Wisconsin political ads (2008)
ad_tone.RdA dataset of 935 Wisconsin congressional campaign advertisements from the
2008 election cycle. Ad texts and pairwise annotations are drawn from
Carlson & Montgomery (2017); five-point tone codes are from the Wisconsin
Advertising Project (WiscAds). The score column contains textscale
negativity scores derived from the pairwise comparisons in
cm_comparisons.
Format
A tibble with 935 rows and 10 columns:
- text
Full text of the advertisement.
- candidate
Coded candidate identifier (factor).
- state
Two-letter state abbreviation. Empty string for ads with no state on record.
- tone
Numeric tone code from the Wisconsin Advertising Project (1 = Promoting, 2 = Mostly promoting, 3 = Contrasting, 4 = Mostly attacking, 5 = Attacking).
NAfor ads without a tone code.- alphas
Continuous negativity score from the SentimentIt model in Carlson & Montgomery (2017).
- score
textscale negativity score derived from the pairwise comparisons in cm_comparisons.
- score_lower
Lower bound of the 95% confidence interval for
score.- score_upper
Upper bound of the 95% confidence interval for
score.- split
Whether the document was in the
"Train Set"or"Test Set"split used to validate the model.- tone_label
Factor version of
tonewith descriptive labels.
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