She Doesn’t Look Presidential: Analyzing Gendered Fashion in Politics

• In Event: Quantitative Study of Race & Gender

Abstract

Past studies show that news media delegitimizes women candidates in multiple ways, including more frequently referring women candidates by their first names (Uscinski and Goren 2011) and overly focusing on their appearance (Cummings and Terrion 2021). Meanwhile, qualitative evidence describes women politicians in the US as having a “husbands, hair, hemline problem” (Duerst-Lahti 2005), though to our knowledge there has been no large-scale quantitative study documenting whether either news media or voters overly focus on women politicians’ personal lives or appearance. This project explores how commentary on women politician’s clothing and appearance dominates online discussions of these candidates. We focus on the 2020 Democratic presidential primary due to the historically large number of women running in that race. Our dataset includes over 25,000 tweets mentioning these candidates and we use novel AI language classification methods to categorize tweets. We have developed a model using Hugging Face Chat API to classify text as either containing references to candidates clothing or not. These methods are similar to handcoding by research assistants but have been shown to be more accurate in recent studies (Giliardi, Alizadeh, and Kubli 2023) and can be performed more quickly on a larger scale. We recently received funding to use a similar model to classify tweets using the “openai” R package, which connects to ChatGPT’s API to conduct this analysis on all 25,000 tweets in our dataset. Findings demonstrate how artificial intelligence language models can aid researchers in text classification on a large scale, as well as gender differences in public discourse surrounding high-profile political candidates. Eventually we plan to apply this method to classifying text of news media transcripts to determine how prominent media personalities may discuss candidates’ appearance differently based on gender.

Authors

• Samantha Register, University of Colorado Boulder

• Katie Glenn, University of Colorado Boulder