Results for the 2014–2015 election cycle are archived here.

Introduction

Research in our laboratory focuses on visualization, the conversion of large collections of strings and numbers into images that viewers can use to explore, analyze, and validate within their data. We are particularly interested in multidimensional visualization techniques. A multidimensional dataset D contains m data elements, D = { e₁, ..., e_m }, representing n data attributes A = ( A₁, ..., A_n ), that is, e_i = ( a_i,1, ..., a_i,n ), a_i,j∈A_j. The challenge is to find effective ways to present even some of this data together in a single image.

U.S. Election Results

(a)

(b)

Fig. 2. (a) Montana (larger image); (b) Massachusetts (larger image)

In order to investigate voting patterns across the United States, we decided to visualize winning candidates for four elected offices: the 2016 Presidential, most recent U.S. Senate, 2016 U.S. House, and most recent state Governor's elections. Results were tabulated by congressional district: for each of the 435 districts spread throughout the 50 United States, we collected or estimated which party's candidate the district's voters selected for each of the four offices. Incumbent party losses are particularly important, since they can change the balance of power throughout the country. We therefore wanted to highlight where an incumbent lost during an election cycle for President, U.S. Senate, U.S. House, and state Governor races.

Given these requirements, we built a dataset with two types of data elements representing congressional district results and state-wide results, respectively. Congressional district data elements contain nine data attribute values, and state-wide data elements contain eleven data attributes:

Fig. 3. Pennsylvania's 4th district (larger image), showing Republican choices in the 2012 Presidential, U.S. Senate, U.S. House, and 2010 state Governor elections, with an Democratic incumbent loss of the U.S. House seat

Fig. 1 shows results for North Carolina's 13 congressional districts, as well as state-wide results presented in a small disc floating over the state. The disc is subdivided into the same quadrants as the congressional districts, showing a state-wide switch to Republican for the 2012 Presidental candidate (a red upper-left quadrant textured with X's), a new Republican U.S. Senator in 2014 (a red upper-right quadrant textured with X's), a Republican majority of U.S. House seats (a red lower-right quadrant), and a switch to a Republican state governor in 2012 (a red lower-left quadrant textured with X's). Finally, the height of the state represents the number of electoral college votes it controls. This can be seen by comparing North Carolina's height (Fig. 1, with 15 electoral college votes) to Montana's or Massachusett's (Fig. 2, with 3 and 11 electoral college votes, respectively).

Trend Maps

(a)	(b)
Fig 4. (a) Presidental trend map showing voting directions during the 2008 election cycle, blue regions trended Democratic, red regions trended Republican (larger image); (b) House trend map for the 2010 election cycle (larger image)

Given data between two election cycles, we can build "trend maps" that show which direction a congressional district or state trended. For example, Figure 4a shows a trend map between the 2004 and 2008 Presidental elections. If a district trended more Democtratic, it is colored blue, with saturation representing the strength of the trend (more saturated for stronger trends). If a district trended Republican, it is colored red. The aggregrate discs show the trend for the state as a whole, blue if the state trends Democratic, red if it trends Republican.

Note that the color of the district or state does not define the winning party. It only shows which direction the region moved during the 2008 election cycle. For example, a blue district could have been held by the Democrats in 2004, and re-held but with a stronger majority in 2008. Alternatively, a blue district could have been held by the Republicans in 2004, and re-held but with a weaker majority in 2008. That is, blue indicates either more Democratic or less Republic. Similarly, red indicates either more Republican or less Democratic.

The Presidental trend map is striking in the fact that most of the country is colored blue. Regions in the southern U.S., Arizona, Alaska, and a small number of additional areas trended Republican. Most other regions trended Democratic, including many states that are normally considered Republican strongholds (e.g. Idaho, Nebraska, Indiana, Virginia, North Carolina, and Florida).

An opposite trend occurred during the 2010 election cycle. Figure 4b shows the trend for U.S. House seats between the 2008 and 2010 elections. Here, the majority of the map is red, representing the strong swing towards Republican candidates during the 2010 election.

Data Collection

(a)

(b)

Fig. 5. (a) South Carolina's seven congressional districts (larger image); (b) South Carolina's 46 counties (larger image)

Presidential, U.S. Senate, and state Governor results are normally not reported by congressional district. The most common format is either a single, state-wide result (e.g. as reported by Politico for the 2016 U.S. Senate races), or by individual state counties (e.g. as reported by Politico for Virginia, a key Senate race where Democrat incumbent Mark Warner held a slight lead over Republican challenger Ed Gillespie after the initial vote count).

For 2012 Presidental, U.S. Senate, and state Governor races, we were unable to locate any existing breakdown of votes by congressional district. If each county fell entirely with a single congressional district (i.e. if congressional district boundaries always fell along county boundaries), the problem would be fairly simply. Unfortunately, many congressional district boundaries are now drawn to try to produce a specific breakdown of expected voters, so an individual county can overlap multiple congressional districts (e.g. as in Fig. 5 for South Carolina). In order to estimate the district results, we applied the following strategy:

1. Collect vote tabulations by county from any of the available data sources (e.g. CNN or Politico).
2. Determine which counties overlap each Congressional district, and the percentage of that overlap.
3. Use the county-district percentage overlap to assign a corresponding percetange of each county's results to the appropriate Congressional district(s), then aggregate to produce estimates of President, U.S. Senate, and Governor vote totals by district.

Another complication is that certain New England states (specifically, Connecticut, Massachusetts, Maine, New Hampshire, and Rhode Island) report results by community rather than by county. This requires one additional step to determine which county each community belongs to. Numerous online sources are available to determine how communities map to counties. Community results can then be aggregated by county, and finally by district as we require.

Since U.S. Senate and Presidential elections occur together with the U.S. House elections (i.e. Presidential elections at four-year intervals, and U.S. Senate elections spread over six years in two-year intervals), voting data is readily available. Unfortunately, although some state Governor elections overlap with U.S. House elections, others do not (e.g. the most recent Gubernatorial elections in New Jersey and Virginia occurred in 2013). "Off-year" state Governor results by county were collected from various state agencies, for example, from the Louisiana Secretary of State for Louisiana's 2007 gubernatorial election.

Election Visualizations

Fig. 6. District and state-wide visualizations for all 50 United States (larger image or high resolution PDF)

Alabama	Alaska	Arizona	Arkansas	California
Colorado	Connecticut	Delaware	Florida	Georgia
Hawaii	Idaho	Illinois	Indiana	Iowa
Kansas	Kentucky	Louisiana	Maine	Maryland
Massachusetts	Michigan	Minnesota	Mississippi	Missouri
Montana	Nebraska	Nevada	New Hampshire	New Jersey
New Mexico	New York	North Carolina	North Dakota	Ohio
Oklahoma	Oregon	Pennsylvania	Rhode Island	South Carolina
South Dakota	Tennessee	Texas	Utah	Vermont
Virginia	Washington	West Virginia	Wisconsin	Wyoming

Fig. 7. District and state-wide visualiztions for each of the 50 United States (click on the state or its legend to see a larger image)

Future Work

Although this project describes visualizing U.S. election results, our technique is not restricted to only this type of data. For example, we are now looking at data from the 2010 U.S. Census, in the hope that a similar strategy can be used to visualize census data for individual counties and as state-wide aggregates. We are also studying whether we can apply our election visualization system to different countries, for example, the current minority government in Canada, or the contested Presidental elections in Mexico.