Abstract
Social media has had a transformative effect on politics and governance. While research into the consequences of social media use on politics and governance is mixed, elected officials and political candidates commonly use multiple social media channels to grow constituencies and draw attention to their agenda. A well-crafted social media strategy can further the perception of a candidate or elected official as credible, trustworthy, and able to effectively govern. However, research into social media has suggested that engagement is often driven by toxic and negative content, limiting the possibility of productive political discourse. Using a dataset of posts from 58 U.S. Senate candidates on Twitter, this study seeks to determine whether social media-based political communication during the 2022 election aligns with preexisting perceptions of toxicity. The results suggest that the toxicity of candidates’ Twitter-based political communication is mostly low, though temporal and group differences were found. This case study provides a basis for further investigation into candidates’ strategic use of social media.
Intoduction
Evidence suggests that many citizens use social media as a primary source of political information. This trend has not gone unnoticed by political communication scholars, as online political communication has long been an area of investigation (Rossini, 2020). The increasing consumption of political information via social media has led to fears that citizens only receive information consistent with their existing beliefs. Much of this information comes from media outlets and/or candidate surrogates rather than the candidates themselves. This situation, coupled with the rise of “bots” and the common presence of misinformation, calls into question whether political communication via social media is a positive force.
As a semi-direct line of communication between themselves and the electorate, candidates use social media to bypass traditional media outlets and communicate their message. Questions remain, however, about whether candidates’ digital communications follow the toxic patterns often associated with social media dialogue and whether those communications change over time. This study explores the Twitter feeds of 2022 U.S. Senate candidates to determine toxicity patterns associated with social media-based political communication. The brevity of social media messaging means that candidates must choose their words with an eye toward voter engagement; as prior research (e.g., Rafail et al., 2024; Rathje et al., 2021) has shown, social media messaging involving more negative, emotional, and moral content is more likely to be engaging. Candidates may therefore be compelled to craft messages that inflame rather than educate. This case study explores whether the content of social media-based political communication aligns with preexisting perceptions of toxicity.
Social media use in electoral politics
Social media has had a transformative effect on governance and politics. As a critical component of e-government, social media can be used as a tool to build trust and confidence in both government institutions and those who govern (Park et al., 2016; Warren et al., 2014). While research into the consequences of social media use on politics and governance is mixed, elected officials and candidates commonly use multiple social media channels to grow their constituencies and draw attention to themselves and their agenda. Prior research exploring the link between candidates’ social media use and electoral outcomes has shown it to be a complex, nuanced relationship (Afonso et al., 2025).
Social media offers politicians the ability to influence the political agenda independent of traditional media outlets. Traditional media, once the gatekeeper of the political agenda, has in many ways been supplanted by the hybrid media systems created by the rise of the Internet and social media (Chadwick, 2017; Jungherr et al., 2019). As a freely available tool for promotion, social media allows politicians to define the issues that matter to them and to keep them in the public consciousness. Prior research has shown the impact of social media on issue salience and political reach (Fazekas et al., 2021; Feezell, 2018).
Social media content allows politicians to shape their own image through unfiltered communication with followers. Acting as a “parasocial proxy for direction interaction” (Carr, 2020: 2), social media can promote greater engagement and greater perceptions of electability. A well-crafted social media strategy can further the perception of a candidate or elected official as credible, trustworthy, and able to effectively govern. Higher perceptions of information source credibility have been shown to promote the chances of electoral success (Housholder and LaMarre, 2014). Candidates’ attempts at using social media to construct a positive “brand” may be differentiated by gender and/or other characteristics. For example, an exploration of tweets from 2020 to 2022 U.S. congressional candidates found men and women presented themselves differently in terms of tone and content, and that party affiliation revealed intragender differences (Horan et al., 2024).
Social media and toxicity
Social media engagement has been shown to be largely driven by out-group animosity and/or the expression of moral outrage, providing politicians with a rationale for sharing more negative content. Research into content features which promote social media engagement have shown that content which stokes emotions and/or includes moral and emotional content are more likely to be shared within the in-group (Brady et al., 2017; Rathje et al., 2021). Recent research by Rafail et al. (2024) found that social media can create a feedback loop of sorts, where polarizing or extreme content triggers greater engagement, which in turn fosters a greater amount of this content. Morality is central to human social behavior, and social media can serve to amplify negative aspects of morality (e.g., harassment) as well as positive aspects (e.g., mobilizing collective action for social good). Expressing moral outrage can signal group affiliation (Van Bavel et al., 2024). This encourages users—including politicians and political organizations—to utilize such content to gather a large following and promote their agenda.
Besides moral outrage, negativity alone can be used to strengthen partisan identity. Exposure to excessively negative content about the in-group can promote polarization by fostering anger toward the opposing side (Rathje et al., 2021). In negating the out-group, in-group members enhance their own identity through differentiation; in the political sense, identity-driven communication serves to build hostility between people on opposing sides (Wojcieszak et al., 2022). This perception aligns with social psychology research suggesting that stronger identification with one group is often paired with more negative perceptions of the opposing group (Nordbrandt, 2021). The strong level of partisanship in the United States may also drive these phenomena on social media. U.S. partisanship has previously been called a “mega identity” along with other identities such as race, structure, citizens, and social networks (Mason, 2018). This identity is said to be aligned with U.S. media consumption—including social media—which in turn helps foster polarization, partisanship, and mega identities (Lane et al., 2025). The unique nature of the U.S. political-media ecosystem offers potential for gaining meaningful insight into modern political communication and political identity, even though the insights and results may not be immediately transferable to other contexts.
Increasing political animosity in the United States and abroad has been linked to incivility toward political opponents (Mamakos and Finkel, 2023) with uncivil attitudes especially prominent among those identifying as strongly partisan (Miller and Conover, 2015). Concerns over civility in online political discourse often focuses on toxic political comments, broadly defined by Kim et al. (2021) as “expressing disrespect for someone by using insulting language, profanity, or name-calling; by engaging in personal attacks; and/or by employing racist, sexist, and xenophobic terms” (p. 924). As mutual respect is a natural precursor to productive political discussion between parties, maintaining civility and adhering to group norms in online discourse is considered necessary to prevent discussions from devolving into a toxic exchange (Cheng et al., 2017; Shmargad et al., 2022). Prior research has confirmed the existence of toxicity in online discussions (e.g., Chen, 2017; Ventura et al., 2021) but researchers have drawn varied conclusions on the severity of this phenomena (Kim et al., 2021; Rossini, 2020).
This study explores the political communication of major party U.S. Senate candidates during the height of the 2022 election to uncover patterns associated with toxicity. The first research question focuses on how the level of toxicity changes during the 100-day period leading up to election day. Temporal changes were expected as negative campaigning is a more common strategy for candidates who trail in the polls (Auter and Fine, 2016), and it is reasonable to assume candidates may become more desperate as the campaign nears completion:
This study will also explore gender differences in candidates’ social media content. Prior research by Haferkamp et al. (2012), Horan et al. (2024), and others has suggested gender-based differences in candidates’ social media presentation strategies:
As toxicity has associated with fostering partisan identity, this study will investigate toxicity differences based on party affiliation:
This study will also investigate toxicity differences based on incumbent status, as this has been identified as an under-researched variable which can impact candidates’ social media strategies (Afonso et al., 2025). Incumbent candidates often possess greater name recognition and visibility, whereas non-incumbents must work to draw attention to themselves and their policies:
Methods
Candidates included in data collection.
*Incumbent.
Descriptive statistics for full dataset.
This investigation used the Perspective API to measure the toxicity for each tweet. This API has been used extensively in research to assess the toxicity of written communications (e.g., Mamakos and Finkel, 2023). Analysis of text data by the Perspective API produced a predefined toxicity attribute for each tweet. Scores range from 0.0 to 1.0, with scores representing a probability that the text in question will be perceived as toxic by readers (Perspective, nd). Scores above 0.7 are thought to represent “high” scores (more likely to be perceived as toxic), and those less than 0.3 representing “low” (less likely to be perceived as toxic) scores (Pascual-Ferrá et al., 2021). Analysis utilized ANOVA and Independent Samples t-Tests for group and temporal differences.
Results
Toxicity of U.S. Senate candidate posts
Descriptive statistics for toxicity.
*Partial month.
The Independent Samples t-Tests for gender differences (RQ2) results showed that tweets from male candidates (mean = 0.058, SD = 0.085) had statistically significantly higher overall toxicity scores compared to those from female candidates (mean = 0.045, SD = 0.068), t (14,998.008) = 12.333, p < 0.001, Cohen’s d = −0.164. In terms of political differentiation (RQ3), tweets from Republican candidates (mean = 0.060, SD = 0.083) had significantly higher overall toxicity scores compared to those from Democratic candidates (mean = 0.051, SD = 0.079), t (20,412.192) = 8.175, p < 0.001, Cohen’s d = 0.110. In terms of incumbent status (RQ4) tweets from non-incumbent candidates (mean = 0.062, SD = 0.087) had significantly higher toxicity scores compared to those from incumbents (mean = 0.036, SD = 0.054), t (18,286.208) = 27.441, p < 0.001, Cohen’s d = 0.331. These results show that tweets from males, Republicans, and non-incumbents were more likely to be perceived as toxic than tweets from females, Democratic candidates, and incumbents, respectively. Effect sizes in all cases were small.
Discussion
The analysis of Twitter/X communication during the 2022 U.S. Senate election season (August–November) indicates that candidate posts exhibited little of the toxic character so often associated with social media. The results suggest that, overall, U.S Senate candidate posts during this period were generally low in toxicity. Over the course election season, tweets from later stages of the campaign were significantly more likely to be perceived as toxic than earlier tweets, perhaps suggesting a more intentional approach among some candidates closer to election day. Tweets from female candidates, Democrats, and incumbents were all found to be significantly less likely to be perceived as toxic than tweets from male candidates, Republicans, and non-incumbents, though effect sizes were small. It is possible that the group differences found in this study may be influenced by the demographics of the platform; prior research has shown that Twitter/X users have tended to be younger and more politically liberal (Bestvater et al., 2022). As a strategic communication tool, focusing Twitter/X messaging toward its audience would be logical, though it is not uncommon to find identical messages broadcast across different social media platforms. Future research should delve into whether politicians utilize differential, targeted messaging based on the predominant demographics of the platform.
The relatively limited toxic content stands in contrast to existing research which has shown that the use of moral and emotional content, especially expressions of outrage, results in greater sharing and engagement. Though this study does not assess engagement, the results show that tweets from these candidates exhibit limited amounts of toxic content overall. This generally neutral messaging may indicate that traditional forms of social media engagement are not a primary consideration for the campaigns. Further research is necessary to determine the link between traditional engagement measures and tweet content. Research into subsequent threaded discussions (i.e., replies) may provide clarity on whether candidates’ social media communications promote meaningful discourse or simply devolve into partisan toxicity.
The acquisition of Twitter by Elon Musk in late 2022 brought with it a number of platform changes, including a shift in content moderation away from a human-based system to one more reliant on both community-driven moderation and algorithmic/AI filtering of false or potentially harmful content. This policy change has necessarily impacted the discussion dynamics of Twitter/X conversations in general, and it is reasonable to assume that political communication on the platform has been impacted as well, especially given the evolution of the U.S. political landscape. As a result, the present study can be considered a point-in-time case study of social media-based U.S. political communication. Future research should compare long-term differences in the toxicity of candidate communications between the current environment and the pre-2022 time period. As mentioned previously, the focus on the U.S. as a case study may not provide results immediately transferable to other national contexts. Future comparative research in international contexts would help to more clearly identify whether common patterns of toxic political communication exist.
The dataset is the primary limitation of this study. The end of academic research access to Twitter content in mid-2023 resulted in a limited dataset; additional candidates were intended to be part of the sample. Future research in this area will require significant funding resources and/or free access to the data found on social media platforms.
Conclusion
This study explored the Twitter (now X) feeds of 2022 U.S. Senate candidates to determine patterns of toxicity. The results suggest that these candidates’ Twitter/X-based political communication is mostly neutral, though group differences were found based on political traits (e.g., political party and incumbent status). This study provides a basis for further investigation into the nature of candidates’ social media-based communications. Future research should incorporate more international, comparative, platform-specific, and long-term analysis techniques to uncover whether toxicity patterns exist in political candidates’ social media communications.
Footnotes
Ethical considerations
The methodology for this study was deemed exempt by the Office for Research Protection of Penn State University. The Office for Research Protections determined that the proposed activity, as described in this submission, does not meet the definition of human subject research as defined in 45 CFR 46.102(e) and/or (l).
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Data Availability Statement
The dataset(s) generated during and/or analyzed during the current study are not publicly available due to Twitter licensing policies prohibiting distribution of large numbers of tweets but are available from the corresponding author on reasonable request.
The tweet texts analyzed during this study are not available in the provided repository (https://doi.org/10.17605/OSF.IO/6CDPK) due to Twitter/X licensing policies prohibiting distribution. The repository includes the study data necessary to replicate all statistical analyses for this study. The repository also includes the numerical ID of each tweet analyzed. The tweets IDs may be used by interested parties to retrieve the full tweet text content via the Twitter/X API.
Carnegie Corporation of New York Grant
This publication was made possible (in part) by a grant from the Carnegie Corporation of New York. The statements made and views expressed are solely the responsibility of the author.