Skip to contents

Human-annotated pairwise comparisons from the Wisconsin congressional ads study of Carlson & Montgomery (2017). Each row records which of two ads was judged more negative by human annotators. Document indices reference row positions in ad_tone.

Usage

cm_comparisons

Format

A tibble with 9,528 rows and 4 columns:

doc_id_a

Integer row index of the first ad in ad_tone.

doc_id_b

Integer row index of the second ad in ad_tone.

winner

"A" if the first ad was judged more negative; "B" if the second ad was.

split

Whether the pair belongs to the "train" or "test" split used to validate the textscale model in ad_tone_validation.

Source

Carlson, T. N., & Montgomery, J. M. (2017). A pairwise comparison framework for fast, flexible, and reliable human coding of political texts. American Political Science Review, 111(4), 835–843. doi:10.1017/S0003055417000302