Examining spread of emotional political content among Democratic and Republican candidates during the 2018 US mid-term elections

Social media, particularly Twitter, has become a crucial tool for political candidates to reach voters. While previous research has focused on the valence (positive or negative) of political messaging, this study explores the impact of specific emotions (joy, anger, fear, sadness, confidence) on the spread of candidates' messages on Twitter. The 2018 US mid-term elections serve as a case study, characterized by antagonism and emotional appeals. This study aims to address two questions: (a) Do certain emotions increase message diffusion (retweets and favorites)? (b) Do these patterns differ between Democrats and Republicans? Existing research suggests a negativity bias, with negative content being more effective, but doesn't fully explore differences among specific negative emotions or across political parties. This study uses a large dataset of tweets from Senate candidates to analyze these questions.

Prior research on political messaging has largely focused on valence, finding that negative content tends to be more effective than positive content in increasing voter turnout or choice. Studies have shown that negative rhetoric, particularly anger, can attract significant attention on social media, possibly due to lower thresholds for reacting to negativity online. However, research examining the impact of emotions like fear and sadness is limited, especially in the context of political communication on social media. Existing work often relies on self-reported surveys or lab experiments, which may have limitations. This study seeks to address these gaps by analyzing a large dataset in a natural setting.

The study used a dataset of 7310 original tweets from 65 Senate candidates (3711 from Democrats, 3599 from Republicans) in the four weeks leading up to the 2018 US mid-term elections. Data was collected using the Twitter API and Tweepy. Text pre-processing was performed to handle colloquialisms, symbols, and special characters. IBM Watson's Tone Analyzer, a natural language processing tool, was used to identify the emotional tone (joy, anger, fear, sadness, confidence) of each tweet, providing a confidence score for each emotion. Tweets were included in the analysis only if the confidence score was above 0.5 for at least one emotion. The study focused on direct retweets and favorites as measures of engagement. Linear mixed-effects models were used to analyze the data, incorporating random effects to account for the hierarchical structure of the data (tweets nested within candidates). Variables were rescaled to a 0-1 range to account for differences in scales. Model comparison was done using AIC and BIC to select the best model. Conditional R² and the squared correlation coefficient between observed and fitted values were used to assess model validity.

The analysis revealed that joy-signaling tweets were less likely to be retweeted or favorited by both Democrats and Republicans. However, the relationship between negative emotions and engagement varied by party. For Democrats, fear-based tweets were significantly associated with increased retweets and favorites, whereas anger-based tweets were not. Conversely, for Republicans, anger-based tweets were significantly associated with increased retweets and favorites, while fear-based tweets were not. Neither sadness nor confidence significantly impacted engagement. The models showed that 74-80% of the variation in retweets and favorites could be explained by the variables included in the model. Further analysis showed that tweets containing URLs or user mentions were less likely to be retweeted or favorited, particularly for Democratic candidates. Hashtags had a negative impact on engagement for both parties.

The findings support the idea of a positive-negative asymmetry in the spread of political messages, but also show that different negative emotions have different effects and that these effects vary across the political spectrum. The association of anger with Republican engagement and fear with Democratic engagement suggests that the motivational bases of these emotions may be relevant. For Democrats, fear-based messages might reflect a group defensive strategy in response to feeling threatened. Republicans, being in power, might have been more responsive to approach-related emotions like anger. However, these are speculative explanations and require further investigation. The context of political power may play a significant role in shaping responses to different emotions.

This study demonstrates that specific emotions, beyond simple valence, significantly influence the spread of political messages on Twitter. The findings highlight the nuanced relationship between emotional content, political party affiliation, and online engagement. Future research should replicate this study in different contexts, use a more balanced dataset, and examine other social media platforms. Further research could also explore the role of shared media content or mentions of specific individuals in shaping engagement.

The study focused on the 2018 US mid-term elections, which may limit the generalizability of the findings. The relatively small proportion of tweets conveying anger or fear might have inflated their impact on engagement. While mixed-effects models addressed the hierarchical nature of the data, a more balanced dataset would strengthen the results. The study also focused on Twitter, and findings may differ for other social media platforms.