Fig. 4.

Predictors of Being First-time Ch-word Users. Note: The figure presents the relationship between being a first-time ch-word user and one’s Twitter activity and user profile keywords. Panels A and B plot the coefficients and 95% confidence intervals from regressing an indicator for being a first-time ch-word user on user’s pre- and mid-pandemic Twitter activity, and user profile keywords, respectively. Both regressions control for account age, log number of followers, and log number of followings. Regressors in panel A are defined as follows: “Anti-Asian user” is one if an user has interacted with other ch-word users before the pandemic; “Anti-minority” is one if an user has tweeted racial epithets against non-Asian minorities (the n-word, w-word, and k-word) before the pandemic; “Trump” is one if an user has ever mentioned #trump or @realDonaldTrump before the pandemic; “McCarthy”, “McConnell”, “Pelosi”, “Schumer”, “Fox”, “CNN”, and “CBS” are similarly defined using @kevinomccarthy, @McConnellPress (or @LeaderMcConnell), @SpeakerPelosi, @SenSchumer, @cnn, @foxnews, @cnn, and @cbsnews as keywords, respectively; “COVID consp.” is one if an user has ever tweeted keywords related to COVID-19 conspiracies (i.e., plandemic, fakepandemic, scamdemic, film your hospital, 5gcoronavirus, or coronavirustruth) by the end of our sample period. Regressors in panel B are the 25 most common user profile words used by first-time ch-word users and the 25 most common user profile words by control users. There is an overlap between the two sets of words, so the number of words included in the regression is less than 50. Standard errors are heteroscedasticity-consistent. Regression results are reported in Tables A6 and A6.