Classifying Candidate Evaluations: Policy, Traits, and Partisanship in the American National Election Studies

Voters’ evaluations of political candidates often center on a limited set of considerations. We identify three dimensions that recur in such evaluations: policy-based (e.g., agreement with the candidate’s positions), trait-based (e.g., perceptions of character or competence), and partisan (e.g., expressions of in-group loyalty). Using approximately 50,000 open-ended like/dislike responses from the American National Election Studies (ANES), 1980–2020, we employ large language models (LLMs) and in-context learning techniques to classify each response into one of these three categories. We then examine how the prevalence of each type of evaluation varies across election years, candidates, voter demographics, and political orientations. We expect systematic temporal patterns such as an increasing emphasis on personal and partisan attributes in recent decades, consistent with broader trends in polarization and candidate-centered politics, as well as clear individual-level differences related to education and partisanship. Finally, by linking these responses to individuals’ Most Important Problem (MIP) answers, we assess whether voters who emphasize policy considerations in candidate evaluations also articulate issue-based concerns when discussing national problems.