"}\nTwitter allows political candidates to broadcast messages directly to \nthe public, some of which spread virally, potentially reaching new \nsupportive audiences. During the 2014 U.S. gubernatorial election, 74 \ncandidates for State Governor posted 20,580 tweets, of which 10,946 were\n retweeted almost 140,000 times. By analyzing a collection of tweets \nposted by gubernatorial candidates that were classified by machine \nlearning into categories of message types, we find that while candidates\n tend to post tweets that advocate for themselves the most, the public \nis more likely to retweet attack messages and messages labeled as \ncall\u2010to\u2010action. As measured by number of retweets, call\u2010to\u2010action tweets\n tend to reach the broadest audience. We also find that middle\u2010level \ngatekeepers, those with between 1,800 and 26,000 followers, tend to have\n the most influence over the flow of political information. Since \nretweets tend to bring new followers, these findings suggest that \npoliticians wishing to grow their audience may benefit from posting more\n call\u2010to\u2010action and attack messages, and that candidates may wish to \nfind ways to actively enlist the support of middle\u2010level gatekeepers.
More influencer theory, although the real step I wonder about is the flow of political information to offline people.