Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Oct 27.
Published in final edited form as: Pediatrics. 2026 Nov 1;158(5):e2025075060. doi: 10.1542/peds.2025-075060

Problematic Social Media Use and Depressive and Anxiety Symptoms in Adolescents

Jason M Nagata 1, Christopher D Otmar 1, Elizabeth J Li 1, Kyle T Ganson 2, Jason M Lavender 3,4, Alexander Testa 5, Jinbo He 6, Dimitri A Christakis 7,8, Fiona C Baker 9
PMCID: PMC13700007  NIHMSID: NIHMS2217785  PMID: 42893794

Abstract

OBJECTIVE

Given limited longitudinal data on social media use and mental health, this study examined within-person lagged associations between problematic social media use and depressive and anxiety symptoms across 5 annual assessments among adolescents.

METHODS

This prospective cohort study used data from the Adolescent Brain Cognitive Development Study, years 2 to 6. The analytic sample included 11 279 participants (aged 12.01 ± 0.66 at year 2; 47.6% female) with at least 1 measure of problematic social media use and depressive or anxiety symptoms across years 2 to 6.

RESULTS

Across 5 years, random-intercept cross-lagged panel models (RI-CLPMs) indicated positive bidirectional lagged associations between problematic social media use and depressive symptoms from year 3 to year 4 and from year 4 to year 5. No significant cross-lagged associations were observed from year 2 to year 3 or from year 5 to year 6. For anxiety symptoms, no significant cross-lagged associations were observed across years. In multigroup RI-CLPMs, sex differences were observed in both models, but stronger cross-lagged associations among female than male participants were specific to depressive symptoms.

CONCLUSIONS

Problematic social media use and depressive symptoms showed year-specific bidirectional lagged associations, whereas within-person prospective associations between problematic social media use and anxiety symptoms were not observed. Sex differences were more apparent for depressive symptoms, for which lagged associations were stronger and more consistently bidirectional among female participants. Clinical and policy efforts should prioritize problematic features of social media use and consider developmental timing and sex-specific risk profiles in screening and intervention.

Introduction

Recent reports have found escalating rates of depression and anxiety among adolescents, prompting the US Surgeon General to issue a 2023 advisory.1–3 Social media has been implicated in this crisis,3,4 but frequency-based metrics such as total screen time alone offer an incomplete picture of mental health risk.5,6 In contrast, problematic or addictive social media use—characterized by withdrawal, conflict, tolerance, and relapse—appears to be a more salient predictor of internalizing symptoms.7,8 The US Surgeon General has reframed the discourse, emphasizing that risk lies less in duration of use and more in design-facilitated dependency, urging policy shifts toward platform accountability, warning labels, and independent audits of digital products that may exploit neurodevelopmental vulnerabilities in adolescent users.3 Consistent with the differential susceptibility model,1 recent reviews of social media and self-esteem report that associations are typically small in magnitude and highly heterogeneous across individuals.9 These person-specific effects suggest that problematic patterns of engagement, rather than overall frequency of use, may serve as more reliable markers of psychosocial vulnerability.

Problematic social media use (PSMU) may be particularly consequential during adolescence, a sensitive developmental window marked by heightened reward sensitivity, ongoing prefrontal maturation, and intensified social-evaluative concerns.7,10,11 Contemporary developmental models posit that adolescent mental health and digital media use are linked via self-reinforcing feedback loops.12 Youth with elevated internalizing symptoms—characterized by inwardly directed distress, including anxiety and depression9—may turn to social media for mood regulation.13 Although such engagement may provide temporary relief, it is also associated with disrupted sleep,14,15 reduced face-to-face social interaction,16,17 and increased upward social comparison.18 These processes may exacerbate depressive and anxious symptoms, perpetuating a cycle in which worsening affect drives further engagement. Socially vulnerable adolescents may be especially susceptible, as online interactions can both reflect and amplify emotion dysregulation, peer detachment, and exclusion.19–22 Developmental cascade frameworks23 suggest that these mutually reinforcing processes accumulate over time, eroding self-regulatory capacity, heightening sensitivity to peer feedback, and displacing health-promoting behaviors such as sleep and physical activity. Moreover, algorithms that drive social media engagement may exacerbate symptoms by tailoring content to users’ prior activity patterns.24,25 Examining bidirectional, time-ordered associations between PSMU and mental health symptoms may help identify maladaptive engagement patterns as early behavioral signs of escalating psychosocial distress.

Existing research on social media use and adolescent mental health has primarily relied on between-person comparisons (eg, contrasting adolescents with higher vs lower average symptom levels). This investigation extends prior work by leveraging data from the Adolescent Brain Cognitive Development (ABCD) Study, a multisite longitudinal US cohort, to examine within-individual changes over time, with a focus on potential bidirectional relationships between PSMU and internalizing symptoms.26–28 Earlier analyses from the same cohort showed that time spent on social media was associated with later depressive symptoms in early adolescence.29 Building on this, the present study focuses on PSMU, which reflects adolescents’ own reports that social media use has become distressing and difficult to regulate. Sex-related differences in reward sensitivity and emotion regulation become particularly salient during adolescence, coinciding with an observed increase in mood disorder risk among female compared with male individuals.30,31 These differences also align with higher rates of social media use among adolescent female individuals, raising questions about sex-specific susceptibility to problematic use.32 Accordingly, we also examined whether lagged associations differed by sex using multigroup models. We hypothesized that PSMU would be bidirectionally associated with depressive and anxiety symptoms across time, with increases in one construct corresponding to subsequent increases in the other within individuals.

Methods

Study Participants and Analytic Sample

Data for this analysis were drawn from adolescents (N = 11 279) in the ABCD Study (mean [SD] age at year 2, 12.01 [0.66] years; 47.6% female), a US population-based, multisite cohort that initially enrolled approximately 11 800 youth aged 9 to 10 years from 21 research centers in 17 states between 2016 and 2018. This investigation used data release 6.0 (July 2025). The primary analytic sample included participants with at least 1 measure of self-reported PSMU and parent-reported depressive or anxiety symptoms across years 2 to 6. Models were estimated using full information maximum likelihood (FIML), which uses all available repeated measures to estimate model parameters.33 Sample characteristics are shown in Supplemental Table 1; details on missing data and FIML procedures are provided in Supplemental Methods 1. The ABCD protocol was approved by a central institutional review board at the University of California, San Diego, with written informed consent from parents or caregivers and assent from youths. Participants received monetary compensation.

Measures

Problematic Social Media Use (Youth-Reported)

Adolescents completed the 6-item Social Media Addiction Questionnaire34 annually from year 2 to 6. The measure, adapted from the Bergen Facebook Addiction Scale, assesses 6 core components of behavioral addiction: salience, tolerance, mood modification, withdrawal, conflict, and loss of control.35 Items were rated on a 6-point Likert-type scale ranging from 1 (“never”) to 6 (“very often”), with higher scores indicating more problematic use. In the present analyses, mean scores (possible range 1–6) were treated as continuous observed variables. The measure was only administered to adolescents who reported having at least 1 social media account at the time of the survey. Psychometric analyses in the ABCD Study have demonstrated that the measure shows a consistent factor structure across sex and race and ethnicity.36

Depressive and Anxiety Symptoms (Parent-Reported)

Depressive and anxiety symptoms were assessed annually from year 2 to 6 using parent-reported raw scores (see Supplemental Methods 2) from the Diagnostic and Statistical Manual of Mental Disorders (Fifth Edition)–Oriented Affective Problems (Depressive) and Anxiety Problems subscales of the Child Behavior Checklist (CBCL).37 Each item was rated 0 (Not True), 1 (Somewhat or Sometimes True), or 2 (Very True or Often True) and summed to yield subscale totals. The Depressive Problems subscale includes 13 items (possible range 0–26) and the Anxiety Problems subscale includes 9 items (possible range 0–18). Higher scores indicate greater symptom severity. The raw summed scores were treated as continuous observed variables. The Depressive Problems subscale has been validated in the ABCD cohort as a reliable continuous measure of depression severity in children,38 and the Anxiety Problems subscale has demonstrated acceptable internal consistency, convergent validity with other anxiety measures, and discriminant validity in distinguishing anxious from nonanxious youth.39

Sex Assigned at Birth

Sex assigned at birth was also reported by caregivers and coded as female (0) or male (1). Gender identity was not assessed in the present analyses.

Statistical Analysis

We modeled longitudinal coupling between PSMU and depressive and anxiety symptoms across 5 years (years 2 to 6) using structural equation modeling in R, version 4.5.1, using the ‘lavaan’ package,40 version 0.6–19. Models were estimated using robust maximum likelihood estimation, which provides robust SEs and a χ2 test statistic; missing data were handled with FIML under a missing at random assumption. Primary models used all available repeated measures from participants with at least 1 measure on the core variables relevant to a given model.

In the primary random-intercept cross-lagged panel models (RI-CLPMs), each observed score was decomposed into 2 orthogonal components: (1) a between-person random intercept capturing time-invariant, trait-like variance and (2) a within-person latent state capturing occasion-specific deviations from each adolescent’s typical level. This decomposition allows the separation of stable interindividual differences from intraindividual processes.41 Within-person autoregressive paths were interpreted as temporal inertia in state-level deviations, whereas cross-lagged paths estimated whether a departure from an individual’s own average level in one construct at time t predicted a similar departure in the other construct at t +1, net of all time-invariant confounds and prior deviations. These effects are directional but not causal, as unmeasured time-varying confounding may persist.42

We estimated 5 models per outcome pairing (PSMU with depression; PSMU with anxiety): (1) traditional cross-lagged panel models (CLPMs), both constrained and unconstrained, as reference models that conflate within- and between-person variance; (2) RI-CLPMs, both constrained and unconstrained, to isolate within-person processes and test whether autoregressive and cross-lagged paths were constant across years; and (3) second-order autoregressive RI-CLPMs (AR[2]), as sensitivity models allowing for longer-lag temporal dependencies. Same-year residual covariances were freely estimated in all models. Because the unconstrained RI-CLPMs allowed lagged paths to vary across adjacent intervals, these models served as the primary models for inference in the main manuscript, whereas constrained RI-CLPMs were used to evaluate whether those paths could reasonably be held equal across years.

To explore sex differences, sex was examined as a moderator of within-person paths in multigroup RI-CLPMs.43 Moderation was evaluated using scaled Δχ2 comparisons of models constraining vs freeing autoregressive and cross-lagged paths across sex. Model fit was evaluated using the comparative fit index (CFI), Tucker-Lewis index (TLI), root mean square error of approximation (RMSEA; 90% CI), standardized root mean square residual (SRMR), and Akaike and Bayesian information criteria. Scaled χ2 difference tests were used for comparisons of constrained vs unconstrained CLPMs and RI-CLPMs, including multigroup moderation models. Given that RI-CLPMs control for both prior levels of the outcome (autoregressive effects) and stable between-person differences (random-intercept factors), cross-lagged coefficients represent prospective within-person effects on change and are therefore substantially smaller than concurrent correlations. Empirical benchmarks derived from a quasi-representative sample of more than 1300 CLPM/RI-CLPM effect sizes44 indicate that 0.03, 0.07, and 0.12 correspond to small, medium, and large effects, respectively. These anchors provide a more appropriate interpretive frame than traditional correlation-based conventions.

Results

Descriptive Statistics

The analytic cohort included 11 279 adolescents (Supplemental Table 1). Across years 2 to 6, mean PSMU scores ranged from 1.86 (SD, 0.90) to 2.10 (SD, 0.98). Parent-reported CBCL scores indicated mean depressive symptoms ranging from 1.50 (SD, 2.30) to 1.93 (SD, 2.92) and mean anxiety symptoms ranging from 1.51 (SD, 2.20) to 1.81 (SD, 2.36). Full descriptive statistics and bivariate correlations are presented in Supplemental Table 2 and Supplemental Figure 1.

Sequential Model Testing

Full model fit indices and χ2 difference tests are reported in Supplemental Table 3. The RI-CLPMs consistently demonstrated superior model fit; thus, we focus our interpretation on the unconstrained RI-CLPMs. The unconstrained RI-CLPMs were retained as the primary models because they allowed autoregressive and cross-lagged paths to vary across years, whereas the constrained RI-CLPMs were used as comparison models to evaluate whether those paths could reasonably be held equal across time. Primary unconstrained RI-CLPM estimates are presented in Table 1, sex-specific multigroup estimates are presented in Supplemental Table 4, cross-lagged estimates across longitudinal model families are presented in Supplemental Table 5, and sensitivity analyses are presented in Supplemental Tables 6 and 7. All models were estimated across 5 years (years 2–6), and results are presented using standardized coefficients, 95% CIs, and robust model comparisons (see Supplemental Methods 3 for more information).

TABLE 1.

Standardized Estimates from the Unconstrained RI-CLPMs of Internalizing Symptoms and Problematic Social Media Use

Parameter A Parameter B Depression β [95% CI] Anxiety β [95% CI]
Between-Person Association (Random Intercepts)
Problematic Social Media Use (RI) ↔ Internalizing Symptoms (RI) .11 [.031, .193] .15 [.081, .210]
Within-Person Residual Covariances
Problematic Social Media Use (Y2) ↔ Internalizing Symptoms (Y2) .00 [−.074, .078] .01 [−.049, .066]
Problematic Social Media Use (Y3) ↔ Internalizing Symptoms (Y3) .11 [.055, .163] .05 [.007, .089]
Problematic Social Media Use (Y4) ↔ Internalizing Symptoms (Y4) .12 [.073, .169] .06 [.011, .112]
Problematic Social Media Use (Y5) ↔ Internalizing Symptoms (Y5) .07 [.043, .104] .05 [.016, .089]
Problematic Social Media Use (Y6) ↔ Internalizing Symptoms (Y6) −.01 [−.047, .033] −.03 [−.069, .013]
Within-Person Autoregressive Paths
Problematic Social Media Use (Y2) → Problematic Social Media Use (Y3) .18 [.139, .226] .18 [.135, .222]
Problematic Social Media Use (Y3) → Problematic Social Media Use (Y4) .17 [.127, .212] .17 [.129, .214]
Problematic Social Media Use (Y4) → Problematic Social Media Use (Y5) .32 [.277, .352] .32 [.279, .354]
Problematic Social Media Use (Y5) → Problematic Social Media Use (Y6) .35 [.305, .392] .35 [.306, .392]
Internalizing Symptoms (Y2) → Internalizing Symptoms (Y3) .09 [.014, .168] .26 [.220, .309]
Internalizing Symptoms (Y3) → Internalizing Symptoms (Y4) .24 [.184, .296] .21 [.152, .264]
Internalizing Symptoms (Y4) → Internalizing Symptoms (Y5) .43 [.385, .471] .28 [.224, .338]
Internalizing Symptoms (Y5) → Internalizing Symptoms (Y6) .48 [.432, .521] .29 [.219, .352]
Within-Person Cross-Lagged Associations
Internalizing Symptoms (Y2) → Problematic Social Media Use (Y3) −.01 [−.055, .032] −.01 [−.054, .025]
Problematic Social Media Use (Y2) → Internalizing Symptoms (Y3) .03 [−.009, .077] .01 [−.024, .053]
Internalizing Symptoms (Y3) → Problematic Social Media Use (Y4) .06 [.003, .124] .00 [−.048, .054]
Problematic Social Media Use (Y3) → Internalizing Symptoms (Y4) .06 [.012, .102] .02 [−.025, .064]
Internalizing Symptoms (Y4) → Problematic Social Media Use (Y5) .05 [.011, .097] .03 [−.010, .076]
Problematic Social Media Use (Y4) → Internalizing Symptoms (Y5) .07 [.025, .116] .04 [−.012, .097]
Internalizing Symptoms (Y5) → Problematic Social Media Use (Y6) .04 [−.010, .082] .03 [−.019, .075]
Problematic Social Media Use (Y5) → Internalizing Symptoms (Y6) .02 [−.020, .064] .02 [−.031, .067]
Open in a new tab

Abbreviations: CFI, comparative fit index; RI-CLPM, random-intercept cross-lagged panel model; RMSEA, root mean square error of approximation; SRMR, standardized root mean square residual; TLI, Tucker-Lewis index.

Estimates are standardized coefficients with 95% CIs from the unconstrained RI-CLPMs. Internalizing symptoms refer to depressive symptoms in the depression model and anxiety symptoms in the anxiety model. Depression model: CFI = .983, TLI = .961, RMSEA = .059, SRMR = .026, scaled χ2(19) = 231.0, N = 11 279. Anxiety model fit: CFI = .988, TLI = .971, RMSEA = .052, SRMR = .026, scaled χ2(19) = 189.0, N = 11 279.

Depressive Symptoms Models

The unconstrained RI-CLPM, which allowed parameters to vary across years, demonstrated good fit, χ2(19) = 231.0, CFI = 0.983, TLI = 0.961, RMSEA = 0.059, and SRMR = 0.026. At the between-person level, trait-like PSMU and depressive symptoms were positively correlated (β = 0.11; 95% CI, 0.031–0.193). Autoregressive effects increased over time for both constructs (depressive symptoms βs = 0.09–0.48; PSMU βs = 0.18–0.35). Cross-lagged paths were strongest between years 3 and 5. PSMU was positively associated with depressive symptoms from year 3 to year 4 (β = 0.06; 95% CI, 0.012–0.102) and from year 4 to year 5 (β = 0.07; 95% CI, 0.025–0.116), whereas depressive symptoms were positively associated with PSMU over the same intervals (year 3 to year 4, β = 0.06; 95% CI, 0.003–0.124; year 4 to year 5, β = 0.05; 95% CI, 0.011–0.097). Other lags were nonsignificant. Same-year residual covariances were positive during years 3 to 5 and were largest at year 4 (β = 0.12; 95% CI, 0.073–0.169). Figure 1 displays the unconstrained random-intercept cross-lagged panel model for PSMU and depressive symptoms.

FIGURE 1.

FIGURE 1.

Open in a new tab

Unconstrained RI-CLPM of problematic social media use and depressive symptoms in the Adolescent Brain Cognitive Development Study.

Standardized estimates from the unconstrained RI-CLPM of problematic social media use and depressive symptoms among adolescents from the Adolescent Brain Cognitive Development Study (N = 11 279) across 5 years. Model fit is presented at the bottom of the figure.

Abbreviations: CFI, comparative fit index; RI-CLPM, random-intercept cross-lagged panel model; RMSEA, root mean square error of approximation; SRMR, standardized root mean square residual; TLI, Tucker-Lewis index.

Anxiety Symptoms Models

The unconstrained RI-CLPM, which allowed parameters to vary across years, demonstrated good fit, χ2(19) = 189.0, CFI = 0.988, TLI = 0.971, RMSEA = 0.052, and SRMR = 0.026. At the between-person level, trait-like PSMU and anxiety symptoms were positively correlated (β = 0.15; 95% CI, 0.081–0.210). Autoregressive paths were positive across years for both constructs (anxiety symptoms βs = 0.21–0.29; PSMU βs = 0.17–0.35). No cross-lagged associations between PSMU and anxiety symptoms were observed across years, as all 95% CIs included 0. Same-year residual covariances were positive during years 3 to 5 and were largest at year 4 (β = 0.06; 95% CI, 0.011–0.112). Figure 2 displays the unconstrained random-intercept cross-lagged panel model for PSMU and anxiety symptoms.

FIGURE 2.

FIGURE 2.

Open in a new tab

Unconstrained RI-CLPM of problematic social media use and anxiety symptoms in the Adolescent Brain Cognitive Development Study.

Standardized estimates from the unconstrained RI-CLPM of problematic social media use and anxiety symptoms among adolescents from the Adolescent Brain Cognitive Development Study (N = 11 279) across 5 years. Model fit is presented at the bottom of the figure.

Abbreviations: CFI, comparative fit index; RI-CLPM, random-intercept cross-lagged panel model; RMSEA, root mean square error of approximation; SRMR, standardized root mean square residual; TLI, Tucker-Lewis index.

Sex Differences and Sensitivity Analyses

Cross-lagged associations in the depressive symptoms model were stronger and more consistent among female than among male individuals (Supplemental Table 4). In sensitivity analyses assigning adolescents without social media accounts the lowest possible PSMU value, the depressive-symptom findings remained consistent with the primary models, whereas the anxiety-symptom models continued to show no statistically significant prospective within-person associations (Supplemental Methods 5). Additional sensitivity analyses can be found in Supplemental Tables 6 to 10.

Discussion

In line with our aim to assess within-person associations between PSMU and internalizing symptoms, we identified bidirectional relationships between PSMU and adolescent depressive symptoms but no prospective cross-lagged associations between PSMU and anxiety symptoms in the primary unconstrained models. The magnitudes of observed cross-lagged effects in this study generally fell within the modest range,44 consistent with magnitudes commonly reported in longitudinal within-person research. For example, Boer et al found that PSMU was associated with subsequent increases in depressive symptoms (β = 0.09 to 0.18) and decreases in life satisfaction (β = −0.12 to −0.16),45 whereas Orben et al observed similarly sized effects of self-reported social media time on life satisfaction (β = −0.07 to −0.13) that emerged only within specific age- and sex-defined developmental windows, with effects otherwise near zero.6 These findings suggest that social media–related effects vary across developmental stages rather than occurring uniformly across adolescence.

Building on our earlier ABCD analysis showing that time spent on social media was prospectively associated with depressive symptoms in early adolescence,29 the present findings suggest a bidirectional coupling between PSMU and depressive symptoms in mid-adolescence. The observed associations were also more specific than the broader internalizing symptom framing of our original hypotheses, with sex differences more evident for depressive symptoms than for anxiety symptoms. In the multigroup models, cross-lagged associations involving depressive symptoms were stronger and more consistent among female than among male individuals, whereas all cross-lagged associations involving anxiety symptoms remained nonsignificant for both groups. This pattern aligns with the argument that social media–related effects vary across adolescents and developmental windows.9 At the same time, the present data do not permit identification of the specific features of online engagement responsible for these differences, so interpretations regarding content, style of use, or platform-specific mechanisms remain speculative.

As previously noted, developmental cascade frameworks describe how difficulties in one dimension of functioning can spill into others across time. This framework provides a useful lens for interpreting the patterns we observed.23 The bidirectional association between PSMU and depressive symptoms is consistent with self-reinforcing feedback loops, in which symptom escalation and problematic engagement amplify one another over time. For example, adolescents experiencing low mood may increasingly turn to social media for mood regulation or distraction; however, such use may coincide with experiences including social comparison, sleep disruption, or peer strain that intensify depressive symptoms.46–48

In contrast, the primary models did not identify prospective cross-lagged associations between PSMU and anxiety symptoms. This null pattern suggests that PSMU may be more tightly linked to depressive symptom fluctuations than anxiety symptom fluctuations after accounting for stable between-person differences and prior symptom levels. However, rather than implying that anxiety is irrelevant to adolescents’ online experiences, these results suggest that anxiety-related associations may be weaker, more transient, or more dependent on shorter-term processes than can be captured in annual assessments.

The sex differences were also concentrated in the path from depressive symptoms to subsequent PSMU rather than the reverse, consistent with research on rumination-driven platform use in which low mood motivates social comparison and reassurance-seeking.49–51 That this pattern was specific to the depression model and absent in the anxiety model might suggest the sex difference reflects something particular to the depression-engagement feedback loop rather than greater female emotional reactivity overall.

From a policy perspective, these findings support public health calls, including those from the US Surgeon General, for social media platforms to take responsibility for and mitigate features that encourage addictive use, such as implementing age-appropriate design changes, warning labels for exploitative features, and tools for parental oversight.3 Additionally, parents may recognize early warning signs, initiate conversations about healthy social media use, and collaborate with clinicians to support at-risk adolescents, complementing broader policy and platform-level interventions.52

A key strength of this study is the large cohort of adolescents observed prospectively during a pivotal developmental period. By using statistical models that differentiated within-person from between-person effects, this study also improved upon prior work that largely relied on cross-sectional or between-person approaches. However, several limitations warrant consideration. First, although the cross-informant design, which paired adolescent-reported PSMU with caregiver-reported symptoms, reduced same-source inflation, caregivers may underrecognize internalizing symptoms that are less observable. Second, the annual timing of assessments limited our ability to capture shorter-term processes through which PSMU and emotional distress may unfold. Third, the cross-lagged effects were modest in absolute size. Fourth, PSMU was assessed only among adolescents who reported having at least 1 social media account, which limits the population to which these findings apply. Attrition and selective missingness may also limit generalizability, although FIML was used to retain participants with partial data. Finally, gender identity was not assessed in the present analyses, so these findings cannot be generalized to gender-diverse adolescents.

In a large, national sample of US adolescents observed for 5 years, year-specific bidirectional associations between PSMU and depressive symptoms emerged during mid-adolescence, whereas no prospective cross-lagged associations were observed for anxiety symptoms. Sex differences were more apparent for depressive symptoms, with stronger and more consistent lagged associations among female individuals. These associations were most pronounced during mid-adolescence, suggesting this developmental period may be especially important for prevention and clinical attention, although prevention beginning earlier in adolescence may help mitigate escalation of PSMU.

Supplementary Material

Supplementary Data

What’s Known on This Subject:

Prior studies on social media use and adolescent mental health have largely focused on between-person comparisons (eg, contrasting adolescents with higher vs lower average symptom levels).

What This Study Adds:

The present investigation extends this work by leveraging data from a multisite longitudinal cohort of US youth to examine within-individual changes over time, with a focus on potential bidirectional relationships of problematic social media use with depression and anxiety symptoms.

FUNDING:

JM Nagata was supported by the National Institutes of Health (K08HL159350) and the Doris Duke Charitable Foundation (2022056). The National Institutes of Health and the Doris Duke Charitable Foundation had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Glossary

ABCD

Adolescent Brain Cognitive Development

CBCL

Child Behavior Checklist

CFI

comparative fit index

CLPMs

cross-lagged panel models

FIML

full information maximum likelihood

PSMU

problematic social media use

RI-CLPMs

random-intercept cross-lagged panel models

RMSEA

root mean square error of approximation

SRMR

standardized root mean square residual

TLI

Tucker-Lewis index

Footnotes

CONFLICT OF INTEREST DISCLOSURES: DA Christakis is currently serving as an expert witness in social media litigation. The remaining authors have no conflict to declare. The opinions and assertions expressed herein are those of the author(s) and do not reflect the official policy or position of the Uniformed Services University of the Health Sciences or the Department of War.

References

  • 1.McGorry PD, Mei C, Dalal N, et al. The Lancet Psychiatry Commission on youth mental health. Lancet Psychiatry. 2024;11 (9):731–774. doi: 10.1016/S2215-0366(24)00163-9 [DOI] [PubMed] [Google Scholar]
  • 2.Verlenden JV, Fodeman A, Wilkins N, et al. Mental health and suicide risk among high school students and protective factors — Youth Risk Behavior Survey, United States. MMWR Suppl. 2024; 73(4):79–86. doi: 10.15585/mmwr.su7304a9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Office of the Surgeon General. Social Media and Youth Mental Health. May 28, 2024. Accessed February 19, 2025. https://www.hhs.gov/surgeongeneral/reports-and-publications/youth-mental-health/social-media/index.html
  • 4.Weigle PE, Shafi RMA. Social media and youth mental health. Curr Psychiatry Rep. 2024;26(1):1–8. doi: 10.1007/s11920-023-01478-w [DOI] [PubMed] [Google Scholar]
  • 5.Panayiotou M, Black L, Carmichael-Murphy P, Qualter P, Humphrey N. Time spent on social media among the least influential factors in adolescent mental health: preliminary results from a panel network analysis. Nat Ment Health. 2023;1(5):316–326. doi: 10.1038/s44220-023-00063-7 [DOI] [Google Scholar]
  • 6.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]
  • 7.Xiao Y, Meng Y, Brown T, Keyes K, Mann J. Addictive screen use trajectories linked to youth suicidal behaviors, suicidal ideation, and mental health. JAMA. 2025. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Shannon H, Bush K, Villeneuve PJ, Hellemans KG, Guimond S. Problematic social media use in adolescents and young adults: systematic review and meta-analysis. JMIR Ment Health. 2022; 9(4):e33450. doi: 10.2196/33450 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Cingel DP, Carter MC, Krause HV. Social media and self-esteem. Curr Opin Psychol. 2022;45:101304. doi: 10.1016/j.copsyc.2022.101304 [DOI] [PubMed] [Google Scholar]
  • 10.Sisk LM, Gee DG. Stress and adolescence: vulnerability and opportunity during a sensitive window of development. Curr Opin Psychol. 2022;44:286–292. doi: 10.1016/j.copsyc.2021.10.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Dahl RE. Adolescent brain development: a period of vulnerabilities and opportunities. Keynote address. Ann N Y Acad Sci. 2004; 1021(1):1–22. doi: 10.1196/annals.1308.001 [DOI] [PubMed] [Google Scholar]
  • 12.Flannery JS, Maza MT, Kilic Z, Telzer EH. Cascading bidirectional influences of digital media use and mental health in adolescence. Adv Child Dev Behav. 2023;64:255–287. doi: 10.1016/bs.acdb.2022.10.003 [DOI] [PubMed] [Google Scholar]
  • 13.Kerr B, Garimella A, Pillarisetti L, Charlly N, Sullivan K, Moreno MA. Associations between social media use and anxiety among adolescents: a systematic review study. J Adolesc Health. 2025; 76(1):18–28. doi: 10.1016/j.jadohealth.2024.09.003 [DOI] [PubMed] [Google Scholar]
  • 14.Brautsch LA, Lund L, Andersen MM, Jennum PJ, Folker AP, Andersen S. Digital media use and sleep in late adolescence and young adulthood: a systematic review. Sleep Med Rev. 2023;68:101742. doi: 10.1016/j.smrv.2022.101742 [DOI] [PubMed] [Google Scholar]
  • 15.Lin S, Fabris MA, Longobardi C, Mastrokoukou S. The association between social media addiction and aggressive behaviors: a longitudinal and gender-specific analysis. J Adolesc. 2025;97(3): 798–807. doi: 10.1002/jad.12454 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Twenge JM, Spitzberg BH. Declines in non-digital social interaction among Americans, 2003–2017. J Appl Soc Psychol. 2020;50(6): 363–367. doi: 10.1111/jasp.12665 [DOI] [Google Scholar]
  • 17.Twenge JM, Spitzberg BH, Campbell WK. Less in-person social interaction with peers among U.S. adolescents in the 21st century and links to loneliness. J Soc Pers Relat. 2019;36(6):1892–1913. doi: 10.1177/0265407519836170 [DOI] [Google Scholar]
  • 18.Goldberg X. Social comparison links social media use and well-being. Nat Rev Psychol. 2023;2(11):658–658. doi: 10.1038/s44159-023-00244-2 [DOI] [Google Scholar]
  • 19.Arora S, Mehta M. Love it or hate it, but can you ignore social media? - a bibliometric analysis of social media addiction. Comput Human Behav. 2023;147:107831. doi: 10.1016/j.chb.2023.107831 [DOI] [Google Scholar]
  • 20.Çiftci N, Yıldız M, Çiftci K. The mediating role of social ostracism in the effect of social media addiction on loneliness in adolescents. J Pediatr Nurs. 2023;73:177–183. doi: 10.1016/j.pedn.2023.09.005 [DOI] [PubMed] [Google Scholar]
  • 21.Yue H, Yue X, Zhang X, Liu B, Bao H. Exploring the relationship between social exclusion and smartphone addiction: the mediating roles of loneliness and self-control. Front Psychol. 2022;13: 945631. doi: 10.3389/fpsyg.2022.945631 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Vogel EA, Ramo DE, Prochaska JJ, Meacham MC, Layton JF, Humfleet GL. Problematic social media use in sexual and gender minority young adults: observational study. JMIR Ment Health. 2021;8(5):e23688. doi: 10.2196/23688 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Masten AS, Cicchetti D. Developmental cascades. Dev Psychopathol. 2010;22(3):491–495. doi: 10.1017/S0954579410000222 [DOI] [PubMed] [Google Scholar]
  • 24.Costello N, Sutton R, Jones M, et al. Algorithms, addiction, and adolescent mental health: an interdisciplinary study to inform state-level policy action to protect youth from the dangers of social media. Am J Law Med. 2023;49(2–3):135–172. doi: 10.1017/amj.2023.25 [DOI] [PubMed] [Google Scholar]
  • 25.Metzler H, Garcia D. Social drivers and algorithmic mechanisms on digital media. Perspect Psychol Sci. 2024;19(5):735–748. doi: 10.1177/17456916231185057 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Hoffman L, Stawski RS. Persons as contexts: evaluating between-person and within-person effects in longitudinal analysis. Res Hum Dev. 2009;6(2–3):97–120. doi: 10.1080/15427600902911189 [DOI] [Google Scholar]
  • 27.Hamaker EL. Why researchers should think “within-person”: A paradigmatic rationale. In: Mehl MR, Conner TS, eds. Handbook of Research Methods for Studying Daily Life. The Guilford Press; 2012:43–61. [Google Scholar]
  • 28.Curran PJ, Bauer DJ. The disaggregation of within-person and between-person effects in longitudinal models of change. Annu Rev Psychol. 2011;62(1):583–619. doi: 10.1146/annurev.psych.093008.100356 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Nagata JM, Otmar CD, Shim J, et al. Social media use and depressive symptoms during early adolescence. JAMA Netw Open. 2025;8 (5):e2511704. doi: 10.1001/jamanetworkopen.2025.11704 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Harden KP, Mann FD, Grotzinger AD, et al. Developmental differences in reward sensitivity and sensation seeking in adolescence: testing sex-specific associations with gonadal hormones and pubertal development. J Pers Soc Psychol. 2018;115(1): 161–178. doi: 10.1037/pspp0000172 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Breslau J, Gilman SE, Stein BD, Ruder T, Gmelin T, Miller E. Sex differences in recent first-onset depression in an epidemiological sample of adolescents. Transl Psychiatry. 2017;7(5):e1139. doi: 10.1038/tp.2017.105 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Nagata JM, Singh G, Sajjad OM, et al. Social epidemiology of early adolescent problematic screen use in the United States. Pediatr Res. 2022;92(5):1443–1449. doi: 10.1038/s41390-022-02176-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Enders CK, Bandalos DL. The relative performance of full information maximum likelihood estimation for missing data in structural equation models. Struct Equ Modeling. 2001;8(3):430–457. doi: 10.1207/S15328007SEM0803_5 [DOI] [Google Scholar]
  • 34.Andreassen CS, Torsheim T, Brunborg GS, Pallesen S. Development of a Facebook Addiction Scale. Psychol Rep. 2012;110(2):501–517. doi: 10.2466/02.09.18.PR0.110.2.501-517 [DOI] [PubMed] [Google Scholar]
  • 35.Griffiths M A ‘components’ model of addiction within a biopsychosocial framework. J Subst Use. 2005;10(4):191–197. doi: 10.1080/14659890500114359 [DOI] [Google Scholar]
  • 36.Bagot KS, Tomko RL, Marshall AT, et al. Youth screen use in the ABCD® study. Dev Cogn Neurosci. 2022;57:101150. doi: 10.1016/j.dcn.2022.101150 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Achenbach TM. The Child Behavior Checklist and related instruments. In: Maruish ME, ed. The Use of Psychological Testing for Treatment Planning and Outcomes Assessment. 2nd ed. Lawrence Erlbaum Associates Publishers; 1999:429–466. [Google Scholar]
  • 38.Zelenina M, Pine DS, Stringaris A, Nielson DM. Validation of CBCL depression scores of adolescents in three independent datasets. JCPP Adv. 2025;5(3):e12298. doi: 10.1002/jcv2.12298 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Ebesutani C, Bernstein A, Nakamura BJ, Chorpita BF, Higa-McMillan CK, Weisz JR; The Research Network on Youth Mental Health. Concurrent validity of the Child Behavior Checklist DSM-oriented scales: correspondence with DSM diagnoses and comparison to syndrome scales. J Psychopathol Behav Assess. 2010;32(3):373–384. doi: 10.1007/s10862-009-9174-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Rosseel Y. Lavaan: An R package for structural equation modeling. J Stat Softw. 2012;48(2):1–36. doi: 10.18637/jss.v048.i02 [DOI] [Google Scholar]
  • 41.Hamaker EL, Kuiper RM, Grasman RPPP. A critique of the cross-lagged panel model. Psychol Methods. 2015;20(1):102–116. doi: 10.1037/a0038889 [DOI] [PubMed] [Google Scholar]
  • 42.Rohrer JM, Murayama K. These are not the effects you are looking for: causality and the within-/between-persons distinction in longitudinal data analysis. Adv Methods Pract Psychol Sci. 2023;6(1):25152459221140842. doi: 10.1177/25152459221140842 [DOI] [Google Scholar]
  • 43.Mulder JD, Hamaker EL. Three extensions of the random intercept cross-lagged panel model. Struct Equ Modeling. 2021;28(4):638–648. doi: 10.1080/10705511.2020.1784738 [DOI] [Google Scholar]
  • 44.Orth U, Meier LL, Bühler JL, et al. Effect size guidelines for cross-lagged effects. Psychol Methods. 2024;29(2):421–433. doi: 10.1037/met0000499 [DOI] [PubMed] [Google Scholar]
  • 45.Boer M, Stevens GWJM, Finkenauer C, de Looze ME, van den Eijnden RJJM. Social media use intensity, social media use problems, and mental health among adolescents: investigating directionality and mediating processes. Comput Human Behav. 2021; 116:106645. doi: 10.1016/j.chb.2020.106645 [DOI] [Google Scholar]
  • 46.Verduyn P, Gugushvili N, Massar K, Täht K, Kross E. Social comparison on social networking sites. Curr Opin Psychol. 2020;36:32–37. doi: 10.1016/j.copsyc.2020.04.002 [DOI] [PubMed] [Google Scholar]
  • 47.Giordano AL, Prosek EA, Watson JC. Understanding adolescent cyberbullies: exploring social media addiction and psychological factors. J Child Adolesc Couns. 2021;7(1):42–55. doi: 10.1080/23727810.2020.1835420 [DOI] [Google Scholar]
  • 48.Nagata JM, Singh G, Yang JH, et al. Bedtime screen use behaviors and sleep outcomes: findings from the Adolescent Brain Cognitive Development (ABCD) Study. Sleep Health. 2023;9(4):497–502. doi: 10.1016/j.sleh.2023.02.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Wang P, Wang X, Wu Y, et al. Social networking sites addiction and adolescent depression: a moderated mediation model of rumination and self-esteem. Pers Individ Dif. 2018;127:162–167. doi: 10.1016/j.paid.2018.02.008 [DOI] [Google Scholar]
  • 50.Nesi J, Miller AB, Prinstein MJ. Adolescents’ depressive symptoms and subsequent technology-based interpersonal behaviors: a multi-wave study. J Appl Dev Psychol. 2017;51:12–19. doi: 10.1016/j.appdev.2017.02.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Nesi J, Prinstein MJ. Using social media for social comparison and feedback-seeking: gender and popularity moderate associations with depressive symptoms. J Abnorm Child Psychol. 2015;43(8): 1427–1438. doi: 10.1007/s10802-015-0020-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Hernandez JM, Ben-Joseph EP, Reich S, Charmaraman L. Parental monitoring of early adolescent social technology use in the US: a mixed-method study. J Child Fam Stud. 2024;33(3):759–776. doi: 10.1007/s10826-023-02734-6 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

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

Supplementary Materials

Supplementary Data
NIHMS2217785-supplement-Supplementary_Data.pdf (672.7KB, pdf)

RESOURCES