Small effects, big questions: the unfinished business of social media restriction science

Introduction

A large and growing body of research has sought to examine the relationship between social media use and indicators of mental wellbeing and distress, but the largely observational nature of this work has raised concerns regarding firm conclusions on causality. Randomized controlled trials (RCTs) evaluating effects of interventions to restrict (i.e., reduce or pause) social media use compared against control conditions with as-usual use hold the potential for more definitive answers. In this issue, Burnell et al. (2025) meta-analyzed thirty-two such RCTs, summarizing the state of the evidence regarding the effects social media restriction interventions on several domains of wellbeing, including but not limited to depressive and anxiety symptoms. The authors advance the field by presenting convincing evidence that reducing or pausing social media use improves, to a relatively small degree, some domains of wellbeing for some people. However, as Burnell et al. carefully note, much about the effects of social media use and of social media use restriction on mental wellbeing and distress remains incompletely understood.

In this commentary, we situate Burnell et al.’s study within the context of an ongoing scientific controversy about the wellbeing-effects of social media restriction and highlight the unique importance of this latest contribution, discuss remaining gaps in the evidence base on effects of social media restriction interventions, and highlight broader questions about the societal impacts of social media use.

An important contribution in the context of ongoing controversy

There is ongoing debate and controversy in the scientific community over the contribution of rising social media use to deteriorating population-level mental health since the early 2010s, particularly among young people (Capraro et al., 2025; Office of the Surgeon General, 2023; Sanders et al., 2024; Thorp, 2024). As part of this debate, competing appraisals of the extant evidence have resulted in conflicting conclusions about the harm, or absence of such harm, that is attributable to social media. One such controversy directly set the stage for the current meta-analysis by Burnell et al.

In 2024, Ferguson published what we believe to be the first meta-analysis of the effects of social media experiments on mental health (2024). The pre-registered meta-analysis found no average effect of social media interventions on mental health, but was subsequently heavily scrutinized—most notably by Rausch and Haidt who conducted their own post-hoc analyses supporting alternative conclusions (Rausch & Haidt, 2024a; Rausch & Haidt, 2024b; Stein, 2024). Notable critiques included decisions related to study inclusion and exclusions; averaging across effect sizes from different types of experiments (i.e., social media restriction experiments such as those examined by Burnell et al., but also lab-based social media exposure experiments and within-person trajectory analyses); reliance on a summary outcome measure of heterogeneous mental health domains; alleged inconsistencies in calculations of study sample and effect sizes; and absence of moderation analyses by intervention length (Ferguson, 2025; Rausch & Haidt, 2024a; Rausch & Haidt, 2024b; Rausch & Haidt, 2025; Stein et al., 2024; Thrul et al., 2025). The discourse extended to a rigorous methodological critique of the meta-analytic methods employed both by Ferguson in the original analysis and by Rausch and Haidt in their post-hoc re-analysis (Jané, 2024).

Ultimately, the debate resulted in little clarity about the effects of social media interventions on mental health. Nevertheless, the episode was of immense scientific value. Namely, it laid bare the key conceptual and methodological gaps that prevented the field from drawing firm conclusions. On the basis of this important debate, Burnell et al. built the current study, systematically addressing the identified gaps, thereby allowing for increased confidence in the aforementioned conclusion: reducing or pausing social media use improves some domains of wellbeing for some people, with generally small effect sizes. This trajectory represents an example of how iterative science should work—and credit is due to all contributing to the debate. Nevertheless, we emphasize that the average effects identified by Burnell et al. were small and that the included RCTs were heterogeneous and with limitations that raise questions about broad generalizability, rendering the debate over the effects of social media use on mental wellbeing and distress far from settled.

Remaining questions about the effects of social media restriction

Beyond consolidating the existing evidence on the effects of social media restriction, the Burnell et al. meta-analysis also crystalized outstanding questions for which additional research is needed. Foremost, additional attention needs to be given to the characteristics of the target populations that are recruited for social media restriction trials. Many of the studies included in the meta-analysis recruited volunteers who were sufficiently motivated to reduce their social media use that they joined a study on the topic. Further, our review of the baseline measures of average amount of social media use indicated generally modest levels of use (though there are exceptions). Indeed, among 26 studies that reported mean or median baseline screentime or hours of social media use, 15 (58%) reported a value of only two hours of use per day or less, and social media ‘addiction’ scores that were relatively low. Further, much of the attention on the potential problems of social media has focused on children and adolescents and those with pre-existing mental health conditions, yet none of the trials focused specifically on children or adolescents, and few focused on those with pre-existing mental health conditions or distress. It is unclear how these results may generalize to the group of people we might be most interested in intervening among to reduce use: those with very high levels of social media use and those who report high levels of distress due to continued use or when unable to access platforms, as well as young people with mental health conditions.

Furthermore, given that Burnell et al. present a convincing case that reducing social media has at least some benefits for some people, we can now open additional lanes of inquiry that probe these effects further. An important area for future work is to clarify which mental health outcomes are actually of most interest. Given that the authors conducted stratifies analyses for ten sub-domains of mental wellbeing, Burnell et al.’s results offer some, albeit preliminary, insight regarding this question. When considering effects that were robust across the main and sensitivity analyses, social media restriction had a significant but small effect on five of the ten evaluated sub-domains, including depressive symptoms, anxiety symptoms, stress, self-esteem, and self-rated wellbeing. Of note, while 14 out of 16 (88%) studies with effect sizes for depressive symptoms and 9 out of 10 (90%) studies with effect sizes for anxiety symptoms employed a validated scale, these scales often captured general psychological distress at non-pathological levels. For other sub-domains, including loneliness, positive and negative affect, happiness, and life-satisfaction, the authors found no statistically significant or robust effects of social media restriction interventions. For positive and negative affect and life-satisfaction, this null finding is further corroborated by an independent recent meta-analysis (Lemahieu et al., 2025). Future work will need to consider: are we interested in social media restrictions to influence all these outcomes? Some? Perhaps improving all of these outcomes is an important goal of reducing social media use, but we currently do not yet understand the extent to which social media restrictions may influence clinical outcomes, or more serious episodes of depression and anxiety that might rise to the level of clinical attention. Theorizing about what and how we want to improve mental health through social media interventions would clarify how future studies should approach outcome measurement.

Further, the often-short duration of social media restriction trials—with most (22 out of 32, 69%) studies intervening for a week or two—underscore the need to think carefully about not only the actual phenomena that are of most interest for developing and scaling interventions, but also the potential mechanisms through which social media reductions work. To reduce onset of more serious mental health problems, which is an important public health goal, short breaks from social media are unlikely to have sustained impact. The meta-analysis revealed no significant difference in effectiveness between complete abstinence and limited-use interventions, despite theoretical predictions that moderate use might be optimal for maintaining social connections while avoiding overconsumption. It may be the case that total amount of time on platforms is an inefficient target for intervention, and instead, a focus on which social media activities and types of content are particularly salient for mental health may yield more appropriate intervention targets.

Social media research outside of randomized controlled trials

It is worth considering what we can and cannot learn from RCTs of social media restrictions. Social media use has permeated virtually all aspects of modern society, creating complex webs of indirect effects that extend far beyond individual usage patterns, including shifts in social norms, changes in offline social interactions, alterations in information ecosystems, and broader cultural transformations around self-presentation. These societal-level phenomena cannot be captured in individual-randomized trials but are crucial for understanding social media’s full impact on population mental health. Furthermore, there is increasing momentum to implement social media regulations at state and federal levels across many countries, including minimum age verification requirements, content moderation mandates, or usage time restrictions (Fardouly, 2025; Maaß, Wortelker & Rott, 2024; Schneiders & Gilbert, 2024; Thimm-Kaiser & Keyes, 2025). There is an urgent need to develop infrastructure to evaluate the effects of these policies. While group-randomized trials of policy interventions are theoretically possible, practical and ethical constraints mean that such evaluations will likely rely primarily on quasi-experimental methods comparing pre-post implementation and comparing across jurisdictions with different regulatory approaches. Finally, there remains a continued need for well-designed observational research that can identify mechanisms linking specific social media exposures to well-theorized outcomes, as this mechanistic understanding can point to developing more targeted interventions that can subsequently be tested through more experimental approaches.

In summary, Burnell et al.’s meta-analysis exemplifies how iterative science should work, building systematically on prior debates and critiques to provide increasingly rigorous evidence to support small positive effects of social media restriction on some domains of wellbeing for some people. The study also crystalizes multiple crucial pathways for future research on social media restriction. It is clear that next steps in this line work need to test interventions among high-risk populations, target specific content types, clarify clinically meaningful outcomes, and continue to integrate methods beyond, but including, individual randomized trials to evaluate ongoing population-level impacts of new technology.