Computer-Mediated Communication,Social Media,and Mental Health: A Conceptual and Empirical Meta-Review

Abstract

Computer-mediated communication (CMC), and specifically social media, may affect the mental health (MH) and well-being of its users, for better or worse. Research on this topic has accumulated rapidly, accompanied by controversial public debate and numerous systematic reviews and meta-analyses. Yet, a higher-level integration of the multiple disparate conceptual and operational approaches to CMC and MH and individual review findings is desperately needed. To this end, we first develop two organizing frameworks that systematize conceptual and operational approaches to CMC and MH. Based on these frameworks, we integrate the literature through a meta-review of 34 reviews and a content analysis of 594 publications. Meta-analytic evidence, overall, suggests a small negative association between social media use and MH. However, effects are complex and depend on the CMC and MH indicators investigated. Based on our conceptual review and the evidence synthesis, we devise an agenda for future research in this interdisciplinary field.

Keywords

computer-mediated communication social media mental health well-being meta-analysis

Computer-mediated communication (CMC), the Internet, and now social and mobile media have repeatedly been characterized as a blessing or a curse for users’ mental health (MH). Widely different claims about the impact of CMC on MH have been reiterated for decades and across disciplines (e.g., Burke & Kraut, 2016; Chan, 2015; Meier et al., 2020; Orben & Przybylski, 2019; Twenge et al., 2018). Research on this relationship has accumulated particularly rapidly in recent years, with a strong focus on social media (Meier et al., 2020). Yet, the fast-paced, interdisciplinary, and fragmented nature of the field requires researchers to keep track of a staggering, ever-growing, and seemingly incompatible evidence base (for initial meta-reviews, see Appel et al., 2020; Orben, 2020).

A key driver as well as consequence of this state of the field is the conceptual diversity of researchers’ approaches to CMC and MH. Many studies and reviews seem to work from narrow, unsystematic approaches to CMC and MH, investigating widely different technology indicators (e.g., “screen time,” self-presentation on SNS, intensity of Facebook use) and a disconnected smorgasbord of MH indicators (e.g., self-esteem, loneliness, depression, life satisfaction) (e.g., Huang, 2017; Twomey & O’Reilly, 2017). Recent specification curve analyses demonstrate that the relationship between CMC and MH can differ drastically, depending on how researchers operationalize them (e.g., Orben & Przybylski, 2019). Hence, without a higher-level conceptual and empirical integration, the bigger picture of associations between CMC and MH cannot be systematically evaluated. In addition, the choice of indicators and identification of research gaps remain largely idiosyncratic.

This study addresses the need for such higher-level integration twofold. We first develop two organizing frameworks that specify how CMC and MH are conceptualized and operationalized in the literature. These frameworks allow researchers to navigate the field more reliably and facilitate systematic identification of patterns, gaps, and conceptual conflation. Moreover, they provide the background of our empirical analysis, a meta-review of systematic reviews and meta-analyses on CMC and MH. This empirical meta-review aims to (a) synthesize the main findings on the relationships between CMC and MH indicators from existing reviews. In addition, we seek to (b) apply the two organizing frameworks to the primary studies included in these reviews, to systematically identify the conceptual foci of prior research, potential conceptual conflation, and research gaps.

To this end, we first develop the theoretical frameworks of CMC and MH based on conceptual reviews and relevant empirical literature. Using these frameworks as organizing principles for the empirical meta-review, we then synthesize the findings from 34 systematically identified meta-analyses and systematic reviews as well as 594 publications included in these syntheses. By reflecting on the empirical meta-review findings through the lens of the new organizing frameworks, we conclude with an agenda for future research.

The Hierarchical CMC Taxonomy

We understand computer-mediated communication (CMC) as an inclusive umbrella term for multimodal human-to-human social interaction mediated by information and communication technologies (ICTs). Social interaction here includes all forms of interpersonal message exchange, encompassing everything from mere social attention (e.g., browsing through the Instagram feed) to deep communication (e.g., a conversation via WhatsApp voice call; Hall, 2018). This meta-review also limits ICTs to those whose primary and original—though not exclusive—function is the facilitation of CMC as social interaction (e.g., email, mobile texting, instant messenger, social network sites, but not, e.g., games). These ICTs have been at the center of recent public concern and research regarding MH effects (e.g., Twenge et al., 2018), thus representing a reasonable focus for this meta-review.

A first step of our synthesis is a systematization of the conceptual and operational approaches to CMC. A key question guided this conceptual review: How can we organize as many CMC indicators with as few levels of analysis as possible? Since no such framework existed, we used concept mapping (Booth et al., 2012) on all CMC indicators included in literature reviews on CMC and MH (see Method for the sample of included reviews). That is, we iteratively mapped out existing CMC measures in a conceptual space to reveal their key conceptual and operational similarities, hierarchies, and differences. This was done until theoretical saturation was reached, meaning that no further levels were needed to encompass all available indicators. Additionally, we grounded the identified levels and approaches in literature that theorizes CMC (see next section). This approach was advantageous over adopting, for instance, an affordances approach (Evans et al., 2017), since none of the reviewed empirical studies on CMC and MH used measures that explicitly operationalized distinct affordances as commonly defined in the literature.

Instead, by breaking down CMC measures into their basic levels of analysis, we build a parsimonious taxonomy that applies not just to one single or a few ICTs (e.g., Facebook, smartphones), but remains useful even in the face of technological change (Ellison & boyd, 2013). This taxonomy should further be exhaustive enough to encompass a wide range of CMC variables and hence facilitate navigation through the entire research landscape. With both analytical parsimony and conceptual inclusivity as our guiding principles, we propose The Hierarchical CMC Taxonomy (see Figure 1).

Figure 1.

The hierarchical CMC taxonomy.

Channel-Centered Versus Communication-Centered Conceptual Approaches to CMC

To explicate the taxonomy, we first distinguish two overarching conceptual approaches to CMC: the channel-centered and the communication-centered approach (e.g., Carr & Hayes, 2015; Ledbetter, 2014). The channel-centered approach aligns with classic (mass) media uses and effects research that studies the channel as a whole but treats the communication within the channel largely as a black box. Typical examples for the channel-centered approach are investigations of “screen time” spent on a device (e.g., the smartphone) in relation to MH (e.g., Twenge et al., 2018). The communication-centered approach, on the contrary, opens up the channel black box and investigates communication as a complex social process of interaction via messages that enfolds within (Walther, 2010).

We propose that channels—and, hence, the channel-centered approach—can be further differentiated into four main levels of analysis: (1) device, (2) type of application, (3) branded application, and (4) feature. Likewise, the communication-centered approach can be differentiated into (5) an interaction and (6) a message level. These levels of analysis are crucial to reflect upon for at least two reasons. First, each level focuses on unique aspects of CMC. For instance, studies at the device level imply that the presence, absence, or usage of the device (e.g., the smartphone) itself has implications for MH, irrespective of the specific applications or features used, or the exact nature of the communication via the device (e.g., Gonzales & Wu, 2016). In contrast, studies at the message level may, for instance, assume that certain message content is the crucial driver of CMC effects on MH (e.g., Holland & Tiggemann, 2016). Studies differing in the levels at which they operationalize CMC are likely to differ drastically in how they can explain effects of CMC on MH. They will thus differ in their implications for users, ICT developers, and MH practitioners.

Second, depending on the level of CMC analysis, studies may differ in the effects they find. For instance, studies at the interaction level may find that CMC and face-to-face communication reinforce one another and, thus, CMC can be beneficial for MH. However, this does not preclude that studies at the device level come to the conclusion that CMC is negatively related to MH, for instance, because the device can distract from other activities. Researchers wishing to draw conclusions about the bigger picture of relationships between CMC and MH need to consider the multiple levels of analysis at which CMC can be studied. In the following, we therefore briefly illustrate how each level is conceptualized.

Six Levels of CMC Analysis

(1) Devices represent the physically palpable ICTs (e.g., laptops, smartphones, or tablets) that enable CMC. Research at the device level, for instance, investigates how the number of devices used to connect to strong and weak ties (i.e., media multiplexity; Chan, 2015), smartphone use during face-to-face interactions (“phubbing”; Gonzales & Wu, 2016), or overall “screen time” (Twenge et al., 2018) relates to MH.

(2) Devices enable CMC because they allow access to types of applications built around mediated social interaction and user-generated content. As unique applications often share a specific set of core characteristics and features, they are studied under a common label (Ellison & boyd, 2013). For instance, classic types of CMC applications include email, chat rooms, or discussion boards, later joined by texting and instant messengers. More recently, applications allowing users to engage in interactions with both broad and narrow audiences have been defined under the labels of social media, with social network sites (SNS) often considered a sub-type (see Bayer et al., 2020, for a detailed discussion). Studying such types of applications is typically more precise than the device level, as it avoids conflating CMC and non-CMC device uses in a simplistic overall measure of “screen time”.

(3) The branded application level refers to variables that focus only on one or several branded application(s), such as Facebook or Instagram. While these branded applications can be subsumed under the broader types outlined above (e.g., SNS), they are frequently studied individually as key exemplars (e.g., Meier & Schäfer, 2018). It is important to distinguish this level of analysis from the previous one, as unique applications may have properties and user cultures that diverge from related applications or their broader types. For instance, while both Facebook and Twitter are considered SNS, Facebook currently affords more diverse uses (e.g., closed groups formed around specific interests). Finally, whether research investigates types of applications, or just single brands, affects the generalizability of findings.

(4) CMC channels, at their most detailed level of analysis, are constituted by individual features, the building blocks of applications. We understand a feature as “a technical tool [. . .] that enables activity on the part of the user” (Smock et al., 2011, p. 2323). Facebook users, for instance, may use the site for status updates, comments, private messages, groups, the news feed, or any combination of these features, resulting in a unique user experience, with unique relations to MH (e.g., Burke & Kraut, 2016). Crucially, research investigating the feature level specifies in more detail the kind of interactions a specific channel enables. It thus allows researchers to test channel effects even while channels change in design (i.e., lose or gain certain features).

(5) Moving from the channel-centered to the communication-centered approach, we specify the interaction level. In contrast to previous levels, this level goes beyond the mere technological properties of channels and instead clarifies the process of how and with whom users communicate within a channel. Early on, CMC research conceptualized the configuration of interaction partners (e.g., one-to-one, one-to-many, many-to-one), clarifying the source and audience size of a communication episode, and distinguished between synchronous and asynchronous communication (e.g., Morris & Ogan, 1996). Beyond their configuration, the characteristics of communication partners (e.g., their tie strength) can be specified and studied in relation to MH (Burke & Kraut, 2016). If either the sender or receiver of a mediated communication is a group of individuals (“many”), the characteristics of the network structure of this group (e.g., network size, diversity) can also be considered at this level. Interactions may further differ in their interaction functions, such as self-disclosure or self-presentation (Walther, 2010). Another key concept clarifying the how of communication is the directionality of interaction. Rafaeli’s (1988) definition of interactivity specifies interaction as a continuum of contingent responsiveness between communication partners, reaching from two-way truly interactive (e.g., a continuous message exchange), over two-way reactive (e.g., an Instagram like), to one-way non-interactive communication (e.g., browsing through the Instagram feed). Similarly, in research on CMC (specifically, SNS) and MH, usage is often grouped into “active” and “passive”. While active usage, in its broadest sense, refers to “activities that facilitate direct exchanges with other(s)” (Verduyn et al., 2017, p. 281), passive usage refers to the mere consumption of messages from status updates, comments, profiles, or stories without any direct response to the sender, akin to classic mass media usage (e.g., watching TV). Thus, passive usage is entirely non-interactive and instead can be thought of as one-way communication from the recipient’s perspective, solely entailing non-directed messages (i.e., messages not sent in reaction to a previous message) (Burke & Kraut, 2016). Active usage, in contrast, may entail both non-interactive one-way communication from the sender’s perspective (e.g., posting a status update without getting any response), as well as two-way reactive, and fully interactive communication (Rafaeli, 1988). In conclusion, the interaction level focuses on social interaction as the process of message exchange, including instances in which this “exchange” is one-sided (i.e., sending or receiving without any response).

(6) While interactions have specific properties, each individual message within an interaction can be considered as the final level of analysis (Ledbetter, 2014). A first distinction is made between different modes of messages (e.g., text, image, voice, video, or one-click reactions such as likes or emojis; Burke & Kraut, 2016; Walther, 2010). While originally a property of separate (types of) applications (e.g., email vs. video-conferencing), many modes of communication can now be readily switched within a single application or even a message exchange (e.g., receiving a text message in WhatsApp and replying with a short voice recording). The mode of a communication is thus best placed at the message level. Along with the mode varies bandwidth (i.e., the available cues) and social presence (Walther, 2010). Similarly, the persistence versus ephemerality of a message used to be a fixed channel characteristic but can now often be modified from message to message (e.g., on Snapchat). The same applies to the accessibility of a message, varying on a continuum from private to public (O’Sullivan & Carr, 2018). The content of a message is another key variable at this level, with multiple possible specifications (e.g., concerning topic or valence).

Note that the taxonomy organizes the six CMC levels in a hierarchy, emphasizing that each lower level (e.g., a single message) can be nested in a higher level (e.g., an interaction). Thus, necessarily, properties of lower levels (e.g., whether an interaction is active or passive) can be incorporated at higher levels (e.g., active vs. passive use of Instagram). The six levels of analysis are conceptualized as rigorously distinguishable ideal types. However, empirical research may often (inadvertently) conflate hierarchical analytical levels, that is, combine properties of several levels in a single CMC indicator. For instance, “passive usage of the Facebook news feed” entails information on a unique branded application, a feature, and the directionality of an interaction process. Finding that such an indicator affects MH raises the question whether this is caused by Facebook (but not other applications), the news feed (but not other features), or passive usage (but not other forms of engaging with the Facebook news feed). We hope that by reflecting on the conceptualization of CMC more systematically through the taxonomy, researchers will be better able to identify at which level(s) of analysis their explanatory focus is located, hence avoiding conflation and increasing construct validity.

Technology-Centered Versus User-Centered Operational Approaches

Beyond the two conceptual approaches (channel- vs. communication-centered) and the six levels of analysis, we supplement our taxonomy with two operational approaches to separate measurement from level of analysis. Prior research on CMC and MH has used a staggering number of measures, ranging from time spent with a device, over types of self-presentation on Facebook, to the content of messages encountered on SNS (e.g., Holland & Tiggemann, 2016; Twenge et al., 2018; Twomey & O’Reilly, 2017). We contend that the operationalizations of CMC differ crucially in whether they are technology-centered or user-centered. Technology-centered operationalizations are descriptive measures that capture some aspect of technology usage, such as its volume (time spent, frequency) or message content, which can principally be observed (e.g., digitally tracked), though they are often measured via self-report. User-centered operationalizations, in contrast, have a psychological-perceptual component that qualifies how a person processes using a CMC technology or why they uses it, which is often most validly captured by self-reports (e.g., attitudes about technology, motivations for usage, perceptions of message content). This distinction is critical, because the two approaches imply drastically different explanatory foci when relating a CMC variable to MH. Essentially, the technology-centered approach argues that the mere exposure to some aspect of a technology itself is related to MH, whereas the user-centered approach explains any relation between CMC and MH through the user’s psychology in interaction with the technology. We note that, in principle, both operational approaches can be applied to all six levels of analysis.

The Extended Two-Continua Model of Mental Health

Mental health (MH), according to the World Health Organization, is more than the absence of mental disorders, but “a state of well-being in which every individual realizes his or her own potential, can cope with the normal stresses of life, can work productively and fruitfully, and is able to make a contribution to her or his community” (World Health Organization, 2005, p. 2). Although this comprehensive understanding of MH is now widely recognized and implemented in policy and practice (e.g., Saxena et al., 2013), research on MH is still mostly divided into two distinct perspectives, psychopathology and psychological well-being. Psychopathology (PTH) refers to “any pattern of behavior—broadly defined to include actions, emotions, motivations, and cognitive and regulatory processes—that causes personal distress or impairs significant life functions, such as social relationships, education, work, and health maintenance” (Lahey et al., 2017, p. 143). While well-being, in contrast, means “how well individuals are doing in life, including social, health, material, and subjective dimensions of well-being” (Diener et al., 2018, p. 3), psychological well-being (PWB), specifically, is understood as “optimal psychological functioning and experience” (Ryan & Deci, 2001, p. 142).

The present study builds on a two-continua model of mental health that integrates these two perspectives into a single coherent framework (Greenspoon & Saklofske, 2001; Keyes, 2005). Several arguments call for such a twofold perspective on MH. First, PTH and PWB represent different psychological states. PTH indicates severe disturbance of a person’s psychological functioning (i.e., dysfunction). PTH narrows an individual’s attention towards the source(s) of disturbance and inhibits normal functioning until the disturbance has been mitigated or eliminated (Lahey et al., 2017). PWB, in turn, indicates how well a person is doing and how much they thrives psychologically. Higher PWB is associated with a variety of positive outcomes such as longevity and prosocial behavior (Diener et al., 2018). Thus, PWB is not the absence of PTH, just as PTH is not the absence of PWB. Second, PTH and PWB are sensitive to different individual and environmental influences (e.g., genes, age, life events) and their indicators fluctuate in unique patterns and timeframes (Diener et al., 2018; Lahey et al., 2017). Third, PTH and PWB are sometimes empirically dissociated. That is, individuals can show high levels on some aspects of PWB while also reporting moderate to high levels on indicators of PTH, or vice versa (e.g., Greenspoon & Saklofske, 2001; Hides et al., 2020). In conclusion, researchers should understand and assess MH as two continua, PTH and PWB, and reflect upon which of these continua is relevant for their research.

Since researchers in the field of CMC and MH employ a variety of so far disconnected MH indicators (e.g., loneliness, self-esteem, life satisfaction, depression, or anxiety; see, e.g., Huang, 2017; Liu et al., 2019), we refine and explicate the classic two-continua model by integrating main dimensions and manifestations of both PTH and PWB, as well as risk and resilience factors, in an Extended Two-Continua Model of Mental Health (see Figure A1 in Supplemental Appendix I). In doing so, we enable researchers to locate and reflect upon MH indicators within the broader context of MH research, both clinical and non-clinical. This should not only facilitate integration of future research on CMC and MH but also lays the foundation for our empirical meta-review. In the following, we will outline how PTH and PWB are further differentiated into main dimensions and manifestations.

Dimensions and Manifestations of Psychopathology

PTH research and practice traditionally distinguishes categorically separable disorders from symptoms (e.g., Lahey et al., 2017). Clusters of symptoms represent the (more or less) manifest basis for the categorical diagnosis of disorders, which are described in diagnostic manuals such as the DSM-5 (American Psychiatric Association, 2013). A disorder comprises a set of symptoms relevant for a specific diagnosis (e.g., depressive symptoms for major depressive disorder). While clinical disorders are categorically diagnosed as either present or absent (American Psychiatric Association, 2013), symptoms are often measured via self- or other-report on a continuum (e.g., depressive symptomatology). This reflects that PTH is “continuously distributed in the population” (Conway et al., 2019, p. 428) and individuals healthy from a clinical point of view can show sub-clinical levels of PTH symptomatology.

Recently, MH research increasingly (re-)discovers that categorical distinctions between PTH disorders are largely artificial, as symptoms across disorders show high systematic covariation (i.e., comorbidity; Lahey et al., 2017). Specifically, researchers now believe PTH manifestations (symptoms and, hence, disorders) to be expressions of several underlying latent dimensions (see Conway et al., 2019, for a detailed mapping of disorders onto PTH dimensions). In the context of CMC research, we focus on the internalizing and externalizing dimensions of PTH, as these (a) are most widely recognized, especially in Clinical Psychology research on children and adolescents (e.g., Lahey et al., 2017), and (b) show the clearest connections to CMC (e.g., Sarmiento et al., 2018). While internalizing PTH refers to overcontrolled behavior, cognitions, and emotions (e.g., anxiety, social phobia, and depression), externalizing PTH refers to undercontrolled behavior, cognitions, and emotions (e.g., hyperactivity, aggression, delinquency, and substance abuse; Conway et al., 2019; Lahey et al., 2017). We thus extend the dual-factor model of MH by clustering PTH manifestations in the two dimensions of internalizing and externalizing PTH. Crucially, instead of investigating disconnected PTH indicators, this allows for the recognition of effect patterns between CMC and higher-level dimensions and manifestations of PTH (see Conway et al., 2019, for additional arguments supporting a dimensional approach to PTH). However, as the research integration of major PTH dimensions is still ongoing (Conway et al., 2019), future revisions of the MH model may include additional PTH dimensions. Moreover, categorical diagnoses are expedient for clinical practice and thus remain relevant.

Dimensions and Manifestations of Psychological Well-Being

Research on PWB distinguishes two key dimensions, hedonic well-being and eudaimonic well-being (Huta & Waterman, 2014; Martela & Sheldon, 2019; Ryan & Deci, 2001). According to the hedonic view, happiness and well-being are defined purely by a subjective experience of pleasure and contentment. A prominent operationalization of this approach is Diener et al.’s subjective well-being (Diener et al., 2018; Huta & Waterman, 2014), consisting of the two interrelated components affective well-being (high positive and low negative affect) and cognitive well-being (satisfaction with life overall and specific life domains). In contrast, the eudaimonic view understands well-being as more than just pleasure and satisfaction. Instead, it propagates the realization of a “true self” (i.e., the daímōn), a concept often associated with striving for meaning and purpose, personal growth, authenticity, and excellence (Huta & Waterman, 2014). At its core, hedonic well-being is about “feeling well,” whereas eudaimonic well-being is about “doing well” (Martela & Sheldon, 2019). While appearing somewhat “elitist” at first glance, eudaimonic well-being is present in the everyday lives of the general population (for recent reviews, see Huta, 2017; Huta & Waterman, 2014; Martela & Sheldon, 2019). Individuals may experience eudaimonic well-being by pursuing their personal or professional goals, engaging in meaningful social interactions, or living autonomously (Martela & Sheldon, 2019). Importantly, experiencing hedonic well-being does not have to be associated with increased eudaimonic well-being and vice versa (Huta, 2017). From this, it follows that investigations into the relationship between CMC and PWB should consider both sides of well-being, hedonic and eudaimonic.

The two dimensions of hedonic and eudaimonic well-being can be further distinguished by their manifestations in daily life. Huta (2017; Huta & Waterman, 2014) proposes that PWB concepts can be differentiated by their (1) category of analysis and (2) level of measurement (trait vs. state). The category of analysis specifies what exactly the well-being indicator measures: orientations (i.e., values, motives, and goals), behaviors (i.e., overt activities such as socializing or writing a diary), experiences (i.e., subjective cognitive and affective states), and functioning (i.e., how well a person is doing, e.g., concerning abilities, accomplishments, or healthy habits; see Huta, 2017, for a detailed description).

Finally, the level of measurement distinguishes between traits that are relatively stable over time, though not immutable, and states that capture the construct of interest with regard to a specific timeframe (e.g., in the moment, the last week, or the last month). As these distinctions crucially specify what exactly researchers are studying when they employ PWB measures, we incorporate Huta’s distinctions into the MH model (see Figure A1 in Supplemental Appendix I). We refer readers interested in the multitude of potential PWB indicators and their place in this model to Huta (2017), as a detailed mapping of all indicators goes beyond the scope of this paper.

Risk and Resilience Factors

As a final extension of the original two-continua model of MH (Greenspoon & Saklofske, 2001; Keyes, 2005), we complement it with risk and resilience factors. Adding these factors appears necessary, as they comprise several variables that have been studied extensively in relation to CMC and are often interpreted as directly indicative of MH (e.g., Huang, 2017; Liu et al., 2019). However, they do not distinctively map onto underlying dimensions of PTH (internalizing, externalizing) or PWB (hedonic, eudaimonic) as defined in the MH literature (see the sections above). Instead, risk factors are here defined as sub-clinical aspects of psychosocial functioning that are (a) non-specific to PTH or PWB dimensions and (b) may increase an individual’s vulnerability to develop PTH symptomatology or decrease PWB (and vice versa for resilience factors). Risk factors may include perceived loneliness, actual social isolation, perceived stress, or poor sleep quality, among many others. Resilience factors include, for instance, social capital, social support, self-esteem, or high sleep quality.

The Present Study

With these two newly developed organizing frameworks as theoretical background, we turn our attention to the evidence on the relationship between CMC and MH. Currently, researchers, practitioners, and members of the general public (e.g., parents, teachers, policy makers, or entrepreneurs) are left with a disconnected and fast-growing review literature that lacks higher-level conceptual and empirical integration. We thus aim to move this field forward by conducting a meta-review—a review of reviews.

First, we aim to synthesize the main findings on the relationship between CMC and MH, considering all available evidence that matches the definitions of CMC and MH. Based on this evidence, we can arrive at (1) more reliable conclusions about the associations between CMC and MH and (2) the current state of the field as well as (3) discover higher-level patterns of results. These efforts are guided by the following research question:

RQ1: What are the main findings of research syntheses on the relationship between CMC and MH?

Beyond reviewing the findings of research syntheses, we also aim to apply the two newly developed organizing frameworks to the empirical studies conducted on CMC and MH so far. Specifically, we seek to explore which levels of CMC analysis and which dimensions of MH have been primarily investigated so far. In doing so, we will be able to systematically identify patterns of prior research focus, discuss their implications, and uncover where research attention may be particularly needed. This is guided by the following question:

RQ2: Which (a) indicators of CMC and (b) indicators of MH have been studied by prior research and (c) which gaps can be identified based on this assessment?

The literature proposes multiple theoretical links and boundary conditions for CMC and MH effects. These include displacement or disruption of activities beneficial for well-being, such as face-to-face communication or sleep (e.g., Sbarra et al., 2019); social comparison (Verduyn et al., 2017); or relational maintenance (Burke & Kraut, 2016), among many others. While these mechanisms currently lack higher-level integration, as well, this is outside the scope of the present study. Instead, we prioritize conceptual approaches to the key variables, CMC and MH, and their empirical association.

Method

Meta-Review as a Method of Research Synthesis

Meta-reviews, also called overviews or umbrella reviews, “compile information from multiple systematic reviews to provide a comprehensive synthesis of evidence” (Ballard & Montgomery, 2017, p. 92), focusing “on breadth rather than depth of coverage” (Thomson et al., 2010, p. 198). Therefore, they typically investigate broader constructs (here: CMC and MH) and include a range of operationalizations. They allow comparisons between research foci, results, and conclusions from multiple reviews. Thus, meta-reviews help identify inconsistencies and discord in the literature and point to future directions (Polanin et al., 2017).

While the methodology of meta-reviews is still developing (Ballard & Montgomery, 2017), researchers can generally apply the steps undertaken in systematic reviews of primary research to conduct a meta-review (Polanin et al., 2017). Accordingly, we (1) state pre-defined eligibility criteria, (2) use a systematic, multi-step literature search, and (3) systematically synthesize and present the characteristics and findings of included reviews (Booth et al., 2012). As a deviation from common meta-review methodology, we also synthesize information from the primary research included in all reviews to answer RQ2.

Eligibility Criteria

To be eligible, a review had to meet seven inclusion criteria concerning scope (i.e., meet our definitions of (1) CMC and (2) MH, (3) their conceptual independence, and (4) include investigations of non-clinical samples) and methodology (i.e., synthesis articles had to be (5) systematic (Booth et al., 2012), (6) synthesize empirical evidence, and be (7) written in English and published). A more detailed description as well as exclusions resulting from these criteria can be found in Supplemental Appendix II. Note that we purposefully excluded research on problematic or addictive ICT usage, as this research, by default, defines and measures CMC as a pathological behavior that impairs MH. Similarly, we excluded clinical samples since we were interested in CMC and MH in the general population. In case a review included evidence on excluded constructs (e.g., pathological usage) or populations (e.g., clinical participants) next to evidence matching our inclusion criteria, we included it and synthesized only eligible evidence (e.g., on non-pathological usage or non-clinical participants).

Systematic Literature Search and Selection

Following recommendations from research synthesis literature (e.g., Polanin et al., 2017), we combined several methods to maximize recall of eligible reviews. As part of a larger effort to review literature on CMC and MH, we conducted standardized academic database searches, citation searches, and reference searches. This was complemented by a Google Scholar title search, targeted specifically at finding systematic reviews and meta-analyzes. A detailed description of all steps undertaken in the literature search and selection, including reliability analysis, can be found in Supplemental Appendix III. The search was first completed in December 2017 and then updated during peer review in September 2019. The final sample consisted of 34 reviews, described in detail in Supplemental Appendix IV.

A common issue in meta-reviews is overlap, meaning that more weight is given to publications included in more than one review (Pieper et al., 2014). Our sample of reviews included 1313 unique publications. Based on the formula provided by Pieper et al. (2014), overlap can be characterized as “slight,” with a corrected covered area (CCA) of 1.3%. Bias due to overlap is thus very unlikely.

Methods of Synthesis

Synthesis was conducted in two stages. In stage one, we descriptively synthesized the information (i.e., narrative conclusions, investigated constructs, effect sizes) from the 34 reviews to answer RQ1. In stage two, we synthesized the CMC and MH indicators investigated in all relevant primary research publications included in the 34 reviews to answer RQ2. For this stage, a coding protocol was developed. We first determined whether a publication was eligible for our meta-review (see eligibility criteria 1–4 and 7) and then coded all relevant CMC and MH indicators. A description of the coding protocol and results of inter-coder reliability tests can be found in Supplemental Appendix III.

Results

Main Findings of Reviews and Meta-Analyzes

To answer RQ1, we first summarize the narrative conclusions about the relationship between CMC and MH from all 34 reviews. Since the reviews included 14 meta-analyses and these provide more informative and conclusive evidence synthesis than narrative reviews, we then summarize the meta-analytic effects, effect heterogeneity, and moderator analyses.

Narrative Conclusions

First results on RQ1 (see Supplemental Appendix IV for details) show that 14 out of 34 reviews concluded the relationship was mixed, finding evidence for positive, negative, and non-significant associations between CMC and MH. Notably, these were mostly narrative reviews rather than meta-analyses. While an additional 11 reviews concluded that negative relationships prevailed, 6 found predominantly positive relationships between CMC and MH. However, these six reviews exclusively synthesized evidence on social resources (capital or support) and/or older adults. Notably, 24 of 34 reviews qualified the investigated effects as conditional, emphasizing that their strength or direction depended on moderators or mediators. Finally, 7 reviews qualified the evidence as insufficient for a definitive conclusion.

Meta-Analytic Effects

We collected all meta-analytic effect sizes on relationships between CMC and MH indicators that matched our conceptual definitions. Almost all meta-analyses focused on indicators of global SNS use (i.e., time spent, frequency, and/or intensity). We refer to these simply as SNS use below and summarize all effects of SNS use in Figure 2. As meta-analyses mostly assessed the type of or branded application levels, we organize this section along the MH dimensions. Wherever available, we highlight findings on CMC indicators other than SNS use, if they rely on k > 2 effect sizes. If multiple effects for the same relationship are available, we only report the one relying on the largest number of k effect sizes within the text. For details on all effect sizes, meta-analyses, and references, see Supplemental Appendix V.

Figure 2.

Forest plot of effect sizes for global SNS use (i.e., time spent, frequency, and/or intensity) and mental health.

Resilience factors

Consistent with narrative conclusions, all meta-analyses on social resources (capital and support) showed small to moderate positive associations with SNS use. While general Internet use, blogs, chat, and email were not significantly associated with perceived social resources, SNS (r = .30, 95% CI [.14; .46]) and forum use (r = .14, 95% CI [.09; .20]) were. Notably, user-centered attitudinal measures of “intensity” (e.g., the Facebook intensity scale) consistently generated larger effect sizes than technology-centered ones (time spent or frequency). Almost all SNS features and interaction properties of SNS use were positively associated with increased social resources, albeit at varying strength (see Supplemental Appendix V for details and references). Only few meta-analyses specifically investigated self-esteem, and none reported findings on other resilience factors. General time spent online was unrelated to self-esteem, but SNS use was slightly negatively related to self-esteem in three meta-analyses finding similar effect sizes (e.g., r = −.05, 95% CI [−.09; −.01]).

Psychological well-being

The only meta-analyzed indicator tapping into hedonic well-being was life satisfaction. No meta-analytic results on eudaimonic well-being were found (see also RQ2 below). General time spent online showed a small negative association with life satisfaction (r = −.05, 95% CI [−.12; −.01]). SNS use, however, showed no significant association with life satisfaction in two meta-analyses. One meta-analysis reported an overall effect size of SNS use (i.e., global use, number of friends, active and passive use) on “positive indicators of MH,” comprising life satisfaction, well-being, self-esteem, and positive affect (r = .05, 95% CI [.01; .08]). However, when separated by SNS indicators, only the number of SNS friends showed a small positive association with “positive MH” (r = .13, 95% CI [.05; .21]). Three other meta-analyses reported effects on “well-being” that included reverse coded negative indicators (e.g., depressive symptoms or loneliness) alongside resilience factors (e.g., self-esteem) and life satisfaction. Time spent online was found to be slightly negatively associated with such “overall well-being” (r = −.04, 95% CI [−.07; −.01]), though this relationship was nonsignificant for social Internet use. SNS use, however, was slightly negatively associated with overall well-being in two meta-analyses (e.g., r = −.06, 95% CI [−.09; −.03]). Differentiating between SNS uses revealed that “self-presentational” use (status updates, photos) was unrelated to overall well-being, “content consumption” (browsing, searching, monitoring) was negatively (r = −.14, 95% CI [−.20; −.08]), and “interactions” (replying, commenting, liking) were positively related (r = .14, 95% CI [.08; .20]). While phone calls showed a small positive association with overall well-being (r = .10, 95% CI [.06; .15]), texting and instant messenger use were not related to overall well-being.

Risk factors

Findings on risk factors are limited to loneliness and stress. While two smaller meta-analyses (both k = 23) found a small positive association between SNS use and loneliness, a considerably larger one (k = 196) found no association (r = .01, 95% CI [−.02; .05]). Phone calls, texting, and instant messaging showed small negative associations with loneliness, though based on only a few studies each (see Supplemental Appendix V for details). SNS use showed a small positive association with stress (r = .13, 95% CI [.05; .21]).

Psychopathology

The most commonly meta-analyzed indicator of internalizing PTH was depressive symptoms. No meta-analyses of externalizing PTH were found (see also RQ2 below). Five meta-analytic effect sizes for the relationship between SNS use and depressive symptoms existed, all showing a small positive association (e.g., r = .11, 95% CI [.08; .14]). In addition, one meta-analysis reported a small positive association between general social comparison on SNS and depressive symptoms (r = .23, 95% CI [.12; .34]), and a somewhat higher one for upward comparison (r = .33, 95% CI [.20; .47]). General time spent online was slightly negatively associated with reverse-coded depressive symptoms (r = −.05, 95% CI [−.07; −.02]), while instant messaging was not associated. SNS use further showed a small positive relation to social anxiety (r = .10, 95% CI, [.05; .15]) and to anxiety symptoms in general (r = .10, 95% CI [.03; .18]). Time spent online, instant messaging, texting, or email use were not related to (social) anxiety. However, social comfort experienced online (r = .34, 95% CI [.25; .41]) and comfort specifically due to reduced non-verbal cues online (r = .27, 95% CI [.23; .31]) showed moderate positive associations with social anxiety.

One meta-analysis focused on SNS and body image disturbance, which can be considered an indicator of internalizing PTH. Combining all measures of SNS use (general use and appearance-focused use), there was a small positive association with disturbed body image (r = .17, 95% CI [.13; .21]). When analyzed separately, similar effects were found for using multiple SNS or Facebook, but not for Instagram or other SNS (though based on k ≤ 5). Notably, technology-centered measures of SNS use showed about a third of the effect (r = .11, 95% CI [.08; .15]) of appearance-focused use (r = .31, 95% CI [.22; .39]), which included upward comparison and appearance-related interactions on SNS. Finally, one meta-analysis reported an overall effect of SNS use (i.e., global use, number of friends, active and passive use) on “negative indicators of MH,” comprising depression, anxiety, and loneliness (r = .06, 95% CI [.03; .09]). However, when separated by indicators, only global SNS use (time spent, frequency) showed a small association with negative MH (r = .11, 95% CI [.06; .15]).

Effect Heterogeneity, Moderator Analyzes, and Publication Bias

All meta-analyses tested for effect size heterogeneity based on the Q statistic, and nearly all concluded that there was “significant heterogeneity,” with I² often exceeding 75%. We thus synthesized findings from moderator analyses on three key sample characteristics (i.e., age, gender, and culture/country) as well as on publication bias.

Age

Of the 14 meta-analyses, 11 reported moderation effects of sample age. Two found that with increasing age the effects of SNS use on MH became less negative (concerning body image disturbance) or more positive (concerning social support), respectively. Two others found that the relationship between several CMC measures and social anxiety was stronger in older samples. Seven meta-analyses found no effect of age. Overall there is little evidence for age effects, but age had a range restricted to young users in most analyses.

Gender

Ten meta-analyses reported moderation effects of the proportion of females in study samples. Three meta-analyses found some evidence for a moderation by gender, albeit with no consistent overall trend for who benefited more or less from SNS use. Seven meta-analyses found no gender effects. Overall, there is little meta-analytic evidence for gender effects.

Culture/country

Seven meta-analyses reported moderation effects of culture or country. Only one found no moderation effect. However, the evidence from the remaining six is incoherent, with two finding more positive effects in Western/individualistic countries, two in Eastern/collectivistic countries, and two finding mixed results. Overall, culture seems to be an important moderator, but yields complex effect patterns.

Publication bias

Seven meta-analyses concluded that there was “no bias” at all. Three meta-analyzes concluded there was “little bias” and two found “some bias” for specific CMC indicators. Accordingly, meta-analysts overall found little evidence of publication bias.

CMC and MH Indicators

One key source of the high heterogeneity of effects in previous meta-analyzes may be the diversity with which CMC and MH are operationalized in studies. To systematize this diversity and answer RQ2, we turn to the primary research included in all 34 reviews. Of the 1,313 publications coded, 594 (45%) met our eligibility criteria 1 to 4 and 7. The remaining 719 publications were excluded due to lack of a relevant MH (30%) or CMC variable (15%), the manuscript being unpublished (16%) or its full text unavailable (10%), or because the publication exclusively investigated addictive or problematic usage (17%). Moreover, 7% contained only qualitative research, which was unsuitable for this stage of synthesis.

Regarding CMC, most publications included either one (23%) or two (24%) indicators, followed by three (19%), four (16%), or more (18%) (M = 3, SD = 2.1). Of the 1,829 CMC indicators in total, 51% addressed more than just one of the six CMC levels of analysis. This demonstrates considerable conflation of analytical levels within many CMC measures. Turning to the four levels of the channel-centered approach, 16% of all indicators addressed the device level (of which 68% mobile/smartphone, 19% computer, 8% various,¹ 5% other), 27% the types of application level (43% SNS, 15% various, 13% texting, 12% social media, 6% email, 4% instant messenger, 7% other), 54% the branded application level (78% Facebook, 4% Instagram, 9% various, 9% other), and 15% the feature level (37% various, 24% status update, 15% profile, 8% comment, 16% other). With regard to the two levels of the communication-centered approach, 39% of all indicators addressed the interaction level (27% network characteristics, 18% sending messages one-to-one or one-to-many, 9% self-disclosure, 8% passive usage, 38% other) and 9% the message level (51% content, 24% content of images, 9% accessibility, 6% various, 10% other).

Overall, most indicators (55%) were exclusively channel-centered, in contrast to only 5% being exclusively communication-centered. A high number of indicators (35%), however, addressed aspects of both channel and communication, suggesting that lower levels (interaction or message) were often studied in the context of a specific channel (e.g., a branded application). Six percent of indicators assessed generalized “Internet use,” neither specifying channel nor communication aspects. Concerning operationalization approaches, most indicators followed the technology-centered (69%) rather than the user-centered approach (28%). Three percent of indicators included aspects of both.

Concerning MH, most publications included only one MH indicator (43%), followed by two (28%), three (16%) or more (13%) (M = 2, SD = 1.5). Of the 1,258 MH indicators in total, 28% addressed internalizing PTH (of which 39% depressive symptoms, 22% social anxiety/social phobia, 14% anxiety symptoms, 11% eating disorder symptoms, 14% other), 3% externalizing PTH (e.g., substance abuse, aggression, AD/HD), 18% hedonic PWB (36% life satisfaction, 25% domain-specific satisfaction, 21% affect, 10% discrete emotions, 8% other), 2% eudaimonic PWB (e.g., meaning, authenticity, mastery), 17% risk factors (53% loneliness, 20% poor sleep, 19% stress, 8% other) and 31% resilience factors (38% self-esteem, 24% social support, 22% social capital, 8% good sleep, 8% other). Thus, the most studied indicators overall were risk and resilience factors (47%), followed by PTH (31%) and PWB (20%). A majority of PTH (57%) and PWB (79%) indicators as well as risk (84%) and resilience factors (91%) were measured at the trait level, without specifying a timeframe.

Discussion

Extending prior work (Appel et al., 2020; Orben, 2020), this study synthesized the fast-growing—yet conceptually and empirically fragmented—literature on CMC, social media, and MH through a meta-review. Our contribution to the literature is twofold. First, we contribute to theory building by presenting two parsimonious frameworks that offer increased organizing power, harmonize conceptual overlaps, and allow comparisons between conceptual approaches to CMC and MH. Second, we contribute to evidence synthesis by connecting and comparing review findings (RQ1) as well as units of analysis (RQ2).

Evidence on the Association Between CMC and MH

In a first step, we synthesized main findings of prior reviews (RQ1). This offers several key insights. (1) Meta-analyses condensing various CMC and MH measures into one overall effect size find a (very) small negative association (r ≈ −.05 to −.15). Yet, when associations are investigated by CMC and MH indicators separately, effect patterns become more complex. (2) There is consistent evidence that those who use SNS more intensely perceive moderately (r ≈ .20 to .40) increased social resources (social capital and support). However, there is little evidence for other positive associations between CMC and MH. (3) The remaining evidence consistently suggests those who use SNS more intensely experience slightly (r ≈ .05 to .20) more internalizing PTH (e.g., depressive symptoms), stress, and lower self-esteem. (3) Meta-analyses show no evidence for an association between SNS use and life satisfaction, the only meta-analyzed PWB indicator. Thus, SNS use is not associated with the cognitive side of hedonic well-being. The largest available meta-analysis also revealed no association between SNS use and loneliness. (4) There was little indication of publication bias across meta-analyzes. Nonetheless, effect sizes should be interpreted in light of evidence that meta-analyses produce almost three-times larger effects than preregistered replication studies (Kvarven et al., 2020).

(5) For applications other than SNS, the evidence base is small and, overall, shows little to no association with MH. There is narrative review evidence for a negative association between the device level and MH, specifically for mobile CMC. However, this requires further meta-analytic synthesis. (6) The meta-analytic evidence for the feature or interaction level (e.g., active vs. passive use) is scarce and inconsistent (cf. Supplemental Appendix V). However, it currently suggests that effects are more nuanced than for higher levels of the CMC taxonomy (i.e., types of or branded applications). The clearest pattern for the message level is a positive association between appearance-focused content and body image disturbance. Overall, findings suggest the need for more systematic research relating the feature, interaction, and message levels to MH. (7) Across several meta-analyses, there was consistent evidence that user-centered measures (e.g., attitudes toward Facebook, social comparison on SNS) resulted in two- to three times larger effect sizes than technology-centered ones (e.g., time spent, frequency). Whether this suggests that user-centered measures reveal stronger, potentially more relevant effects or produce artifacts due to, for instance, common method variance of self-report scales remains an important question.

(8) Among all 34 reviews, the most common narrative conclusion was that effects depended on moderators and/or mediators. However, meta-analyses revealed little evidence for moderating effects of age and gender—despite popular concerns about more negative effects particularly among younger and female users (e.g., Twenge et al., 2018). It should be noted, however, that the age-range was quite restricted (participants were mostly adolescents or young adults) and that narrative reviews on CMC among older adults highlighted mostly positive effects, specifically on social resources. Thus, future research needs to sample across the life span (e.g., Chan, 2015). The culture or country a study was conducted in did emerge as a relevant moderator in meta-analyses, yet showed no consistent trend. Future research should thus compare cultures more systematically. Overall, research needs to test additional moderators (e.g., personality) to explain the large heterogeneity found in meta-analyses.

Conceptual and Operational Approaches to CMC and MH

Given the range of average effects across meta-analyses (i.e., r ≈ .00 to |.40|), how researchers measure CMC and MH seems to matter considerably for the conclusions drawn in this field (see also Orben & Przybylski, 2019). In a second step, we thus synthesized conceptual and operational approaches (RQ2). Based on the detailed analysis of 1,829 CMC and 1,258 MH indicators from 594 publications, we arrive at several implications.

Measuring CMC and Social Media Use

(1) Research has largely relied on the channel-centered (e.g., devices, applications) rather than the communication-centered (e.g., interactions, messages) approach. Notably, the default approach of the field has been to study individual branded applications, specifically Facebook. This limits the evidence base severely, as findings on single applications may demonstrate little generalizability over time (e.g., due to changes in design or popularity). Instead, identifying key features used for CMC in numerous applications (e.g., status updates, profiles, private messages) should be a more future-proof way to study channels (Bayer et al., 2020).

(2) The measures of CMC in this field show considerable conflation of analytical levels, thus potentially resulting in misattribution of effects to the wrong causes (e.g., to “screen time” on a device rather than to a certain type of interaction). Research on the communication-centered approach (i.e., the interaction and message level), specifically, has conflated most measures with individual channels (e.g., “passive Facebook use”). Given that users now communicate via a multitude of channels simultaneously (i.e., media multiplexity; Chan, 2015) and the dynamic design changes of these channels, the low generalizability of the channel-centered approach also applies to most of the available evidence at the interaction and message level. Research should thus strive to develop measures that capture interaction and message characteristics independently of users’ devices or applications. In addition to (a) the low generalizability of the current channel-centered approach (Bayer et al., 2020), studying characteristics of interactions or messages (b) avoids technological determinism (i.e., social media as overall “good” or “bad”); (c) helps clarify whether one assumes effects to result from mass communication, interpersonal communication, or masspersonal communication (O’Sullivan & Carr, 2018) rather than unspecific “screen time”; and (d) allows for more nuanced conclusions about the causes of any effects, hence facilitating the development of effective interventions, if necessary.

(3) Beyond illuminating conceptual approaches, our analysis shows that researchers have largely relied on technology-centered (e.g., time spent, frequency) rather than user-centered operational approaches (e.g., how technology use was processed). However, both approaches have their pitfalls. Technology-centered measures of exposure, especially self-reports, are notoriously unreliable (Orben, 2020) and risk conflation of distinct phenomena such as interpersonal and mass communication (O’Sullivan & Carr, 2018). User-centered measures, in contrast, may artificially inflate the association between outcome (i.e., perceptions of MH) and predictor (i.e., perceptions of CMC). Moreover, they may result in misattributing outcomes of psychological processing to technology. For instance, a study finding upward comparison on Instagram negatively affects well-being cannot inform upon whether this is an effect of upward comparison, characteristics of Instagram, or both (e.g., Meier & Schäfer, 2018). Our recommendation for future research is therefore a combination of the technology- and user-centered approaches. Studies should strive to measure technology use descriptively, ideally via digital tracking (e.g., Bayer et al., 2018) and at multiple levels of the taxonomy, to allow level comparisons. Additionally, studies should assess key motivations and psychological processes that occur across channels (e.g., social comparison or social support seeking), and test how these processes are modulated by channel features and their affordances (Evans et al., 2017).

(4) Finally, we observe a discrepancy between the CMC measures meta-analyzed so far and the measures identified in our conceptual synthesis. Meta-analytic evidence is mostly limited to “global SNS use”, that is, time spent on, frequency of, or intensity of using a SNS, while many more CMC measures exist. More research on the other levels (i.e., devices, features, interactions, messages), and meta-analyses comparing these levels, are needed to ground conclusions about the role of CMC for MH in a more comprehensive evidence base.

Measuring Mental Health

(1) Existing research focuses largely on internalizing PTH, the cognitive side of hedonic PWB (i.e., life satisfaction and domain-specific satisfaction), and risk and resilience factors. Research has paid less attention to eudaimonic and affective PWB as well as externalizing PTH. Yet, these dimensions capture relevant and unique aspects of MH. Ignoring them in empirical research on CMC may thus overlook crucial effect patterns. Recent research, for instance, suggests that conclusions about the effects of social comparison on SNS partly depend on whether one investigates internalizing PTH (e.g., depression) or outcomes such as inspiration (eudaimonic PWB) and positive affect (hedonic PWB) (Meier & Schäfer, 2018). Externalizing PTH (e.g., aggression) could, in turn, be affected by online incivility and may be a more relevant PTH indicator among men (e.g., Kramer, Krueger, & Hicks, 2008). The field should thus broaden its empirical approach in order to cover the two continua of MH more completely.

(2) A second finding is a strong reliance on risk (e.g., loneliness) and resilience factors (e.g., self-esteem). These factors tap into important aspects of psychosocial functioning, relevant to MH in multiple ways. They are crucial predictors or boundary conditions for MH (e.g., social support as a buffer that increases PWB; Burke & Kraut, 2016) or link CMC indirectly to more central MH indicators (e.g., stress as a risk factor for depressive symptoms; Aalbers et al., 2019). However, our review of MH literature reveals that none of the prominent risk and resilience factors (e.g., self-esteem, loneliness, social capital) are integrated into current conceptual models of PTH or PWB. This remains an important task for MH research at large. For researchers interested in effects of CMC on MH, this suggests, however, that to truly measure MH, studies should include indicators more central to our current understanding of PWB and PTH, next to risk and resilience factors.

(3) We identified a great diversity of MH indicators across the empirical literature on CMC, which hinders research synthesis. The field should thus agree on a core outcome set of MH indicators (Brunton et al., 2020). If studies were to measure a set of the same indicators, tapping into core aspects of MH, this would greatly enhance evidence accumulation and research integration (e.g., meta-analyses). Our tentative proposal for such an outcome set would be a selection of cross-culturally validated scales covering the most central internalizing and externalizing PTH symptoms (Conway et al., 2019); cognitive and affective well-being (Diener et al., 2018); meaning as the most useful “proxy for eudaimonic experience” (Huta, 2017, p. 22); and competence, autonomy, and relatedness need satisfaction as a self-determination theory approach to eudaimonia (Martela & Sheldon, 2019). This set may, of course, be complemented by key risk and resilience factors (e.g., self-esteem, loneliness, social resources, or perceived stress) or limited to only a subset.

(4) Finally, findings show that most evidence on CMC and MH relies on trait level assessments of MH, that is, measures that do not specify a timeframe. This is problematic for several reasons. First, individual MH constructs (e.g., affective well-being) fluctuate in specific timeframes (e.g., Diener et al., 2018), which the measurement should reflect. Second, MH constructs may be temporally connected to each other—and to CMC—in unique ways. For instance, from a network perspective on PTH, phenomena such as depression are “a complex, dynamic network of symptoms that cause each other” (Aalbers et al., 2019, p. 1454). Thus, risk factors such as stress, and depressive symptoms such as sad mood, may cause other, increasingly more severe symptoms (e.g., suicidal ideation; Aalbers et al., 2019). Identifying at which points of this temporal symptom network CMC is particularly relevant is thus a crucial direction for future research. More generally, MH research should theorize and test the dynamic interplay between PTH and PWB indicators over time. For instance, individuals suffering from internalizing PTH may lack the energy necessary to pursue eudaimonic PWB. Finally, a temporal perspective on MH and CMC would also sensitize for prospective or reciprocal effects of MH on CMC (e.g., Aalbers et al., 2019).

Limitations

Several limitations need to be considered. First, evidence on the relationship between CMC and MH is largely based on small-scale, cross-sectional studies. The findings on the association, let alone causal order, of CMC and MH should be treated as preliminary (for an extended discussion, see Orben, 2020). In addition, our review, while relying on comprehensive conceptual approaches to CMC and MH and an extensive evidence base, is limited. First, we excluded some research areas, particularly on “addictive” usage of CMC and cyberbullying. These may come to different conclusions about the relationship between CMC and MH. Second, we excluded evidence from clinical samples, as research on these populations differs markedly from the evidence reviewed here. Third, we did not review theoretical mechanisms on the relationship between CMC and MH. Several reviews provide crucial syntheses of such mechanisms (e.g., Bayer et al., 2020; Liu et al., 2019; Sbarra et al., 2019). However, a comprehensive theoretical review of all relevant mechanisms and boundary conditions is outside the scope of our work. Fourth, our conceptual framework of MH by no means reflects and integrates all approaches to, and dimensions of, MH. For instance, there may be several additional dimensions of PTH beyond the internalizing and externalizing spectra (see Conway et al., 2019). Rather, our proposed MH model presents a working model covering the most relevant aspects of PTH and PWB that current theorizing from Clinical and Positive Psychology can agree on. We call on future researchers to revise the MH model based on new developments in MH research. Fifth, a necessary limitation of any literature review is a time lag between the available evidence and the evidence included in the review. Thus, there may be conceptual and empirical approaches to CMC and MH this meta-review does not include. However, given the scope of our evidence base, spanning nearly 20 years of research, we are confident that this meta-review is reasonably representative of the field’s conceptualization of and findings on CMC and MH.

Conclusion

Public concern and research attention on the impact of CMC, specifically social media, on the mental health and well-being of (young) users has dramatically increased in recent years. This study offers a conceptual and empirical review of reviews. Findings suggest an overall (very) small negative association between using SNS, the most researched CMC application, and mental health. Findings further show, however, that associations partly depend on the choice of MH indicators. On both conceptual and empirical grounds, research thus needs to develop and measure a more comprehensive set of MH outcomes, so as not to overlook effects. Moreover, associations become more complex when research addresses not just the channels used for CMC (i.e., “screen time” spent on devices or applications), but the types of interactions and messages transmitted via those channels. Instead of investigating “screen time” monolithically, the new decade of research on CMC, social media, and MH should operationalize channels through their core features, tease apart the types of interactions users engage in across channels, and consider the characteristics of messages they send and receive. Ideally, research tests how these interactions and messages are modulated by the core features and affordances of social media. By reflecting on the CMC taxonomy proposed here, specifically by avoiding conflation of its levels in measures, future research can more rigorously test which uses of social media contribute to, impair, or are irrelevant for mental health.

Supplemental Material

Meier_Reinecke_Meta_Review_Online-Appendix_R3_no_fieldcodes – Supplemental material for Computer-Mediated Communication, Social Media, and Mental Health: A Conceptual and Empirical Meta-Review

Supplemental material, Meier_Reinecke_Meta_Review_Online-Appendix_R3_no_fieldcodes for Computer-Mediated Communication, Social Media, and Mental Health: A Conceptual and Empirical Meta-Review by Adrian Meier and Leonard Reinecke in Communication Research

Footnotes

Acknowledgements

We are grateful for the financial support of this research by the Forschungsschwerpunkt Medienkonvergenz (Research Center for Media Convergence) at Johannes Gutenberg University Mainz, Germany. We extend our thanks to Catalina Toma and four anonymous reviewers, who have given invaluable feedback to earlier versions of this manuscript. Our deep gratitude also goes to Alicia Gilbert, Robin Riemann, Mareike Weiß, Lea Wilke, and Alicia Ernst for their assistance in various stages of this project.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship and/or publication of this article: This research was funded by the Forschungsschwerpunkt Medienkonvergenz (Research Center for Media Convergence) at Johannes Gutenberg University Mainz, Germany.

ORCID iDs

Adrian Meier

Leonard Reinecke

Supplemental Material

Supplemental material for this article is available online.

Notes

Author Biographies

Adrian Meier, PhD, is a research associate at the Department of Communication, Johannes Gutenberg University Mainz, Germany. His research focuses on the positive and negative effects of media and communication technology on mental health and well-being. He investigates this relationship through the lens of self-regulation and social comparison processes, using short-term longitudinal and experimental designs as well as systematic literature reviews.

Leonard Reinecke, PhD, is an associate professor for Media Psychology at the Department of Communication, Johannes Gutenberg University Mainz, Germany. His research interests include media uses and effects, media entertainment, and online communication, with a special focus on the interplay of media use, self-control, and psychological well-being.

References

Aalbers

McNally

R. J.

Heeren

de Wit

Fried

E. I.

(2019). Social media and depression symptoms: A network perspective. Journal of Experimental Psychology: General, 148(8), 1454–1462. https://doi.org/10.1037/xge0000528

American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders: DSM-5 (5th ed.). American Psychiatric Publishing.

Appel

Marker

Gnambs

(2020). Are social media ruining our lives? A review of meta-analytic evidence. Review of General Psychology, 24(1), 60–74. https://doi.org/10.1177/1089268019880891

Ballard

Montgomery

(2017). Risk of bias in overviews of reviews: A scoping review of methodological guidance and four-item checklist. Research Synthesis Methods, 8(1), 92–108. https://doi.org/10.1002/jrsm.1229

Bayer

J. B.

Ellison

Schoenebeck

Brady

Falk

E. B.

(2018). Facebook in context(s): Measuring emotional responses across time and space. New Media & Society, 20(3), 1047–1067. https://doi.org/10.1177/1461444816681522

Bayer

J. B.

Triệu

Ellison

N. B.

(2020). Social media elements, ecologies and effects. Annual Review of Psychology, 71, 471–497. https://doi.org/10.1146/annurev-psych-010419-050944

Booth

Papaioannou

Sutton

(2012). Systematic approaches to a successful literature review. Sage.

Brunton

Webbe

Oliver

Gale

(2020). Adding value to core outcome set development using multimethod systematic reviews. Research Synthesis Methods, 11(2), 248–259. https://doi.org/10.1002/jrsm.1391

Burke

Kraut

R. E.

(2016). The relationship between Facebook use and well-being depends on communication type and tie strength. Journal of Computer-Mediated Communication, 21(4), 265–281. https://doi.org/10.1111/jcc4.12162

10.

Carr

C. T.

Hayes

R. A.

(2015). Social media: Defining, developing, and divining. Atlantic Journal of Communication, 23(1), 46–65. https://doi.org/10.1080/15456870.2015.972282

11.

Chan

(2015). Multimodal connectedness and quality of life: Examining the influences of technology adoption and interpersonal communication on well-being across the life span. Journal of Computer-Mediated Communication, 20(1), 3–18. https://doi.org/10.1111/jcc4.12089

12.

Conway

C. C.

Forbes

M. K.

Forbush

K. T.

Fried

E. I.

Hallquist

M. N.

Kotov

Mullins-Sweatt

S. N.

Shackman

A. J.

Skodol

A. E.

South

S. C.

Sunderland

Waszczuk

M. A.

Zald

D. H.

Afzali

M. H.

Bornovalova

M. A.

Carragher

Docherty

A. R.

Jonas

K. G.

Krueger

R. F.

. . . Eaton

N. R.

(2019). A hierarchical taxonomy of psychopathology can transform mental health research. Perspectives on Psychological Science, 14(3), 419–436. https://doi.org/10.1177/1745691618810696

13.

Diener

Lucas

R. E.

Oishi

(2018). Advances and open questions in the science of subjective well-being. Collabra: Psychology, 4(1), 1–49. https://doi.org/10.1525/collabra.115

14.

Ellison

N. B.

boyd

d. m.

(2013). Sociality through social network sites. In Dutton

W. H.

(Ed.), The Oxford handbook of Internet studies (pp. 151–172). Oxford University Press.

15.

Evans

S. K.

Pearce

K. E.

Vitak

Treem

J. W.

(2017). Explicating affordances: A conceptual framework for understanding affordances in communication research. Journal of Computer-Mediated Communication, 22(1), 35–52. https://doi.org/10.1111/jcc4.12180

16.

Gonzales

A. L.

(2016). Public cellphone use does not activate negative responses in others. . . unless they hate cellphones. Journal of Computer-Mediated Communication, 21(5), 384–398. https://doi.org/10.1111/jcc4.12174

17.

Greenspoon

P. J.

Saklofske

D. H.

(2001). Toward an integration of subjective well-being and psychopathology. Social Indicators Research, 54(1), 81–108. https://doi.org/10.1023/A:1007219227883

18.

Hall

J. A.

(2018). When is social media use social interaction? Defining mediated social interaction. New Media & Society, 20(1), 162–179. https://doi.org/10.1177/1461444816660782

19.

Hides

Quinn

Stoyanov

Cockshaw

Kavanagh

D. J.

Shochet

Deane

F. P.

Kelly

P.J.

Keyes

C. L. M.

(2020). Testing the interrelationship between mental well-being and mental distress in young people. The Journal of Positive Psychology, 15(3), 314–324. https://doi.org/10.1080/17439760.2019.1610478

20.

Holland

Tiggemann

(2016). A systematic review of the impact of the use of social networking sites on body image and disordered eating outcomes. Body Image, 17, 100–110. https://doi.org/10.1016/j.bodyim.2016.02.008

21.

Huang

(2017). Time spent on social network sites and psychological well-being: A meta-analysis. Cyberpsychology, Behavior, and Social Networking, 20(6), 346–354. https://doi.org/10.1089/cyber.2016.0758

22.

Huta

(2017). An overview of hedonic and eudaimonic well-being concepts. In Reinecke

Oliver

M. B.

(Eds.), The Routledge handbook of media use and well-being: International perspectives on theory and research on positive media effects (pp. 14–33). Routledge.

23.

Huta

Waterman

A. S.

(2014). Eudaimonia and its distinction from hedonia: Developing a classification and terminology for understanding conceptual and operational definitions. Journal of Happiness Studies, 15(6), 1425–1456. https://doi.org/10.1007/s10902-013-9485-0

24.

Keyes

C. L. M.

(2005). Mental illness and/or mental health? Investigating axioms of the complete state model of health. Journal of Consulting and Clinical Psychology, 73(3), 539–548. https://doi.org/10.1037/0022-006X.73.3.539

25.

Kramer

M. D.

Krueger

R. F.

Hicks

B. M.

(2008). The role of internalizing and externalizing liability factors in accounting for gender differences in the prevalence of common psychopathological syndromes. Psychological Medicine, 38(1), 51–61. https://doi.org/10.1017/S0033291707001572

26.

Kvarven

Strømland

Johannesson

(2020). Comparing meta-analyzes and preregistered multiple-laboratory replication projects. Nature Human Behaviour, 4(4), 423–434. https://doi.org/10.1038/s41562-019-0787-z

27.

Lahey

B. B.

Krueger

R. F.

Rathouz

P. J.

Waldman

I. D.

Zald

D. H.

(2017). A hierarchical causal taxonomy of psychopathology across the life span. Psychological Bulletin, 143(2), 142–186. https://doi.org/10.1037/bul0000069

28.

Ledbetter

A. M.

(2014). The past and future of technology in interpersonal communication theory and research. Communication Studies, 65(4), 456–459. https://doi.org/10.1080/10510974.2014.927298

29.

Liu

Baumeister

R. F.

Yang

C.-C.

(2019). Digital communication media use and psychological well-being: A meta-analysis. Journal of Computer-Mediated Communication, 24(5), 259–274. https://doi.org/10.1093/ccc/zmz013

30.

Martela

Sheldon

K. M.

(2019). Clarifying the concept of well-being: Psychological need satisfaction as the common core connecting eudaimonic and subjective well-being. Review of General Psychology, 23(4), 458–474. https://doi.org/10.1177/1089268019880886

31.

Meier

Domahidi

Günther

(2020). Computer-mediated communication and mental health: A computational scoping review of an interdisciplinary field. In Yates

Rice

R. E.

(Eds.), The Oxford handbook of digital technology and society (pp. 79–110). Oxford University Press.

32.

Meier

Schäfer

(2018). Positive side of social comparison on social network sites: How envy can drive inspiration on Instagram. Cyberpsychology, Behavior, and Social Networking, 21(7), 411–417. https://doi.org/10.1089/cyber.2017.0708

33.

Morris

Ogan

(1996). The Internet as mass medium. Journal of Computer-Mediated Communication, 1(4), 39–50. https://doi.org/10.1111/j.1083-6101.1996.tb00174.x

34.

Orben

(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

35.

Orben

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

36.

O’Sullivan

P. B.

Carr

C. T.

(2018). Masspersonal communication: A model bridging the mass-interpersonal divide. New Media & Society, 20(3), 1161–1180. https://doi.org/10.1177/1461444816686104

37.

Pieper

Antoine

S.-L.

Mathes

Neugebauer

E. A. M.

Eikermann

(2014). Systematic review finds overlapping reviews were not mentioned in every other overview. Journal of Clinical Epidemiology, 67(4), 368–375. https://doi.org/10.1016/j.jclinepi.2013.11.007

38.

Polanin

J. R.

Maynard

B. R.

Dell

N. A.

(2017). Overviews in education research: A systematic review and analysis. Review of Educational Research, 87(1), 172–203. https://doi.org/10.3102/0034654316631117

39.

Rafaeli

(1988). Interactivity: From new media to communication. Sage Annual Review of Communication Research: Advancing Communication Science, 16, 110–134.

40.

Ryan

R. M.

Deci

E. L.

(2001). On happiness and human potentials: A review of research on hedonic and eudaimonic well-being. Annual Review of Psychology, 52(1), 141–166. https://doi.org/10.1146/annurev.psych.52.1.141

41.

Sarmiento

I. G.

Olson

Yeo

Chen

Y. A.

Toma

C. L.

Brown

B. B.

Bellmore

Mares

M.-L.

(2018). How does social media use relate to adolescents’ internalizing symptoms? Conclusions from a systematic narrative review. Adolescent Research Review, 36(5), 517. https://doi.org/10.1007/s40894-018-0095-2

42.

Saxena

Funk

Chisholm

(2013). World health assembly adopts comprehensive mental health action plan 2013–2020. The Lancet, 381(9882), 1970–1971. https://doi.org/10.1016/S0140-6736(13)61139-3

43.

Sbarra

D. A.

Briskin

J. L.

Slatcher

R. B.

(2019). Smartphones and close relationships: The case for an evolutionary mismatch. Perspectives on Psychological Science, 14(4), 596–618. https://doi.org/10.1177/1745691619826535

44.

Smock

A. D.

Ellison

N. B.

Lampe

Wohn

D. Y.

(2011). Facebook as a toolkit: A uses and gratification approach to unbundling feature use. Computers in Human Behavior, 27(6), 2322–2329. https://doi.org/10.1016/j.chb.2011.07.011

45.

Thomson

Russell

Becker

Klassen

Hartling

(2010). The evolution of a new publication type: Steps and challenges of producing overviews of reviews. Research Synthesis Methods, 1(3–4), 198–211. https://doi.org/10.1002/jrsm.30

46.

Twenge

J. M.

Martin

G. N.

Campbell

W. K.

(2018). Decreases in psychological well-being among American adolescents after 2012 and links to screen time during the rise of smartphone technology. Emotion, 18(6), 765–780. https://doi.org/10.1037/emo0000403

47.

Twomey

O’Reilly

(2017). Associations of self-presentation on Facebook with mental health and personality variables: A systematic review. Cyberpsychology, Behavior, and Social Networking, 20(10), 587–595. https://doi.org/10.1089/cyber.2017.0247

48.

Verduyn

Ybarra

Résibois

Jonides

Kross

(2017). Do social network sites enhance or undermine subjective well-being? A critical review. Social Issues and Policy Review, 11(1), 274–302. https://doi.org/10.1111/sipr.12033

49.

Walther

J. B.

(2010). Computer-mediated communication. In Berger

C. R.

Roloff

M. E.

Roskos-Ewoldsen

D. R.

(Eds.), The handbook of communication science (2nd ed., pp. 489–505). Sage.

50.

World Health Organization. (2005). Promoting mental health: Concepts, emerging evidence, practice. http://www.who.int/mental_health/evidence/MH_Promotion_Book.pdf

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.38 MB