Social media use and its impact on adolescent mental health: An umbrella review of the evidence

Literature reviews on how social media use affects adolescent mental health have accumulated at an unprecedented rate of late. Yet, a higher-level integration of the evidence is still lacking. We fill this gap with an up-to-date umbrella review , a review of reviews published between 2019 and mid-2021. Our search yielded 25 reviews: seven meta-analyses, nine systematic


Introduction
The past years have witnessed a staggering increase in empirical studies into the effects of social media use (SMU) on adolescents' mental health (e.g. [1e3]), defined as the absence of mental illness and the presence of well-being [4]. This rapid increase may be due to at least two reasons. First, SMU occupies an ever-growing part of adolescents' daily lives, whereas, at the same time, adolescents do not easily accept parental regulation of this use [5]. Second, adolescence is the stage in life in which well-being shows the most fluctuations [6], in which risk-taking is at its peak [7], and in which mental disorders, such as depression, typically emerge [8]. As social media (SM) offer adolescents ample opportunities to engage in risky behaviors, join dubious communities, and interact with strangers outside of parental oversight, it is imaginable that parents, policymakers, and researchers alike want to understand the effects of adolescents' avid SMU on their mental health.
The rapid increase in empirical studies into the effects of SMU on mental health has been paralleled with a comparable increase in literature reviews. Therefore, instead of adding another review of empirical studies, we decided to conduct an umbrella review, also called a meta-review, which is a synthesis of existing literature reviews [9]. Three earlier umbrella reviews have summarized the effects of SMU on mental health [10e12], but two of them did not focus on adolescents, and none included the 19 reviews published in 2020 and 2021. The aims of our umbrella review were to identify and discuss (1) general characteristics of existing reviews, such as the type of review (meta-analytic, systematic, narrative); (2) the conceptualization of SMU and mental health or its constituent outcomes; (3) the interpretation of the effects of SMU on these outcomes (e.g. weak, inconsistent, strong); and (4) the gaps in the evidence base and directions for future research.
Our operational definition of mental health included indicators of well-being (i.e. happiness, positive affect, This research was funded by an NWO Spinoza award to the first author. life satisfaction) and two levels of ill-being, clinical (i.e. depression, anxiety disorder) and non-clinical illbeing (i.e. depressive and anxiety symptoms, distress, negative affect). Because of space restrictions, other indicators and precursors of mental health, such as selfesteem, self-harm, suicidality, loneliness, sleep quality, or externalizing problems, were not considered. We defined SMU as the active (e.g. posting) or passive (e.g. browsing), private (one-to-one) or public (e.g. oneto-many) usage of SM platforms, such as Instagram, Snapchat, Facebook, WeChat, and WhatsApp. Studies focusing on overall 'screen time' were excluded to avoid conceptual conflation of SMU with, for example, television viewing and/or gaming (e.g. [13]).

Results
Our search yielded 25 reviews, seven meta-analyses, which either included only adolescents [14] or used age as a moderator [15e20]; nine systematic reviews (which reported a systematic search and a synthesis of included studies in tables) [21e29]; and nine narrative reviews [30e38]. Fourteen of these reviews were published in medical/psychiatric journals, eight in psychology journals, and three in social science journals.

Conceptualizations of SMU and well-being
Tables 1e3 at the end of this paper list the predictors and outcomes that each of the meta-analytic (Table 1), systematic (Table 2), and narrative reviews (Table 3) mention in their title or abstract. Although all reviews largely relied on the same evidence base, some studies used SMU in the title or abstract, others 'digital media use,' and yet others '(digital) technology use.' Six out of the 25 reviews did not define their predictor. Likewise, 15 reviews failed to define their outcome variables. Some reviews considered well-being as an aspect of mental health [31], whereas others perceived mental health as an aspect of well-being [23]. In addition, several reviews used a broad and sometimes even boundless (operational) definition of mental health, which led to the inclusion of a multitude of outcomes, including marihuana use, identity development, social support, (cyber)bullying, and/or academic performance [22,23,27,30].

Main findings of the reviews
As Table 1 shows, five meta-analyses yielded associations of general use of social network sites (SNS use) with higher levels of adolescent ill-being that ranged from very small to moderate (r = .05 to r = .17) [14,17e 20], and one did not find such an association (r = .02 ns, 15). As for well-being, one meta-analysis found that SNS use was weakly associated with higher levels of wellbeing (r = þ.05) [19], whereas another found that it was weakly related to lower levels of well-being (r = À.06) [17]. However, the latter study aggregated well-being outcomes (e.g. happiness, life satisfaction) with ill-being outcomes (e.g. reversed depression and anxiety scores) in a composite 'well-being' score. When this meta-analysis analyzed happiness, life satisfaction, and depression separately, it found that SNS use was associated with both higher levels of well-being and illbeing [17].
In all, the available meta-analytic evidence suggests that SNS use is weakly associated with higher levels of illbeing [14,17e20] but also with higher levels of wellbeing [17,19], a result that suggests that ill-being is not simply the flip-side of well-being and vice versa, and that both outcomes should be investigated in their own right [11,39]. Finally, all meta-analyses reported considerable variability in the reported associations. For example, in the meta-analysis by Ivie et al. [14], the reported associations of SMU with depressive symptoms ranged from r = À.10 to r = þ.33.
While the meta-analyses interpreted the effect sizes predominantly in statistical terms (e.g. small or moderate effect size), the systematic and narrative reviews left more room for diverging interpretations. As Tables 2 and  3 show, most of the conclusions of the 18 systematic and narrative reviews agreed that the effects of SMU are small, and the findings are inconsistent across studies. However, some reviews were less nuanced in their conclusions and used qualifications of the effect sizes such as 'substantial,' 'detrimental,' and 'deleterious' [25,30,38]. Some of these reviews also confounded the associations of general time spent with SM with problematic SMU [21,22,25], which is questionable because problematic SMU is a complex phenomenon that entails more than spending a great deal of time with SM. In fact, time spent with SM explains only 6% of problematic SMU [40]. Problematic SMU is characterized by an enduring preoccupation with SM, an inability to stop using SM, persistent neglect of one's health (e.g. lack of sleep) and important life areas (e.g. family, friends, schoolwork) [40]. For further conclusions of the systematic and narrative reviews, see Tables 2 and 3.

Identified gaps in the literature and proposed avenues for future research
As Tables 1e3 show, 21 out of the 25 reviews agreed that the evidence on which their conclusions are based is primarily cross-sectional so that causal conclusions are not warranted. Other identified gaps involved the lack of attention to mediators to explain the association of SMU with mental health (e.g. [24,32,37]), and the lack of attention to risk and protective factors that may uncover which adolescents are particularly susceptible to the effects of SMU (e.g. [28,32,37]). Most reviews, therefore, called for longitudinal studies to determine the causal direction of the effects of SMU on mental health (e.g. [14,15], and [22]), as well as for research designed to investigate why and for whom SMU is associated with mental health (e.g. [15], [26], [33]). Many reviews observed an over-reliance on self-report measures of SMU and its outcomes (e.g. [21], [33], [37]), which may have introduced various biases. This may necessitate a shift toward more objective measures of SMU, such as log-based measures. Some reviews also noted the typically small and homogenous samples (e.g. [21], [33], [41]) and the lack of attention to the content of SM interactions (e.g. [27], [34], [35]), which is likely a more important predictor than time spent with SM [11]. Another future avenue was to use research methods that distinguish between-person associations from withinperson associations of SMU with mental health [14, 27,31]. Finally, more research needs to investigate how SMU can be used to promote mental health among youth [27,34] (see Tables 1e3, for further gaps in the literature).

Discussion
In this umbrella review, we synthesized the results of 25 recent reviews into the effects of SMU on adolescent mental health. Given that adolescents' SMU is continually changing, it is important to provide regular research updates on this use and its potential effects. In addition to the many important future directions raised in earlier reviews, we discuss three crucial avenues for future research.
Defining SMU, defining mental health First, future research needs to consistently define the predictors and outcomes under investigation. Several reviews regularly switched between terms such as digital media use, technology use, and SMU without specifying to which media activities these terms refer. In some studies, emailing and gaming were part of the definitions of SMU, whereas others covered only time spent on SNSs. Such imprecise definitions may greatly hinder our understanding of the effects of SMU on mental health because different types of SMU may lead to different effects on mental health outcomes. For example, time spent on SNS is associated with higher levels of depression [17], whereas emotional connectedness to SNS ('intensity of use') [15] and the number of friends on SNS [16] are unrelated to depression. In the world of SM, everything is rapidly new and rapidly old, and, therefore, it is all the more important to define the specific types of SMU under investigation and to hypothesize how and why these types of SMU could affect mental health outcomes.
Likewise, in several reviews, both mental health and well-being were used as catchall terms that were left undefined, which sometimes led to the discussion of a potpourri of cognitive and affective outcomes that each deserve to be investigated in their own right. Our umbrella review confirmed that similar types of SMU can lead to opposite associations with different mental health outcomes [17]. Both SMU and mental health are highly complex constructs. Although most studies have focused on the associations of SMU with depression or depressive symptoms, all other constituent mental health outcomes, including their risk (e.g. loneliness) and resilience factors (e.g. selfesteem), also deserve our full research attention, provided that they are clearly defined and demarcated from other mental health outcomes.
Capturing the content and quality of SM interactions Several reviews have pointed at a need to move away from possibly biased self-report measures toward more objective measures of SMU use, such as log-based measures of time spent with SM. Indeed, self-report measures of time spent with SM correlate only moderately with similar log-based measures [42,43]. However, although log-based measures are often seen as the gold standard, they have their own validity threats, such as technical errors and the erroneous tracing of SM apps running in the background when the screen is turned off [42,43]. This means that the modest correlations between self-reports and log-based measures could be due to validity issues of self-reports but also of objective measures. More importantly, though, most log-based measures only capture time spent with SM apps, which is just as crude a predictor of mental health as comparable self-report measures. If logging measures only reiterate the 'screen time' approach of most selfreport research, they provide only a limited way forward.
To arrive at a true understanding of the effects of SMU on mental health, future research needs to adopt measures that capture adolescents' responses to specific content or qualities of SM interactions. In experimental settings, this can be realized by using mock SM sites, such as the Truman Platform (https://socialmedialab. cornell.edu/) or the mock SM site developed by Shaw et al. [44]. In non-experimental settings, there are three approaches that can be combined with survey or experience sampling studies: (1) The 'Screenomics' approach developed by Reese et al. [45], which entails end-to-end software that randomly collects screenshots of adolescents' smartphones, and extracts text and images; (2) phone-based mobile sensing [46], which captures sound via the microphone and text entered via the keyboard; and (3) analysis of SM 'data download packages' [47], the archives of SM interactions that each SM user is allowed to download. While each of these methods is promising, they require sophisticated technical skills and specific expertise. Therefore, they can best be achieved in collaborative interdisciplinary projects, which are also better equipped to realize larger samples.

Understanding inconsistent interpretations
Although the majority of the reviews concluded that the reported associations of SMU with mental health were small to moderate, some others interpreted these associations as serious [30], substantial [48] or detrimental [25]. Such disagreeing interpretations can also be Table 1 Meta-analyses on the associations of social media use (SMU) with adolescent mental health.   Table 3 Narrative reviews on the associations of social media use (SMU) with adolescent mental health.  [51,52]). Oftentimes, the involved scholars do not disagree that much about the size of the reported effects but just on how to interpret them.

Study
What has often been ignored in such debates is that the effect sizes are just what they are: statistics observed at the aggregate level. Such statistics are typically derived from heterogeneous samples of adolescents who may differ greatly in their susceptibilities to the effects of environmental influences in general [53] and media influences in particular [54]. After all, each adolescent is subject to unique dispositional, social-context, and situational factors that guide their SMU and moderate its effects [55]. Such person-specific antecedents and effects of SMU cannot be captured by the aggregatelevel statistics that have been reported in the majority of empirical studies and reviews, including the current one.
If we accept the propositions of media-specific susceptibility theories [54], it is plausible to assume that both optimistic and pessimistic conclusions about the effects of SMU are valid d they just refer to different adolescents. In fact, recent studies that have adopted an idiographic (i.e. N = 1 or person-specific) media effects paradigm [56] have found that a small group of adolescents experienced negative effects of SMU on wellbeing (around 10e15%) and another small group experienced positive effects (also around 10%e15%). Reassuringly though, most adolescents experienced no or negligible effects [57].
A person-specific approach to media effects requires a large number of respondents and a large number of withinperson observations per respondent. Indeed, statistical power is expensive. However, due to rapidly advancing technological (e.g. phone-based experience sampling methods) and methodological developments (e.g. N = 1 time series analyses), such approaches are increasingly within everyone's reach, especially when researchers pool resources in interdisciplinary teams. A person-specific media effects paradigm may not only help academics resolve controversies between optimistic and pessimistic interpretations of aggregate-level effect sizes, but it may also help us understand when, why, and for whom SMU can lead to positive or negative effects on mental health. And above all, it may help us facilitate personalized prevention and intervention strategies to help adolescents maintain or improve their mental health.