Fig. 6.

Predictors of the number of RTs received by messages. Panels A1 and A2 show estimates for the Yellow Vests data; panels B1 and B2 show estimates for the Catalan Referendum data. The estimates result from message-level mixed-effects models (with 95% bootstrapped CI). Upper row: controlling by structure and content (and random variability at the account level), messages published by media accounts receive more RTs than messages published by unverified bots (the base category). Messages posted by human accounts, however, receive less diffusion. Lower row: Tweets published by the subset of news organizations for which we have web reach data also receive less attention than verified media accounts, an effect that is particularly strong and significant during the Catalan referendum.