In January 2011, I sent an email to the Washington State Democratic Party asking about precinct results from the 2008 presidential caucus race. State party chairman Jaxon Ravens replied simply with the question, “Why would you want them?” As he reminded me, the precinct results had no bearing on delegate allocation or the eventual party nominee. My nineteen-year-old self struggled to articulate a satisfying answer.
What I failed to put into words was that election results show so much more than simply who won and lost a constitutionally-legitimized popularity contest. Election results lay bare the souls of its voters, translating millions of individual hopes, dreams, fears, aspirations, and biases into tangible, observable quantities. No census or survey can truly capture that singular moment of personal truth which occurs in the ballot box. We can identify your race, your income, a list of a thousand other measurable values which statistically imply the outcome of this moment, but as deterministic as we might try to make it seem, it always comes down to a final act of free will. These individual acts sum to make manifest the inner milieu of a people at a particular moment in time, a secular sacrament ordaining to our political priesthood.
Election results tell us who we are while other statistics only hint at what we are; maps visually organize this information in a way both informative and artful. Living on the West Coast my entire life, I’ve never had the opportunity to travel to many of the great population centers of the East, the veiled gulches and hollows of Appalachia, or the wide open country of the Heartland, but through election maps I could learn many things about the people who live there, how they vary, and how they view themselves. Using precinct-level data, I could almost walk the streets of neighborhoods and communities thousands of miles away. They made me feel connected to and in solidarity with a world I have never experienced; they remind me that I too am a part of the American democratic experiment, even if it all too often seems a petty game for out-of-touch East Coast elites.
In 2008, I started collecting General Election county data and creating rudimentary maps in MS Paint going back to the 1840s using the University of Richmond’s Voting America application. I began a project the following year expanding into Presidential Primary Election county data, and I took up learning GIS mapping for my high school culminating project. As an undergraduate, I continued to dabble in GIS, even though I never took a course in the subject, and at the start of the 2012 Republican Presidential Primaries I even briefly toyed with the idea of a national precinct map before realizing the enormity of such a task.
Precinct data, despite providing the clearest available picture of how areas vote, can be quite difficult to both come by and to visualize. There isn’t a singular, unified source of precinct-level data nationally nor even at the statewide level in many cases. Precinct boundaries frequently shift over time, especially during the decennial redistricting process following each Census. Only on the eve of the 2012 General Election did Stanford University’s Spatial Social Science Lab produce the first ever publicly-available national precinct map, depicting the 2008 General Election with the aid of Census voter precinct shapefiles.
To my knowledge, no such national precinct map was ever made for the 2012 General Election for President. The closest was the New York Times’s national ZIP-code map published exclusively in print form on the eve of the 2016 General Election. While this represented a clear improvement over county maps, there are only 43,000 ZIP-codes in the United States compared to almost 175,000 voter precincts, over four times fewer. Further, a static, print-only map leaves much to be desired in dense urban centers depicted far too small to be legible.
After spending most of my spare time in 2015 working on a global religion map, the 2016 Presidential Primaries rolled around, and I decided to go for it: I would do everything in my power to create a national precinct map. I didn’t have a team of researchers. I didn’t have aides. I didn’t have much extra money. I didn’t have connections. But for some reason, I thought I could do it anyway.
Hundreds of emails and phone calls and months of work later, here’s what I came up with:
There’s a saying in economics and statistics that “all models are wrong, but some are useful.” So too it is with maps. It’s not perfect. While I have precinct results for every 2016 county, there were three rural counties I simply could not get precinct maps or shapefiles for without physically visiting. At times I had to omit some empty precincts or aggregate to the township or ward level due to out-of-date shapefiles. In several states, especially in the parts of the South and Northeast, I had no way of objectively assigning absentee, early, and provisional ballots reported at the county level to individual precincts. Despite all this, the map still displays the returns of more than 131 million ballots cast, over 95% of all Presidential votes in the election.
For a number of states, I had to create precinct shapefiles almost entirely by hand. States such as Rhode Island and Oregon didn’t participate in the Census’s 2010 Voting Tabulation District shapefile program while only about half of Montana’s counties supplied spatial data, meaning I barely had a base to begin from in these states. I was left with no option but to create them myself, but once I had the shapefiles, I was able to map 2012 as well:
And since I knew where to find 2008 data (Harvard Election Data Archive and the work of the wonderful David Bradlee) and shapefiles, I figured it might be interesting to compare the start and end of the Obama presidency, so I also mapped 2008:
With so many factions realigning themselves for the 2016 General Election, a swing map depicting the change in percentage margin of victory would reveal many unique attributes of the underlying population. A “true” swing map simply can’t be created in most cases, but I employed ArcMap’s Spatial Join feature to match precincts based on their centroids and degree of overlap in order to estimate a close approximation of the true swing. For the vast majority of precincts, this methodology works quite well, though not as well in areas where drastic precinct changes have been made. Here’s what it looks like:
This map tells many particularly interesting stories on which I’ll elaborate in future posts, but suffice to say that most of the precinct swing can be explained by one variable: education level, perhaps augmented somewhat by race and ethnicity. While we can’t conclusively say anything about the nature of underlying voter preference changes due to the ecological fallacy, the data is certainly suggestive of broader themes. As they say, “correlation doesn’t imply causation, but it does waggle its eyebrows suggestively and gesture furtively while mouthing ‘look over there’.”
I’d like to thank a number of people and groups who’ve been instrumental in this endeavor, whether they know it or not: Brandon Finnigan and J. Miles Coleman of Decision Desk HQ, Dave Bradlee of Dave’s Redistricting App, Derek Willis and OpenElections, David Leip of U.S. Election Atlas, Tom Giratikanon and the New York Times, the Harvard Election Data Archive, UC Berkeley’s California Statewide Database, Kevin Rancik of Portland State University for his work on Oregon shapefiles, countless county clerks and boards of election who were time and again generous and helpful, and my wife Hally who put up with far too many nights and weekends consumed by this project.
Below the full interactive GIS version of all four maps.
Note that the color scheme for the interactive 2012-16 swing map differs from the above static swing map due to hosting limitations. Similarly, precincts with zero votes are colored the lightest pro-Trump swing due to how the host’s binning works.