Online Sentiment Toward Harris Declined—More So than that for Trump—Vice-Presidential Debate Study Shows

While presidential campaigns and political pundits await the next round of polling in order to gauge the impact of this week's vice-presidential debate, a new study finds that online sentiment toward Vice President Kamala Harris declined more steeply than that for former President Donald Trump during and after the exchange between Senator JD Vance and Governor Tim Walz.

"Post-debate conversations often center on who won the contest," says Anasse Bari, a clinical professor in computer science at the Courant Institute and head of NYU's Predictive Analytics and AI Research Group, which conducted the study. "But a more significant question is what impact did a vice-presidential debate have on the two candidates not on the stage. By measuring sentiment of online users' posts of both presidential nominees during and after the event, we can get a sense of how it affected the views of the candidates at the top of the ticket."

To understand how online users felt, overall, about the candidates, the researchers examined Reddit posts. "Positive sentiment" indicated that the text surrounding the candidate conveyed positive emotions, support, or approval (e.g., using words such as "improving" when referring to inflation). "Negative sentiment" suggested that the words surrounding the candidate conveyed negative emotions, resistance, or criticism. "Neutral sentiment" marked an absence of either positive or negative sentiments. The researchers also considered where sentiment stood before the debate-in order to understand how it shifted as a result of the event.

Harris's sentiment score decreased by approximately 18% in the period leading up to the debate to the period during the exchange, becoming less positive overall, then declined even further after the event. By contrast, Trump's sentiment score decreased by approximately 23% from before to during the event, but rebounded by 10% from the period during to after the event, indicating a partial recovery in sentiment score.

Interestingly, the sentiment scores of two issues-immigration and healthcare-that received significant attention from both candidates were unaffected as a result of the debate, even though these scores changed during and afterward. Immigration's sentiment score decreased by approximately 51% from before to during the event, becoming much less positive, but in the during-to-after-the-debate period returned to its initial sentiment level. Healthcare's sentiment score decreased by approximately 35% from before to during the event, becoming less positive overall, but then increased from the period during to after the event, recapturing the overall positive level it had before the debate.

The researchers also examined online topical searches before, during, and after the debate, finding that, as with the September 10 presidential debate, abortion-related searches were the top-searched topic-and, in fact, increased by 22% compared to the presidential debate. Other highly searched topics were the following:

• "Economy"-a 94% increase from the period before the debate to during the debate and another 168% increase from the period during the debate to the two hours after it.
• "Israel"-an 81% increase from the period before the debate to during the debate and another 131% increase from the period during the debate to the two hours after it.
• "Tax"-a 41% increase from the period before the debate to during the debate and another 185% increase from the period during the debate to the two hours after it.
• "Carbon emissions"-a modest 9% increase from the period before the debate to during the debate, but a dramatic jump of 441% from the period during the debate to the two hours after it.

The study's other authors included researchers at the Courant Institute's Predictive Analytics and AI Research Group: Charles Wang, Harrison Gao, Naman Lalit, Atmaj Koppikar, Kartik Kanotra, Suryavardan Suresh, Yifei Xu, Dev Pant, Anway Agte, and Tomisin Adeyemi.

Methodology
To gauge online users' interests and sentiments three hours before the debate, during the debate, and two hours after the debate, the researchers analyzed and measured sentiment in Reddit posts, Google search trends, and YouTube comments and also tracked trending tags on X related to the debate and each candidate. They also created indices for each candidate and key debate topics, analyzing data in near real-time, starting three hours before and continuing during the debate. To measure the mood of these users, the researchers applied natural language processing and sentiment analysis algorithms to the online posts.

Press Contact