WITHDRAWN: Who Influences Whom about What?

Abstract This study examines the dynamics of opinion formation influenced by US senators on Twitter from 2018 to 2022, focusing on nine politically charged issues such as climate change, immigration, and race relations. Leveraging advanced topic and stance detection methods with Large Language Models (LLMs) and an event study design that exploits the intricacies of Twitter's recommendation algorithm to introduce exogenous variation, the research explores both agenda-setting and stance-setting effects. Key findings reveal that senators' tweets significantly shape public discourse, with variations in influence based on topic, partisan identity, and rhetorical style. While Democrats exhibit a progressive influence across issues, Republicans show a neutral/centrist stance on cultural topics, underscoring affective polarization on immigration and abortion rights. Cognitive language is most effective for economic topics, while emotional and egocentric language enhance engagement on climate action. Conversely, toxic language generally backfires, especially on climate topics. Additionally, political communication fosters inclusive over egocentric language and emotional over cognitive expression. These insights are critical for policymakers, social media platforms, and opinion leaders aiming to leverage digital political communication to foster informed and constructive public discourse.