A systematic review of worldwide causal and correlational evidence on digital media and democracy

The paper addresses the contentious question of whether and how the global spread of digital media affects democracy. Motivated by the dual-use nature of communication technologies and the unique affordances of digital platforms that empower users and platform governance alike, the authors aim to provide an evidence-based synthesis on links between digital media and politically relevant variables (for example, participation, trust, knowledge, polarization, populism). The pre-registered research question asks if, to what degree, and in which contexts digital media have detrimental effects on democracy, encompassing both effect directions and contextual moderators (for example, regime type). Given that most existing studies are correlational, the authors undertake a broad review of associations and an in-depth appraisal of the smaller subset of studies that credibly identify causal effects. Digital media is defined broadly (internet access, online news, social media platforms and content exposure), and political outcomes are mapped to core democratic principles (trust in institutions, informed citizenry, active civil society, exposure to diverse opinions) as well as potentially detrimental phenomena (polarization, populism, discrimination, homophily). The study seeks to integrate disparate literatures to inform public debate and future platform design and policy.

The authors situate their work among prior reviews on specific topics such as radicalization, polarization, hate speech, participation, echo chambers, and Twitter campaigning. They note that existing reviews typically address subsets of media–politics relationships and often omit a cross-variable, cross-method synthesis spanning multiple political outcomes and media types. Unlike reviews focused on elites, institutions, or traditional media, this work concentrates on citizens’ behaviors and attitudes in relation to digital media and excludes purely traditional media effects and non-digital consumption behaviors. It also departs from lab-only micro-mechanism studies to emphasize external validity, prioritizing field experiments and large-scale observational designs. This comprehensive approach is intended to reveal patterns and gaps that are not visible within siloed literatures.

The systematic review followed MOOSE guidelines and was pre-registered on OSF (https://osf.io/7ry4a/). Searches were conducted in Scopus and Web of Science using a broad query targeting digital media, political outcomes, and both causal and descriptive methods (final update: 09/15/2021). Inclusion criteria: original empirical, peer-reviewed journal articles in English reporting associations between a digital media variable and a political outcome (or single-variable studies when measuring a digital-media feature like misinformation prevalence). Exclusions included conceptual/theoretical and simulation studies, preprints, small-N laboratory experiments and student convenience samples (N<100), studies on digitization of government, and outcomes not specific to digital media. Screening proceeded via title and abstract review with dual coding; disagreements were resolved through discussion. Inter-coder reliability after abstract screening was Krippendorff’s alpha = 0.66 (87% agreement). The final sample comprised N=496 articles. Variables were inductively categorized into digital media types, content features (for example, misinformation, selective exposure), and political outcome categories, with iterative refinement. For ten top outcome categories, direction of association with digital media use was coded conceptually as beneficial or detrimental to liberal democracy (irrespective of statistical sign). Causal inference studies (N=24) were examined in-depth, summarizing treatments, outcomes, mediators, moderators, settings, and primary identification strategies, including instrumental variables, matching, panel designs (fixed effects, DiD), natural experiments, and large-scale field experiments. Due to heterogeneity in measures and methods (surveys, social media trace data, web-tracking, field experiments), meta-analytic effect sizes were not computed. Additional coding documented sampling methods, sample sizes, transparency indicators (competing interests, open data, pre-registration) and assessed potential risks of bias. Data and code are publicly available on OSF (tables of all retrieved studies, included studies, and coded effects; R scripts).

- Scope: N=496 articles reviewed; N=354 associations coded across the ten most frequent political outcome categories. Causal evidence subset: N=24 studies. - Beneficial associations: Digital media use is most consistently associated with increased political participation/mobilization (from low-cost online actions to high-cost protest, particularly in emerging democracies and autocracies), with suggestive positive links to political knowledge and more diverse news exposure. Some evidence indicates enhanced political expression, though findings are mixed due to spiral-of-silence dynamics and context. - Detrimental associations: Strong and consistent associations link digital media use to decreased trust in institutions and media; increased political polarization (affective and other forms); growth of populism and right-wing radicalization; greater exposure to misinformation; and more homophily in online networks (echo chamber-like structures), often documented on social media (notably Twitter). - Causal evidence: Across established democracies, causal studies generally find that social media/internet use increases political participation and information seeking/knowledge, but also increases polarization and reduces trust in mainstream media and social trust. In authoritarian contexts, greater internet access often decreases trust in political institutions, which may be democratically enabling in those settings. Some field experiments show countervailing or null effects (for example, counter-attitudinal news on Facebook reducing affective polarization; partisan news exposure not shifting opinions but eroding mainstream media trust). - Context heterogeneity: Beneficial effects (participation, knowledge) appear more commonly in emerging democracies (South America, Africa, South Asia), whereas detrimental outcomes (trust erosion, polarization, populism) are more pronounced in Europe and the United States. Effects often vary by subgroups (political interest, ideology/partisanship) and media system quality (public service journalism potentially mitigating polarizing effects). - Method-patterns and bias: Behavioral trace data are more prevalent in studies of detrimental outcomes; surveys (probability/quota) are more common in studies of beneficial outcomes. Few null effects are reported. Open data and pre-registration are relatively rare; a slight increase in reported detrimental and null effects over time is observed. No systematic bias attributable to specific authors or time trends is evident from the reported analyses.

The synthesis indicates that digital media constitute a double-edged sword for democracy. They can mobilize participation and, in many contexts, support knowledge acquisition and exposure to diverse viewpoints, but they also appear to erode institutional trust and contribute to polarization, populism, hate, and homophily. These trends broadly converge across correlational and causal evidence, despite methodological heterogeneity. Context matters: reductions in trust may undermine established democracies but can weaken authoritarian regimes; polarizing or populist effects vary with media systems and preexisting sentiments. The apparent tension between diversified news diets and homophilic network structures suggests that exposure and network dynamics affect different facets of the information ecosystem. The findings underscore the need for more credible causal identification, better linkage of trace and survey data, and greater transparency and data access from platforms to understand mechanisms and design interventions. The results directly address the research question by mapping where digital media’s effects align with or threaten democratic functioning and by delineating how these effects differ across political contexts.

This review integrates correlational and causal evidence across 496 studies to characterize how digital media relate to key democratic outcomes. It identifies robust beneficial links to participation (and suggestive benefits for knowledge and diverse exposure), alongside consistent detrimental links to institutional trust, polarization, populism, misinformation exposure, and network homophily—patterns especially pronounced in established democracies. The work highlights the centrality of context and the dual-use nature of digital platforms. It calls for methodological advances (more causal designs; multi-measure, linked data approaches), improved access to platform data, and potentially regulatory measures to enable independent research. Future research should resolve contradictions between network homophily and exposure diversity, deepen understanding of trust dynamics, and evaluate platform design and governance changes that curb corrosive effects while preserving emancipatory potential.

- Broad classification: Substantial methodological and measurement heterogeneity required grouping diverse constructs (for example, polarization variants), limiting granularity and preventing meta-analytic effect size estimation. - Uneven coverage: The search strategy yielded unequal numbers of studies across outcomes (for example, more on participation than expression or trust); strict inclusion criteria excluded some studies covered in other reviews. - Baseline ambiguity: Absence of clear offline benchmarks for constructs like exposure diversity, homophily, misinformation, and hate complicates interpretation of digital-era levels and trends. - Contextual skew: Overrepresentation of right-wing populist content online may bias observed links to populism; generalizing from US-centric studies is risky. - Data access and selection: Behavioral data depend on platform-provided samples, introducing potential selection biases; representativeness of active users is uncertain. - Transparency: Few studies used open data or pre-registration; potential file-drawer issues given relatively few null effects, though no clear systematic bias was detected in the authors’ assessments.