Content Analysis and Topic Modeling of Trump/Clinton Tweets
These days the world, or at least the media, is all a twitter over the tweets emanating from the President of the United States. It seems as if every morning we wake up to some new missive, observation or attack the President has unleashed against real or perceived foes. More rarely, he arranges 280 or fewer characters to announce or to pronounce on some aspect of government policy. Today, it seems safe to say that these tweets are a major influence on the setting of the daily media agenda. This is in large part because they come from the President. Was it always this way? Did candidate Trump’s tweets have a significant impact in setting or changing the agenda of Hilary Clinton during the presidential primary campaign?
Mr. Trump developed his Twitter strategy during his campaign for the Presidency in 2016. At the same time the presumptive Democratic nominee Hilary Clinton was deploying a Twitter strategy of her own. When we look at the perceived impact of the President’s tweets today, we might assume his tweets had a similar influence on the campaign. In their paper, “Trump vs Clinton, Twitter Communication during the U.S. Primaries,” Fromm, Melzer, Ross and Stieglitz look at the Twitter record of the two main candidates during the primary campaign and ask two questions. What is the nature of the Twitter communication by the main presidential candidates during the 2016 US primaries? To what extent does agenda setting take place on Twitter between these candidates?
The authors hypothesized that, because of their status, current and previous positions and the way they use Twitter, “there are good reasons to believe that the candidates should inﬂuence each other’s agenda. Examining whether this is indeed the case helps understand how Twitter was used by the candidates before the election.”
To conduct their research the authors collected more than 6,000 tweets split about evenly between the two candidates. The tweets were from the primary season Nov. 15, 2015 to June 4, 2016. The tweets were subjected to content analysis, using QDA Miner and WordStat topic modeling. The authors built a dictionary for their specific subject matter using an existing one as a base. The resulting dictionary included 21 categories: employment, environment, guns, healthcare, military and defense, terrorism, slogans, media, family, rights, meetings, thank-you messages, campaign funding, parties and politicians, caucus, foreign politics, education, economics, justice, Trump family and minorities. The dictionary was tested against randomly collected, manually coded tweets. The authors also subjected the candidate’s tweets to a time series analysis to see how they impacted each other’s agenda.
The paper shows how topic modeling, the construction and use of a dictionary can help you research large amounts of text, in this case more than 6,000 tweets. The results of the research did not fit neatly into the authors hypothesis, raised other questions and prompted areas of additional research. You can read the complete paper here