546; minimum, 41; maximum, 10 563). There were high levels of heterogeneity (I2 = 94.85) (eAppendix 13 in Supplement 1).

We included 65 studies with 236 effect sizes in our combined meta-analysis. Across all sample types (n = 68 807; mean, 1,058; SD, 1605; median, 551; minimum, 41; maximum, 10 563), there was a positive meta-correlation between social media engagement and internalizing symptoms (r, 0.14; 95% CI, 0.10 to 0.17; P < .001) with high heterogeneity (I2, 94.63). There was no evidence of small study bias (β, 0.54; SE, 0.67; P = .22) also confirmed by visual inspection of the funnel plot (eAppendix 11 in Supplement 1).

Sample type (clinical vs community) was not considered a significant moderator of the overall association between social media engagement and internalizing symptoms (β, 0.01; SE, 0.02; t, 0.72; 95% CI, −0.05 to 0.08; P = .52), and heterogeneity remained high (I2, 94.70) (Table). Our additional moderation analyses, summarized in the Table, showed that neither mental health measure (anxiety vs internalizing: β, 0.03; SE, 0.04; t, 0.78; 95% CI, −0.11 to 0.17; P = .49; depression vs internalizing: β, 0.02; SE, 0.03; t, 0.74; 95% CI, −0.08 to 0.12; P = .53) nor COVID-19 (β, −0.06; SE, 0.05; t, −1.23; 95% CI, −0.15 to 0.04; P = .24) explained heterogeneity in the meta-correlation between social media engagement and mental health symptoms. There was also no moderation for the type of social media measures (Table), age, or sex (eAppendix 14 in Supplement 1).

Discussion

This systematic review and meta-analysis synthesized data from 16 years of research examining the association between social media use and internalizing mental health in more than 1 million adolescents. We found that 11% of studies examined clinical populations, while 88% recruited adolescents from the general population. There was a small, positive, and significant meta-correlation between social media use and internalizing mental health in clinical samples, regardless of whether time- or engagement-based social media metrics were studied. Notably, these meta-correlations did not substantially differ from those found in community samples.

Our first finding highlights a lack of research on clinical populations. Notably, adolescents affected by clinical-level anxiety and depression can face higher social withdrawal, sleep problems, low self-esteem, increased susceptibility to peer influence, and excessive rumination compared to adolescents from the general population.45 These symptoms may alter their social media interaction and its impact on their mental health.36,46,47 Hence, the lack of an evidence base in these high-risk populations, resulting in limited investigation of clinically relevant mechanisms, limits our capacity to draw accurate inferences about the relationship between social media use and mental health.

In contrast to the common assumption that clinical populations might show a stronger association between social media use and mental health declines than community samples,37 we found no substantial differences. This result could be explained by the increasing occurrence of clinically significant symptoms in community samples3,4 and the diminishing divide between these groups. Alternatively, adolescent clinical populations might adjust their social media use based on their mental health needs, leading to comparable usage patterns and correlations. Lastly, clinical groups could also be experiencing less variability in mental health symptomatology (eg, ceiling effects), lessening the observable correlations between social media use and mental health symptoms.

Limitations

We underscore some limitations of this work. First, inaccurate self-report measures of social media use48 might have decreased our ability to locate differences between clinical and community samples even if they existed. Second, while we summarized studies with longitudinal effect sizes and selected control variables as part of our systematic review (eAppendix 11 in Supplement 1), our meta-analysis only included correlations. Hence, no causal inferences can be drawn from the pooled meta-correlation about whether increased social media use leads to higher symptoms or vice versa.

Further, we categorized social media engagement with 5 predefined categories, which are not exhaustive and could mask important nuances. For example, the role of social media content will depend on its nature, which could be positive, negative, or neutral. In addition, we focused on internalizing mental health only. Hence, conclusions cannot be generalized to other conditions. Limiting the inclusion of studies to those published in English may introduce language bias and exclude valuable research conducted in other languages.

Conclusions

The findings in this study demonstrated the moderate to high levels of heterogeneity common to this research area.13 This variation could potentially be explained by individual differences in demographic characteristics among participants that we did not test, due to the lack of data or statistical power. However, when conducting exploratory moderation analyses for age and sex, we found that neither of those factors explained heterogeneity. We also found no evidence of publication bias.

Many worry about social media’s role in increased clinical-level mental health symptoms among adolescents. However, current research falls short of adequately targeting the specific populations required to draw accurate inferences about this matter. Despite our initial findings of a similar association across clinical and community samples, there is still a real risk that we are incorrectly generalizing results from the general population to young people with mental health conditions. The potential impact of this extends beyond research to clinical practice and policymaking. For clinicians, more research on clinical populations could enrich strategies for patient consultations and family education, allowing for the integration of social media management into treatment plans. For policymakers, it could shape policies for safer social media platforms and funding allocation toward mental health programs. In a world increasingly saturated by digital technology, we cannot afford to design prevention programs, interventions, and regulations without knowing that they work for everyone, especially those who are most vulnerable.

Supplement 1.

eAppendix 1. Search strings

eAppendix 2. Selection criteria

eAppendix 3. Identification and screening

eAppendix 4. Data extraction

eAppendix 5. Quality of evidence

eAppendix 6. Conversion of effect sizes to Pearson’s r

eAppendix 7. Sensitivity analysis for longitudinal effect sizes

eAppendix 8. Outliers

eAppendix 9. Systematic review descriptive results

eAppendix 10. Clinical studies

eAppendix 11. Systematic review of longitudinal studies

eAppendix 12. Funnel plots

eAppendix 13. Distribution of effect sizes in community samples

eAppendix 14. Meta-analysis moderation for age and gender

eAppendix 15. Clinical studies enhanced forest plot

eReferences

jamapediatr-e242078-s001.pdf (2.3MB, pdf)
Supplement 2.

Data sharing statement

jamapediatr-e242078-s002.pdf (19.7KB, pdf)

References

  • 1.Sawyer SM, Azzopardi PS, Wickremarathne D, Patton GC. The age of adolescence. Lancet Child Adolesc Health. 2018;2(3):223-228. doi: 10.1016/S2352-4642(18)30022-1 [DOI] [PubMed] [Google Scholar]
  • 2.NHS Digital . Rate of mental disorders among 17 to 19 year olds increased in 2022, new report shows. Accessed May 20, 2024. https://digital.nhs.uk/news/2022/rate-of-mental-disorders-among-17-to-19-year-olds-increased-in-2022-new-report-shows
  • 3.NHS Digital . Mental health of children and young people in England, 2017. Accessed May 20, 2024. https://digital.nhs.uk/data-and-information/publications/statistical/mental-health-of-children-and-young-people-in-england/2017/2017
  • 4.World Health Organisation . Mental health. Published 2021. Accessed May 20, 2024. https://www.who.int/health-topics/mental-health.
  • 5.Ofcom. Ofcom Report 2022. https://www.ofcom.org.uk/__data/assets/pdf_file/0024/234609/childrens-media-use-and-attitudes-report-2022.pdf
  • 6.Gunnell D, Kidger J, Elvidge H. Adolescent mental health in crisis. BMJ. 2018;361:k2608. doi: 10.1136/bmj.k2608 [DOI] [PubMed] [Google Scholar]
  • 7.Twenge JM, Joiner TE, Rogers ML, Martin GN. Corrigendum: “Increases in depressive symptoms, suicide-related outcomes, and suicide rates among US adolescents after 2010 and links to increased new media screen time.” Clin Psychol Sci. 2019;7(2):397. doi: 10.1177/2167702618824060 [DOI] [Google Scholar]
  • 8.Orben A, Przybylski AK. The association between adolescent well-being and digital technology use. Nat Hum Behav. 2019;3(2):173-182. doi: 10.1038/s41562-018-0506-1 [DOI] [PubMed] [Google Scholar]
  • 9.Marciano L, Driver CC, Schulz PJ, Camerini AL. Dynamics of adolescents’ smartphone use and well-being are positive but ephemeral. Sci Rep. 2022;12(1):1316. doi: 10.1038/s41598-022-05291-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Huang C. Time spent on social network sites and psychological well-being: a meta-analysis. Cyberpsychol Behav Soc Netw. 2017;20(6):346-354. doi: 10.1089/cyber.2016.0758 [DOI] [PubMed] [Google Scholar]
  • 11.Steinsbekk S, Nesi J, Wichstrøm L. Social media behaviors and symptoms of anxiety and depression. a four-wave cohort study from age 10-16 years. Comput Hum Behav. 2023;147:107859. doi: 10.1016/j.chb.2023.107859 [DOI] [Google Scholar]
  • 12.Odgers CL, Jensen MR. Annual research review: adolescent mental health in the digital age: facts, fears, and future directions. J Child Psychol Psychiatry. 2020;61(3):336-348. doi: 10.1111/jcpp.13190 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Valkenburg PM, Meier A, Beyens I. Social media use and its impact on adolescent mental health: an umbrella review of the evidence. Curr Opin Psychol. 2022;44:58-68. doi: 10.1016/j.copsyc.2021.08.017 [DOI] [PubMed] [Google Scholar]
  • 14.Ferguson C, Kaye L, Branley-Bell D, et al. Like this meta-analysis: screen media and mental health. Prof Psychol-Res Pract. 2022;53(2):205-214. doi: 10.1037/pro0000426 [DOI] [Google Scholar]
  • 15.Orben A, Przybylski AK, Blakemore SJ, Kievit RA. Windows of developmental sensitivity to social media. Nat Commun. 2022;13(1):1649. doi: 10.1038/s41467-022-29296-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Ghai S, Fassi L, Awadh F, Orben A. Lack of sample diversity in research on adolescent depression and social media use: a scoping review and meta-analysis. Clin Psychol Sci. 2023;11(5):759-772. doi: 10.1177/21677026221114859 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Liu M, Kamper-DeMarco KE, Zhang J, Xiao J, Dong D, Xue P. Time spent on social media and risk of depression in adolescents: a dose-response meta-analysis. Int J Environ Res Public Health. 2022;19(9):5164. doi: 10.3390/ijerph19095164 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Beyens I, Pouwels JL, van Driel II, Keijsers L, Valkenburg PM. The effect of social media on well-being differs from adolescent to adolescent. Sci Rep. 2020;10(1):10763. doi: 10.1038/s41598-020-67727-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Kieling C, Baker-Henningham H, Belfer M, et al. Child and adolescent mental health worldwide: evidence for action. Lancet. 2011;378(9801):1515-1525. doi: 10.1016/S0140-6736(11)60827-1 [DOI] [PubMed] [Google Scholar]
  • 20.Rao PA, Beidel DC, Turner SM, Ammerman RT, Crosby LE, Sallee FR. Social anxiety disorder in childhood and adolescence: descriptive psychopathology. Behav Res Ther. 2007;45(6):1181-1191. doi: 10.1016/j.brat.2006.07.015 [DOI] [PubMed] [Google Scholar]
  • 21.Teachman BA, Allen JP. Development of social anxiety: social interaction predictors of implicit and explicit fear of negative evaluation. J Abnorm Child Psychol. 2007;35(1):63-78. doi: 10.1007/s10802-006-9084-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Kostyrka-Allchorne K, Stoilova M, Bourgaize J, Rahali M, Livingstone S, Sonuga-Barke E. Review: Digital experiences and their impact on the lives of adolescents with pre-existing anxiety, depression, eating and nonsuicidal self-injury conditions—a systematic review. Child Adolesc Ment Health. 2023;28(1):22-32. doi: 10.1111/camh.12619 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Lilienfeld SO. Comorbidity between and within childhood externalizing and internalizing disorders: reflections and directions. J Abnorm Child Psychol. 2003;31(3):285-291. doi: 10.1023/A:1023229529866 [DOI] [PubMed] [Google Scholar]
  • 24.Protogerou C, Hagger MS. A checklist to assess the quality of survey studies in psychology. Methods Psychol. 2020;3:100031. doi: 10.1016/j.metip.2020.100031 [DOI] [Google Scholar]
  • 25.OSF . RMarkdown files for the systematic review and meta-analysis. Accessed May 20, 2024. https://osf.io/gr3xh/?view_only=7b7434c46ad44d6fbbeafcd1481f2e93
  • 26.Wikimedia . International Telecommunications Union classification (2021). List of countries by regional classification. Accessed May 29, 2024. https://meta.wikimedia.org/wiki/List_of_countries_by_regional_classification
  • 27.Fisher Z, Tipton E. robumeta: an R-package for robust variance estimation in meta-analysis. arXiv. Preprint posted online March 2015. https://arxiv.org/abs/1503.02220v1
  • 28.Hedges LV, Tipton E, Johnson MC. Robust variance estimation in meta-regression with dependent effect size estimates. Res Synth Methods. 2010;1(1):39-65. doi: 10.1002/jrsm.5 [DOI] [PubMed] [Google Scholar]
  • 29.R Project . Robumeta package. Accessed May 20, 2024. https://cran.r-project.org/web/packages/robumeta/index.html
  • 30.Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ. 1997;315(7109):629-634. doi: 10.1136/bmj.315.7109.629 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Peters JL, Sutton AJ, Jones DR, Abrams KR, Rushton L. Contour-enhanced meta-analysis funnel plots help distinguish publication bias from other causes of asymmetry. J Clin Epidemiol. 2008;61(10):991-996. doi: 10.1016/j.jclinepi.2007.11.010 [DOI] [PubMed] [Google Scholar]
  • 32.Viechtbauer W. Conducting meta-analyses in R with the metafor package. J Stat Softw. 2010;36(3):1-48. doi: 10.18637/jss.v036.i03 [DOI] [Google Scholar]
  • 33.Xiang M, Zhang Z, Kuwahara K. Impact of COVID-19 pandemic on children and adolescents’ lifestyle behavior larger than expected. Prog Cardiovasc Dis. 2020;63(4):531-532. doi: 10.1016/j.pcad.2020.04.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Marques de Miranda D, da Silva Athanasio B, Sena Oliveira AC, Simoes-E-Silva AC. How is COVID-19 pandemic impacting mental health of children and adolescents? Int J Disaster Risk Reduct. 2020;51:101845. doi: 10.1016/j.ijdrr.2020.101845 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Salzano G, Passanisi S, Pira F, et al. Quarantine due to the COVID-19 pandemic from the perspective of adolescents: the crucial role of technology. Ital J Pediatr. 2021;47(1):40. doi: 10.1186/s13052-021-00997-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Nesi J, Burke TA, Extein J, et al. Social media use, sleep, and psychopathology in psychiatrically hospitalized adolescents. J Psychiatr Res. 2021;144:296-303. doi: 10.1016/j.jpsychires.2021.10.014 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Nesi J, Burke TA, Caltabiano A, Spirito A, Wolff JC. Digital media-related precursors to psychiatric hospitalization among youth. J Affect Disord. 2022;310:235-240. doi: 10.1016/j.jad.2022.05.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Fardouly J, Magson NR, Rapee RM, Johnco CJ, Oar EL. The use of social media by Australian preadolescents and its links with mental health. J Clin Psychol. 2020;76(7):1304-1326. doi: 10.1002/jclp.22936 [DOI] [PubMed] [Google Scholar]
  • 39.Hamilton JL, Biernesser C, Moreno MA, et al. Social media use and prospective suicidal thoughts and behaviors among adolescents at high risk for suicide. Suicide Life Threat Behav. 2021;51(6):1203-1212. doi: 10.1111/sltb.12801 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Uçar HN, Eray Ş, Kocael Ö, et al. Big data in adolescent psychiatry: do patients share their psychiatric symptoms on social networking sites? Psychiatr Danub. 2018;30(4):395-403. doi: 10.24869/psyd.2018.395 [DOI] [PubMed] [Google Scholar]
  • 41.Noel JK, Jacob S, Wensley IA, Rosenthal SR. Subjective smartphone screen time and co-morbid mental illness. J Technol Behav Sci. 2022;7(4):578-587. doi: 10.1007/s41347-022-00276-0 [DOI] [Google Scholar]
  • 42.Kelly Y, Zilanawala A, Booker C, Sacker A. Social media use and adolescent mental health: findings from the UK Millennium Cohort Study. E Clinical Medicine. 2019;6:59-68. doi: 10.1016/j.eclinm.2018.12.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Parry DA, Fisher JT, Mieczkowski H, Sewall CJR, Davidson BI. Social media and well-being: a methodological perspective. Curr Opin Psychol. 2022;45:101285. doi: 10.1016/j.copsyc.2021.11.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Meier A, Reinecke L. Computer-mediated communication, social media, and mental health: a conceptual and empirical meta-review. Commun Res. 2020;48(8). doi: 10.1177/0093650220958224 [DOI] [Google Scholar]
  • 45.Huberty TJ. Emotional and behavioral problems, students with. In: Spielberger CD, ed. Encyclopedia of Applied Psychology. Elsevier; 2004:723-730, doi: 10.1016/B0-12-657410-3/00791-1 [DOI] [Google Scholar]
  • 46.Radovic A, Gmelin T, Stein BD, Miller E. Depressed adolescents’ positive and negative use of social media. J Adolesc. 2017;55:5-15. doi: 10.1016/j.adolescence.2016.12.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Weinstein E, Kleiman EM, Franz PJ, et al. Positive and negative uses of social media among adolescents hospitalized for suicidal behavior. J Adolesc. 2021;87:63-73. doi: 10.1016/j.adolescence.2020.12.003 [DOI] [PubMed] [Google Scholar]
  • 48.Scharkow M. The accuracy of self-reported internet use—a validation study using client log data. Commun Methods Meas. 2016;10(1):13-27. doi: 10.1080/19312458.2015.1118446 [DOI] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement 1.

eAppendix 1. Search strings

eAppendix 2. Selection criteria

eAppendix 3. Identification and screening

eAppendix 4. Data extraction

eAppendix 5. Quality of evidence

eAppendix 6. Conversion of effect sizes to Pearson’s r

eAppendix 7. Sensitivity analysis for longitudinal effect sizes

eAppendix 8. Outliers

eAppendix 9. Systematic review descriptive results

eAppendix 10. Clinical studies

eAppendix 11. Systematic review of longitudinal studies

eAppendix 12. Funnel plots

eAppendix 13. Distribution of effect sizes in community samples

eAppendix 14. Meta-analysis moderation for age and gender

eAppendix 15. Clinical studies enhanced forest plot

eReferences

jamapediatr-e242078-s001.pdf (2.3MB, pdf)
Supplement 2.

Data sharing statement

jamapediatr-e242078-s002.pdf (19.7KB, pdf)

Articles from JAMA Pediatrics are provided here courtesy of American Medical Association

RESOURCES