As of January 2014, all 100 United States senators Twitter accounts. And, it seems to me, all but one senator, Idaho’s Jim Risch, have a Facebook page. I found the account names and/or numbers mainly The Sunlight Foundation and then added a handful that I discovered to be missing or wrong.
Next, I took the text of those posts, discarding any images or videos, and used Latent Dirichlet allocation to classify the topics found in posts. After tying a few different options, I settled on twelve topic groups. Of course, there are still some posts that don’t fit easily into any category but these seemed reasonable. The names of the groups I created myself.
You can view my viz here. It’s made in D3 and shows the data for the period of July 1, 2014 to October 31, 2014. Hover over a senator’s bar to see their details or hover over the category names to see prominent words in the cluster and example tweets.
One of the first things that jumps out is that New Jersey Senator Cory Booker tweets/facebooks (is facebooks a verb?) more than twice the next highest senator. He is also by far the most likely to reply to another user on Twitter, replying 1784 times. This is a big reason such a high percentage of his tweets fall into the “Thank yous and Personal Appeals” category. He had the third most Facebook replies however, behind Mark Begich and Mitch McConnell.
Facebook replies turned out to be the most predictive factor in being able to tell if a person was currently running for election. My analysis didn’t look at who the users a senator was responding to were but it seems to make sense that Facebook replies would be more likely to be constituents whereas Twitter users could either regular citizens or other politicians. For example, I noticed some replying to @BarackObama.
Among those Senators running for re-election, there is an average 26% increase in the “Press and News Release” category compared to four months prior, the clearest topical indicator that he or she is actively campaigning. There is also a average 30% decrease in the “Violence and Women” category among those running in the midterm elections compared to a 19% increase among those Senators not running which seems a bit counter-intuitive given the attention female voters have received running up to the midterm elections.
Want to see more? Check out the source code on github.