American democracy under strain—and the human side of changing information environments
My research has two equal points of entry: changing information environments and American political behavior. I study both the systems that organize political information and the citizens who encounter, interpret, and act within them. At the system level, I study how political institutions, media systems, and digital platforms shape the production and circulation of political information, redistribute authority, and condition citizens’ ability to orient themselves. At the individual level, I study how Americans’ identities, grievances, and psychological dispositions shape what they trust, how they interpret political threats, and their willingness to compromise on democratic principles and embrace illiberal politics. AI enters this agenda in two ways: as a political technology that changes what citizens encounter and what institutions can hear, and as a research tool whose outputs require careful validation.
Methods: Surveys · measurement · audience analysis · computational content analysis
Current funded project
INFLUPOL: Influencers and Politics in the Digital Age
INFLUPOL examines how political organizations work with social media influencers, how influencers cultivate political authority, and how citizens respond across seven countries. As a PI collaborator, I contribute to its US research. The project combines social media analysis, interviews, and surveys and is jointly funded by the French National Research Agency and the Austrian Science Fund.
Selected publications
Trust but Verify: AI Models for Coding Political Content in Social Media
When can AI models be trusted to classify American political speech? Using 4,500 posts from members of Congress and political influencers, we evaluate BART and LLaMA against trained human coders across 18 concepts. We propose a three-step validation procedure and find that performance varies sharply by concept and architecture: BART is more precise, LLaMA has higher recall, and both require concept-specific validation combining reliability and error analysis.
Request paper →Epistemic Vulnerability: Theory and Measurement at the System Level
Why, despite its sophisticated news system, are so many Americans distrustful of political information and unsure what to believe? This article develops the concept of epistemic vulnerability for understanding how political and media institutions make citizens more or less able to trust and evaluate political information. Using a new index across 20 Western democracies, the study identifies the United States as one of the more vulnerable cases and identifies structural features of political and media systems that contribute to greater vulnerability.
Read →Media Use, Feelings of Being Devalued, and Democratically Corrosive Sentiment in the US
Does media use explain the current rise in populist attitudes, conspiratorial beliefs, and distrust towards US elections? Different platforms assemble different publics. Using a 2022 survey of 2,000 US adults, we distinguish mainstream and right-wing news from strong-tie, weak-tie, and Alt-Tech platforms. We also introduce a measure of feeling socially devalued that is more strongly associated with these attitudes than authoritarianism, social dominance, or ideology.
Read →French Fox News? Audience-Level Metrics for the Comparative Study of News Audience Hyperpartisanship
What makes Fox News distinctive—and can its audience profile identify analogous outlets elsewhere? I develop audience-level measures of ideological lean and voting behavior, then use France’s CNEWS as a comparative test. Fox News and CNEWS attract equivalently right-leaning audiences and disproportionate support for far-right candidates, showing how a prominent feature of the US media system can be studied across national contexts.
Read →Beyond Large Language Models: Rediscovering the Role of Classical Statistics in Modern Data Science
How much can an AI classification change with the model, topic, or wording of a prompt? This collaboration combines experimental design and classical statistics to determine when those differences are meaningful, providing an early methodological foundation for my later work on validating AI-assisted analysis of political content.
Read →Working papers & current projects
The Three Logics of Illiberalism in the American Public
Why do some Americans accept presidential lawbreaking or endorse granting the president permanent emergency powers? Americans do not embrace illiberal power for a single reason. Some excuse it for partisan advantage, others seek authority and order, and still others are attracted to the transgression itself. Using a late-2024 survey of 1,059 eligible US voters, we distinguish three logics behind illiberal attitudes: opportunistic illiberalism, order-seeking illiberalism, and transgressive illiberalism. The transgressive logic, rooted in antisocial personality traits, emerges as the leading mechanism behind the endorsement of illiberal propositions and is especially powerful even among non-Trump voters.
Request paper →Political Social Media Influencers and the Parasocial Bridge to Democratically Corrosive Attitudes in the US
How do political influencers become consequential for American democracy? The democratic significance of political influencers lies not simply in how many people they reach, but in the relationships they create. An original survey of 1,055 eligible US voters fielded after the 2024 election distinguishes exposure from parasocial attachment. Exposure is associated with populist and conspiratorial attitudes; more troubling outcomes emerge among Americans who feel personally connected to an influencer. Attachment is also associated with feelings of being devalued and tolerance for political violence, with the latter relationship clearest among Trump voters and politically like-minded audiences.
Request paper →Feelings of Being Devalued: Negative Affect and Democratically Corrosive Attitudes in the US and Beyond
Research on American democratic discontent often begins with specific grievances—racial status threat, material decline, or cultural displacement. This paper theorizes Feelings of Being Devalued as a mid-level affective construct capturing perceived losses of personal and collective worth without prespecifying their social source. It validates two parsimonious measures, FBD-3 and FBD-5, in the United States, then demonstrates that FBD-3 travels across France, the United Kingdom, and Poland. Across these settings, FBD is strongly associated with populist and conspiratorial attitudes; in the United States, it is also associated with tolerance for political violence.
Request paper →Synthetic Campaign Communication and Election Confidence in the Age of Artificial Intelligence
Generative AI entered the 2024 US election against an already fragile background of partisan distrust in electoral legitimacy. Rather than asking only whether individual deepfakes deceive voters, this paper distinguishes two broader democratic questions: whether the perceived profusion of synthetic campaign material is associated with confidence in election integrity, and which citizens tolerate candidates’ use of it. Across two quota-based surveys fielded during and after the election, perceived exposure consistently tracks lower election confidence, while tolerance follows a separate logic: those with greater AI familiarity are most likely to reject the use of synthetic material by political candidates, while those who distrust the professional news media are more accepting.
Request paper →Tolerance for Political Violence Before and After the 2024 US Presidential Election
Support for political violence is neither purely dispositional nor simply partisan. This paper develops a three-stage account: deep psychological dispositions determine whether violence is morally available; political orientations supply possible motives; and electoral circumstances determine when violence appears defensible. Across three surveys of US voters spanning the 2024 election, presidential transition, and opening months of Trump’s second administration, antisocial traits, interpersonal victimhood, and social dominance remain stable correlates. The partisan pattern does not: an initial Trump–Harris gap narrows after Trump’s victory and disappears after his inauguration. Partisan asymmetries observed at one moment, the paper argues, should not be mistaken for fixed features of the American electorate.
Ask about this project →AI and Information Integrity
This project examines how AI-mediated political information shapes citizens’ ability to evaluate claims, assign credibility, and form democratic judgments. It is organized around a supply-side computational project and a demand-side multifactorial experimental survey design.
Ask about this project →Selected reports & public scholarship
Alongside my journal research, I contribute collaborative and public-facing work that places the United States in comparative perspective and translates research on information environments into usable diagnostics.
Resilience and Vulnerability to Misinformation and Disinformation in Political Information Environments: A Comparative Review of Ten Indexes
How should researchers compare countries’ resilience to misinformation when the available indexes measure different things? This synthesis compares ten leading indexes, clarifies their theoretical and methodological tradeoffs, and identifies the foundations for a more coherent diagnostic of political information environments.
Read report →The Media, Technology, and Resentment in the 2024 US Election
Written after the 2024 US election, this essay examines how political resentment, fragmented media authority, and emerging AI practices converged during the campaign—and what their interaction reveals about the democratic vulnerabilities of the American information environment.
Read essay →Unconventional Voices: Alternative Media Trends in Europe and the US
This 18-country study maps who uses alternative media and how non-mainstream outlets communicate on social platforms. It places the United States’ partisan alternative-media sphere in comparative context, distinguishing patterns that are distinctly American from those shared across democracies.
Read report →