Something I’ve become interested in lately is the interplay between moral psychology, language, and politics. For my final data visualization project, I decided to focus on one place in the world where this interplay is perhaps at its most bizarre: Donald Trump’s Twitter feed.  

There’s a tight link between moral psychology and politics; for example, an individual’s set of moral values is a good predictor of their political ideology. There also appears to be a tight link between word choices and moral values; changing the morally-charged words one uses in an argument, for example, can make it more or less effective depending on the moral values of the the audience. The question underlying my project, then, is this: What can we learn about Trump’s moral values—or at least the moral values of the people he seeks to influence—by analyzing the words he chooses to use in his tweets?

To classify moral language, I used Moral Foundations Theory, which divides morality into five categories, or foundations: authority/subversion (we value order, tradition, and hierarchy), care/harm (we feel compassion for the suffering and vulnerable), loyalty/betrayal (we keep track of who is “us” and who is “them”), fairness/cheating (we make sure people are getting what they deserve), and sanctity/degradation (we believe certain things are elevated and pure and shouldn’t be tarnished). 

This information is essential to understanding the visualization, and I found it one of the biggest challenges of this project to figure out the best way to get it across. Of course, it’s asking a lot of an audience to throw a wall of exposition at them before showing them the actual visualizations they’ve come to see, and I’m not sure I got the theory across in the most effective way. Perhaps a more graphical representation of the theory would prevent attrition among the tl;dr-inclined crowd.

For the viewers who do read and scroll down, however, the first thing I wanted them to see was, in large print, two numbers that I think set the scene for the rest of the visualization: the absurd number of tweets Trump has sent since 2009, and the (also high, I think, but maybe not absurd) percentage of those tweets that were deemed by Linguistic Inquiry and Word Count (LIWC) to contain moral language. From there, the viewer gets, via line chart, a general overview of Trump’s moral language from his first tweet to the present, broken down by positive (blue) and negative (red) moral words and by each individual moral foundation. I annotated three events that I think were drivers of the trends in the data, though I could be credibly accused of editorializing a bit.

After the line charts comes a bar chart that illustrates the prevalence of each moral foundation in all of his tweets, split further by the positive and negative dimension of each foundation. In the part of the story I wanted to emphasize the contrast between how vicious and petty Trump is with his enemies and how unfailingly positive he is about his family and allies. So I created small multiples of the graph according to the different people he mentions in his tweets, divided by friends and foes. The final set of charts is meant to show how different types of moral language attract different levels of engagement.

As for next steps, I’d like to figure out a better way to visually communicate theory. For data visualizations that require some niche background knowledge to really appreciate, what’s the best way to provide that knowledge in an engaging way?