Attention Segregation: The Anatomy, Reach, and Origins of Partisan Division in Private Online Life

Dissertation

This dissertation asks one question of American political life: what organizes the segregation of citizens’ private information worlds — information or animus — how far beyond politics does it reach, and does it flow from elites or from citizens themselves?

Three decades of scholarship have produced divergent answers about how divided those worlds are. I argue that the divergence is largely diagnostic rather than contradictory: the major evidentiary traditions measure different objects — public expression rather than private consumption, the outlet rather than the article, two camps rather than the full structure of partisan identity — and each answers faithfully for the object it measures. What the theoretical debate actually concerns, however, is none of these objects. It is attention segregation: the uneven allocation of private attention across groups, a quantity I define, measure, and trace across three faces (ideological, affective, and cultural) using multigroup information-theoretic indices on the National Internet Observatory, a longitudinal panel that links what Americans actually read, and for how long, to who they are.

Paper 1 establishes the anatomy: whether the echo chamber exists in private attention, whether it is a bubble (omission) or a chamber (animus), and how consumers cluster behaviorally. Paper 2 establishes the reach: whether cultural segregation extends from the affective face, or from demography, geography, or party. Paper 3 establishes the origins: whether elite and media supply lead private attention or follow it, pairing the panel with MediaCloud and GDELT.

Throughout, the aim is to complement the existing traditions with the measurement they jointly presuppose — and to test whether, so measured, the field’s longest-standing disputes become decidable.

Committee
Nick Beauchamp (advisor), David Lazer, Risa Kitagawa
Concept
Attention segregation — ideological, affective, and cultural
Methods
Multigroup information-theoretic indices, computational text analysis, network analysis
Data
National Internet Observatory panel · MediaCloud · GDELT
Status
In progress · expected Spring 2027