Work

Who Speaks Matters More Than How: Automated Accounts, Account Reach, and Message Features in COVID-19 Vaccine Tweets Around the Delta Wave

Year: 2026
Type: preprint
Abstract: Studies of crisis communication on social media often infer which message features drive engagement from large public tweet corpora, yet such corpora can contain automated accounts and are rarely adjusted for who is posting. We re-examine Media Richness Theory and the feature-based operationalization of Dialogic Communication Theory using 180,278 deduplicated English-language COVID-19 vaccine tweets from a public Kaggle corpus, covering windows before (April–June 2021), during (July–September 2021) and after (October–November 2021) the Delta wave. Automated vaccine-slot notification accounts produced 34.0% of Delta-window tweets and attracted almost no engagement. A naive negative binomial model of all Delta tweets suggested that a link or attached media multiplied engagement 4.71-fold. After removing automated accounts, adjusting for account reach and clustering standard errors by account, the association fell to 1.40 (95% CI 1.19–1.64). Positive sentiment remained positively associated with engagement (IRR = 1.48), questions (0.85) and hashtags (0.91 per tag) were modestly negative, and @mentions were unrelated to engagement. Follower count was the strongest predictor: doubling an account's followers was associated with about 36% more engagement. Across windows, the link and sentiment associations were stable in direction, whereas the dialogic cues shifted, and we found no evidence that the link effect reverses after the peak. The results suggest that who speaks matters more than how a message is designed, that counting dialogic features captures little of what dialogue theory describes, and that auditing corpora for automation is a precondition for drawing communication lessons from public tweet data.
Language: English
Subfield: Communication
Open Access status: green
OpenAlex ID: W7225999115
Cites:
Cited by:
Related to: 10