Setting the Tone or Supplying the Words? Donald Trump's Facebook Posts and the Hostility of His Comment Section in the 2016 Republican Primaries

Do politicians set the tone of the conversations beneath their posts? I link 730 posts on Donald Trump's Facebook page during the 2016 Republican primaries to the 5.1 million comments that could be retrieved beneath them, and measure toxicity with an open classifier on a sample of 87,600 comments and with a dictionary on every comment. Comparing posts published on the same day, comments were more toxic beneath posts that named Hillary Clinton and, on the dictionary, beneath posts that named a Republican rival or used negative words. The 35 critical posts, which named an opponent or the media and used negative words, were followed by comment sections about a quarter more toxic and by about a quarter more comments. Much of that difference is accounted for by comments using the same kind of words the posts used, such as "liar". When those words are removed and the comments re-scored, the remaining difference is between about half a point and one and a half points. Posts that named Clinton were the exception: they were followed by more toxic comments even with such words removed. The subject of the comments also followed the post: when a post named Clinton, 13% of comments mentioned her, against 3% otherwise. The results are associations, not causal effects. The files include the preprint and an aggregate post-level dataset (post features, comment-section rates and counts). No post text, comment text or commenter codes are shared.