Effects of dehumanization and disgust-eliciting language on attitudes toward immigration: a sentiment analysis of Twitter data

Abstract

Attitudes towards immigration may be influenced by dehumanization and disgust: The more people dehumanize immigrants and the more disgusted they feel, the more negative their attitudes toward immigrants tend to be. Despite the fact that the majority of U.S. adults are on social media, however, little is known about how exposure to social media that links dehumanization, disgust and immigration influences users’ attitudes on this issue over time. We used Twitter data, machine learning and sentiment analysis to explore this question. Results showed that dehumanizing and/or disgust-eliciting language appeared in 66% of sampled immigration-relevant tweets. Surprisingly, exposure to both types of language in such tweets related to increases in positive sentiment about immigration over time. There was evidence of Granger-causality only for dehumanizing language, however, and only when controlling for communicators’ political affiliation. These findings suggest social media exposure may influence public attitudes toward immigration in unexpected ways.

Keywords:

• Attitudes
• dehumanization
• disgust
• immigration
• sentiment analysis
• social media
• Twitter

Acknowledgements

We thank Pablo Barberá from Meta and the University of Southern California for generously sharing his corpus of Twitter users for use in this research. We also are grateful for the suggestions and insight into how to improve this paper provided by Alissa Worden and Frankie Bailey from the University at Albany as well as Anamika Twyman-Ghoshal from Brunel University London. We also wish to recognize the significant research contributions made to this project by Ashley Tatis, Camilla Marzella, Isabella De La Vega, Ashley Coakley, Kayla Ashchmutat, Victoria Giorgio, Healy Turesky, Carlie Cegielski, Kristale Abdulla, Manshuen Yu, Brianna McKernan and Jayonna Treacy in the PULSE Lab at the University at Albany, all of whom assisted in operationalizing variables. We also acknowledge the funding provided for this project by the University at Albany’s Benevolent Association Research and Creative Activity Grant, without which the analyses for this paper would not have been possible.

Ethical standards

Declaration of conflicts of interest

Katherine S. Wahrer has declared no conflicts of interest.

Cynthia J. Najdowski has declared no conflicts of interest.

John V. Passarelli has declared no conflicts of interest.

Ethical approval

This article does not contain any studies with human participants or animals performed by any of the authors.

Disclosure statement

The authors have no competing or incentivizing interests at stake in the results of this research. Questions may be directed to the corresponding author, Cynthia J. Najdowski (at [email protected]).

Notes

1 Since this research was conducted, Twitter was acquired by Elon Musk and rebranded as ‘X’. We refer to the platform as it was known at the time of our study.

2 Barberá et al. (2015) collected data from 3.8 million active Twitter users in the United States and estimated their ideological preferences and social-network structures. The researchers examined whether users selectively exposed themselves to ideologically similar others and whether ideological preferences predicted users’ social media behavior. Their results showed that users’ networks and behavior varied across political and nonpolitical issues, but across topics conservative users were more likely than liberal users to select into echo chambers.

3 To ensure political affiliation could be included as a control variable in analyses, we checked the assumption of homoscedasticity by verifying that communicators’ political affiliation did not interact with the level of dehumanizing or disgust-eliciting language they exhibited in their immigration-relevant tweets to influence the sentiment of consumers’ tweets. Linear regression analyses indicated that neither of the interaction effects reached significance, β = 0.00, t(116) = 0.02, p = .99, and β = 0.26, t(116) = 1.40, p = .16, respectively.

Additional information

Funding

Funding for this project was provided by the University at Albany’s Benevolent Association Research and Creative Activity Grant. Any opinions, findings, conclusions, and recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the funding organization.