Fig. 1: The center panel shows a conceptual model of potential strategic diversion by Donald Trump via Twitter (where his handle is @realDonaldTrump).
From: Using the president’s tweets to understand political diversion in the age of social media
The word cloud on the left contains the 50 most frequent words from all articles in the NYT that contained “Russia” or “Mueller” as keywords. The NYT articles captured by the word cloud contained a total of 146,307 unique words. We excluded “president” and “Trump” because of their outlying high frequency. In addition to the 50 words shown here, the top 1% of high-frequency items included terms such as “collusion,” “impeachment,” “conspiracy,” and numerous names of actors relevant to the investigation, such as “Mueller,” “Putin,” “Comey,” “Manafort,” and so on. The word cloud on the right represents the 50 most frequent words occurring in Donald Trump’s tweets that were chosen, on the basis of keywords, to represent his preferred topics. The expected sign of the regression coefficient is shown next to each path and is identical for both approaches (OLS and 3SLS). The delay between media coverage and diversionary tweets is assumed to be shorter than the subsequent response of the media to the diversion. This reflects the relative sluggishness of the news cycle compared to Donald Trump’s instant-response capability on Twitter.
