The Valence Paradox: Positive Policy, Political Outrage, and Engagement on X in the 2024 U.S. Presidential Election

Abstract

This study engages in a machine learning analysis of the relationship between valence under digital messages (positive vs. negative messages), engagement actions (likes and retweets), over the performance of presidential election contenders during the United States Presidential Election-2024 on Twitter/X. Using Natural Language Processing (NLP) and natural language sentiment analysis (VADER) as well as term frequency inverse document frequency (TF-IDF), Using a clean dataset of 26,351 presidential election posts collected from June 4 to November 4, 2024 cleaned for readers to help clarify the relationship between these variables on X. Results from these analyses reveal a valence paradox whereby posts with both positive and policyfocused content received more likes/retweets than those that were negative in focus and devoted to attacking the political opponent, contradicting most of what political communication theory suggests regarding the place of outrage and negative communications in an environment where social media is involved. Trump used a low-engagement strategy in social media posting 4,812 tweets and only generating an average number of likes per tweet (52.64) versus highengagement tweets from Kamala Harris with 118.23 likes on average per tweet.The results seem to show that online voters tend to support constructive political ideas rather than negative political attacks. This has important implications for future political communication and voter activation across social media