The preliminary title for my dissertation is “Seeing Polarization through a Fresh Set of AIs.” Consequently, my research interests revolve around different processes behind how people develop political views, how this impacts society at large, and how we can measure them in new ways.
I am a big believer in the new opportunities computational methods bear for social scientists. However, they are by no means silver bullets and validity of results remains key. Therefore, I try to complement and contrast results with “classic” quantitative social science approaches such as surveys if possible.
Quoting Against the Out-Group: How Negative Campaigning Might Fuel Vertical Affective Polarization
(with Florian Dittrich)
Abstract: Affective polarization (AP) describes the emotional divide between political groups, manifesting in hostility toward opposing parties (vertical affective polarization, vAP) and loyalty to one’s own. While previous studies link negative campaigning (NC) by political elites to increased AP, they often lack temporal and relational granularity, particularly in multiparty systems. This study examines the role of NC in shaping vAP during the German parliamentary elections of 2021 by analyzing over 1,500 candidates’ Twitter communications and pairing these data with biweekly survey measures of vAP from the German Politbarometer. Using Natural Language Processing techniques, we identify NC rhetoric in Tweets and assess its relationship with shifts in public sentiment. Our findings suggest that vAP is influenced by election phases, coalition dynamics, and ideological proximity. Moreover, candidates’ rhetoric toward opponents significantly shapes voters’ perceptions of rival parties, demonstrating how elite cues drive intergroup dynamics in a multiparty system. This research contributes to understanding how NC affects democratic norms and inter-group relations by offering a granular, real-time perspective on elite-voter interactions.
Do Liberals Drive Volvos Everywhere? Assessing Cultural Bundles in Sweden
(with Anastasia Menshikova, Elida Izani Binti Ibrahim, and Miriam Hurtado Bodell)
Abstract: This study investigates lifestyle polarization in Sweden’s multi-party political context, extending research on politicized lifestyle bundles beyond the U.S. case. We examine the extent to which clusters of lifestyle preferences emerge among politically active users on Swedish Twitter and analyze the factors explaining these clusters. Using a sample of 12,230 politically active Swedish Twitter users, we employ seeded Latent Dirichlet Allocation (LDA) models to measure political preferences and cultural consumption patterns in Twitter co-following data. Our findings reveal distinctive corresponding patterns between political views and non-political spheres. Partisan belonging significantly drives lifestyle polarization in areas such as media consumption, cultural institutions, religion, politics, and humor. This research contributes to understanding polarization dynamics in a multi-party system, explicating lifestyle polarization patterns on Swedish Twitter over a decade (2010-2020). By using behavioral digital traces, we complement previous survey-based work and uncover lifestyle polarization structures that account for relationships between individuals and their cultural preferences. Our study provides a contribution to the broader discussion on how political polarization may impact social cohesion in diverse political contexts.
It’s in the News! How Elite Polarization is Transported Through the Media
(with Väinö Yrjänäinen and Måns Magnusson)
Abstract: This study examines the relationship between elite polarization and opinion constraint among voters in France, addressing limitations in previous research on ideological polarization. We employ a novel dataset of 25 years of newspaper reporting and longitudinal survey data to investigate how political elites’ stances and citizens’ views on 26 political issues co-vary over time. Our methodology leverages Contextualized Dynamic Word Embeddings to measure associations between political actors and issues in media coverage, hence providing a more empirically grounded and temporally granular alternative to expert surveys. Our approach helps us to capture the media’s role in transmitting political cues to voters, a key mechanism in the formation of opinion constraint. We analyze the strength of associations between issues and political leanings among both elites and voters, and subsequently compare the predictive power of elite polarization in the media on voter constraint and vice versa. Overall, the analysis aims to determine whether elite polarization precedes and predicts voter constraint. Our findings offer limited support for this mechanism, with only few models showing that elite-issue associations in media predict voter-issue associations better than the reverse. These results suggest that voters follow elite cues only on certain issues, indicating that elite polarization may only partially drive structural ideological polarization.
Media Slant as Political Refraction. Measuring Political Media Slant and Polarization in the French Media Landscape
(with Rubing Shen, Arnault Chatelain, and Etienne Ollion)
Abstract: This study introduces a novel method for measuring media slant in newspaper content, addressing limitations of existing approaches. We develop a content-based measure grounded in Bourdieu’s field theory concept of refraction, which posits systematic differences between political and journalistic language due to field-specific norms. We use word embeddings to identify the political bias of micro-frames, which allows for scalable and fine-grained analysis of media slant. To demonstrate the efficacy of our approach, we examine evidence of increasing polarization among mainstream French national daily newspapers from 2000-2010. We analyze a comprehensive sample of articles from Le Monde, Libération, Le Figaro, Les Echos, and La Croix using our semi-supervised technique. Thereby, we contribute to the literature on media polarization and its potential impacts on democratic discourse. Our method enables precise estimation of media slant at the paragraph level, hence enabling nuanced comparisons between newspapers, articles, and journalists. By providing a valid and reliable content-based measure of media slant, this study offers a valuable tool for researchers and policymakers concerned with media polarization and its societal effects. Our findings have implications for understanding the role of traditional media in shaping public opinion.
Perceptions of Intergenerational Mobility in Germany, Sweden, and the UK: Insights from Machine-Learning Text Analysis
(with Alexi Gugushvili and Patrick Präg)
Abstract: Intergenerational social mobility research has traditionally focused on objective markers of socioeconomic position. In this study, we argue that the subjective aspects of intergenerational mobility deserve greater attention and empirically explore what individuals report they compare when they gauge their intergenerational mobility trajectories. Drawing on representative survey data from Germany, Sweden, and the United Kingdom, as well as machine-learning-driven text analyses of open-ended survey responses, we reveal that, in addition to conventional measures such as education, occupational status, and income, individuals consider a diverse array of factors, including family life, home ownership, and lifestyle choices. Our findings highlight the heterogeneity of these comparisons across different countries, genders, and generations. We identify significant variations in the dimensions of intergenerational comparisons, such as the prominence of education in Sweden and the focus on housing in the United Kingdom. Furthermore, gender differences reveal that females are more likely to emphasize education and family life, while males focus on income and occupational status. These insights provide a deeper understanding of the subjective dimensions of intergenerational mobility and contribute to ongoing debates in social stratification research and general social theory.