What the projects on this page share is a measurement problem: the things that matter most in politics are the hardest to count. Two lines of work follow from it. Political fragmentation, the largest, traces where attention, affect, and identity pull apart — in citizens’ private information worlds, across national publics, in expert networks, and in legislative speech. AI interpretability & evaluation examines the instruments themselves — whether LLM annotations measure the constructs we think they measure, and how ideology is organized inside model representations.
The map in the sidebar travels the same terrain: every node is a page.
Political fragmentation
- Attention Segregation Dissertation — what organizes the segregation of citizens’ private information worlds, and how far beyond politics it reaches
- Cross-National Affective Segregation — anti-Americanism in public discourse
- Network Segregation — who talks to whom among transnational epistemic elites, and how networks respond to shocks
- Ideological Segregation — ideological division in legislative speech on technology
AI interpretability & evaluation
- AI Interpretability & Evaluation — validity gaps in LLM annotation, and the political geometry of ideology in models
Selected & recent work
-
Stance Is Not a Construct: LLM Validity Gaps in Annotation Working Paper
-
How Do Networks Respond to Shocks? Working Paper
-
The Political Geometry of Ideology in LLMs Working Paper
-
Who Talks to Whom among Transnational Epistemic Elites Online? Working Paper
-
What Drives Anti-Americanism in Turkish Social Media? Working Paper
The full record, including talks and earlier publications, is in the CV (PDF).