By Ryan Burge, Eastern Illinois University
In the canonical work on social science research methods (Designing Social Inquiry), King and friends argue that scholarship be theory driven and that analysis should come after strong theorizing. I have to admit, I don’t do that. I always start with a little bit of a question in my mind, do some analysis, make a good chart and then stare at it for a while, and work my way backwards. To be completely honest most of the things that interest me are trying out a new kind of visualization. That’s how this post came about. I really wanted to try out an R package for creating smoothed animations called tweenr. So, I had to find some data. I had stumbled upon a dataset of county level presidential election results dating back to 2000, and I had also been working with county level religious census data that goes back decades. That’s an ideal combination. In order to aid interpretation I aggregated everything up to the state level. On the y-axis is the percentage of vote for the Republican candidate and the x-axis contains the rate of evangelicals per thousand in the state.
I was just going to make the graph, tweet it, and forget about it, but the more I stared at it the more things popped out to me that are worthy of some discussion.
- The first thing that people see is Utah, which makes sense because it takes a huge dive between 2012 and 2016. I have written about this in other places, but Mormon support for Trump was about 30% lower than it was for Mitt Romney in 2012. That’s the largest move of any state in any of the five elections.
- The general trend of the LOESS line is also worth some reflection. For the first two election cycles the relationship between evangelicals and GOP support was positive but relatively small. However, after 2004 the line goes from slightly tilted upward to significantly sloped in a positive direction. Obviously there are many factors at play here and this is just a simple linear regression. States that have more evangelicals tend to be away from the coasts of the United States. Either way, there’s a strengthening relationship here.
- The 2008 election was truly an aberration, that is clear from the animation. From one election cycle to the next there’s a slow drift as a state moves up or down a few percentage points. However from 2004 to 2008, gravity takes over and many many states moved toward Barack Obama’s candidacy. That’s also largely evident by the fact that 4 out of 5 of the most recent presidential elections have been incredibly close.
- I only have religious census data from 2000 and 2010 so there is only horizontal movement at one time in the animation. What is noteworthy about this is that most states are becoming more evangelical, although there are a handful that have seen a modest decline.
5. States that used to be directly on the trend line have now become more Republican. The three examples that stand out to me are Pennsylvania, Ohio, and Iowa. All three of those states were very much on the line until the 2016 election when they move upwards (pro-Trump) on the y-axis.
6. Other states have gone in the opposite direction. The two biggest examples of this for me are Texas and Virginia. Both of these states used to be much more supportive of the GOP but in recent elections have become true tossups. I wrote last week about the changing demography of Texas in regards to the Senate race between Cruz and O’Rourke there. That’s a state that is becoming less and less reliably red by the election cycle.
I know that animated visualizations are usually eye candy and I tend to agree, but sometimes they can help detect movement in a way that static graphs don’t emphasize. I don’t think academic journals will be accepting animated gifs as part of submissions anytime soon, but it does speak to the need for academics to begin thinking about how things like color and movement can help readers understand what the data is saying.
Full coding syntax is available here.