My research covers topics related to the media, algorithms, and their relation to political and market institutions. Methodologically, I specialize in developing new approaches to quantify politically and economically relevant concepts from unstructured data, such as text, images, and videos, and implementing them in open-source software.
I am an Associate Professor at UniDistance Suisse and an Affiliated Researcher at the Paris School of Economics. Before, I was an Assistant Professor at the Paris School of Economics (2024-2026) and a Post-Doctoral researcher at ETH Zurich. I completed my Ph.D. in Economics at the University of St.Gallen (summa cum laude, 2023), supervised by Roland Hodler.
News
- Nov 2026 A new version of Science under Threat? A Natural Experiment in Economics (with Dominic Rohner and Oliver Vanden Eynde) is out.
- Nov 2026 I am looking forward to joining the CEPR Research Policy Network on AI and Democracy in December.
Selected Publications
-
The Political Effects of X’s Feed Algorithm
with Germain Gauthier, Roland Hodler, and Ekaterina Zhuravskaya
Nature (2026)
Ranked #1 on Altmetric for several months, covered by 100+ news outlets worldwide. Explained in a video produced by the Paris School of Economics. Funded by SNSF Grant No. 100018_215554.We examine the impact of Twitter’s/X’s feed algorithm on political attitudes. We randomly assigned U.S.-based users to an algorithmic or chronological feed in an experiment. Users who switch from the chronological to the algorithmic feed shift their political opinions toward more conservative positions. By analyzing users’ feeds, we confirm that the algorithm promotes more conservative content. We do not find symmetrical effects among users who switched from the algorithmic feed to the chronological feed.
-
Media Slant is Contagious
with Clémentine Abed Meraim, Sergio Galletta, and Elliott Ash
The Economic Journal (forthcoming)We examine the diffusion of media slant. We document the influence of Fox News Channel (FNC) on the partisan slant of local newspapers in the U.S. from 1995-2008. We measure the political slant of newspapers by scaling their articles to Republicans’ and Democrats’ speeches in Congress. Using channel positioning as an instrument for viewership, we find that higher FNC viewership shifts local newspaper content to the right.
-
Aligning Large Language Models with Diverse Political Viewpoints
with Dominik Stammbach, Eunjung Cho, Caglar Gulcehre, and Elliott Ash
Proceedings of the Conference on Empirical Methods in Natural Language Processing (2024)Large language models such as ChatGPT exhibit political biases. If users query them about political information, they often take a normative stance. To overcome this, we align LLMs with diverse political viewpoints from 100,000 comments written by candidates running for national parliament in Switzerland. Models aligned with these data can generate more accurate political viewpoints from Swiss parties, compared to commercial models.
-
Text Semantics Capture Political and Economic Narratives
with Elliott Ash and Germain Gauthier
Political Analysis (2024)
Python package relatioWe develop a new text-as-data method. Social scientists’ interest in how narratives influence politics or the economy has surged. Yet, it remains challenging to identify narratives in large text corpora. Our unsupervised method achieves this goal by identifying entities and mapping the relations between them.
-
Ministers Engage in Favoritism Too
with Noémie Zurlinden
Journal of Public Economics (2022)We provide large-scale evidence for favoritism by African cabinet members. Using disaggregated health surveys and novel data on cabinet members’ birthplaces (from newspapers and other online text sources), we compare siblings to get causal estimates. Neonates’ and infants’ mortality is lower when the current health minister was born in their province.
All publications: Google Scholar profile
Working Papers
-
Science under Threat? A Natural Experiment in Economics
with Dominic Rohner and Oliver Vanden EyndeWe exploit a natural experiment, the U.S. government’s “blacklist” of undesirable words released in early 2025, to study whether academics react to political pressure. We find a sharp reduction in the use of banned words among economists at universities relying heavily on NSF funding, particularly for content related to gender, race, and environment.
-
Who Owns the Online Media?
with Ulrich MatterWho owns a newspaper is relevant from a political viewpoint, since newspaper ownership comes with benefits beyond the financial returns, such as influence on democratic decision-making. We map detailed ownership networks behind thousands of online news websites in the U.S., Canada, and Europe. We show that ownership matters for content and that outlets tend to have one ultimate owner or, otherwise, highly diversified ownership.
-
Buried in the Back-Pages: Subtle Censorship through News Placement
single-authoredWhen information is abundant, attention is the binding constraint. Censors can then limit what most people read by making some content less accessible. We study one such tool: within-outlet news placement. Using one million articles from 53 Chinese online news outlets (2020-2022), we find that front-page stories are systematically less likely to feature potentially sensitive content than back-page stories, and we provide evidence that this reflects censorship rather than catering to reader preferences.
Work in Progress
-
Towards a Cheap Personalization of Political Messaging with Large Language Models?
draft on request
with Tiffany Kreutschy
An experiment with U.S. respondents on whether personalizing LLM-generated political posts increases persuasion, and how LLM biases affect it.
Funded by SNSF Spark Grant No. 221092 -
Deep Latent Variable Models for Unstructured Data
draft on request
with Elliott Ash and Germain Gauthier
Topic models that use language representations and document metadata, for exploration, causal inference, and prediction on large multimodal corpora. Python package deeplatent. -
The Neural Ideal Point Model
with Germain Gauthier and Hugo Subtil
A deep learning framework to estimate ideal points from texts, images, audio, and video, for example political ad videos on Facebook and Instagram in the 2024 US elections. -
AI Chatbots as a Source of Political Information: Evidence from Benin
with Joel Eke, Eugenia Frezza, Arnault Gombert, Sergei Guriev, and Ekaterina Zhuravskaya
A randomized controlled trial with university students in Benin around the April 2026 presidential election. Fieldwork concluded in July 2026. -
The Politics of Netflix: Political and Social Dimensions of Streaming Content
with Roland Hodler, David Kündig, and Julius Reinhold
Political beliefs and gender norms in streaming content, combining web-mined data on available titles with computer vision and NLP annotations.
Funded by SNSF Grant No. 10008028 -
Gendered Social Media Consumption and its Consequences
with Germain Gauthier, Roland Hodler, and Katharina Kaeppel
Gender differences in social media consumption and their consequences for political beliefs and gender norms, studied in a field experiment with US-based Instagram users.