A Literature Review on Social Media Use and Adolescent Mental Health

Introduction

Few questions in contemporary developmental psychology have attracted as much public and scholarly attention as whether social media harms the mental health of young people. The rise of smartphone ownership and platform use among adolescents has coincided with reported increases in anxiety, depression and self-harm across several high-income countries, prompting an intuitive causal narrative that has since been widely rehearsed in the popular press. Yet the empirical picture is considerably more contested than headlines suggest, and the strength, direction and even the existence of the association remain matters of active debate.

This review synthesises the existing literature on the relationship between social media use and adolescent mental health, with a particular focus on wellbeing, internalising symptoms and the psychological mechanisms proposed to link the two. Its aim is threefold: to summarise what quantitative research has established about the association between screen-based social media use and wellbeing; to examine the mechanisms through which harm, or benefit, might plausibly operate; and to appraise critically the methodological debates that have reshaped the field since 2019. The scope is deliberately restricted to adolescence, broadly defined as ages 10 to 19, and to non-clinical populations, though findings concerning vulnerable subgroups are considered where relevant. The review draws on observational cohort studies, experimental and quasi-experimental work, and the large-scale secondary analyses that have proved especially influential. No primary data are presented; the contribution is one of synthesis and critical evaluation.

A brief note on terminology is warranted, because much of the confusion in this literature is definitional. “Screen time”, “digital technology use”, “social media use” and “smartphone use” are frequently treated as interchangeable, yet they denote overlapping but distinct constructs, and the choice between them materially affects the conclusions drawn. Where possible, this review distinguishes general screen exposure from social media specifically, and passive consumption from active, interactive use, since the evidence increasingly suggests that these distinctions carry psychological weight.

Screen Time, Social Media and Wellbeing

Early research on this topic tended to report modest but statistically significant negative correlations between time spent on social media and indicators of adolescent wellbeing. Twenge and colleagues were among the most prominent voices arguing that the mental health of adolescents deteriorated markedly around the same time that smartphones and social media became ubiquitous, and that these trends were unlikely to be coincidental (Twenge et al., 2018). Drawing on large repeated cross-sectional surveys in the United States, they reported that heavier users of screen-based media were more likely to report low wellbeing and depressive symptoms than lighter users, and interpreted the cohort-level rise in adolescent distress as consistent with a displacement of psychologically protective activities. The argument drew rhetorical force from its timing: the post-2010 inflection in adolescent depression and suicide-related outcomes mapped neatly onto the diffusion of the smartphone, and the coincidence lent the causal story an air of near-inevitability.

However, the magnitude of these associations has been repeatedly called into question. In a widely cited analysis of three large representative datasets from the United Kingdom and United States, Orben and Przybylski (2019) demonstrated that the association between digital technology use and adolescent wellbeing was, at best, very small. Applying specification curve analysis, a technique that computes the outcome across the full range of defensible analytical choices, they found that the negative relationship explained approximately 0.4% of the variance in wellbeing. They noted, memorably, that the association was comparable in size to the wellbeing effect of wearing glasses and smaller than that of regularly eating potatoes. This work has become a touchstone precisely because it exposed how sensitive earlier conclusions were to arbitrary analytical decisions, and how easily small effects can be inflated through selective reporting. The specification curve approach mattered methodologically because it removed the researcher’s degrees of freedom from the equation: rather than presenting a single favoured model, it laid bare the entire distribution of plausible estimates, most of which clustered close to zero.

Subsequent syntheses have tended to reinforce this picture of a weak and inconsistent aggregate association. A number of meta-analyses and systematic reviews have concluded that cross-sectional correlations between social media use and depressive or anxiety symptoms are typically small and heterogeneous, with effect sizes varying substantially according to how use is measured and which outcomes are assessed (Orben, 2020). A systematic review by Keles et al. (2020) similarly reported that, while associations between social media use and depression, anxiety and psychological distress were frequently observed, they were generally modest and inconsistently replicated, and were confounded by wide variation in measurement and design. Population-cohort evidence points in the same direction. Analysing the UK Millennium Cohort Study, Kelly et al. (2018) found that greater social media use was associated with higher depressive symptoms, particularly among girls, but that the pathways implicated online harassment, poor sleep, low self-esteem and body image rather than screen time as such. Importantly, the reliance on self-reported screen time, which correlates only moderately with objectively logged use, introduces measurement error that may attenuate or distort observed relationships. The field’s gradual movement away from crude “screen time” measures towards more granular consideration of what adolescents actually do online reflects a growing recognition that duration alone is a poor proxy for psychological impact.

Not all theoretical framings predict a simple monotonic harm. The “Goldilocks hypothesis” advanced by Przybylski and Weinstein (2017) proposed that the relationship between screen use and wellbeing is non-linear, with moderate use being benign or even mildly beneficial and only very high levels of engagement associated with lower wellbeing. Such curvilinear framings help to reconcile the coexistence of genuine concern about heavy users with the reassuringly flat average associations reported in the large secondary analyses. They also caution against the dose-response logic that implicitly underpins much popular commentary, in which every additional hour is assumed to carry a comparable psychological cost.

The magnitude of the debate is easier to appreciate visually. Figure 1 juxtaposes illustrative effect-size estimates against the everyday comparators that Orben and Przybylski invoked, underscoring how small the average association is relative to its prominence in public discourse.

Proposed Mechanisms

If the aggregate association is small, attention naturally turns to the mechanisms that might produce harm in some circumstances and for some individuals. A mechanism-focused reading reframes the question from “does social media harm adolescents?” to the more tractable “through what pathways, and for whom, might it do so?” Three mechanisms recur in the literature: social comparison, displacement, and cyberbullying. These are not mutually exclusive, and in practice they are likely to operate together and to reinforce one another. Figure 2 sets out the conceptual structure adopted here, in which social media use acts on wellbeing indirectly through these mediating processes, with the strength of each pathway conditioned by individual and contextual moderators.

Social comparison

Social comparison theory, originally articulated by Festinger (1954), provides one of the most frequently invoked explanatory frameworks. Social media platforms are argued to intensify upward social comparison because they present curated, idealised portrayals of peers’ appearance, achievements and social lives. Adolescents, whose sense of identity and self-worth is still forming, may be particularly susceptible to the corrosive effects of comparing their unedited inner experience with the polished external presentations of others. Empirical work has linked appearance-focused comparison on image-based platforms with body dissatisfaction and lowered self-esteem; experimental exposure studies indicate that even brief engagement with idealised imagery can depress state body image, and that the effect is concentrated among those already prone to appearance comparison (Fardouly et al., 2015). There is also evidence that this pathway is more pronounced among adolescent girls, though the interpretation of such sex differences is contested. A related mechanism concerns technology-based feedback-seeking: Nesi and Prinstein (2015) reported that seeking reassurance and validation through social media was associated with depressive symptoms, particularly among adolescents low in perceived popularity, suggesting that it is the interpersonal function of the behaviour, rather than exposure per se, that carries risk. The social comparison account is theoretically coherent and increasingly well specified, yet much of the supporting evidence remains cross-sectional and cannot by itself establish the direction of causation.

Displacement

The displacement hypothesis holds that time spent on social media crowds out activities known to support wellbeing, notably sleep, physical activity and face-to-face social interaction. Sleep displacement has attracted particular concern, given that adolescents already experience a physiological shift towards later sleep timing, and that late-night device use may further curtail sleep duration and quality. Because insufficient sleep is itself a robust predictor of low mood and emotional dysregulation, displacement offers a plausible indirect route from heavy use to poorer mental health, and mediation analyses in cohort data are consistent with sleep and self-esteem carrying part of the association between use and depressive symptoms (Kelly et al., 2018). The displacement account has the attraction of specifying a concrete, measurable intermediary. Its logic is also intuitive: hours are finite, and time online is time not spent elsewhere. Yet the empirical evidence remains mixed. The strong form of the hypothesis, in which online interaction simply substitutes for offline connection, is difficult to sustain, since for many adolescents digital communication supplements rather than replaces in-person friendship and can extend and consolidate offline relationships. The more defensible version of the account is therefore selective: it is late-night, sleep-displacing and solitary use, rather than social use in aggregate, that is most plausibly harmful.

Cyberbullying

Cyberbullying represents a more direct route to psychological harm. Online environments can extend the reach and persistence of peer aggression beyond the school gates, removing the respite that physical distance once afforded and creating a sense that victimisation is inescapable. Victims of cyberbullying report elevated rates of depression, anxiety and, in the most serious cases, suicidal ideation, and the association appears more robust and clinically significant than the diffuse effects attributed to general use. This distinction is analytically important: it suggests that the aggregate small-effect findings may obscure a smaller group for whom specific negative experiences, rather than time spent per se, carry substantial risk. It also reframes the policy question, since interventions targeting online victimisation are conceptually distinct from, and may be more tractable than, blanket efforts to reduce screen time. The mechanism further illustrates why average exposure is a poor guide to harm: two adolescents may spend identical amounts of time online while having entirely different experiences, one supported and one harassed.

Vulnerable Groups and Individual Differences

A recurring theme across recent scholarship is that population-average effects are likely to mask meaningful individual differences. The same platform activity may be protective for one adolescent and harmful for another, depending on their pre-existing vulnerabilities, motivations for use and the nature of their online experiences. Valkenburg and colleagues have argued forcefully that the field’s preoccupation with average effects has obscured this heterogeneity, and that person-specific analyses reveal substantial variation in how individual adolescents respond to social media use (Valkenburg et al., 2021). Their work suggests that a minority of young people experience notable declines in wellbeing following use, a comparable minority experience improvements, and the majority are largely unaffected. If this pattern holds, then the small average effect is not merely a weak signal but a statistical artefact of aggregation, in which opposing individual trajectories cancel out to produce a near-null mean. Such a finding has profound methodological implications: it implies that the between-person designs dominating the field are, in principle, incapable of detecting the effects that matter, and that within-person, intensive-longitudinal methods are required instead.

Several groups appear more susceptible to adverse outcomes. Adolescent girls, as noted, may be more exposed to appearance-related comparison, and some analyses report stronger associations between social media use and depressive symptoms in girls than boys, a pattern documented in large cohort studies although the interpretation of these interactions is contested (Kelly et al., 2018). Adolescents with pre-existing mental health difficulties may also be more vulnerable, both because they may use platforms in more maladaptive ways, such as compulsive checking, rumination or seeking out distressing content, and because such content may exacerbate existing symptoms. This raises the possibility of a reinforcing loop in which vulnerability shapes use and use in turn deepens vulnerability. LGBTQ+ young people present a more nuanced case: while they may encounter elevated online harassment, social media can also offer vital access to community, identity affirmation and support that is unavailable offline, a duality that resists any simple verdict. This dual potential underscores a broader point running through the literature, namely that social media is neither uniformly harmful nor uniformly benign, and that its effects are conditioned by who is using it, why, and in what circumstances.

Socioeconomic context adds a further layer of complexity that is easily overlooked in average estimates. Adolescents from more disadvantaged backgrounds may rely more heavily on online spaces for social connection where offline opportunities are constrained, and may also be less well positioned to draw on the offline resources that buffer against online harms. The interaction between platform experience and offline circumstance is therefore likely to matter as much as either in isolation, yet it is rarely modelled explicitly. Attending to such moderators is not merely a refinement; it is central to translating a small and heterogeneous average effect into actionable knowledge about who is genuinely at risk.

The Debate on Effect Sizes and Causation

The most consequential development in this field over the past several years has been methodological rather than substantive. Orben and Przybylski’s (2019) demonstration that headline conclusions were highly sensitive to analytical flexibility prompted a broader reckoning with how the evidence base had been constructed. Their subsequent work extended this critique, showing that longitudinal associations between social media use and wellbeing were similarly small and often inconsistent across datasets and specifications (Orben et al., 2019). Using random-intercept cross-lagged panel models, which separate stable between-person differences from within-person change, they found little evidence of an enduring within-person effect of social media use on later life satisfaction, and what evidence there was proved small and unstable. The implication is not that social media is harmless, but that the confident causal claims common in public discourse outrun what the correlational evidence can support. Odgers and Jensen (2020), reviewing the field, reached a comparable verdict, cautioning that fears about digital technology have consistently run ahead of the evidence and that the most alarming claims rest on the weakest designs.

Two limitations dominate the causal debate. First, most studies are correlational, leaving the direction of any association ambiguous. The plausible reverse pathway, in which adolescents already experiencing low mood turn to social media as a form of withdrawal or self-soothing, is at least as consistent with the data as the harm hypothesis, and the two processes may operate simultaneously in a reciprocal cycle. Distinguishing these possibilities requires designs capable of isolating within-person change over time, yet the cross-lagged models that attempt this remain sensitive to the interval between measurements and to how stable individual differences are modelled. Second, the near-universal reliance on self-reported use introduces systematic measurement error, since retrospective estimates of screen time correlate only modestly with objectively logged behaviour, and the direction of that error may itself be confounded with mood. Experimental and quasi-experimental designs, which could in principle adjudicate questions of causation, remain relatively scarce. Those that exist, such as short-term abstinence or “digital detox” studies, often involve brief interventions of uncertain ecological validity, small and self-selected samples, and difficulties in blinding participants to the manipulation, all of which limit the inferences that can be drawn.

The distinction between cross-sectional and longitudinal designs deserves particular emphasis, because it is frequently elided in public commentary. Cross-sectional studies, however large, can establish only that use and distress co-occur; they are structurally silent on temporal order. Longitudinal designs improve on this by ordering measurements in time, but the conventional cross-lagged panel model conflates within-person processes with stable between-person differences, and can generate spurious cross-lagged paths when such trait-like stability is not modelled. The migration towards random-intercept and related within-person models represents a genuine advance, since it asks the more pertinent question of whether an individual’s own change in use predicts their own subsequent change in wellbeing. When the literature is re-examined through this lens, the associations that appeared substantial in cross-sectional snapshots tend to shrink towards, and often to, negligibility.

This methodological scrutiny has been productive. It has shifted the field away from simplistic dose-response framing towards more sophisticated questions about content, context and individual susceptibility. It has also encouraged greater transparency, including pre-registration and the reporting of full specification curves, which makes selective analytical practices harder to sustain. The cost, however, is a more uncomfortable and less quotable conclusion: that the relationship between social media and adolescent mental health is genuinely small on average, highly variable between individuals, and not yet well understood at the level of causal mechanism. Table 1 summarises the key studies underpinning this assessment, highlighting the recurring pattern of small aggregate effects alongside growing recognition of heterogeneity.

Table 1. Summary of key studies on social media use and adolescent mental health

Critical Appraisal and Gaps

Several limitations constrain the existing evidence base and point towards priorities for future research. The dominance of cross-sectional and correlational designs is the most obvious. While large secondary analyses have brought welcome rigour to the estimation of effect sizes, they cannot resolve questions of causal direction, and the longitudinal studies that might do so are comparatively few and often limited to short follow-up windows. Even where longitudinal data exist, the choice of analytic model exerts a decisive and under-appreciated influence on the conclusions, as the divergence between conventional and random-intercept cross-lagged panel models illustrates. There is a pressing need for well-powered, pre-registered longitudinal and experimental research capable of testing specific mechanisms rather than global associations, and for the routine reporting of within-person as well as between-person estimates.

Measurement presents a second, pervasive weakness. The continued reliance on self-reported duration of use is difficult to defend given evidence of its poor correspondence with objective logs, and given that the measurement error may be systematically related to the very outcomes under study, so that distressed adolescents over-report use and inflate the apparent association. Advances in passive sensing and platform-provided usage data offer a route to more accurate measurement, though these raise their own ethical and access challenges, not least the reluctance of platforms to share granular behavioural data with independent researchers. Relatedly, the field has often treated “social media” as a monolith, when platforms differ markedly in their affordances, and active, connection-oriented use may have very different consequences from passive, comparison-laden scrolling. Finer-grained distinctions between types of use, and between the specific experiences use affords, are likely to prove more illuminating than aggregate exposure.

A third gap concerns the neglect of individual differences and moderating context. If, as the person-specific evidence suggests, effects are concentrated among a vulnerable minority, then population-average estimates are of limited practical value for identifying who is at risk and why. Future work should prioritise the characterisation of susceptible subgroups and the conditions under which use becomes harmful, ideally combining intensive-longitudinal measurement with theoretically derived moderators. Finally, the literature remains skewed towards high-income, Western and predominantly Anglophone populations, limiting the generalisability of findings across cultural and socioeconomic contexts. The mechanisms and moderators identified in these settings may operate differently elsewhere, and cross-cultural research is conspicuously underdeveloped. A further, more reflexive limitation deserves mention: the field is not immune to the incentives that reward striking findings, and the very salience of the topic in public and policy debate creates pressure towards confident conclusions that the evidence does not warrant.

Conclusion

The synthesis presented here supports a more measured position than either the alarmist or the dismissive extremes that characterise public debate. At the population level, the association between social media use and adolescent mental health is consistently small, and the strong causal claims frequently made in its name are not well supported by the largely correlational evidence available. The methodological interventions of Orben and Przybylski (2019) and others have been salutary in exposing how earlier conclusions were inflated by analytical flexibility and imprecise measurement. At the same time, the small average effect should not be mistaken for the absence of harm. Specific experiences such as cyberbullying carry substantial and well-evidenced risk, plausible mechanisms including social comparison and sleep displacement warrant continued investigation, and the marked heterogeneity between individuals implies that a vulnerable minority may be meaningfully affected even when most are not.

The most defensible reading of the literature is therefore one of conditional and contingent effects rather than uniform impact. What harms a susceptible adolescent late at night, exposed to appearance-based comparison or online victimisation, is not well captured by an average computed across a population for whom social media is, on balance, neither markedly harmful nor markedly beneficial. Progress will depend on moving beyond crude measures of exposure towards mechanism-focused, methodologically rigorous and individually sensitive research, and towards designs that ask whether a given young person’s own change in use predicts their own change in wellbeing. Until such evidence accumulates, both scholars and policymakers would be well advised to resist the appeal of simple stories, and to treat the relationship between social media and adolescent mental health as an open and genuinely complex empirical question.

References

Fardouly, J., Diedrichs, P. C., Vartanian, L. R., & Halliwell, E. (2015). Social comparisons on social media: The impact of Facebook on young women’s body image concerns and mood. Body Image, 13, 38–45. https://doi.org/10.1016/j.bodyim.2014.12.002

Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7(2), 117–140. https://doi.org/10.1177/001872675400700202

Keles, B., McCrae, N., & Grealish, A. (2020). A systematic review: The influence of social media on depression, anxiety and psychological distress in adolescents. International Journal of Adolescence and Youth, 25(1), 79–93. https://doi.org/10.1080/02673843.2019.1590851

Kelly, Y., Zilanawala, A., Booker, C., & Sacker, A. (2018). Social media use and adolescent mental health: Findings from the UK Millennium Cohort Study. EClinicalMedicine, 6, 59–68. https://doi.org/10.1016/j.eclinm.2018.12.005

Nesi, J., & Prinstein, M. J. (2015). Using social media for social comparison and feedback-seeking: Gender and popularity moderate associations with depressive symptoms. Journal of Abnormal Child Psychology, 43(8), 1427–1438. https://doi.org/10.1007/s10802-015-0020-0

Odgers, C. L., & Jensen, M. R. (2020). Annual research review: Adolescent mental health in the digital age — Facts, fears, and future directions. Journal of Child Psychology and Psychiatry, 61(3), 336–348. https://doi.org/10.1111/jcpp.13190

Orben, A. (2020). Teenagers, screens and social media: A narrative review of reviews and key studies. Social Psychiatry and Psychiatric Epidemiology, 55(4), 407–414. https://doi.org/10.1007/s00127-019-01825-4

Orben, A., Dienlin, T., & Przybylski, A. K. (2019). Social media’s enduring effect on adolescent life satisfaction. Proceedings of the National Academy of Sciences, 116(21), 10226–10228. https://doi.org/10.1073/pnas.1902058116

Orben, A., & Przybylski, A. K. (2019). The association between adolescent well-being and digital technology use. Nature Human Behaviour, 3(2), 173–182. https://doi.org/10.1038/s41562-018-0506-1

Przybylski, A. K., & Weinstein, N. (2017). A large-scale test of the Goldilocks hypothesis: Quantifying the relations between digital-screen use and the mental well-being of adolescents. Psychological Science, 28(2), 204–215. https://doi.org/10.1177/0956797616678438

Twenge, J. M., Joiner, T. E., Rogers, M. L., & Martin, G. N. (2018). Increases in depressive symptoms, suicide-related outcomes, and suicide rates among U.S. adolescents after 2010 and links to increased new media screen time. Clinical Psychological Science, 6(1), 3–17. https://doi.org/10.1177/2167702617723376

Valkenburg, P. M., Beyens, I., Pouwels, J. L., van Driel, I. I., & Keijsers, L. (2021). Social media use and adolescents’ self-esteem: Heroes or villains? Journal of Youth and Adolescence, 50(12), 2440–2455. https://doi.org/10.1007/s10964-021-01530-z