The Conditional Echo Chamber: A Computational Analysis of Elite Partisan Rhetoric and Domain Specific Selective Exposure Effects

Studies on selective exposure to partisan media has produced inconsistent findings on whether ideologically congruent media consumption polarizes political attitudes. This study argues that the inconsistency reflects a domain specific phenomenon and inattention to the elite information environment. Combining longitudinal panel data from Pew Research Cen- ter's American Trends Panel with a corpus of over one million congressional tweets from the 117th Congress, this study examines whether selective exposure on Twitter predicts atti- tude change across four policy domains, namely racism, economic inequality, immigration, and climate change. Partisan language divergence within each domain is measured using regularized logistic regression classifiers trained on feature representations of party-labeled tweets, and individual level attitude change is modeled with OLS regressions. Selective exposure predicts polarization only on economic inequality, the one domain where elite partisan rhetoric was both substantively divergent and actively increasing during the observation window. Each of the three null findings corresponds to a distinct feature of the elite environment, namely a ceiling on immigration, mild convergence on climate, and a heavily skewed Democratic party share on racism. These findings reframe selective exposure as a moderator that conditions elite cue effects rather than as a solely independent driver of ideological polarization.