From alternative conceptions of honesty to alternative facts in communications by US politicians

The global rise of misinformation poses a significant threat to democratic institutions and societal cohesion. While the role of social media in disseminating misinformation is widely acknowledged, the contribution of political leaders remains under-researched. This study examines a potential link between the changing conception of honesty among US politicians and the spread of misinformation. The researchers hypothesize that a shift towards prioritizing "belief speaking" (statements based on personal beliefs regardless of factual accuracy) over "fact speaking" (statements grounded in evidence) is associated with increased dissemination of misinformation. This hypothesis is grounded in the observation that some segments of the public value overt honesty, even if it entails blatant falsehoods, as a signal of authenticity and opposition to the establishment. The study uses Twitter data as a primary source to investigate this hypothesis due to its role as a major avenue of public-facing discourse by US politicians. The specific focus is on US Congress members' tweets, considering the US's position as a key site of conflict between populism and liberal democracy, and Twitter's significant influence on political communication and agenda setting. The study aims to identify patterns of belief speaking and fact speaking in Congressional tweets, analyze their evolution over time and across party lines, and assess the relationship between these conceptions of honesty and the quality of information shared.

The paper reviews existing research on the retreat of democracy worldwide and the role of misinformation in this process. It notes the established evidence that exposure to misinformation can influence behavior and contribute to phenomena such as voting for populist parties and increased ethnic hate crimes. The psychological aspects of misinformation, including its persistence in memory even after correction and its potential valorization as a signal of authenticity, are discussed. The paper introduces the concepts of "belief speaking" and "fact speaking" as two distinct ontologies of truth and honesty. Belief speaking prioritizes the speaker's subjective beliefs and feelings, while fact speaking emphasizes the search for accurate information and evidence-based belief updates. The authors connect "belief speaking" to historical examples of radical constructivist truth, such as in 1930s fascism and contemporary populism, and contrast it with the evidence-based approach essential for effective democratic governance. The difference between truth (the accuracy of information) and honesty (a virtuous human quality) is clarified. The paper highlights the limited empirical evidence regarding the shift towards belief speaking in US political discourse despite significant concerns.

The study employed a computational analysis of a comprehensive dataset of tweets posted by members of the US Congress between January 1, 2011, and December 31, 2022. The dataset, obtained through the Twitter API, initially contained over 5.9 million tweets. After removing retweets and duplicates, the final corpus consisted of approximately 3.9 million tweets. The tweets were categorized by party affiliation. To measure belief speaking and fact speaking, the researchers created two dictionaries of keywords associated with each concept. A computational grounded theory approach was used, combining expert knowledge with computational pattern recognition. The dictionaries were validated in three stages: (1) a survey assessing keyword representativeness; (2) a survey comparing computed similarity scores to human ratings of tweets; and (3) analysis of New York Times articles across different sections (opinion, politics, science) to examine the dictionaries' ability to discern between contexts. The dictionaries included 37 keywords each. The cosine similarity between tweet embeddings and dictionary embeddings was used to calculate belief-speaking (Db) and fact-speaking (Df) similarity scores for each tweet. To account for the correlation between tweet length and similarity scores, the scores were centered and length-corrected (Db', Df'). The quality of shared information was assessed using NewsGuard scores, which provide trustworthiness ratings for websites. A linear mixed-effects model was employed to analyze the relationship between belief speaking, fact speaking, party affiliation, and NewsGuard scores (for both tweet texts and article texts scraped from linked websites). An independent dataset of domain trustworthiness was also used for validation.

The analysis revealed a bipartisan increase in both belief-speaking and fact-speaking language over time, particularly noticeable after the 2016 presidential election. This parallel increase might reflect the growing centrality of "fake news" in political discourse, leading to a proliferation of claims and counterclaims. However, a significant difference emerged between the parties when information quality was considered. Republicans tended to share lower-quality information compared to Democrats. Critically, for Republicans (but not Democrats), a 10% increase in belief-speaking similarity was associated with a 12.8-point decrease in NewsGuard score. Conversely, an increase in fact-speaking similarity was linked to higher-quality sources for both parties. This finding held even when analyzing the text of articles linked in tweets. The analysis also showed that the voting patterns in the 2020 presidential election in the politicians' home states did not influence the quality of news shared by members of Congress. Further analyses explored topic-specific variations in belief and fact speaking, showing greater use of belief speaking in contentious topics (except for vaccine-related discourse). Individual-level analysis revealed substantial variation in belief-speaking and fact-speaking patterns among politicians. The study found no relationship between belief speaking and electoral outcomes, suggesting that voters were not deterred by the spread of misinformation.

The findings support the hypothesis that a shift towards belief speaking is associated with the dissemination of lower-quality information, particularly among Republicans. This suggests that the public might be sensitive to cues from political elites regarding the credibility of information sources. The authors offer alternative explanations for this partisan difference, considering the possibility that belief speaking among Republicans might stem from a desire to derogate political opponents. A mediating effect of negative emotion is proposed and investigated in the supplementary material, suggesting a potential link between belief speaking, negative emotionality towards the out-group, and the sharing of lower-quality content. The authors acknowledge the correlation nature of the findings and the limitations of solely focusing on the "political class" in the US. The study's results highlight the need for further research in different cultural and political contexts to establish the generalizability of the observed patterns.

This study provides valuable insights into the evolving relationship between conceptions of honesty, political communication, and the spread of misinformation. The strong correlation between belief speaking and the sharing of low-quality information among Republicans highlights the potential for political rhetoric to influence the public's information diet and contribute to political polarization. Future research should explore the temporal stability of these trends, consider the unique role of social media versus traditional media, and extend the analysis to other political systems to assess the generality of these findings. The results underscore the critical need for fact-based political discourse and media literacy to protect the integrity of democratic processes.

The study's correlational nature prevents causal inferences. The focus on US politicians and Twitter data limits the generalizability of findings to other contexts. The reliance on NewsGuard scores, while widely used, involves limitations inherent in any fact-checking system. The study's time frame ends before significant changes in the Twitter platform under Elon Musk's ownership, potentially impacting the long-term stability of the observed trends. Finally, the analysis of scraped website texts has a success rate of 65% due to various issues, impacting representativeness of the data used for analysis.