It’s a new soundtrack:

Candidates, Campaigns, and Playlists

Jean Clipperton, University of Chicago

Agenda

• Role of campaigns and elected officials
• Data on campaign music
• Content of music: emotions and themes
• Framework for analysis: PCA
• Playlist stability over time
• Conclusions

Background

Political campaigns provide an opportunity for candidates to provide information about themselves and the representatives they will become.

Campaigns provide

• Opportunities to signal their qualities (delegate model (Mansbridge 2003, 2011))
• Information about their ‘brand’ (gyroscopic model (Mansbridge 2003, 2011), ‘brandidate’ (Kaneva 2016, Harrison 2022))
• Context for voters (both informed and uninformed) (Bonilla 2022, Roumanias 2005)

Theory: models of representatives

• Delegate models center on the notion of a representative as an agent on behalf of their constituents
  • Collective-focused
  • ‘We’ over ‘I’
  • Principled: what is ‘right’ from an ethical or moral perspective

• Brand model (Mansbridge’s gyroscopic model) is centered upon the representative as the commodity
  • Self-focused
  • ‘I’ over ‘we’
  • Determines what is ‘right’ from own perspective
  • Typically resolute

Context: Campaign music

Campaign songs used for hundreds of years to connect with voters and send messages about candidates (Schoening and Kasper 2011, Peterson 2018). Technology makes it easy to curate and share playlists.

Source: BBC

Focus and expectations

• Themes within the playlist music related to candidate archetypes, for example, themes of loyalty and specific personality characteristics.

• Candidates’ playlists incorporate different emotions and characteristics, with a focus on trust

Theory

• Different models of representation borne out in types of songs chosen:
  • Delegate models center on the notion of a representative as an agent on behalf of their constituents
    • ‘We’ over ‘I’
    • Morality-focused

• Brand model (Mansbridge’s gyroscopic model) is centered upon the representative as the commodity {.fragment}
  • ‘I’ over ‘we’
  • ‘Signature’ elements, such as songs

Data

Data based on Trax on the Trail Database, focusing upon candidate-focused events (campaign launch, rally, stop, playlist, speech, town hall). Lyrics sourced from a combination of and .

• Approximately 2000 songs for candidates with more than five songs on a playlist and more than five campaign-related events.
• Approximiately 900 distinct songs represented
• 19 unique campaigns in 2016, 2020, and 2024

Methods

Use of three primary methods:

• Lexicon-based text analysis (NRC Emotion Lexicon and Linguistic Inquiry Word Count (LIWC)))
• Topic Models (LDA (Latent Dirichlet allocation))
• Principal Component Analysis (PCA) of relevant terms

NRC Emotion Lexicon

• The NRC emotion lexicon is a dictionary-based approach to text analysis.

• Lists of words and their associated emotion

  • Example: ‘cheerful’ is coded as a joyful word while ‘idiotic’ is an angry word

NRC use

Songs have tallies for each emotion and these are aggregated by candidate in the analysis

Sentiment and emotions: NRC

Using the NRC emotion lexicon, I incorporate eight emotions:

• Disgust
• Anger
• Fear
• Sadness
• Trust
• Anticipation
• Surprise
• Joy

Sentiment and emotions: Overall

Sentiment and emotions: by candidate

Overall, candidates from both parties incorporate songs with ‘trust’ as the most common emotion.

Language and themes: LIWC

Linguistic Inquiry Word Count (LIWC) incorporates measures of linguistic use, which I use for its indicators of the following:

• Linguistic terms:
  • pronoun use, such as I, we, she/he, and they
• Analytic language: use of logical language, including causal language
• Social terminology:
  • including prosocial language, moral language, and conflict
  • gendered terms/social references: male vs female

Themes: social behavior

All candidates increased the use of songs with prosocial language over the elections with Democratic candidates employing more moral language in their song choices and Trump nearly doubling the ratio of prosocial language in playlists.

Topics: approach

• Incorporated a topic model on the lyrics from all campaign songs.

• Set topic numbers from 1 to 20 to find the best number of topics, incorporating two measures from CaoJuan2009 and Deveaud2014.

• Latent Dirichlet Allocation model: model maps documents to fixed number of topics (five for this setup)

Topics

Rank	Energizing music	Golden era	Feel good celebration	Love and happiness	Empowerment anthems
1	shake	time	babi	love	girl
2	high	life	love	day	back
3	gotta	gonna	wanna	peopl	happi
4	babi	heart	night	world	make
5	tip	back	stop	burn	run
6	jump	golden	feel	man	work
7	power	make	danc	give	thing
8	low	livin	good	make	woo
9	bout	long	tonight	start	gon
10	song	born	music	dream	money
11	play	home	alright	train	bitch
12	home	love	walk	woman	hand
13	worri	stand	gonna	higher	rock
14	gonna	freedom	bodi	live	hold
15	thing	light	talk	come	feel

Topics

Trump has leaned toward more ‘feel good’-type music while Harris had empowerment anthems.

Principal Component Analysis (PCA): explanation

• Principal Component Analysis offers a way to preserve the variation in the underlying data while reducing the number of variables present.

• Variables are reduced through identifying relevant dimensions and collecting variables together.

• Researchers name the dimensions thematically based on the variables present within the dimensional space

PCA: Candidate ‘spaces’

There are three distinct spaces:

1. Linguistically focused (particular language and pronouns present)

2. Emotional

3. Analytical or logical language

Candidate ‘spaces’

Distinct landscape: Trump and Sanders and Years of Harris

Overlapping songs

Conclusions and future research

• Campaign music relevant to candidates and key differences across parties and campaigns:
  • Pro-social language: all candidates increased, but particularly Trump
  • Moral language: reduction by Trump and increase by Democratic candidates
  • Generally more negative emotion by Democratic candidates but all candidates incorporated trust