with Arthur Müller, Jasmin Riedl, Johannes Steup, and Andreas Neumeier
Real-time large-scale data streams from Twitter/X provide valuable information for political scientists to analyze publicly expressed stances during election campaigns. However, the fast-moving nature of emerging events and topics on social media leads to concept shifts and consequently degrades model performance. Our research explored continuous retraining techniques for temporal model adaption on differently sized XLM-RoBERTa models. We applied these models to multi-target stance detection in a real-time setting, focusing on Twitter/X discussions of the campaign for the 2022 German state election in North Rhine-Westphalia. Our analysis considered stances towards political parties and their leading candidates. We initially trained models with different scales and tested them on future tweets. These models were then continuously retrained on new data over the final eight weeks of the election campaign to test the effects of different strategies on three model scales. In addition, we scrutinized the effect of pre-fine-tuning these models on a content-related stance dataset from the 2021 German federal election. We show that the effect of continuous retraining became less pronounced for larger models. On the contrary, pre-fine-tuning on the related dataset showed little positive effect on smaller models. Furthermore, we provide evidence that the language used in tweets becomes more emotional as election day approaches.
with Jasmin Riedl, Andreas Dafnos, and Andreas Neumeier
This study investigates the structural persistence of polarised digital communities on X (formerly Twitter) across German-language discourse on three major crises: the COVID-19 pandemic (2021), the Russo-Ukrainian War (2022), and the Gaza War (2023). Drawing on over four million posts, we triangulate the Echo Chamber Score (ECS) – a metric that captures homophilic user interactions and linguistic alignment in repost networks – with qualitative community analysis to trace how ideologically aligned publics evolve over time and across issues. Our findings highlight two key dynamics: within-topic temporal stability and cross-topic re-emergence of polarised communities. Echo chambers remain stable over multi-week periods within a single issue, and, more notably, re-emerge across conceptually distinct crises with similar internal cohesion, actor constellations, and framing logics. A small group of highly visible users – including (political) influencers, journalists, (alternative) media, and political actors – plays a key role in sustaining these formations by reactivating issue-specific discourse along persistent ideological lines. Our findings challenge event-specific explanations of polarisation and suggest that divisions in digital publics are not merely episodic or algorithmically driven but structurally reinforced by recurring actor networks.
with Philipp Darius, Andreas Neumeier, and Jasmin Riedl
Social media platforms play an increasingly important role in political campaigning, enabling parties to bypass traditional media and mobilize support directly. While prior research highlights the online prominence of far-right and radical populist actors, most studies are limited to single platforms or national contexts. This study presents the first cross-platform and cross-national analysis of digital campaign communication by 401 parties across all 27 EU member states during the 2024 European Parliament election. Using data from Facebook, Instagram, TikTok, X/Twitter, and YouTube, we examine party activity, audience engagement, and electoral outcomes. By linking digital trace data with expert surveys, we test whether populist radical right parties disproportionately succeed in raising support online. Our findings confirm strong platform-specific advantages of radical populist parties, particularly on TikTok, YouTube and Facebook. We also observe high engagement for far-left populist parties with similar emotional and anti-elitist communication strategies. The more Eurosceptic positions a party holds, or the more frequently experts describe them to use emotional appeals or anti-elitist communication, there more audience engagement they received across several platforms. Overall the findings emphasize a disproportionate online support for radical populist parties across the European Union.
with Joris Frese, Hilke Brockmann, Pedro Fierro Zamora, and Daniel Triana
Early research on echo chambers described them as (i) a problem affecting the masses and (ii) mainly driven by social media algorithms. Recently, there have been some pushbacks against these ideas. In this article, we test alternative ways in which echo chambers form and grow. We investigate whether they also pose a problem for political elites, and if their occurrence and intensity are affected by non-algorithmic factors such as strategic elite behavior and external shocks. To do this, we analyze an original dataset of all Tweets sent by members of the European Parliament and the European Commission between October 2021 and July 2022 (N = 520.871). We classify Tweets by topic using BERTopic and by sentiment using VADER, and combine these results with a novel measure of echo chambers based on embedding distances. Our analyses, based on pairwise comparisons, fixed-effects regressions, and difference-in-differences estimations, reveal that echo chambers are prevalent among political elites, particularly those at the extreme ends of the political spectrum. We also find that politicians strategically benefit from producing extreme content and that external shocks, such as the Russian invasion of Ukraine, further amplify this extremity. Our findings suggest that efforts to counter echo chamber dynamics need to go beyond focusing solely on platform algorithms, as many actors and factors outside of big tech can influence their formation and spread.
with Jasmin Riedl and Pauline Kielbassa
Social media have become indispensable arenas of political communication, yet they also function as resonance chambers for digital violence. This paper examines digital, sexualized violence against women politicians as a structural form of communicative inequality that undermines democratic participation. Drawing on the 2025 German federal election, we analyze 31 million German-language posts from X/Twitter collected by the SPARTA project. Using large language models for stance and toxicity detection, we find that 44.2% of posts mentioning women candidates contained toxic speech compared to 37.5% for men, with the disparity widening once reposts are considered. Violent posts about women are not only more frequent but also disproportionately amplified, embedding gendered hostility into the communicative infrastructure of electoral competition. Theoretically, we situate these practices within the framework of communication inequalities, arguing that sexualized digital violence operates not merely as individual hostility but as a disciplinary practice rooted in authoritarian-nativist ideologies that seek to silence women’s voices and restrict their political participation. By combining large-scale behavioral data with a conceptual discussion of digital violence as both communicative practice and ideological strategy, the paper contributes to debates on gender, democracy, and digital platforms. Our findings demonstrate that tolerating digital, sexualized violence against women politicians means tolerating the erosion of democratic equality itself, highlighting a broader challenge: communication infrastructures that should enable inclusion simultaneously reproduce and intensify inequalities.
with Jasmin Riedl and Constantin Wurthmann
This study investigates the Twitter discourse of Die Linke and BSW members of parliament from January 2017 onwards to analyze the evolution of political communication leading up to and following the split of BSW members from Die Linke. Using the Echo Chamber Score (ECS) method, we identify online communities formed around specific vocabulary and content, tracing how these communities reflect and amplify the political schism. By leveraging neural network-based stance detection, we determine whether individual members—and their associated communities—express positive, neutral, or negative stances toward Die Linke or BSW over time. Our findings map the formation and consolidation of BSW as distinct from Die Linke, providing insight into how political fragmentation manifests in online discourse. Furthermore, this method offers a framework for detecting early signals of defection and potential splits within political parties, with implications for understanding party cohesion and factionalism in the digital era.
with Jasmin Riedl and Marius Sältzer
This paper examines the incentives and dynamics of intra-party competition under Bavaria’s mixed-member electoral system, which combines single-member districts (SMD) and multi-member districts (MMD) with preferential voting. The system creates unique strategic tensions, particularly for candidates unlikely to win their SMD, as they may benefit from the loss of their own party peers in the MMD to secure preferential list positions. Conversely, candidates in safe districts have incentives to support peers to maximize party-wide gains through overhang and balancing mandates. Using Twitter data from the 2023 Bavarian state election, we analyze intra-party attacks on social media to uncover patterns of negative campaigning and party fragmentation. Our findings highlight how electoral systems shape intra-party dynamics and how unmediated platforms like social media facilitate intra-party conflict by circumventing party discipline. The study provides broader insights into the interaction between electoral incentives and the personalization of political campaigns in proportional systems.
Traditional after all? ‘Protective-conservative’ attitudes among populist right and Green supporters on (seemingly) non-partisan issues
with Julia Schulte-Cloos
Scholars concerned with understanding the growing salience of the cultural dimension of politics agree that the New Right and the New Left are divided by their positions related to questions such as immigration, European integration, minority rights, or the environment. While previous scholarship has convincingly demonstrated that party positions of populist right and Green parties are indeed characterized by a sharp cleavage on such issues, it has neglected the similarity of the two political opponents when it comes to the ‘protective-conservative’ attitudes of their core constituencies. Integrating socio-psychological theories of attitude formation with accounts on political conflict in Western Europe, we contend that populist right and Green partisans display similar, seemingly non-partisan preferences to preserve ways of living, protect local communities, and cultural ideas. To test our argument, we draw on a novel text corpus that contains discursive data from all Twitter followers of the German Greens (Bündnis 90/Die Grünen), the populist radical-right Alternative für Deutschland (AfD) and more than 500 political representatives of the two political parties, both at the national and the state level. Relying on a Naïve Bayesian classifier, we first demonstrate that the two groups of partisans show great similarities on issues related to ‘preserving-conservatism’: the discourse related to questions of lifestyle of health and sustainability, agriculture and spatial development, as well as community responses to global challenges, is similar enough that the algorithm cannot successfully predict the partisan group of the respective sender. We cross-validate our findings with linear Support Vector Machines. Based on a set of dictionaries, we also show, however, that both groups differ markedly in articulating a diverse set of responses when seeing their traditional visions of lifestyles challenged. The results of our study have important implications for our understanding of the role of new cultural politics in seemingly non-partisan domains.
Drews, W., Riedl, J., & Steup, J. (2025). Topical Negative Campaigning Under Spatial Pressure: Party-Level Strategies for Attacks Across Multiple Issues. German Politics, 1–27. https://doi.org/10.1080/09644008.2025.2561584
Darius, P., Drews, W., Neumeier, A., & Riedl, J. (2024). The EUDigiParty data set: The digital campaigning presence of 401 political parties during the European Parliament election 2024 including websites and social media handles on Facebook, Instagram, TikTok, X/Twitter, and YouTube. Harvard Dataverse V1. DOI: https://doi.org/10.7910/DVN/U6UWPN,
Riedl, J., Drews, W., & Richter, F. (2024). Avoiding the Elephant in the Room: Echo Chambers and the (De-)Politicization of COVID-19 during the 2021 German Election on Twitter. Front. Polit. Sci. Sec. Politics of Technology 6. DOI: https://doi.org/10.3389/fpos.2024.1509981
Christlmaier, R., Drews, W., Müller, A., Neumeier, A., Riedl, J., & Steup, J. (2023). Twitter/X Accounts of the Candidates in the 2023 German State Election of Bavaria. Data File Version 1.0.0. DOI: https://doi.org/10.7802/2608
Drews, W. (2022). E-Expression in a Comparative Perspective: Contextual Drivers and Constraints of Online Political Expression. Political Research Exchange 4(1). DOI: https://doi.org/10.1080/2474736X.2022.2083520
Drews, W., Müller, A., Neumeier, A., Riedl, J., & Steup, J. (2022). Twitter Accounts of the Candidates in the 2022 German State Election of North Rhine-Westphalia. Data File Version 1.0.0. DOI: https://doi.org/10.7802/2468
Müller, A., Riedl, J., & Drews, W. (2022). Real-Time Stance Detection and Issue Analysis of the 2021 German Federal Election Campaign on Twitter. In M. Janssen et al. (Eds), Electronic Government, EGOV 2022, Lecture Notes in Computer Science, vol. 13391 (pp. 125-146). Cham: Springer. DOI: https://doi.org/10.1007/978-3-031-15086-9_9
Ceron, A., Curini, L., & Drews, W. (2022). Short-Term Issue Emphasis on Twitter During the 2017 German Election: A Comparison of the Economic Left-Right and Socio-Cultural Dimensions. German Politics 31(3), 420-439. DOI: https://doi.org/10.1080/09644008.2020.1836161
Brockmann, H., Drews, W., & Torpey, J. (2021). A Class for Itself? On the World Views of the New Tech Elite. PLoS ONE 16(1): e0244071. DOI: https://doi.org/10.1371/journal.pone.0244071
Drews, W. (2020). Digital Politics Across Contexts, Social Media, Parties and Citizens: Technological Opportunities and Challenges in Modern Democracies. [Doctoral dissertation, European University Institute]. DOI: https://doi.org/10.2870/188696
Riedl, J., Drews, W., Jager, A., & Kurze, K. (2016). Krisen- und Risikokommunikation bei Hochwasser- und Unwetterereignissen. ZA6440 Datenfile Version 1.0.0. Köln: GESIS Datenarchiv.
Jager, A., Riedl, J., Kurze, K., & Drews, W. (2016). Krisen- und Risikokommunikation im Baulichen Bevölkerungsschutz. Mehr als ein rationaler Diskurs. 2. aktualisierte Auflage. Neubiberg: Universität der Bundeswehr München.
Drews, W., Jager, A., & Kurze, K. (2016). Das Hochwasser in Deutschland 2013. Eine qualitative Analyse der Krisen- und Risikokommunikation. Neubiberg: Universität der Bundeswehr München.
Drews, W., Jager, A., & Münch, U. (2015). Paradigmenwechsel im Hochwasserschutz: Die europäische Hochwasserrisikomanagement-Richtlinie und ihre Umsetzung in der BRD. In Europäisches Zentrum für Föderalismus-Forschung Tübingen (Ed.), Jahrbuch des Föderalismus 2015 (458-472). Baden-Baden: Nomos.
Drews, W., Jager, A., & Münch, U. (2014). Land (und Föderalismus) unter? Chancen und Grenzen des Hochwasserschutzes im föderalen System der Bundesrepublik Deutschland. In Europäisches Zentrum für Föderalismus-Forschung Tübingen (Ed.), Jahrbuch des Föderalismus 2014 (159-173). Baden-Baden: Nomos.
Drews, W. (2013). A Functional Perspective on Post-Communist Civil Society: Contentious Activities and Internet Activism in Latvia. Abgerufen über: http://cbs.ut.ee/index.php/current/defended