Wishing for better maps shortly after the election, I took matters into my own hands and analyzed the election results by county using the same Mapping Toolbox that we sell to customers. After a few evenings of work I had produced some maps to help me better understand the so-called red/blue electoral divide.
The maps spawned some interesting discussion on our company-internal, recreational “talk” newsgroup—sadly it has become rather partisan and is at risk of being shut down. Some of my map-based analysis from that forum appears below. I was prompted to post this now, more than a year after the election, by a recent comment on data visualization guru Edward Tufte’s web site.
(Note to cartographers and others: I went for the easiest maps that I could make. I know that there are better map projections that show area more accurately. In general, better maps would include a legend, too. Shape data for the maps came from the U.S. Census Bureau. I obtained most election results from the USA Today web site, though New England reported results by town (!) which lead me to Polidata.com, The Bangor Gazette, and various Secretaries of State’s offices. Thanks to Kelly Luetkemeyer for Mapping Toolbox assistance.)
This map shows the absolute difference in votes won by Bush (red) and Kerry (blue). It’s a logarithmic scale designed to accomodate the wide variance in county populations. White counties were decided by fewer than 100 votes for either candidate; light pink/blue, 101 – 1,000 votes; medium red/blue, 1,001 – 10,000 votes; and deep red or blue counties were decided by more than 10,000 votes.
The purple lines are major urban areas according to the Census Bureau. The result is interesting to me. Conventional wisdom holds that Kerry did well in urban areas and Bush did well everywhere else. The truth seems to be that Kerry did well in urban “centers” but couldn’t penetrate out into the suburbs, which went overwhelmingly for Bush. In fact, Kerry’s largest margin (L.A. county) is right next to Bush’s best county (Orange).
The big pockets of blue throughout the country are also interesting. They are mainly in rural areas — Rocky Mountain towns, Indian reservations, border communities, agricultural areas in the Midwest and South — and the faltering industrial belt of the Cotton Kingdom. While scattered, these pockets or ribbons suggest that both parties could pick up votes by going to out-of-the-way areas traditionally ignored by the candidates.
It’s also interesting to note that Mono County, California was a dead heat: 2,200 to 2,200. Kerry did not win any counties in Utah, Nebraska, and Oklahoma. Massachusetts and Rhode Island were the only states where Bush didn’t carry any counties, though the total vote margin shows that Bush took most other counties with a relatively modest number of votes.
Another map tells the electoral story from a different angle, the percent difference between Kerry and Bush. White counties have a 52-48% or closer result; light pink or blue 56-44% or closer; medium red and blue 60-40% or closer; and deep red or blue represents 64-36% or better for the winning candidate, roughly 2/3 of the votes.
While the same pockets of diversity appear, the Northeast looks much weaker for Kerry and the rest of the country much stronger for Bush — except for the suburbs in major urban areas, which are closer to 55-45 or 60-40 — than the other map might suggest.
Bush’s strength throughout most of the country was the overwhelming thing I took away from the percent map. Imagine taking a road trip across the country and asking people everywhere you stopped how they voted. On just about any route you picked, the result would be that most people would say, “I voted for Bush and 2/3 or more of my neighbors did, too.” To me, that sounds like a mandate . . . of a sort.
I believe in the U.S., you can consider counties to be a surrogate for communities. Small counties with high population densities and large counties with low population densities might spatially distort community distributions; nevertheless, in both cases communities within counties tend to resemble each other.
Of the 3,399 counties, Bush had a majority in 2,553 (75%) and a large 2/3 supermajority in 1,129 (33%) of them. Kerry on the other hand had a straight majority in only 598 (18%) and a supermajority in 80 (2%).
Entire population areas of this country wanted Bush to win (or didn’t want Kerry to win). Now Kerry did well in a select few places, and those places had a large number of votes, but most areas favored Bush by large enough margins for me to say that most communities around the U.S. actually want [or wanted] Bush to govern.
To be sure, if you were to spread voters out over the whole of the country — or shrink them together as some maps do — the results would be a very divided 51% of the ubercommunity voted Bush and 48% Kerry. But these maps, while interesting, distort the power of counties to convey community.
If all of the urban voters were slathered across the West and South, the West and South’s communities would have a different composition and new concerns. If all of the rural/suburban electorate were bunched up against the big cities, these voters would have a much different perspective. Votes would change.
If you take a random, equal sample of voters from each of the nation’s communities (i.e., points of view), Bush wins with a large mandate.
I’m not trying to say that an individual’s vote doesn’t matter if she lives in one place and not in another. In fact, that’s my general opposition to the electoral college: it effectively removes minority opinion voters’ voices from the political process in “solid” or “secure” states and violates the principle of one person, one vote.
As a snapshot of the American zeitgeist in late 2004, it’s interesting to look at deviation from the middle, as represented by a 51-48% election result.
The margin in white counties deviated by less than 4 percent from the norm; light blue/pink, less than 12 percent; medium blue/red, less than 20 percent; deep blue and red counties had deviations of more than 20%.
Perhaps it shows more conservative and liberal areas. Perhaps it shows differences in partisanship. Perhaps it only shows approval patterns for this one election.
Standard deviation — a hefty 13% — might be more interesting in showing “solid” areas for each party. White areas are within the first standard deviation; subsequent shades are another 13% standard deviation each.