More than news! Mapping the deliberative potential of a political online ecosystem with digital trace data

The study investigates to what extent the internet facilitates public deliberation by examining the deliberative potential of websites within Germany’s political online ecosystem. It argues that focusing solely on visible communicative acts (e.g., social media comments) overlooks the largely invisible behavior of passive audiences who consume political information. The paper adopts a systemic perspective that treats deliberation as an emergent property of an ecosystem of arenas with different democratic functions. It emphasizes the need to assess infrastructural affordances (information, communication, participation) and usage patterns (connectivity, inclusivity, heterogeneity) as upstream conditions for deliberation. The study addresses three research questions: (1) Which websites hold potential for online public discourse, including political information consumption and discussion? (2) How is the political online ecosystem structured along infrastructural and usage characteristics? (3) How do user demographics interact with different classes of websites? The purpose is to provide a holistic, data-driven mapping of the online deliberative system, moving beyond the traditional news vs. social media dichotomy, and to inform debates on echo chambers, selective exposure, and the role of public broadcasting in democratic discourse.

The paper reviews debates on online deliberation and polarization, noting mixed evidence on the quality of online discussions and the prevalence of echo chambers. Platform-specific studies often find homophily, while cross-platform work suggests diverse exposure. It highlights systemic theories of deliberative democracy that posit different arenas fulfill different functions (information, communication, participation). Prior research examined specific domains (e.g., news sites, municipal sites) focusing on communicative features and deliberative quality; however, the infrastructural foundations enabling deliberation are underexplored. The study conceptualizes deliberative potential across six dimensions: information (political and administrative content, curation), communication (expression, reciprocity), participation (contact, petitions/polls, organization), connectivity (flows between sites), inclusivity (demographic diversity), and heterogeneity (diversity of political orientations and party preferences). It situates its approach within scholarship on incidental exposure, inter-media agenda setting, and representation (inclusivity and heterogeneity), arguing for a holistic mapping of the online public sphere using digital trace and survey-linked data.

Data: The study uses German YouGov Pulse panel data with passive web tracking linked to surveys over six months (July–December 2017), spanning the federal election. About 1,500 were initially surveyed; 1,282 individuals provided desktop browser histories totaling over 56 million visits to ~200,000 domains. The sample mirrors the German online population on gender, age, and partly education. Tracking was opt-in with pause/opt-out options. Selection of politically relevant sites: A three-step, topic-driven pipeline identified political website visits. (1) A manual dictionary of 2017 German political topics (names, parties, institutions, issues; seeded from newspapers, Wikipedia, and the Federal Agency for Civic Education) was constructed. (2) Automated matching of keywords to full URL strings yielded 516,158 political clicks across 208 domains from 1,198 users. (3) Manual cross-validation refined the dictionary and removed mismatches, resulting in 493,714 political clicks to 69 domains by 1,190 users. Measures of deliberative potential: Six dimensions were operationalized. Supply-side (manual coding at domain level): information (political content, admin/local info, curation; each binary), communication (expression; reciprocity via reply functions; binary), participation (contact to politicians, petitions/polls, organization; binary). Demand-side (computed from trace + survey): connectivity (network in-/out-degrees based on temporally subsequent visits to different domains sharing the same topic by the same user; no self-loops), inclusivity (Shannon–Wiener diversity indices for age, gender, education among visitors), and heterogeneity (Shannon–Wiener diversity indices for political left–right orientation and 2017 first-vote party preferences among visitors). Clustering: After labeling all 69 sites on six dimensions, a latent class analysis (LCA) identified website profiles based on conditional response probabilities. Model selection favored a three-class solution, with robustness checks (alternative inputs, separate clustering of supply vs. demand criteria). Class membership probabilities assigned each domain to a class. Engagement analyses: User-level engagement compared number of clicks vs. duration (time per click) across classes and examined demographic engagement patterns by age, gender, and education.

• Scope and prevalence: From >56 million visits, ~1% (493,714) were politically relevant visits to 69 domains by 1,190 of 1,282 users, indicating broad but sparse engagement with political content. About 52% of politically relevant sites were news outlets; only ~12% of sites featuring political discussions were social media platforms.
• Latent classes (three-class LCA):
  1. Mainstream hubs (34 sites; 49% mixing proportion): High connectivity (central in flows), high inclusivity and heterogeneity; include major newspapers (Zeit, Spiegel, Bild), social media (Facebook, Twitter), Wikipedia, search and tools (Google, Wahl-O-Mat), petitioning (Change). They moderately fulfill information/communication/participation but stand out for engagement-based criteria.
  2. Quality information providers (20 sites; 22%): Predominantly public service broadcasters and journalistic outlets (ARD, ZDF heute, regional broadcasters MDR/WDR/SWR/NDR; local/regional papers such as Südkurier, Kölner Stadt-Anzeiger; municipal information). Strong on curated political information; limited communication features and less diverse audiences than mainstream hubs.
  3. Niche forums (25 sites; 29%): Topic/community-specific forums (e.g., Reddit, Gamestar, Erwerbslosenforum, Esoterikforum, Mamikreisel). High on communication (expression and reciprocity) and some organization, lower on other criteria. Manual checks confirmed substantial political discussions within ostensibly non-political spaces.
• Engagement patterns: Mainstream hubs receive the most clicks. However, time per click is higher on quality information providers (~59s) and niche forums (~48s) than on mainstream hubs (~35s), suggesting more in-depth consumption/discussion outside mainstream hubs.
• Diversity metrics: Inclusivity and heterogeneity correlate with overall engagement; highly frequented sites tend to appear more inclusive and heterogeneous. Nevertheless, there was no evidence of central websites with low heterogeneity, aligning with limited support for pervasive echo chambers.
• Demographic engagement: Across age, gender, and education, no strong selection pattern across classes; minor intuitive tendencies (e.g., higher engagement with quality information providers among higher-education users).

The study’s systemic, infrastructure-plus-usage approach reveals that political deliberation potential is distributed across distinct arenas serving complementary functions. Mainstream hubs act as general-interest intermediaries central to topical flows, facilitating broad access and exposure that counters strong echo-chamber narratives. Quality information providers, notably public service broadcasters and regional/local outlets, maintain the backbone of curated political information, though with less diverse audiences and fewer interactive affordances. Niche forums—often overlooked—host in-depth political discussions within tighter communities, enabling reciprocity and potential organization and highlighting incidental exposure in non-political spaces. These findings answer the research questions by identifying which websites matter, mapping their structural profiles, and demonstrating relatively weak demographic sorting across classes. The significance lies in reframing debates away from a sole focus on toxic visible interactions to include the silent majority’s consumption patterns and the infrastructural preconditions for deliberation. It also underscores that inclusivity and heterogeneity metrics are influenced by engagement volume, warranting careful interpretation. Overall, the results suggest a deliberative system wherein different venues contribute distinct democratic functions rather than a single site fulfilling all roles.

The internet provides multiple arenas with potential for political information, communication, and limited participation. Although political content constitutes a small fraction of overall online activity, most users engaged with it at least once around the 2017 election, indicating that political exposure is widespread despite low volume. The ecosystem organizes into three latent classes: central mainstream hubs facilitating exposure and flows; quality information providers anchoring curated, often public-service, content; and niche forums enabling reciprocal discussion within tighter communities. Users spend relatively more time on quality sites and niche forums than on mainstream hubs when engaging politically. The mapping challenges strong echo-chamber claims and emphasizes the enduring role of public broadcasting in Germany’s deliberative infrastructure. Future research should quantify user-side selection mechanisms (e.g., political attitudes, efficacy, knowledge) and extend coverage to mobile usage and longitudinal changes to track how these arenas evolve and interact in supporting democratic deliberation.

• Measurement dependence on engagement: Inclusivity and heterogeneity (Shannon–Wiener indices) correlate with site size/traffic, so highly frequented sites tend to rate as more inclusive/heterogeneous, complicating causal interpretation.
• Political content identification via URL strings: Not full HTML scraping; may miss or misclassify content, despite manual cross-validation. Keyword dictionary construction favors precise political terms; broader issue terms (e.g., family, housing) risk ambiguity.
• Device coverage: Desktop-only tracking excludes mobile usage, potentially omitting substantial political engagement and creating non-random blind spots.
• Temporal asymmetry: Behavioral data (2017) were coupled with website infrastructure coding conducted in 2021; some sites may have changed structure or features.
• Conservative selection: Emphasis on minimizing false positives likely underestimates political engagement prevalence.
• Aggregation and centralization: Central platforms may still contain siloed sub-communities not captured by aggregate measures of heterogeneity/inclusivity.