RESEARCH DOSSIER
Digital Well-Being and Media Use Outcomes
Abstract. This fundamental research monograph examines digital well-being as a family of measurable psychological and behavioral outcomes associated with specific forms of digital-media use. It reconstructs the historical development of media-effects and well-being research, establishes distinctions among affect, life satisfaction, belonging, autonomy, sleep, functioning, and clinical symptoms, and critically evaluates the explanatory roles of social connection, comparison, self-determination, displacement, exposure selection, and platform design. The evidence synthesis considers primary experience-sampling studies, observational panels, natural experiments, randomized restriction and deactivation trials, measurement-validity studies, umbrella reviews, and a 2024 national consensus report. Across these methods, outcomes vary according to interaction quality, content, developmental context, available alternatives, intervention definition, and time horizon. Randomized restriction studies indicate modest average benefits for some well-being indicators in selected populations; the evidence also identifies heterogeneous associations, differential effects across activity categories, and limitations of aggregate duration measures. The monograph develops an outcome-centered framework for causal interpretation, transparent evidence appraisal, ethical research design, and future studies of adaptive digital environments in the Artificial Era. Its method is a critical narrative scientific synthesis, not an independently conducted systematic review, meta-analysis, or original experiment.
Keywords: digital well-being; media use outcomes; subjective well-being; life satisfaction; social media effects; media exposure; social connectedness; social comparison; digital displacement; adolescent development; ecological momentary assessment; randomized deactivation; media restriction; objective screen time; causal inference; individual differences; platform design; Artificial Era
Introduction: Digital Media as a Context of Psychological Outcomes
Digital well-being and media use outcomes concern the measurable consequences of interactions with networked communication systems for affect, life evaluation, social belonging, autonomy, cognition, everyday functioning, and health-related behavior. The scientific object is an outcome system: changes in psychological states and functioning that occur in relation to differentiated media exposure, use, and nonuse. An outcome may be beneficial, adverse, mixed, delayed, or null within a specified period and population. Accordingly, the appropriate unit of explanation joins an identifiable exposure or activity to a well-defined outcome and a plausible causal or selection pathway. The broad term digital well-being denotes evaluation and experience within digitally mediated life; the research question here is how particular media activities relate to and sometimes change that well-being.
A person may use the same device to exchange supportive messages, encounter hostile comparisons, complete paid work, receive health information, play a cooperative game, or manage an urgent family problem. These activities engage different psychological processes even when aggregated device logs report an identical duration. A meaningful research design therefore specifies the platform or channel, actual content and relationship, activity and agency, timing, baseline state, and counterfactual condition. Reviews of media-effects theory have long emphasized the interaction of message, recipient, and context rather than a uniform technological stimulus (Valkenburg et al. (2016)). The present monograph develops an integrated critical narrative synthesis of this evidence, examining original observational, experimental, quasi-experimental, and intensive longitudinal research alongside systematic reviews and methodological analyses. It does not claim an independently registered systematic review or an original meta-analysis.
Definition, Scope, and Explanatory Boundaries
A media use outcome is an operationally measured psychological or behavioral endpoint associated with a specified encounter with digital media. The endpoint can be immediate affect after viewing a feed, nightly sleep timing following bedtime engagement, a change in loneliness over weeks, or academic functioning following sustained exposure. Each formulation requires an explicit timescale. A momentary increase in positive affect is distinct from durable life satisfaction; reduced anxious symptoms do not by themselves establish a lower incidence of a psychiatric disorder. Causal language is justified by the identification strategy and its assumptions, not by the cultural prominence of the technology. The literature frequently blurs measures of exposure, mechanisms, evaluative responses, and clinical symptoms into a single idea of digital harm. A disciplined research architecture holds them apart while assessing their actual interconnections.
This investigation places quality of life, mood, connectedness, displacement, benefits, and harms at the center of the scientific analysis. Mechanisms such as social comparison, attentional capture, algorithmic ranking, or peer reinforcement enter as explanations for those outcomes. Separate dossiers investigate adolescent susceptibility as a developmental object, problematic use as a potential functional impairment, and notifications as a reinforcement and interruption mechanism. Another planned dossier, Digital Well-Being and Platform-Mediated Experience, studies the wider configuration of digitally mediated experience through a positive-psychology lens. Here the explanatory emphasis is outcome-specific evidence, contrasts between media behaviors, and the strength of causal inference. This boundary permits interdisciplinary synthesis without assigning the same scientific object to several canonical pages.
The Architecture of Well-Being Measurement
Subjective well-being conventionally includes affective experience and cognitive evaluation of life. Diener (1984) offered a foundational conceptual account; Diener et al. (1985) developed the Satisfaction With Life Scale as a measure of the judgment component. Digital-media research inherits the need to distinguish how a participant feels during an interaction from the person's broader appraisal of a life. A brief episode of delight, irritation, or fatigue can be a valid outcome while revealing little about an enduring change in well-being. Measurement claims should state the construct, reference period, response format, validity evidence, and whether baseline variation was modeled.
Psychological well-being also includes perceived autonomy, competence, relationships, purpose, and functioning. Ryan & Deci (2000) provide an influential self-determination framework that can generate predictions about digital experiences that support or undermine the satisfaction of psychological needs. Across studies, mental health, well-being, life satisfaction, negative affect, stress, loneliness, and depressive symptoms are sometimes listed side by side. These measures are related but not interchangeable. A positive association between frequent messaging and belonging does not license a claim about depression prevalence. A composite index may be useful if its constituent outcomes, weights, and reliability are disclosed. When an intervention changes one indicator while leaving others unchanged, that asymmetry is a substantive result rather than a reporting inconvenience.
Historical Foundations: From Mass Media Effects to Digital Life
Media-effects research predates smartphones and social platforms. Earlier research investigated how mediated messages alter attention, interpretation, attitudes, and learning, while audience-centered programs examined why individuals choose certain content. The later growth of computer-mediated communication redirected attention to interpersonal interaction, identity presentation, and the maintenance of social relationships at a distance. Digital platforms integrated communication, entertainment, interpersonal visibility, advertising, and continuous information retrieval. The analytical challenge changed accordingly: exposure increasingly became reciprocal, personalized, and embedded in ongoing daily routines rather than bounded by a single broadcast event. The historical synthesis in Valkenburg et al. (2016) helps locate contemporary claims within a wider theory of selective, contingent media effects.
An early social-networking tradition treated platforms as possible resources for relational connection. Ellison et al. (2007) examined Facebook use and forms of social capital among college students, helping establish questions about maintaining weak ties and social support. Later research addressed detrimental experiences such as comparison and envy, and examined whether time devoted to media displaced other valued activities. The popularity of ubiquitous mobile connectivity then encouraged studies of generalized device time. That aggregate emphasis generated large datasets but also increasingly apparent interpretive problems. Modern research has therefore turned toward activity-specific measures, objective traces, within-person designs, experiments, and individual differences. Historical concepts remain useful when reformulated as operational hypotheses rather than transferred unmodified from one medium or historical platform version to another.
Active Selection, Uses, and Gratifications
People select media in pursuit of information, entertainment, reassurance, escape, coordination, connection, and identity work. Those goals influence both the content encountered and the likelihood of subsequent well-being outcomes. Selection introduces a familiar causal problem: a participant feeling lonely may seek interpersonal contact online; an already happy participant may have more energy for reciprocal communication. Observed correlations between use and mood can reflect either pathway, both pathways, or shared antecedents such as available time, illness, workload, and social resources. A credible outcome study therefore asks what purpose prompted the activity, what alternative action was available, and how prior psychological state shaped use.
Media selection also changes its consequences. A solitary person may encounter friendship or comparative exclusion; a worker may benefit from rapid communication while becoming more vulnerable to interruptions; someone seeking distressing news may become informed and simultaneously less calm. Purpose should be measured before the encounter wherever possible, since retrospective motive reports can be reshaped by the experienced result. A design that compares otherwise similar participants randomly assigned to distinct activities, content categories, or interface conditions can isolate more specific questions than a global daily-hours survey. This methodological progression accords with recent reviews urging attention to differentiated uses rather than mechanically treating exposure volume as the operative mechanism (Parry et al. (2022); Valkenburg (2022)).
Social Comparison, Envy, and Self-Evaluation
Visible profiles, status updates, images, and metrics make selected aspects of other people's lives continuously observable. The material is often curated, and the observer may compare an ordinary daily experience with an unusually favorable depiction. This creates a plausible pathway from media exposure through perceived relative standing, envy, dissatisfaction, and lower momentary affect. Vogel et al. (2014) examined social comparison and self-esteem in the social-media context, while Verduyn et al. (2015) combined experimental and longitudinal evidence concerning passive Facebook use and affective well-being. The findings help characterize a possible mechanism but cannot justify assigning the same consequence to every observer or every passive session.
Comparison effects depend on interpretation, target, domain, prior self-worth, and whether the model evokes aspiration, admiration, threat, or exclusion. Even a single activity category can contain informational, relational, and evaluative episodes. Studies can improve mechanism identification by experimentally varying target similarity, curation cues, comparison framing, and available supportive feedback while holding viewing time constant. Mediators should be measured after exposure and before the endpoint, with appropriate designs to address mediator–outcome confounding. Evidence that comparison mediates a specific effect cannot be generalized automatically to all long-term mental-health outcomes. The distinction between immediate self-evaluation and durable well-being remains central.
Self-Determination, Agency, and Perceived Control
A person may experience digital media as a chosen resource aligned with values or as a sequence of demands that capture attention despite competing goals. Self-determination theory identifies autonomy, competence, and relatedness as psychologically meaningful needs (Ryan & Deci (2000)). Within a media context, autonomy concerns the perceived capacity to direct participation and disengage; competence concerns effective participation and understanding; relatedness concerns meaningful connection. These are conceptual mechanisms whose causal operation in particular platforms requires empirical testing. An interface offering clear settings, predictable interaction, and intelligible feedback may support agency differently from a design that confuses choice or creates persistent social demands.
Researchers should distinguish subjective control from behavioral independence. People can enjoy voluntary immersion while spending substantial time online, and can experience stressful obligation during brief interactions. Measures of problematic loss of control are relevant when behavior causes persistent impairment, but ordinary habits, liking an activity, and long duration do not constitute diagnostic evidence. Interventions that promote deliberate selection, bedtime boundaries, or notification management should assess whether agency actually increases and whether valued communication becomes harder to sustain. The outcome distribution matters: a restriction may help those who feel overburdened and harm those whose primary support network is accessible through the same media channel.
Displacement, Substitution, and the Counterfactual
The displacement hypothesis proposes that time or attention devoted to media may reduce time available for sleep, in-person interaction, physical movement, study, work, or restorative solitude. The hypothesis is testable only when the alternative activity is specified. An hour of digital contact can replace an hour of sleep, an hour of television, an hour of unsupported isolation, or an hour of family conversation. Its well-being implications depend on the comparison. Consequently, a study reporting a reduction in use should measure how time and attention were redistributed. Otherwise the proposed mechanism may be an unobserved assumption.
The randomized Facebook-deactivation experiment of Allcott et al. (2020) is informative precisely because it tracked changes in offline activity and subjective well-being in addition to reduced social-media access. Deactivation increased certain offline activities, including social interaction with family and friends, and increased measured subjective well-being, while other outcomes changed in different directions. The study identifies the effect of an intervention under a particular historical and political context; it does not isolate a single intrinsic property of Facebook content. A general account of media outcomes therefore models substitution, novelty, withdrawal, adaptation, and social network responses, including the possibility that the comparator changes as platforms and everyday obligations evolve.
Evidence Hierarchies and the Identification of Media Effects
Research on media use and well-being contains cross-sectional surveys, prospective cohorts, repeated daily reports, behavioral logs, natural experiments, and randomized interventions. These designs address different estimands. A cross-sectional association describes a difference between people at one point in time. A within-person association describes covariation as the same person uses media more or less than usual. A randomized intervention estimates the consequence of an assigned change in access or behavior under a specified protocol. Natural experiments can identify causal contrasts under assumptions about selection and trends. None of these estimates automatically transfers across populations, platforms, content environments, or durations. Evidence synthesis requires naming the estimand before comparing effect sizes.
The strongest inference often comes from triangulation: do independently designed studies predict consistent consequences for the same construct and mechanism? Even this comparison must account for differential measurement and treatment definition. A deactivation trial, a notification-disabling trial, and a feed-content experiment may each change psychological outcomes but estimate different causal interventions. Some studies randomize instructions rather than verified behavior, introducing imperfect compliance. Others rely on convenience samples, short follow-up, or outcomes collected after participants infer the study hypothesis. Reviews of social media and well-being document why apparently contradictory findings can emerge from this diversity (Valkenburg (2022); Odgers & Jensen (2020)). A comprehensive monograph should preserve these distinctions while evaluating cumulative evidence.
Early Experience Sampling and Everyday Mood Trajectories
Experience-sampling studies ask participants to report recent media activities and psychological states repeatedly during ordinary life. Kross et al. (2013) examined Facebook use and subsequent subjective well-being among young adults using repeated assessments. Their results associated greater use with subsequent declines in particular well-being measures during the observation period. The temporal arrangement offers more information than a single survey, because it records changes over time. Yet the design remains susceptible to time-varying confounders: stressful events, interpersonal circumstances, loneliness, and changes in daily schedules can affect both later use and mood. The study describes a historical Facebook environment, which also constrains extrapolation to current multi-platform behaviors.
Intensive longitudinal evidence is particularly valuable for distinguishing the relatively stable characteristics of participants from fluctuations around their personal baseline. Proper analysis should specify lag length, autoregressive adjustment, missing reports, time of day, and whether exposure occurred before the measurement window. A short association between browsing and negative affect may disappear, strengthen, or reverse when the outcome is assessed later. Experience sampling also imposes burdens and can change self-monitoring. Accordingly, these studies deserve close interpretation as evidence about time-ordered everyday dynamics and possible mechanisms, complemented by experiments that manipulate particular exposures or affordances.
Passive Viewing, Experimental Exposure, and Affective Experience
The idea that passive exposure differs from reciprocal interaction emerged from studies combining survey, field, and experimental evidence. Verduyn et al. (2015) examined passive Facebook use with experimental and longitudinal components, finding evidence of less favorable affective experience in studied contexts. An experimental manipulation can help evaluate whether viewing particular material influences immediate states; the longitudinal portion can address real-world patterns. But passive viewing itself is a broad class. Reading an educational explanation, seeing a friend's joyful announcement, witnessing harassment, and consuming humorous videos all qualify under some definitions while varying substantially in appraisal and outcome.
Later critical synthesis questioned simple active-versus-passive generalizations because the categories conceal content, relationship, motive, and platform design (Verduyn et al. (2017); Valkenburg (2022)). The scientific question becomes which conditions produce which results. Researchers can test whether comparison opportunities, targeted communication, and subjective control change the effect of equivalent exposure time. A stronger model might classify a session simultaneously by content, conversational reciprocity, social tie strength, emotion-regulation goal, and algorithmic presentation. Such multidimensional exposure descriptions improve causal interpretation but require preregistered hypotheses or transparent exploratory analysis to avoid a new version of uncontrolled analytic flexibility.
Experimental Social Media Reduction in University Students
Hunt et al. (2018) randomized 143 University of Pennsylvania undergraduates following a baseline monitoring week. The experimental group was instructed to restrict Facebook, Instagram, and Snapchat to ten minutes per platform per day over three weeks, and the control group continued ordinary use. Relative to controls, the restricted group showed reductions in loneliness and depressive symptoms. Both groups improved on certain other outcomes, including anxiety and fear of missing out, suggesting that measurement, attention to use, or secular changes may also matter. This was a limited intervention in a selected student population rather than a general population estimate.
The trial demonstrates that a feasible behavioral restriction can produce psychological benefits under particular conditions, but interpretation must consider short follow-up, concurrent platform use, compliance verification, and whether benefits persist after the intervention. It does not establish a universal healthy threshold of thirty minutes, nor does it identify precisely which displaced activities or mediated processes produced improvement. Those questions require factorial experiments that distinguish monitoring from restriction and compare different combinations of social and non-social activities. Results also cannot be assumed to generalize to participants for whom online contact is a primary source of community access or practical support. The causal conclusion appropriately attaches to the assigned program, timeframe, measured outcomes, and sampled participants.
Facebook Deactivation and the Welfare Counterfactual
Allcott et al. (2020) conducted a randomized study of Facebook deactivation over the four weeks before the 2018 U.S. midterm elections. The experiment revealed changes in online activity and replacement behaviors, including increased time with family and friends, together with changes in political-information outcomes and improved subjective well-being. This is unusually rich evidence because access was manipulated, alternative activities were observed, and several outcome domains were assessed. It also exposes why an intervention may have simultaneous benefits and costs: reducing exposure can alleviate certain emotional burdens while decreasing access to relevant information or valued social infrastructure.
The assignment effect is the average impact of this deactivation protocol relative to its study control condition. Researchers should resist translating that into an invariant psychological effect of every minute of Facebook. The platform's 2018 social and political ecology, study recruitment, incentive structure, duration, and measures define the scope. Participants may have anticipated the experiment's end, switched to other services, or experienced different reactions depending on their existing social networks. Future trials can build on this design by independently manipulating feed content, social interaction opportunities, and information access, while measuring objective substitution and long-term adjustment after deactivation ends.
Platform Rollout as a Natural Experiment
Braghieri et al. (2022) exploited the staggered introduction of Facebook at U.S. colleges and linked campus-level exposure timing to student mental-health survey outcomes. Their generalized difference-in-differences strategy estimated unfavorable effects on mental health around the rollout and greater reported impairment of academic performance due to poor mental health. Additional evidence in their analysis pointed toward unfavorable social comparison as a potential explanatory process. The study's importance lies in its exposure contrast: rather than asking current users how many hours they spend online, it takes advantage of variation in whether a new social environment became available.
Natural experiments require close attention to parallel-trend assumptions, potential campus-specific changes, spillovers between exposed and unexposed students, timing of survey measures, and the historical novelty of the service. The intervention is introduction of a platform within a collegiate environment, not an incremental adjustment to a mature recommendation system. Even when a rollout estimate is credible for its historical setting, a contemporary digital-life application requires a new transportability argument. This is a general principle of technological outcome science: identification in one period provides a causal result for that setting and an informative mechanism hypothesis for other periods. The evidence should inform an accumulating theory, not serve as a substitute for current measurement.
Disaggregating Smartphone and Social Media Restriction
Walsh et al. (2024) conducted a preregistered eight-day randomized deprivation study with 338 Generation Z participants assigned to restrict broader digital-media use, restrict social-media use, restrict water use as an active comparison, or receive measurement alone. Restrictions on broader digital media were associated with improvements on several outcomes, including life satisfaction, autonomy, competence, self-esteem, loneliness, and stress. The social-media-specific condition produced fewer benefits and some unfavorable emotional results in certain comparisons. Effects differed by measured construct; the authors did not find improvements in depression or emotional well-being in every condition. These distinctions make the trial methodologically consequential.
The result undermines a simplistic model in which all media categories have identical psychological costs. Broader smartphone activity and social communication fulfill different functions, and their restriction may change different alternative behaviors. An active control also clarifies how intervention assignment, unusual instructions, and participants' expectations affect inference. The experiment remains limited by its short duration, specific age group, and the intervention's adherence and implementation. It supports targeted questions about which activities confer benefits when reduced, which needs go unmet, and how effects change when participants have more time to adapt. Outcome-specific interpretation is essential: a gain in autonomy is substantively meaningful even if depressive symptoms do not change.
Large Randomized Deactivation Before a U.S. Election
A later program directly evaluated prolonged platform deactivation on emotional outcomes before the 2020 U.S. election. Allcott et al. (2026) reported two randomized experiments comparing six weeks of deactivation with control participants deactivating for only the first week. The published report estimates an improvement of 0.060 standard deviations in an index of happiness, depression, and anxiety for the Facebook condition, and 0.041 standard deviations for the Instagram condition. Exploratory subgroup analyses suggested possible variation by age and gender across platforms. These values are small intervention-average effects on a particular composite and should remain attached to the exact study design and period.
The study illustrates both the value and limits of direct intervention evidence. Its comparator involved a brief period of deactivation, so the identified contrast is extended versus shorter deactivation, not simply complete abstinence versus uninterrupted platform access. Exploratory subgroup differences are hypotheses requiring replication. The election environment may alter exposure to conflict, information, and social commentary. These results should be considered alongside the earlier Facebook experiment and recent restriction trials because they test related but nonidentical interventions. The cumulative picture supports modest average psychological changes under some deactivation regimes, while leaving open the behavioral mechanisms, individual distribution of benefits, and durability of effects after ordinary participation resumes.
Meta-Analytic Evidence and the Scale of Average Change
Burnell et al. (2025) conducted a meta-analysis of randomized social-media restriction studies and included 32 articles representing 5,544 participants and 91 effect sizes. Their random-effects analysis estimated a small beneficial average for subjective well-being, standardized g = 0.17 with a 95% confidence interval of 0.08 to 0.27. Effects differed across specific outcomes; estimates for some relatively stable indicators such as life satisfaction were less compelling than for some symptom or state measures. Their sample base consisted of college students and adults and was disproportionately female, limiting generalization to younger adolescents and other populations. They also documented variation and methodological limitations across studies.
A pooled estimate integrates studies under analytical assumptions; it does not make the contributing interventions uniform. The reviewers excluded the 2020 Allcott experiment from their main synthesis because its control arm also received a deactivation instruction. That decision highlights how inclusion criteria affect the intended comparison. A complementary examination by Skeggs & Orben (2025) mapped the broader field of social media interventions and emphasized careful intervention definition. The empirical synthesis therefore supports modest average gains from some restrictions, with heterogeneous effects, unresolved intervention mechanisms, and restricted demographic coverage. The most productive next step is experimental comparison of meaningful behaviors and contexts rather than a search for one universally optimal duration.
Between-Person Associations and Within-Person Change
A common statistical error is to interpret a correlation between people as an effect within a person. If people who spend more time on a platform also report lower average well-being, the pattern may arise from preexisting social circumstances, temperament, stress, or differences in measurement. It does not follow that a particular person would become happier by reducing their use by the population-average difference. Within-person analysis instead asks whether departures from an individual's usual behavior predict departures from that individual's usual well-being. Even that association may remain confounded by daily circumstances, but it defines a more precise psychological question.
Beyens et al. (2020) repeatedly assessed adolescents' everyday experiences, obtaining 2,155 real-time observations. For passive social media use, the study reported considerable person-specific variation: 44% of adolescents showed no improvement or decline, 46% showed a more favorable association, and 10% showed a less favorable association under its analytic classification. Those proportions describe this specific sample, measure, and analysis, and are not prevalence estimates for all young people. Their central implication is methodological: an average association may conceal positive, negative, and null individual patterns. Scientific synthesis should report the population estimate, variation across individuals, precision of subgroup estimates, and whether heterogeneous effects replicate in independent samples.
Developmental Timing and Adolescent Susceptibility
Adolescence involves changing peer relationships, identity exploration, social evaluation, school demands, family contexts, and sleep timing. Digital communication may interact with these changes, but the age pattern cannot be inferred merely by comparing adolescents with adults. Developmental effects require repeated observation and attention to sex, puberty, school transitions, cohort, and historical platform exposure. Orben et al. (2022) analyzed two U.K. datasets covering 84,011 people and reported age-dependent associations between reported social-media use and life satisfaction, including evidence consistent with developmental sensitivity in specific periods. The analysis raises mechanistic hypotheses while remaining sensitive to observational confounding and measurement limitations.
The broader developmental evidence provides reasons to study peer visibility, social evaluation, identity formation, and family scaffolding in their own right. Odgers & Jensen (2020) reviewed evidence about adolescent mental health in the digital age and argued for stronger data and less generalized inference. The National Academies (2024) integrated evidence about adolescent health, platform design, harassment, literacy, and research access in a consensus report. The correct conclusion is an agenda of age-sensitive, context-specific tests. Adolescent outcomes deserve a separate scientific program with longitudinal methods, validated measures, and careful safety governance; adult restriction trials cannot be presented as direct estimates of adolescent intervention benefits.
Measurement Error in Screen Time and Reported Activity
Most ordinary surveys ask respondents to estimate hours spent online, smartphone pickups, or frequency of checking. Such recollections can be inaccurate because many interactions are brief, interleaved with other tasks, or poorly distinguished by platform. Parry et al. (2021) systematically evaluated discrepancies between logged and self-reported digital media use. Their findings demonstrate that retrospective usage estimates are not interchangeable with behavioral traces. Measurement error may attenuate genuine associations or create artifacts when reporting errors correlate with affect, beliefs, or self-perception.
Behavioral logs provide event timing, duration, application identity, and sometimes interaction category, but they still do not reveal what an interaction meant. Foreground application time differs from attentive engagement; background media, co-use, and communication through multiple devices complicate totals. A robust measurement system combines validated trace data with contextual momentary reports, content or activity classification, and independent outcomes. Researchers should disclose whether logs were obtained from operating systems, platform servers, participants' exports, or commercial analytics, and whether those sources capture the intended exposure. Logs can support greater temporal accuracy while introducing privacy risks, selective participation, and platform-specific missingness. Better measurement requires both technological validity and psychological construct validity.
Analytic Flexibility and the Problem of Small Associations
Large digital-well-being datasets make many statistically significant results possible, particularly when researchers can choose among multiple exposure definitions, subgroups, adjustment sets, and outcome scales after observing the data. Orben & Przybylski (2019) demonstrated the importance of analytical flexibility in large adolescent technology-and-well-being datasets and helped establish specification-focused approaches to evaluating small associations. Statistical significance at scale is not itself evidence of a substantial or socially consequential effect. Conversely, a small population-average association may coexist with serious harms or meaningful benefits in smaller groups and circumstances.
Pre-registration, specification-curve analysis, transparent code, sensitivity analyses, and independent replication address distinct weaknesses. Preregistration guards against hindsight in selected hypotheses; sensitivity analysis tests the stability of estimates under credible alternatives; replication evaluates transport to new samples and versions of the exposure. Where exposure is endogenous, more statistical controls do not automatically identify a causal effect, and controls measured after exposure can induce bias. Researchers should present standardized and raw-scale estimates, uncertainty intervals, prespecified primary outcomes, and credible minimum effects of practical interest. The scientific priority is explanatory precision rather than emphasizing a universal verdict based on whichever aggregate coefficient happened to cross a significance threshold.
Content, Recommendation, and Exposure Composition
Digital media content is increasingly distributed through personalized ranking, recommendation, search, and social forwarding. As a result, nominally identical platform use can expose different users to substantially different material. Ranking may alter emotional tone, visibility of social feedback, opportunities for upward comparison, and perceptions of social norms. These features introduce additional selection loops: past behavior influences future recommendations, future recommendations alter engagement, and engagement produces further training signals. A simple association between time and well-being averages across these evolving exposure histories. The resulting data-generating process calls for models of sequential treatment, dynamic confounding, and social interference.
Specific platform-affordance experiments may reveal causal pathways, but a change in rank ordering is not equivalent to a change in total time online. Likewise, a politically salient or distressing content category can have consequences through beliefs or emotional responses that cannot be attributed to screen exposure as such. A useful intervention study records actual served impressions and received interactions where ethically and legally possible, rather than assuming exposure from platform membership. Research about political information and algorithmic ranking also reminds investigators to measure outcomes outside self-reported mood, including knowledge, trust, polarization, and offline behavior when relevant (Allcott et al. (2020)). Such outcomes may change in different directions, requiring an explicit normative account of how they are weighed.
Sleep, Circadian Opportunity, and Functional Outcomes
Late-night digital engagement is plausibly connected to sleep through several mechanisms: the displacement of intended bedtime, heightened cognitive or emotional arousal, persistent notifications, and exposure to light. These mechanisms are related but need separate measures. A person can lose sleep because a conversation runs late even when the screen's light has little additional effect; another may experience disrupted rest from alerts that continue after active use has stopped. Sleep outcomes include timing, latency, duration, continuity, next-day alertness, and subjective quality. Their health implications depend on age, occupation, school schedule, and baseline sleep conditions.
Nagata et al. (2024) studied bedtime screen-use behavior and sleep outcomes in an early-adolescent prospective cohort, providing a relevant observational basis for exposure-specific research. Prospective sequencing improves the interpretation of timing but does not eliminate influences from household routines, chronotype, workload, or preexisting sleep problems. An experimental design might randomize notifications, device location, or evening engagement while preserving necessary communication and reporting objective and subjective sleep outcomes. Mechanism-specific interventions can clarify whether an observed improvement comes from fewer disruptions, earlier bedtime, altered content, or participants' expectations. Sleep should be treated as a distinct functional endpoint, not collapsed into a generic digital well-being score.
Harmful Social Encounters and Protective Relationships
Media use outcomes include distress associated with harassment, unwanted contact, threats, social exclusion, and hostile messages. They also include access to peers, identity-affirming communities, informational support, and emergency communication. Exposure to harassment is qualitatively distinct from ordinary scrolling; effects may depend on severity, repetition, audience size, social identity, relationship to the perpetrator, and available support. A person enduring hostile targeting cannot be represented adequately by a dataset containing only total hours online. Correspondingly, an intervention that disconnects a vulnerable person from all online communities can remove both harmful exposures and valued protective relationships.
A scientific account should measure harmful events directly, use ethically appropriate procedures, and evaluate support and reporting mechanisms. Because victimization is connected to substantial distress, researchers must incorporate consent, data protection, adverse-event procedures, and appropriate referral protocols for high-risk participants. The National Academies (2024) emphasized design, accountability, and harassment-related governance in its adolescent social-media assessment. The evidence for particular harms should be interpreted by exposure type and severity; diagnoses and individual treatment recommendations belong to clinical assessment and specialized literature. The appropriate outcome framework includes safety, access, social participation, and recovery, not only frequency of digital engagement.
Cultural Context, Accessibility, and Digital Inequality
The opportunity to benefit from a digital environment varies with language, disability access, affordability, geography, education, political conditions, and the availability of offline substitutes. A restriction that is trivial for someone with extensive local contacts may remove the only practical route to a support community for someone living remotely. Moreover, measurement scales developed in one language or cultural setting may not measure belonging, life satisfaction, or autonomy identically elsewhere. The distribution of benefits and harms therefore requires attention to sampling, context, and measurement invariance.
A cross-cultural outcome design would record the technological infrastructure, cost of access, types of accessible communication, and prevailing social expectations rather than treating national categories as psychological causes. Participants with disabilities may use assistive technologies and online communities to obtain opportunities not available offline; their experience should not be reduced to generic duration categories. Researchers can test platform access interventions, usability improvements, moderation practices, and inclusive design through appropriately stratified samples and accessible procedures. Disaggregated results should not be reported where they would create privacy risks or identify small vulnerable groups. Fairness requires evaluating whether proposed interventions redistribute both psychological benefit and practical burden across populations.
Dynamic Adaptation and the Problem of Temporal Scale
Media-use effects may depend on the novelty of a platform, gradual habit formation, changing social norms, and adaptation after exposure or restriction. A laboratory study of a five-minute feed manipulation is well suited to immediate affect; it cannot demonstrate a stable change in life satisfaction over a year. A six-week deactivation experiment is informative about relatively short-term behavioral adaptation but does not establish consequences across successive developmental stages. Technological systems themselves change, sometimes much faster than academic publication cycles. The object measured during data collection may have acquired new interfaces, ranking systems, and social meanings by the time results appear.
Longitudinal research should therefore date and characterize the technology version and observe multiple outcome timescales. Repeated baseline periods, post-intervention follow-up, and flexible growth models can help distinguish novelty, habituation, rebound, and enduring change. Participants may shift from one platform to another in response to restrictions, changing the total informational and relational environment. Network spillovers may occur when some participants alter their participation and others must adapt. These phenomena are not methodological nuisances alone: they are substantive features of digitally mediated social life. A durable explanatory theory must specify the temporal process through which benefits and harms emerge, dissipate, or accumulate.
Intervention Families and Their Distinct Targets
Digital-well-being interventions comprise different causal strategies. Total abstinence temporarily eliminates an entire platform; partial restriction changes available duration; notification management reduces interruptions; interface adjustment modifies affordances; content curation changes the informational environment; skills-based interventions aim to change appraisal or behavior. These interventions may have different effects even when participants report similar reductions in use. For each intervention, researchers should state the target mechanism, comparison condition, adherence measure, outcome schedule, and plausible unintended consequences. Directly comparing strategies is more informative than accumulating trials with imprecise treatment labels.
The randomized evidence reviewed by Burnell et al. (2025) supports a small average benefit of restriction in the studied populations while leaving substantial room for improved mechanism identification. Walsh et al. (2024) showed that broader digital restriction and social-media-specific restriction need not produce identical effects. Skeggs & Orben (2025) discuss the emerging intervention landscape. The practical implication for research is to replace a generic intervention variable with a taxonomy of purpose, contact, timing, exposure, and design. Outcomes should include beneficial activities lost through restrictions as well as any psychological improvements.
Evidence for Restriction, Its Limits, and Ethical Tradeoffs
Interventions that curtail access may reduce distressing comparison, exposure to conflict, habit-driven checking, or bedtime interference. They may also impair contact with supportive networks, community information, educational resources, and interpersonal coordination. A well-designed trial assesses the full outcome distribution, including negative effects, acceptability, adherence, and unintended substitution. Reporting a significant reduction in loneliness while omitting an unchanged measure of life satisfaction creates a biased understanding of the intervention's overall effects. The combination of small average benefits and heterogeneity observed across randomized studies justifies proportionate evaluation rather than assuming either universal harm or universal benefit.
The 2026 Facebook and Instagram deactivation trials offered comparatively precise estimates for specified emotional composite outcomes (Allcott et al. (2026)). However, the estimated benefits were small, and exploratory subgroup patterns require further testing. Longer follow-up should examine whether changes persist, whether people adapt to different media, and whether psychological functioning improves independently of the temporary novelty of intervention. For children and adolescents, intervention ethics further demand developmental protection, attention to consent and family context, and evidence directly obtained from relevant age groups. Outcome studies should avoid equating one observed clinical symptom measure with a diagnosis or treating population evidence as a personal medical prescription.
An Integrative Model of Exposure, Selection, and Outcome
A scientifically useful conceptual model can be expressed as a sequence of conditional relationships: prior context and psychological state shape media choice; selected platforms and algorithms influence actual content and interaction; individuals appraise that experience through psychological needs, expectations, relationships, and stress; subsequent affect, cognition, behavior, and social feedback update both the person's state and the next exposure. This is a research synthesis of established mechanisms, not the announcement of an independently validated new theory. The model makes clear why reverse causation and feedback loops are expected. It also distinguishes platforms as environments, media activity as behavior, and well-being as measured consequence.
A formal causal graph could include baseline mood, prior social resources, exposure opportunity, activity selection, delivered content, relational responses, sleep displacement, and later affect. Such a graph must be specified before controlling for variables, since some apparent confounders are actually mediators or consequences. Randomization of one node—such as notification timing—may identify a local effect even when the complete graph remains partly uncertain. Cross-study synthesis should ask whether interventions on different nodes converge on a common mechanism. The more informative question is not how much technology a person uses in isolation, but how different exposures and counterfactual opportunities organize psychological functioning in a known population and timescale.
Comparative Evidence Matrix and Confidence Judgments
The literature supports several distinguishable propositions. First, some restrictive and deactivation interventions improve average well-being modestly in particular adult and student samples, as supported by direct randomized evidence (Allcott et al. (2020); Burnell et al. (2025)). Second, specific interaction qualities correlate with different well-being outcomes, with the strongest interpretive caution when the study is observational (Burke & Kraut (2016)). Third, person-specific associations vary considerably within intensive adolescent samples (Beyens et al. (2020)). Fourth, screen-time self-reports contain consequential measurement error (Parry et al. (2021)). These statements concern distinct estimands rather than four rival explanations of one universal coefficient.
Confidence judgments depend on design, sample, exposure validity, outcome specificity, uncertainty intervals, replicability, and external validity. Random assignment increases confidence in a protocol's causal effect but does not eliminate attrition, noncompliance, expectancy effects, or transportability limitations. A large observational study may provide precise population patterns but still lack causal identification. A meta-analysis may estimate the average effect more efficiently while inheriting study-level biases and treatment heterogeneity. The proper scientific synthesis uses evidence grades as a structured assessment of what has been learned and what remains unidentified, rather than treating sample size, publication prestige, or statistical significance as an independent guarantee of truth.
Research Ethics, Platform Data, and Independent Replication
Digital-media outcomes research frequently depends on information generated by private systems that control exposure, rankings, participant access, and logging. This raises questions about researcher independence, reproducibility, informed consent, retention of sensitive trace data, and the incentives that shape which results become available. Platform collaboration can enable experiments at a scale otherwise difficult to obtain; it can also restrict external scrutiny of sampling, preprocessing, and changes in the service during the trial. Independent replication requires adequate access to methods, code, measurement definitions, and ethically shareable data, together with a precise description of historical platform conditions.
Investigators should minimize collection of private communication content, avoid reidentification through granular traces, and evaluate harms to people whose data may appear in networked observations without direct enrollment. Publicly shared posts remain context-sensitive personal information. Researchers should document conflicts of interest and distinguish data access from editorial control. The National Academies (2024) highlighted researcher access, accountability, and design transparency in the adolescent social-media context. Scientific quality and participant protection reinforce each other when exposure categories can be measured without indiscriminate surveillance.
Artificial Mediation and Future Digital Well-Being
The Artificial Era extends media-use research from relatively static content delivery toward persistent interaction with adaptive, conversational, and generative systems. An artificial assistant may support scheduling, social exchange, knowledge acquisition, creative exploration, or emotion regulation, while introducing new uncertainties regarding trust, dependency, privacy, and control. These are distinct psychological hypotheses and operational research targets. They require measures of interaction frequency, system behavior, user expectation, functional autonomy, and longitudinal outcomes. The historical transformation From Homo to Artificial expands the ecology in which psychological outcomes occur, while existing experimental and psychometric standards remain necessary for empirical evaluation.
A study of human–artificial mediation should identify the difference between an automated interaction and the next-best available alternative. Comparator conditions might include a search interface, human-supported service, noninteractive content, or no additional tool. Human–artificial relational experience cannot be collapsed into ordinary entertainment exposure, and positive user ratings do not establish long-term well-being benefits. Mechanism-specific trials should examine whether adaptive systems change perceived agency, information quality, connectedness, and the distribution of cognitive labor, while documenting risks of inaccurate advice or overreliance. This research direction connects the present dossier to the Psychology of Artificial and Psychology of Human–Artificial Relations without altering the evidence-based scope of established media-outcome findings.
Research Priorities for Cumulative Scientific Progress
The next stage of research requires more informative intervention contrasts, reliable and privacy-preserving behavioral traces, repeated outcome assessment, and designs capable of distinguishing individual change from between-person selection. Major priorities include multi-site randomized comparisons of duration reduction versus content change, tests of purposeful interaction versus passive exposure, and long follow-up examining whether displacement and adaptation explain transient benefits. Mechanism-specific investigations should establish whether shifts in autonomy, belonging, comparison, sleep, and perceived control mediate outcomes with appropriate temporal ordering and sufficient statistical power. Such programs would connect theoretical claims to identifiable interventions.
A second priority is sampling. Many restriction trials disproportionately enroll students and adults in affluent connected settings. Age-stratified cohorts, disability-inclusive designs, more culturally diverse populations, and participants with limited offline support are needed to establish generalizability. Preregistration, independent data auditing, transparent outcome hierarchies, and public reporting of null results would counter selective emphasis on striking associations. Where multi-platform social interference prevents conventional individual randomization, cluster experiments, stepped implementation, or credible natural experiments should be considered with explicit assumptions. Progress depends on cumulative identification of particular effects within defined contexts, supported by replication across populations and changing technology versions.
Critical Scientific Discussion
The central disagreement in media-use psychology is often described as a dispute over whether technology makes people happier or less happy. That formulation combines exposures that vary in content, purpose, relationship, and agency; outcomes that vary in temporal stability and clinical significance; and designs that identify different questions. When these elements are separated, many apparently conflicting findings become interpretable. Restrictions can yield modest average improvements while supportive communication confers localized benefits; one subgroup may experience more social comparison while another gains connection; a short-term mood shift can coexist with unchanged life satisfaction. A comprehensive evidence account must allow different endpoints to move in different directions.
Important uncertainties remain. Intervention results may depend on participant expectations, selective adherence, and the specific alternatives available when access is curtailed. Intensive observational studies reveal heterogeneity but are not equivalent to person-specific causal trials. Proprietary recommendation systems complicate measurement and replication, and researchers rarely observe the full network feedback induced by a platform change. Published experimental evidence does not supply a global, age-invariant effect of screen time. The existing literature supports a differentiated, outcome-centered research program whose principal task is identification of circumstances and mechanisms rather than endorsement of one general media verdict.
Conclusion
Digital well-being and media use outcomes form a multidimensional psychological field organized by measurable experiences, causal pathways, and the opportunities available beyond a given medium. The historical development of media-effects research, social-capital theory, self-determination, social comparison, and contemporary causal inference provides a coherent foundation for analyzing this field. The evidence documents small average improvements in well-being under some restriction and deactivation interventions, substantial heterogeneity in everyday associations, meaningful differences between relational activities, and serious limitations of recalled duration as a universal exposure measure. The strongest conclusions attach to well-specified populations, interventions, outcome measures, and timescales.
The resulting research framework treats quality of connection, agency, exposure content, behavioral displacement, sleep opportunity, developmental context, and social inequality as scientifically consequential variables. Future progress will come from ethically conducted, adequately powered, transparent studies that manipulate meaningful mechanisms, validate exposure measures, report the full distribution of outcomes, and test durability across changing digital environments. Within the Artificial Era, new adaptive media and artificial interaction systems extend the range of possible outcomes and research questions. The central scientific discipline remains constant: every claim about human flourishing or harm must be tied to a defined psychological construct and a verifiable chain of evidence.
References
- Allcott, H., Braghieri, L., Eichmeyer, S., & Gentzkow, M. (2020). The welfare effects of social media. American Economic Review, 110(3), 629–676. https://doi.org/10.1257/aer.20190658
- Allcott, H., Gentzkow, M., Wittenbrink, B., Mason, W., Wilkins, A., Stroud, N. J., Tucker, J. A., & US 2020 Facebook and Instagram Election Study Collective. (2026). The effect of deactivating Facebook and Instagram on users' emotional state. American Economic Journal: Economic Policy, 18(4), 188–207. https://doi.org/10.1257/pol.20240806
- Beyens, I., Pouwels, J. L., van Driel, I. I., Keijsers, L., & Valkenburg, P. M. (2020). The effect of social media on well-being differs from adolescent to adolescent. Scientific Reports, 10, Article 10763. https://doi.org/10.1038/s41598-020-67727-7
- Braghieri, L., Levy, R., & Makarin, A. (2022). Social media and mental health. American Economic Review, 112(11), 3660–3693. https://doi.org/10.1257/aer.20211218
- Burke, M., & 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
- Burnell, K., Meter, D. J., Andrade, F. C., Slocum, A. N., & George, M. J. (2025). The effects of social media restriction: Meta-analytic evidence from randomized controlled trials. SSM – Mental Health, 7, Article 100459. https://doi.org/10.1016/j.ssmmh.2025.100459
- Diener, E. (1984). Subjective well-being. Psychological Bulletin, 95(3), 542–575. https://pubmed.ncbi.nlm.nih.gov/6399758/
- Diener, E., Emmons, R. A., Larsen, R. J., & Griffin, S. (1985). The Satisfaction With Life Scale. Journal of Personality Assessment, 49(1), 71–75. https://doi.org/10.1207/s15327752jpa4901_13
- Ellison, N. B., Steinfield, C., & Lampe, C. (2007). The benefits of Facebook “friends”: Social capital and college students’ use of online social network sites. Journal of Computer-Mediated Communication, 12(4), 1143–1168. https://doi.org/10.1111/j.1083-6101.2007.00367.x
- Hunt, M. G., Marx, R., Lipson, C., & Young, J. (2018). No more FOMO: Limiting social media decreases loneliness and depression. Journal of Social and Clinical Psychology, 37(10), 751–768. https://doi.org/10.1521/jscp.2018.37.10.751
- Kross, E., Verduyn, P., Demiralp, E., Park, J., Lee, D. S., Lin, N., Shablack, H., Jonides, J., & Ybarra, O. (2013). Facebook use predicts declines in subjective well-being in young adults. PLOS ONE, 8(8), e69841. https://doi.org/10.1371/journal.pone.0069841
- Nagata, J. M., Cheng, C. M., Shim, J., Kiss, O., Ganson, K. T., Testa, A., He, J., & Baker, F. C. (2024). Bedtime screen use behaviors and sleep outcomes in early adolescents: A prospective cohort study. Journal of Adolescent Health, 75(4), 650–655. https://doi.org/10.1016/j.jadohealth.2024.06.006
- National Academies of Sciences, Engineering, and Medicine. (2024). Social media and adolescent health. National Academies Press. https://doi.org/10.17226/27396
- Odgers, C. L., & Jensen, M. R. (2020). Annual research review: Adolescent mental health in the digital age: Facts, fears, and future directions. Journal of Child Psychology and Psychiatry, 61(3), 336–348. https://doi.org/10.1111/jcpp.13190
- Orben, A., & Przybylski, A. K. (2019). The association between adolescent well-being and digital technology use. Nature Human Behaviour, 3, 173–182. https://doi.org/10.1038/s41562-018-0506-1
- Orben, A., Przybylski, A. K., Blakemore, S.-J., & Kievit, R. A. (2022). Windows of developmental sensitivity to social media. Nature Communications, 13, Article 1649. https://doi.org/10.1038/s41467-022-29296-3
- Parry, D. A., Davidson, B. I., Sewall, C. J. R., Fisher, J. T., Mieczkowski, H., & Quintana, D. S. (2021). A systematic review and meta-analysis of discrepancies between logged and self-reported digital media use. Nature Human Behaviour, 5, 1535–1547. https://doi.org/10.1038/s41562-021-01117-5
- Parry, D. A., Fisher, J. T., Mieczkowski, H., Sewall, C. J. R., & Davidson, B. I. (2022). Social media and well-being: A methodological perspective. Current Opinion in Psychology, 45, Article 101285. https://doi.org/10.1016/j.copsyc.2021.11.005
- Ryan, R. M., & Deci, E. L. (2000). Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. American Psychologist, 55(1), 68–78. https://doi.org/10.1037/0003-066X.55.1.68
- Skeggs, A., & Orben, A. (2025). Social media interventions to improve well-being. Nature Human Behaviour, 9, 1079–1089. https://doi.org/10.1038/s41562-025-02167-9
- Valkenburg, P. M. (2022). Social media use and well-being: What we know and what we need to know. Current Opinion in Psychology, 45, Article 101294. https://doi.org/10.1016/j.copsyc.2021.12.006
- Valkenburg, P. M., Peter, J., & Walther, J. B. (2016). Media effects: Theory and research. Annual Review of Psychology, 67, 315–338. https://doi.org/10.1146/annurev-psych-122414-033608
- Verduyn, P., Lee, D. S., Park, J., Shablack, H., Orvell, A., Bayer, J., Ybarra, O., Jonides, J., & Kross, E. (2015). Passive Facebook usage undermines affective well-being: Experimental and longitudinal evidence. Journal of Experimental Psychology: General, 144(2), 480–488. https://doi.org/10.1037/xge0000057
- Verduyn, P., Ybarra, O., Résibois, M., Jonides, J., & Kross, E. (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
- Vogel, E. A., Rose, J. P., Roberts, L. R., & Eckles, K. (2014). Social comparison, social media, and self-esteem. Psychology of Popular Media Culture, 3(4), 206–222. https://doi.org/10.1037/ppm0000047
- Walsh, L. C., Regan, A., Okabe-Miyamoto, K., & Lyubomirsky, S. (2024). Does putting down your smartphone make you happier? The effects of restricting digital media on well-being. PLOS ONE, 19(10), e0306910. https://doi.org/10.1371/journal.pone.0306910
Social Connection and the Quality of Digital Interaction
The relationship between digital communication and psychological well-being depends on what interaction accomplishes in a person's actual relational life. A personalized message from a trusted friend may provide recognition, emotional support, and continuity of attachment; a broadcast viewed by strangers may provide social information without reciprocal support. Burke & Kraut (2016) linked server-logged interaction to repeated well-being assessments and found that receiving composed, targeted communication from strong ties was associated with more favorable well-being trajectories than low-effort clicks or broadly distributed content. Their design was observational, and participants completing the repeated surveys differed from the wider platform population. It nevertheless demonstrates why equating message exchanges, passive viewing, and one-click reactions under a single usage total obscures relational mechanisms.
Digital communication can extend existing relationships across geography and mobility constraints, help communities maintain continuity, and connect individuals whose offline environments are restrictive. It can also intensify rejection, social surveillance, conflict, and perceived exclusion. The relevant exposure includes relationship quality and perceived reciprocity, not simply an online/offline distinction. More communication is not always more support; burdensome group messaging and unreciprocated contact may create social demands. An outcome model should measure supportive interactions, belonging, loneliness, conflict, and the opportunity costs of communication separately. This relational interpretation can accommodate both beneficial and harmful empirical findings without presuming a uniform effect for media as a category.