# paper.pdf ## Metadata - PDFFormatVersion=1.7 - Language=en-US - IsLinearized=true - IsAcroFormPresent=false - IsXFAPresent=false - IsCollectionPresent=false - IsSignaturesPresent=false - Author=Nandita Vijayakumar, S. Ghazaleh Dashti, Louise Canterford, Susan Ellul, Anthony J. A. Parissi, Anne-Lise Goddings, Russell M. Viner, Paul Moran, Rohan Borschmann, Lisa K. Mundy, Ellie M. Robson, Susan M. Sawyer - CreationDate=D:20260608151245+05'30' - Creator=Adobe InDesign 19.5 (Windows) - ModDate=D:20260608151303+05'30' - Producer=Adobe PDF Library 17.0 - Title=The Effects of Social Media on Adolescent Mental Health: Findings From a Population‐Based Cohort Study in Australia - Trapped=False - Custom.WPS-ARTICLEDOI=10.5694/mja2.70220 - Custom.WPS-JOURNALDOI=10.5694/(ISSN)1326-5377 - Custom.WPS-PROCLEVEL=3 - jav:journal_article_version=VoR - prism:doi=10.5694/mja2.70220 - pdfx:wps-articledoi=10.5694/mja2.70220 - pdfx:wps-journaldoi=10.5694/(ISSN)1326-5377 - pdfx:wps-proclevel=3 - dc:title=The Effects of Social Media on Adolescent Mental Health: Findings From a Population‐Based Cohort Study in Australia ## Contents ### Page 1 Medical Journal of Australia RESEARCH[ ![Open Access logo]() ](https://doi.org/10.5694/mja2.70220) # The Effects of Social Media on Adolescent MentalHealth: Findings From a Population-Based Cohort Studyin Australia Nandita Vijayakumar1,2[ ![ORCID logo]() ](https://orcid.org/0000-0002-5622-9547) | S. Ghazaleh Dashti3,4[ ![ORCID logo]() ](https://orcid.org/0000-0002-1399-7220) | Louise Canterford1 | Susan Ellul3,4[ ![ORCID logo]() ](https://orcid.org/0000-0002-1135-2015) | Anthony J. A. Parissi1,3[ ![ORCID logo]() ](https://orcid.org/0009-0005-0665-5539) |Anne-Lise Goddings5,6 | Russell M. Viner5[ ![ORCID logo]() ](https://orcid.org/0000-0003-3047-2247) | Paul Moran7[ ![ORCID logo]() ](https://orcid.org/0000-0002-9257-8699) | Rohan Borschmann1,4,8 | Lisa K. Mundy1,2,4[ ![ORCID logo]() ](https://orcid.org/0000-0002-7983-5232) |Ellie M. Robson1 | Susan M. Sawyer1,4[ ![ORCID logo]() ](https://orcid.org/0000-0002-9095-358X) 1Centre for Adolescent Health, Royal Children's Hospital and Murdoch Children's Research Institute, Melbourne, Victoria, Australia |2School ofPsychology, Faculty of Health, Deakin University, Melbourne, Victoria, Australia |3Clinical Epidemiology and Biostatistics Unit, Murdoch Children'sResearch Institute, Melbourne, Victoria, Australia |4Department of Paediatrics, The University of Melbourne, Melbourne, Victoria, Australia |5GreatOrmond Street Institute of Child Health, University College London, London, England, UK |6Department of Brain Sciences, Imperial College London,London, England, UK |7Centre for Academic Mental Health, Medical School, University of Bristol, Bristol, England, UK |8Department of Psychiatry,Medical Sciences Division, University of Oxford, Oxford, England, UK Correspondence: Nandita Vijayakumar () Received: 12 November 2025 | Revised: 14 April 2026 | Accepted: 22 April 2026 Keywords: adolescence | longitudinal studies | social media ABSTRACT Objectives: To examine the effects of social media on future mental health problems (depressive symptoms, anxiety symptoms,poor well-being and self-harm) in adolescents aged 12–18 years, overall and stratified by sex and age periods (early, middle andlate adolescence). Study Type: Prospective longitudinal study. Setting: Participants were recruited in 2012 through schools in Melbourne, selected using stratified random sampling. In wave1 (2012), 1239 Grade 3 students participated and have since completed annual surveys. Participants: The analysis used data up to wave 11 (2022). Participants with no data on mental health, social media and con-founders were excluded, leaving a sample of 1195 (552 [46%] male participants). Main Outcome Measures: Exposure was self-reported duration of daily social media use at each wave, from waves 4 to 10(ages 12–18 years). Outcomes (self-reported depressive symptoms, anxiety symptoms, well-being, self-harm) were assessed at thesubsequent annual wave, from waves 5 to 11 (ages 13–19 years). Results: Across adolescence, > 2 h versus < 1 h of daily social media use was associated with a small increase in risk of highdepressive symptoms (risk difference [RD] per 100, 6.3 [95% CI, 2.7–9.9]) and poor well-being (RD, 4.9 [95% CI, 1.1–8.6]) at thesubsequent annual wave. Estimated risks for all mental health problems were greatest in early adolescence (12–13 years), with thelargest effects observed for high depressive symptoms in female participants (> 2 h vs. < 1 h: RD, 10.8 [95% CI, 2.7–18.9]). Conclusions: Higher levels of social media use were associated with small increases in future risk of high depressive symptomsand poor well-being across adolescence. The largest risks for all mental health problems were observed during early adolescencefor both male and female participants, supporting the need to consider policies that mitigate the adverse effects of social mediaon the mental health of younger adolescents. JEL Classification: Paediatric medicine, Mental disorders, Information science, Statistics, epidemiology and research design ![Society logo]() Medical Journal of Australia, 2026; 224:e70220 This is an open access article under the terms of the License, which permits use, distribution and reproduction in any medium,provided the original work is properly cited. © 2026 The Author(s). Medical Journal of Australia published by John Wiley & Sons Australia, Ltd on behalf of AMPCo Pty Ltd. ### Page 2 ## 1 | Introduction Social media has transformed adolescents' lives by providing aplatform for instantaneous and constant connectivity to peers.Many adolescents report positive experiences with social media[], which may help foster key elements of psychosocial develop-ment, including social belonging, identity development and self-expression []. However, high levels of adolescent mental healthproblems have sparked widespread concerns about the negativeimpacts of social media. Social media combines user-generatedcontent with engagement-optimising algorithms, raising con-cerns about potential mental health impacts through mecha-nisms such as idealised social comparisons, cyberbullying andother negative peer interactions, emotionally salient contentand displacement of sleep or offline activities. This has evokedcalls for stricter regulations to mitigate potential harms [], withmany governments discussing or enacting policies to raise theminimum age to access social media []. Notwithstandingpublic and political interest, there is little robust scientific evi-dence available to inform age-based policy decisions. Several reviews and meta-analyses have reported weak or in-consistent associations between adolescents' social media use(SMU) and mental health, including for depression and poorwell-being []. These have noted poor methodological qualityof studies, including an early focus on ‘screentime’ that did notdistinguish different types of digital screen engagement, and aheavy reliance on cross-sectional data that limits inferences re-garding the temporality of associations. Although longitudinalstudies have increasingly examined associations between SMUand future mental health [], many have wide age bands at eachwave, which can introduce cohort effects [], or a limitednumber of waves that capture narrow developmental periods[]. Moreover, longitudinal studies have predominantlyfocussed on within-subject effects of social media [], theresults of which are not easily translated into population-levelpolicy recommendations, or relied on structural equation mod-els that cannot always be interpreted causally []. Studies have also primarily focussed on the effects of socialmedia in adolescents as a homogenous group [], but age andsex differences may contribute to improved understanding of itsimpact on mental health []. One UK study found thatSMU in early adolescence (11–13 years) was related to poorer lifesatisfaction in female adolescents, while similar associations wereevident in middle adolescence (14–15 years) in male adolescents[]. Further research, including consideration of a broader setof mental health outcomes, could inform potential social mediaregulations. We sought to address these issues in a longitudinal population-based cohort of Australians followed annually from childhoodinto early adulthood. We used data from nine annual waves toexamine the effect of SMU at any given wave on multiple mentalhealth outcomes (namely depressive symptoms, anxiety symp-toms, poor well-being and self-harm) 1 year later. We consideredeffects across adolescence (12–18 years) as well as by sex (males,females) and age periods (early, middle and late adolescence). ## 2 | Methods ### 2.1 | Study Sample Data for this study were drawn from the Child to Adult TransitionStudy (CATS), an ongoing longitudinal cohort study that was as-sembled to be representative of Grade 3 students in Melbourne,Australia in 2012 []. A stratified random sample of 43 schoolswas selected (government, Catholic and independent schoolstrata). From these schools, all Grade 3 students (N = 2289) wereinvited to participate (there were no inclusion or exclusion crite-ria), of whom 54% (n = 1239) were recruited into the study. Themain reason for non-recruitment was failure to return a con-sent form (refer to Figure for a flowchart of participants fromrecruitment onwards). Details on the representativeness of thecohort are presented in the. Participants were followed annually from wave 1 (mean [stan-dard deviation (SD)] age, 9.0 [0.4] years; 2012) to wave 11 (mean[SD] age, 18.9 [0.4] years; 2022). Data collection occurredthrough self-reported questionnaires in primary schools (waves1 to 4), with the additional option of survey completion online,by post or by computer-assisted telephone interview (waves 5and 6). From wave 7 (age 15 years) onwards, data collection wasprimarily completed online. At waves 1 and 2, a parent of eachparticipant also completed a brief survey. Participants were included in the current study if they had dataon exposure (SMU), outcome (mental health) and/or confounders(peer victimisation, parent support, sleep duration and physicalactivity) at any wave from 3 to 11. This resulted in an analytic sam-ple of 1195 participants. Similarities in demographic characteris-tics to the full cohort are presented in Table. Further details onmissing data of analytic variables at each wave are presented inTable. For the current analyses, SMU at each wave from 4 (mean [SD]age, 11.9 [0.4] years) to 10 (mean [SD] age, 17.8 [0.4] years) was ex-amined as the exposure, with mental health outcomes assessed atthe subsequent wave (i.e., exposure at wave t preceding outcomeat wave t + 1). Confounders were measured at study baseline or atthe prior wave (t − 1), so they temporally preceded the exposure. Plain Language Summary The Known: Concerns about social media's impact onadolescent mental health have driven policy debate, yetevidence remains limited and largely cross-sectional. The New: Using annual data from a longitudinal cohortof Australian adolescents, we estimated future men-tal health risks from social media use between 12 and18 years of age. More than 2 h of daily social media usewas associated with an increased risk of high depressivesymptoms and poor well-being 1 year later. Estimatedrisks were greater in early adolescence (12–13 years) com-pared to risks in later ages. The Implications: Early adolescence is a key vulnerabil-ity period for social media use, supporting targeted educa-tion and policies aimed at limiting use. ### Page 3 ### 2.2 | Measures #### 2.2.1 | Sex Sex (female, male) was collected at wave 1 by parent report. #### 2.2.2 | SMU At each wave from 3 to 10, participants self-reported their du-ration of SMU for a normal school day using a 6-point scalefrom ‘none’ to ‘4 or more hours’. Based on distributions at earlyand late waves (see Figure), responses were categorised intothree groups of daily duration of use (< 1, 1–2, > 2 h), to ensuresufficient numbers within exposure groups across adolescence.Details on social media sites endorsed at each wave are pre-sented in Figure. #### 2.2.3 | Mental Health Problems From waves 3 to 11, self-reported depressive symptoms over thepast 2 weeks were measured using the 13-item Short Mood andFeelings Questionnaire []. From waves 3 to 10, self-reportedanxiety symptoms over the past 2 weeks were measured using ashortened 8-item Spence Children's Anxiety Scale [], andat wave 11 were measured using the 7-item Generalised AnxietyDisorder assessment []. From waves 3 to 10, self-reportedsubjective well-being was measured using six items from thePaediatric Quality of Life Inventory General Well-Being Scale[]. Data from these measures were dichotomised to definehigh levels of depressive symptoms, high levels of anxiety symp-toms and poor well-being at each wave (see Table for furtherdetail). From waves 4 to 11, participants also self-reported anyself-harming behaviours over the prior 12 months (see Tablefor further detail). #### 2.2.4 | Confounders A causal diagram was developed a priori to guide the selectionof confounders (see Figure). These are listed in the ‘StatisticalAnalysis’ section below, and details on their measurement areprovided in Table. ### 2.3 | Statistical Analysis Demographic characteristics, mental health problems andSMU were described for the retained participants (n = 1195)as frequencies and percentages. Data were structured longitu-dinally such that, for each wave, confounders were measuredat the preceding wave (t–1), SMU at wave t and mental healthoutcomes at the subsequent wave (t + 1). Within this framework,the primary effect of interest was the mean effect of SMU atany given wave (from waves 4 to 10) on mental health problemsat the subsequent wave, expressed on risk ratio (RR) and riskdifference (RD) scales. This was obtained as the mean of RRsor RDs averaged across all consecutive wave pairs (e.g., SMUat wave 4 on mental health at wave 5, SMU at wave 5 on men-tal health at wave 6, and so on). The RD scale was examinedbecause it is more appropriate for estimating the potential publichealth impact of policies aimed at reducing SMU during ado-lescence. Specifically, RD estimates allow us to infer the abso-lute number of adolescents who might experience changes inmental health outcomes if exposed to different levels of SMUresulting from policy interventions. Regression standardisation(g-computation) was used to adjust for confounding. A logisticregression model was fitted for each mental health outcome,with SMU and relevant confounders (see below) as predictors.Using this model, predicted probabilities of the outcomes wereestimated for all individuals under three levels of SMU (< 1, 1–2,> 2 h). Separate models were fitted for each outcome (high de-pressive symptoms, high anxiety symptoms, poor well-being,self-harm). The means of the predicted probabilities were thencalculated across individuals for each exposure level. RDs werecalculated by contrasting the mean of the predicted probabilities(risks) between exposure levels: > 2 h versus 1–2 h, > 2 h versus< 1 h and 1–2 h versus < 1 h. To account for correlated data struc-ture (each participant contributed a separate row for each wave),generalised estimating equations with a logit link function, ex-changeable correlation matrix and robust standard errors wereused as the outcome model. Standard errors and 95% confidenceintervals (CIs) were estimated using the delta method. In keep-ing with current advice for best practice that discourages dichot-omising evidence for an effect [], we interpreted study findingsby considering the strength of point estimates and range of effectsizes with which our data were compatible under modelling as-sumptions (i.e., values within the 95% CI). #### 2.3.1 | Confounders All reported RRs and RDs were adjusted for confounders. Modelswere adjusted for wave 1 measures of sex (female, male), coun-try of birth (Australia, other) and Socio-Economic Indexes forAreas category (based on the Index of Relative Socio-economicAdvantage and Disadvantage). In addition, the following con-founders were drawn from the wave immediately precedingthe exposure: depressive symptoms, anxiety symptoms, poorwell-being, peer victimisation, parental support, sleep duration,physical activity and SMU (e.g., when SMU at wave 5 was theexposure, SMU at wave 4 was included as a confounder). Formodels of self-harm, we also adjusted for self-harm measuredat the wave before SMU. In sensitivity analyses, we includedan additional confounder that indicated the presence of mentalhealth problems at the same wave as SMU, measured as meet-ing the threshold for high depressive symptoms, high anxietysymptoms or poor well-being (or self-harm for the self-harm out-come models). This adjustment aimed to account for potentialconfounding by concurrent mental health status. However, thisvariable was not included in the primary analysis because men-tal health at the same wave may act as a mediator rather than aconfounder. Models were also adjusted for age (years), under an assumed lin-ear relationship with the outcome, and age period (early, middleor late adolescence) at the time of SMU. We defined age periodsas early adolescence (waves 4 and 5, mean ages 12–13 years),middle adolescence (waves 6 to 8, mean ages 14–16 years) andlate adolescence (waves 9 and 10, mean ages 17–18 years). Theage distribution across waves is outlined in Figure. ### Page 4 Based on a priori decisions, interaction terms between SMU andsex, and between SMU and age period, were included in modelsto relax the assumption that the effects of SMU were constantacross strata of these variables. A sex and age period interactionterm was also included for more flexible confounding adjust-ment. We assumed that the effect of SMU on the outcome wasconstant across strata of all other covariates. ### 2.4 | Missing Data Of the 1195 participants, 79.1% (945) had missing data for at leastone variable from waves 1 to 11 (Table). Missing data werehandled using multiple imputation by fully conditional specifi-cation, stratified by sex [] with 80 complete datasets imputed.All analysis variables (Table) were included in imputationmodels. Variables were imputed in the same form in which theywere included in analytic models. Each imputed dataset wasanalysed separately, and final estimates and 95% CIs were com-bined using Rubin's rules. Multiply imputed data were generated using the mice(Multivariate Imputation by Chained Equations) package [] inR 4.4.2 []. All other analyses were performed in Stata 19.0 []. ### 2.5 | STROBE Statement Checklist This study is reported according to the Strengthening theReporting of Observational Studies in Epidemiology (STROBE)Statement for cohort studies (see). ### 2.6 | Ethical Considerations Ethics approval was granted by the Royal Children's HospitalHuman Research Ethics Committee (#31089). Permission wasgranted from the Victorian Department of Education and EarlyChildhood Development and the Catholic Education OfficeMelbourne to recruit through their schools, and from individ-ual independent schools. Participants were recruited throughactive, written, informed parent consent. From wave 11, partic-ipants themselves provided written informed consent as all hadturned 18 years old. ## 3 | Results Table describes the demographic characteristics of the retainedcohort (n = 1195). The proportions of adolescents using 1–2 h or> 2 h of social media increased across waves (Figure; Table).By waves 9 to 10 (ages 17 to 18 years), about 50% of female partici-pants reported > 2 h of use, while most male participants reported1–2 h or > 2 h of use. The proportion of adolescents reportingmental health problems also tended to increase across waves(Figure; Table). The highest proportions were reported atwave 10 (mean age of 18 years), when about 40%–50% of femaleparticipants and 15%–30% of male participants reported high de-pressive symptoms, high anxiety symptoms and poor well-being.Endorsement of self-harm was lower, although sex differenceswere still evident, with higher rates in female participants. ### 3.1 | Estimated Effects of SMU Across Adolescence Estimated RRs and RDs (per 100) for the effects of differinglevels of SMU on mental health problems 1 year later, acrossall adolescents and stratified by sex, are illustrated in Figuresand (point estimates and 95% CIs are presented in Table).Results indicated increased risk for high depressive symptomsin adolescents reporting > 2 h versus < 1 h of SMU (RR, 1.29[95% CI, 1.12–1.49]; RD, 6.3 [95% CI, 2.7–9.9]) and > 2 h versus1–2 h of SMU (RR, 1.25 [95% CI, 1.09–1.43]; RD, 5.6 [95% CI,2.1–9.0]). Sex-stratified analyses revealed a similar pattern ofincreased risk in male and female participants. Relative riskswere higher in male participants (> 2 h vs. < 1 h: RR, 1.49 [95%CI, 1.10–2.02]; > 2 h vs. 1–2 h: RR, 1.47 [95% CI, 1.07–2.02])than in female participants (> 2 h vs. < 1 h: RR, 1.22 [95% CI,1.06 to 1.41]; > 2 h vs. 1–2 h: RR, 1.18 [95% CI, 1.03 to 1.34]), butabsolute risk was similar in male participants (> 2 h vs. < 1 h:RD, 5.8 [95% CI, 1.1 to 10.5]; > 2 h vs. 1–2 h: RD, 5.6 [95% CI, 0.9to 10.3]) and female participants (> 2 h vs. < 1 h: RD, 6.7 [95%CI, 2.0 to 11.4]; > 2 h vs. 1–2 h: RD, 5.5 [95% CI, 1.2 to 9.9]). Those reporting > 2 h versus < 1 h of SMU also had increasedrisk of poor well-being (RR, 1.17 [95% CI, 1.04 to 1.32]; RD,4.9 [95% CI, 1.1 to 8.6]), as did those with 1–2 h versus < 1 hof SMU (RR, 1.12 [95% CI, 1.01 to 1.24]; RD, 3.3 [95% CI, 0.2to 6.4]). This increased risk was observed in sex-stratifiedanalyses for female participants (> 2 h vs. < 1 h: RR, 1.19 [95%CI, 1.04 to 1.37] and RD, 6.5 [95% CI, 1.5 to 11.6]; 1–2 h vs.< 1 h: RR, 1.17 [95% CI, 1.04 to 1.31] and RD, 5.6 [95% CI, 1.2to 10.0]), but risk estimates in male participants were smallerand 95% CIs were inconclusive. Overall, estimated RDs and RRs were smaller and 95% CIswere inconclusive for high anxiety symptoms. However, in sex-stratified analyses, there was increased risk in male participants TABLE 1 | Demographic characteristics (wave 1) of the 1195participants included in the analytic sample. Abbreviations: SD, standard deviation; SEIFA, Socio-Economic Indexes forAreas. an = 36 (3%) of the sample had missing data for this variable. bn = 34 (3%) of the sample had missing data for this variable. ### Page 5 with > 2 h versus < 1 h of SMU (RR, 1.52 [95% CI, 1.09 to 2.13];RD, 5.2 [95% CI, 0.8 to 9.6]). Associations were close to null forfemale participants. There were no clear patterns of risk for self-harm across levels ofSMU on both RD and RR scales. In sex-stratified analyses, therewas increased risk in male participants with > 2 h versus 1–2 hof SMU (RR [1.65 to 2.69]; RD [3.8 to 7.5]), but associations wereclose to null for female participants. ### 3.2 | Estimated Effects of Social Media byAge Period Estimated risk for mental health problems by SMU, stratifiedby age period, are illustrated in Figures and (statistics pre-sented in Table). For both female and male participants, gen-eral patterns of risk were most pronounced in early adolescence(becoming smaller in later stages of adolescence) and with > 2 hversus < 1 h of SMU. ![image]() FIGURE 2 | Prevalence of mental health problems (high depressive symptoms, high anxiety symptoms, poor well-being and self-harm) acrosswaves (spanning 13–19 years of age) among female and male participants (n = 1195). At each wave, the standard deviation of the age distribution was0.4 years. ![image]() FIGURE 1 | Proportions of social media use (SMU) across waves (spanning 12–18 years of age) among female and male participants (n = 1195). Ateach wave, the standard deviation of the age distribution was 0.4 years. ### Page 6 The largest increased risk of high depressive symptoms in bothsexes was for > 2 h versus < 1 h of SMU in early adolescence,with higher absolute risk but lower relative risk for female par-ticipants (RR, 1.72 [95% CI, 1.22 to 2.43]; RD, 10.8 [95% CI, 2.7 to18.9]) compared with male participants (RR, 2.18 [95% CI, 1.32to 3.59]; RD, 7.2 [95% CI, 0.7 to 13.7]). For high anxiety symptoms, the largest increased risk was for> 2 h versus < 1 h of SMU in early adolescence in male partic-ipants (RR, 2.32 [95% CI, 1.38 to 3.88]; RD, 6.5 [95% CI, 0.7 to12.3]). A smaller increased risk was also seen for 1–2 h versus< 1 h in early adolescence in male participants (RR, 1.66 [95%CI, 1.02 to 2.69]; RD, 3.2 [95% CI, −0.6 to 7.0]). Associations wereclose to null for female participants. For poor well-being, the largest increased risk was for > 2 h versus< 1 h of SMU in early adolescence, with greater risks observed forfemale participants both in absolute and relative terms (RR, 1.43[95% CI, 1.13 to 1.82]; RD, 10.6 [95% CI, 2.8 to 18.4]) compared withmale participants (RR, 1.37 [95% CI, 0.98 to 1.91]; RD, 6.2 [95% CI,−1.1 to 13.4]). Among female participants, a smaller but clear in-creased risk was also seen for 1–2 h versus < 1 h in early adoles-cence (RR, 1.34 [95% CI, 1.08 to 1.66]; RD, 8.2 [95% CI, 1.6 to 14.9]). Self-harm risk was also highest for > 2 h versus < 1 h of SMU inearly adolescence, with greater risk in male participants in bothrelative and absolute terms (female participants: RR, 2.26 [95% CI,1.12 to 4.57] and RD, 6.6 [95% CI, −0.1 to 13.2]; male participants:RR, 2.62 [95% CI, 1.19 to 5.77] and RD, 7.1 [95% CI, −1.3 to 15.4]). ![image]() FIGURE 3 | Risk ratios (95% confidence intervals [CIs]) in mental health problems (high depressive symptoms, high anxiety symptoms, poorwell-being and self-harm) across different levels of social media use (SMU) during adolescence. ### Page 7 The patterns of associations from analyses that additionallyadjusted for the combined indicator of mental health problemsat the concurrent wave as the exposure were similar to thosefrom the primary analyses, in terms of both direction andstrength of associations (statistics are presented in Tablesand). ## 4 | Discussion This study sought to inform debate about the potential effect ofSMU on adolescent mental health problems. Using data fromCATS, we found that more than 2 h of SMU per day was associ-ated with a small increase in the risk of high depressive symp-toms and poor well-being at the subsequent annual wave acrossadolescence. Increased risk for all mental health problems wasstrongest in early adolescence, suggesting this period may be animportant period for targeting policies to mitigate the harmfuleffects of SMU. Across adolescents aged 12–18 years, our findings suggestthat compared with those who used social media less than1 h/day, there were six additional individuals with high de-pressive symptoms at the subsequent annual wave for every100 adolescents who used social media more than 2 h/day.Moreover, we identified similar levels of risk for poor well-being, but in female adolescents only. These results broadlyalign with several reviews concluding that the negative effectsof social media on mental health problems are small across alladolescents [], but also with many studies suggesting that ![image]() FIGURE 4 | Risk differences (per 100) (95% confidence intervals [CIs]) in mental health problems (high depressive symptoms, high anxiety symp-toms, poor well-being and self-harm) across different levels of social media use (SMU) during adolescence. ### Page 8 effects for well-being are greater for female relative to maleadolescents []. As most prior studies have usedcorrelation-based analyses rather than risk estimates, directcomparisons are challenging. Our findings also provide somesupport for increased risk for high anxiety symptoms and self-harm in male adolescents in contrast to the few existing lon-gitudinal studies that have reported null effects for self-harm[]. We identified the greatest risk associated with SMU was inearly adolescence (12–13 years) for both female and male adoles-cents. This equated to around 11 additional female adolescentswith high depressive symptoms and poor well-being, and sevenadditional male adolescents with high depressive symptoms andhigh anxiety symptoms, for every 100 with more than 2 h SMUper day compared with less than 1 h of use. In some instances,high levels of use were also associated with increased risk inmiddle and late adolescence, but the estimated effects wereabout half the size of those for early adolescence. Moreover,even a lesser 1–2 h of social media per day was associated withincreased risk of poor well-being in female adolescents andhigh anxiety symptoms in male adolescents during early ado-lescence—an association not evident in later periods. Finally,high levels of SMU also increased risk for self-harm in early ad-olescence but not in later periods for both male and female ad-olescents. These findings collectively provide some support for ![image]() FIGURE 5 | Risk ratios (95% confidence intervals [CIs]) in mental health problems (high depressive symptoms, high anxiety symptoms, poorwell-being and self-harm) across different levels of social media use (SMU) by age period. Early adolescence: waves 4 and 5, mean ages of 12–13 years.Middle adolescence: waves 6 to 8, mean ages of 14–16 years. Late adolescence: waves 9 and 10, mean ages of 17–18 years. ### Page 9 heightened mental health risks associated with SMU for youngadolescents, and suggest that regulation of social media in earlyadolescence may have the largest public health impacts com-pared to later age periods. Our findings broadly align with those from Orben and colleagues[], who showed that female adolescents aged 11–13 years whoincreased their SMU reported poorer life satisfaction 1 year later(within-subject associations), with no such associations observedin 14–18-year-olds. While Orben and colleagues reported simi-lar associations between SMU and life satisfaction for male ado-lescents aged 14–15 years, we observed that the greatest mentalhealth risk for male adolescents was also in early adolescence.It remains to be determined whether this inconsistency reflectsour focus on between-subject effects (i.e., comparing individualswith > 2 h vs. < 1 h of SMU) or other methodological differences.Of note, several prior longitudinal studies have failed to identifyage-related differences when considering social media effectsacross male and female adolescents [], underscoringthe need to consider how sex and age may jointly influence thepotential benefits from social media regulations. Many young people first use social media and learn to navigateits complexities during early adolescence. As certain cognitiveand regulatory capacities are less developed at this time [], thismay make it difficult to manage emotionally intense experiences ![image]() FIGURE 6 | Risk differences (per 100) (95% confidence intervals [CIs]) in mental health problems (high depressive symptoms, high anxiety symp-toms, poor well-being and self-harm) across different levels of social media use (SMU) by age period. Early adolescence: waves 4 and 5, mean ages of12–13 years. Middle adolescence: waves 6 to 8, mean ages of 14–16 years. Late adolescence: waves 9 and 10, mean ages of 17–18 years. ### Page 10 on social media, such as idealised social comparisons (e.g., ofphysical appearance), bullying and exposure to risky behaviours[]. Moreover, peer approval and body image become highly sa-lient (particularly for female adolescents) during early adoles-cence, leading them to engage with more appearance-focussedcontent and associated social comparisons. The combinationof increased social media engagement, heightened social sen-sitivity and immature regulatory capacities may contribute toincreased risk during early adolescence. In the context of sexdifferences in the prevalence of mental health problems duringearly adolescence [], this risk also results in a higher abso-lute number of female adolescents affected by SMU relative tomale adolescents. ## 5 | Limitations The strengths of the current study provide a strong foundationfor understanding causal inference at the population level.The Child to Adult Transition Study is one of few cohorts toannually track the same individuals over the course of ado-lescence. Findings may be broadly applicable to adolescentsin other high-income Western countries that have parallelhigh levels of social media engagement and adolescent mentalhealth problems. We examined adolescents' SMU from 2014 to2021, yet social media platforms, features and patterns of usechanged considerably during this period, and effects on mentalhealth may also have changed. To maintain comparability ofexposure across waves in this long-running cohort, we mea-sured overall duration of SMU rather than platform-specificbehaviours. However, duration is a relatively coarse indica-tor and does not capture differences in the type or context ofengagement, such as “active” versus “passive” engagement orexposure to specific types of content or experiences (e.g., cy-berbullying). Adolescents also tend to under-report their SMU[], reflecting the need for more objective assessments in fu-ture. Moreover, because both SMU and mental health were self-reported, our estimates may also be affected by measurementbias arising from dependent measurement error. We were alsolimited to categorising “high” levels of SMU as more than 2 hper day due to small levels of endorsement in early adolescence.However, there may be important risk differences in durationsabove 2 h per day for older adolescents. Similarly, our study didnot distinguish between no use and low use (up to 1 h) due tolow numbers in these categories (particularly in later adoles-cence), which is an important limitation of the conclusions wecan draw about age-based restrictions. Although we used a lagged exposure–outcome design, adjustedfor prior mental health and a range of relevant confounders in-formed by a causal diagram, and conducted sensitivity analysesadjusting for contemporaneous mental health, alternative ex-planations, including residual confounding, reverse causationand shared underlying vulnerabilities that may influence bothSMU and mental health, cannot be fully excluded. We usedmultiple imputation to address missing data, including miss-ingness due to attrition. This approach can be expected to yieldestimates with negligible bias provided that, conditional onthe variables included in the imputation model, the outcomedoes not cause its own missingness []. This assumptionmay have been violated in our study, which could introduceselection bias. Finally, the late adolescent waves of data collec-tion occurred during the coronavirus disease 2019 (COVID-19)pandemic, which may have influenced adolescents' SMU andmental health. However, the largest associations observed inthis study occurred during early adolescence, well before thepandemic period. It is also likely that the effects of social mediadiffer across other subgroups of adolescents, such as those withpre- existing mental health problems or those from marginal-ised communities, which represents an important avenue forfuture research. ## 6 | Conclusions In summary, more than 2 h of SMU per day was associated witha small increased risk for high depressive symptoms in adoles-cents 1 year later, and a similar increased risk for poor well-being in female adolescents, as well as high anxiety symptomsand self-harm in male adolescents. Increased risk differences forall mental health problems were most pronounced during earlyadolescence, when even 1–2 h of SMU increased the risk of somemental health problems. Our findings suggest that in terms ofthe absolute number of individuals impacted, early adolescencemay represent a key developmental period in which policiesaimed at reducing high levels of SMU could have the greatest po-tential population-level impact. Such policies would also need tobe complemented by developmentally sensitive supports such asparental guidance, school-based media literacy and well-beingprograms, and platform-level safeguards. ## Author Contributions Nandita Vijayakumar: conceptualisation, methodology, originaldraft, funding acquisition. S. Ghazaleh Dashti, conceptualisation,methodology, formal analysis, original draft, funding acquisition.Louise Canterford: methodology, data curation, formal analysis, orig-inal draft. Susan Ellul: methodology, formal analysis, original draft.Anthony J. A. Parissi: methodology, formal analysis, writing – re-view and editing. Anne-Lise Goddings: writing – review and editing.Russell M. Viner: writing – review and editing, funding acquisition.Paul Moran: data curation, review and editing. Rohan Borschmann:data curation, writing – review and editing, funding acquisition. LisaK. Mundy: writing – review and editing, funding acquisition. Ellie M.Robson: writing – review and editing. Susan M. Sawyer: conceptuali-sation, methodology, writing – review and editing, funding acquisition. ## Funding This study was funded by the National Health and Medical ResearchCouncil (NHMRC) of Australia, GNT 1010018 (2011-15) and GNT1122189 (2017–20), as well as a grant from the Royal Children's HospitalFoundation. It was also supported by the Melbourne Children'sLifeCourse Initiative, which has received funding from the VictorianGovernment's Department of Health and Human Services, VictorChiodo Foundation and Morgan Stanley. The following authors are sup-ported by National Health and Medical Research Council InvestigatorGrants: SMS (GNT 1196999), SGD (202171) and RB (GNT 2008073). RBreceives salary support from the Better Health & Care Hub at King'sCollege London, UK. Permission was granted from the VictorianDepartment of Education and Early Childhood Development and theCatholic Education Office Melbourne to recruit through schools. Thefunding sources had no role in the study design; in the collection, anal-ysis and interpretation of data; in the writing of the report; and in thedecision to submit the paper for publication. ### Page 11 ## Disclosure Not commissioned; externally peer reviewed. ## Conflicts of Interest The authors declare no conflicts of interest. ## Data Availability Statement The de-identified data we analysed are not publicly available. Details ofthis data source can be found at. Requests to access data can also be submitted here. ## References 1. M. Faverio, M. Anderson, E. Park, and Pew Research Center, “Teens,Social Media and Mental Health,” (2025),. 2. N. Costello, R. Sutton, M. Jones, et al., “Algorithms, Addiction, and Ad-olescent Mental Health: An Interdisciplinary Study to Inform State-LevelPolicy Action to Protect Youth From the Dangers of Social Media,” Ameri-can Journal of Law & Medicine 49, no. 2–3 (2023): 135–172. 3. Commonwealth Parliament; Parliament House C, “Online SafetyAmendment (Social Media Minimum Age) Bill 2024 [Internet],” ac-cessed September 25, 2025,. 4. Office of the Surgeon General (OSG), Social Media and Youth MentalHealth: The US Surgeon General's Advisory [Internet] (US Departmentof Health and Human Services, 2023),. 5. C. L. Odgers and M. R. Jensen, “Annual Research Review: AdolescentMental Health in the Digital Age: Facts, Fears, and Future Directions,”Journal of Child Psychology and Psychiatry 61, no. 3 (2020): 336–348. 6. MQ Mental Health Research, “A Safe, Healthy and Empowering Dig-ital Future: What Do We Know About Mental Health and the Internetin Young People [Internet],” (2024),. 7. E. J. Ivie, A. Pettitt, L. J. Moses, and N. B. Allen, “A Meta-Analysis ofthe Association Between Adolescent Social Media Use and DepressiveSymptoms,” Journal of Affective Disorders 275 (2020): 165–174. 8. P. M. Valkenburg, A. Meier, and I. Beyens, “Social Media Use and ItsImpact on Adolescent Mental Health: An Umbrella Review of the Evi-dence,” Current Opinion in Psychology 44 (2022): 58–68. 9. J. Arias- de la Torre, E. Puigdomenech, X. García, et al., “Relation-ship Between Depression and the Use of Mobile Technologies and SocialMedia Among Adolescents: Umbrella Review,” Journal of Medical Inter-net Research 22, no. 8 (2020): e16388. 10. J. Course- Choi and L. Hammond, “Social Media Use and AdolescentWell-Being: A Narrative Review of Longitudinal Studies,” Cyberpsy-chology, Behavior and Social Networking 24, no. 4 (2021): 223–236. 11. S. M. Coyne, L. M. Padilla-Walker, H. G. Holmgren, and L. A. Stock-dale, “Instagrowth: A Longitudinal Growth Mixture Model of SocialMedia Time Use Across Adolescence,” Journal of Research on Adoles-cence 29, no. 4 (2019): 897–907. 12. M. Boer, G. W. J. M. Stevens, C. Finkenauer, M. E. de Looze, and R.J. J. M. van den Eijnden, “Social Media Use Intensity, Social Media UseProblems, and Mental Health Among Adolescents: Investigating Direc-tionality and Mediating Processes,” Computers in Human Behavior 116(2021): 106645. 13. K. E. Riehm, K. A. Feder, K. N. Tormohlen, et al., “Associations Be-tween Time Spent Using Social Media and Internalizing and External-izing Problems Among US Youth,” JAMA Psychiatry 76, no. 12 (2019):1266–1273. 14. A. Vannucci and C. McCauley Ohannessian, “Social Media UseSubgroups Differentially Predict Psychosocial Well-Being DuringEarly Adolescence,” Journal of Youth and Adolescence 48, no. 8 (2019):1469–1493. 15. S. Huang, X. Lai, Y. Li, Y. Cui, and Y. Wang, “Beyond Screen Time:The Different Longitudinal Relations Between Adolescents' Smart-phone Use Content and Their Mental Health,” Children (Basel) 10, no.5 (2023): 770. 16. K. Puukko, L. Hietajärvi, E. Maksniemi, K. Alho, and K. Salmela-Aro, “Social Media Use and Depressive Symptoms—A LongitudinalStudy From Early to Late Adolescence,” International Journal of Envi-ronmental Research and Public Health 17, no. 16 (2020): 5921. 17. S. Houghton, D. Lawrence, S. C. Hunter, et al., “Reciprocal Relation-ships Between Trajectories of Depressive Symptoms and Screen MediaUse During Adolescence,” Journal of Youth and Adolescence 47, no. 11(2018): 2453–2467. 18. C. Fitzpatrick, A. Lemieux, J. Smith, G. L. West, V. Bohbot, and M.Asbridge, “Is Adolescent Internet Use a Risk Factor for the Develop-ment of Depression Symptoms or Vice-Versa?,” Psychological Medicine53, no. 14 (2023): 6773–6779. 19. E. Boers, M. H. Afzali, N. Newton, and P. Conrod, “Association ofScreen Time and Depression in Adolescence,” JAMA Pediatrics 173, no.9 (2019): 853–859. 20. D. T. Beeres, F. Andersson, H. G. M. Vossen, and M. R. Galanti, “So-cial Media and Mental Health Among Early Adolescents in Sweden: ALongitudinal Study With 2-Year Follow-Up (KUPOL Study),” Journal ofAdolescent Health 68, no. 5 (2021): 953–960. 21. A. Orben, A. K. Przybylski, S. J. Blakemore, and R. A. Kievit, “Win-dows of Developmental Sensitivity to Social Media,” Nature Communi-cations 13, no. 1 (2022): 1649. 22. A. Orben, T. Dienlin, and A. K. Przybylski, “Social Media's Endur-ing Effect on Adolescent Life Satisfaction,” Proceedings of the NationalAcademy of Sciences of the United States of America 116, no. 21 (2019):10226–10228. 23. Z. J. Kunicki, M. L. Smith, and E. J. Murray, “A Primer on StructuralEquation Model Diagrams and Directed Acyclic Graphs: When andHow to Use Each in Psychological and Epidemiological Research,” Ad-vances in Methods and Practices in Psychological Science 6, no. 2 (2023):25152459231156085. 24. I. E. Thorisdottir, R. Sigurvinsdottir, A. L. Kristjansson, J. P. Al-legrante, C. L. Lilly, and I. D. Sigfusdottir, “Longitudinal AssociationBetween Social Media Use and Psychological Distress Among Adoles-cents,” Preventive Medicine 141 (2020): 106270. 25. S. M. Coyne, A. A. Rogers, J. D. Zurcher, L. Stockdale, and M. Booth,“Does Time Spent Using Social Media Impact Mental Health?: An EightYear Longitudinal Study,” Computers in Human Behavior 104 (2020):106160. 26. L. K. Mundy, J. G. Simmons, N. B. Allen, et al., “Study Protocol: TheChildhood to Adolescence Transition Study (CATS),” BMC Pediatrics13 (2013): 160. 27. H. Thabrew, K. Stasiak, L. M. Bavin, C. Frampton, and S. Merry,“Validation of the Mood and Feelings Questionnaire (MFQ) and ShortMood and Feelings Questionnaire (SMFQ) in New Zealand Help-Seeking Adolescents,” International Journal of Methods in PsychiatricResearch 27, no. 3 (2018): e1610. 28. T. Reardon, S. H. Spence, J. Hesse, A. Shakir, and C. Creswell, “Iden-tifying Children With Anxiety Disorders Using Brief Versions of theSpence Children's Anxiety Scale for Children, Parents, and Teachers,”Psychological Assessment 30, no. 10 (2018): 1342–1355. 29. S. H. Spence, P. M. Barrett, and C. M. Turner, “Psychometric Prop-erties of the Spence Children's Anxiety Scale With Young Adolescents,”Journal of Anxiety Disorders 17, no. 6 (2003): 605–625. ### Page 12 30. R. L. Spitzer, K. Kroenke, J. B. W. Williams, and B. Löwe, “A BriefMeasure for Assessing Generalized Anxiety Disorder: The GAD-7,” Ar-chives of Internal Medicine 166, no. 10 (2006): 1092–1097. 31. J. W. Varni, M. Seid, and P. S. Kurtin, “PedsQL 4.0: Reliability andValidity of the Pediatric Quality of Life Inventory Version 4.0 GenericCore Scales in Healthy and Patient Populations,” Medical Care 39, no.8 (2001): 800–812. 32. S. Greenland, S. J. Senn, K. J. Rothman, et al., “Statistical Tests, PValues, Confidence Intervals, and Power: A Guide to Misinterpreta-tions,” European Journal of Epidemiology 31 (2016): 337–350. 33. M. H. Huque, J. B. Carlin, J. A. Simpson, and K. J. Lee, “A Com-parison of Multiple Imputation Methods for Missing Data in Longitu-dinal Studies,” BMC Medical Research Methodology 18, no. 1 (2018):168. 34. S. van Buuren and K. Groothuis- Oudshoorn, “Mice: MultivariateImputation by Chained Equations in R,” Journal of Statistical Software45, no. 1 (2011): 1–67. 35. R Core Team, R: A Language and Environment for Statistical Com-puting (R Foundation for Statistical Computing—References—Scien-tific Research Publishing [Internet], 2024). 36. StataCorp, Stata Statistical Software: Release 198 (StataCorp LLC,2025). 37. C. L. Booker, Y. J. Kelly, and A. Sacker, “Gender Differences in theAssociations Between Age Trends of Social Media Interaction and Well-Being Among 10-15 Year Olds in the UK,” BMC Public Health 18, no. 1(2018): 321. 38. L. Twigg, C. Duncan, and S. Weich, “Is Social Media Use AssociatedWith Children's Well-Being? Results From the UK Household Longitu-dinal Study,” Journal of Adolescence 80 (2020): 73–83. 39. X. Zhu, H. Griffiths, Z. Xiao, et al., “Trajectories of Screen TimeAcross Adolescence and Their Associations With Adulthood MentalHealth and Behavioral Outcomes,” Journal of Youth and Adolescence52, no. 7 (2023): 1433–1447. 40. Y. Kelly, B. Xue, C. Booker, et al., “What (If Anything) is Going on?Examining Longitudinal Associations Between Social Media Use andMental Ill-Health Among Young People,” [preprint], medRxiv (2022):2022.03.31.22273198,.(viewed on 24 July 2024). 41. P. McNamee, S. Mendolia, and O. Yerokhin, “Social Media Use andEmotional and Behavioural Outcomes in Adolescence: Evidence FromBritish Longitudinal Data,” Economics and Human Biology 41 (2021):100992. 42. B. Larsen and B. Luna, “Adolescence as a Neurobiological CriticalPeriod for the Development of Higher- Order Cognition,” Neuroscienceand Biobehavioral Reviews 94 (2018): 179–195. 43. A. Orben, A. Meier, T. Dalgleish, and S. J. Blakemore, “MechanismsLinking Social Media Use to Adolescent Mental Health Vulnerability,”Nature Reviews Psychology 3, no. 6 (2024): 407–423. 44. E. M. Robson, H. M. Husin, S. G. Dashti, et al., “Tracking the Courseof Depressive and Anxiety Symptoms Across Adolescence (The CATSStudy): A Population-Based Cohort Study in Australia,” Lancet Psychia-try 12, no. 1 (2025): 44–53. 45. T. J. Wade, J. Cairney, and D. J. Pevalin, “Emergence of Gender Dif-ferences in Depression During Adolescence: National Panel ResultsFrom Three Countries,” Journal of the American Academy of Child andAdolescent Psychiatry 41, no. 2 (2002): 190–198. 46. J. D. Alexander, J. Linkersdörfer, K. Toda-Thorne, et al., “PassivelySensing Smartphone Use in Teens With Rates of Use by Sex and AcrossOperating Systems,” Scientific Reports 14, no. 1 (2024): 17982. 47. J. Zhang, S. G. Dashti, J. B. Carlin, K. J. Lee, and M. Moreno-Betancur, “Recoverability and Estimation of Causal Effects UnderTypical Multivariable Missingness Mechanisms,” Biometrical Journal66, no. 3 (2024): 2200326. 48. M. Moreno-Betancur, K. J. Lee, F. P. Leacy, I. R. White, J. A. Simp-son, and J. B. Carlin, “Canonical Causal Diagrams to Guide the Treat-ment of Missing Data in Epidemiologic Studies,” American Journal ofEpidemiology 187, no. 12 (2018): 2705–2715. Supporting Information Additional supporting information can be found online in theSupporting Information section. Figure S1: Flowchart of CATS par-ticipants (N = 1239) from recruitment onwards. Table S1: Wave 1 de-mographic characteristics of the adolescents included in the currentstudy (N = 1195) compared to the full Childhood to Adult TransitionStudy cohort (N = 1239). Table S2: Frequency and percent of missingdata for all variables, overall and by sex. Figure S2: Distribution ofsocial media use (on a normal school day) at each of waves 4–10. Ateach wave, the standard deviation of the age distribution was 0.4 years.Figure S3: Proportion of sample using popular social media sites fromwaves 4 to 10. At each wave, the standard deviation of the age distri-bution was 0.4 years. Table S3: Information about how depressive andanxiety symptoms were measured at waves 3–11 and description of con-founders included in the analytic models. All measures were self-reportapart from SES, which was based on home postcode by parent-report.Figure S4: Causal diagram that guided confounder selection. FigureS5: Distribution of age at each of waves 3–11. Table S4: Number (per-centage) of study participants (N = 1195) in each social media use cate-gory (on a typical school day), at each of waves 4–10, by sex. Table S5:Number (percentage) of study participants (N = 1195) reporting highdepressive symptoms, high anxiety symptoms, and poor general well-being, at each of waves 5–11, by sex. Table S6: Estimated risk ratios (95%CI) and risk differences (95% CI) of social media use (on a typical weekday) on future mental health (high depressive symptoms, high anxietysymptoms, poor well-being, self-harm), over adolescence, overall and bysex. Table S7: Estimated risk ratios (95% CI) and risk differences (95%CI) of social media use (on a typical week day) on future mental health(high depressive symptoms, high anxiety symptoms, poor well-being,self-harm), by period of adolescence and by sex. Table S8: Estimatedrisk ratios (95% CI) and risk differences (95% CI) of social media use (ona typical week day) on future mental health (high depressive symptoms,high anxiety symptoms, poor well-being, self-harm), over adolescence,overall and by sex. Table S9: Estimated risk ratios (95% CI) and riskdifferences (95% CI) of social media use (on a typical week day) on fu-ture mental health (high depressive symptoms, high anxiety symptoms,poor well-being, self-harm), by period of adolescence and by sex.