Wednesday, October 06, 2004

Red and Blue: Mapping the Votes

I've always been very curious about the exact geographical distribution of votes. We've all seen the electoral college maps like this one (or this, this and this). But obviously each state is not a simple red or blue, it is some kind of shade, and it is not a uniform shade either. Is it a light-blue or is it a spot of dark-blue in a sea of gray? Is it flat gray or a mosaic of red and blue? Inquiring minds want to know.

Unfortunately even county-by-county maps like this do not truly answer the question, since we're left to wonder whether a county was won by 1% or by 20%. We need a more, ahem, nuanced approach.

So: here's a map of county-by-county votes of the 2000 election results as shades of color (red=100% Bush, blue=100% Gore, gray=exactly 50-50).



(click for larger image)


As you can see, the western states are basically red and the eastern states are gray with red splotches and the occasional blue-ish stripe, and of course the coasts have a blue fringe here and there. Let this be a caution to anyone who stereotypes states: most of California is basically just like Nevada, and most of Illinois like Missouri, except for the presence of a large metropolitan area that shifts the statistics. California is not a liberal stronghold: Los Angeles is.

Next, and this is the real paydirt: a map of votes as color, and voter density as lightness/darkness (log scale, black=10,000 per sq mile, white=0; the overall colors had to be adjusted slightly so that the details are visible)




(click for larger image)


Now it all becomes clear! There is an excellent correlation between population density and vote. Virtually all blue areas are dense; there are very few if any dense red areas. Salt Lake City and Boise are the closest thing to dense red, along with a number of cities in Texas (but not the largest ones). There are certainly many neutral (Houston) or blue (Atlanta) cities that have a red suburban fringe (a common pattern), but they are not red cities per se.

Numerically, voters in dense ares (over 500 per square mile, about one quarter of all voters) voted 61% for Gore. Voters in sparse areas (under 50 voters per square mile, again about a quarter of the total) voted 58% percent for Bush. Voters in super-dense areas (over 1000 per square mile, 1/8 of all voters) voted 66% for Gore. You can look at a scatter graph of density vs votes that shows this effect perfectly. Density by itself accounts for about half of the variation in voting patterns.

Hypothesis: the perception of the Democrats as the "bi-coastal" party and the Republicans as the "flyover" party is wrong. Rather, the Democrats are a city party and the Republicans are a suburban/rural party.

Why do dense areas vote Democrat? This is not an easy question, and I hope to do more maps that show votes alongside tax rates, government spending, crime rate and more to try to answer that, but I will take a really simple stab at answering anyway. Big cities have problems, in some cases severe (e.g. Chicago). It is tempting to look for a government solution to problems, which is what the Democrats are offering. Of course, it doesn't usually work, but hey.

UPDATE: This post has gotten a huge response - thank you! Due to the high traffic the images may be intermittently unavailable, if you can't see them please try again in a few minutes. I have just obtained a complete database of demographic data by county and will have some more interesting maps for you very soon. This has also prompted a good discussion on why we see the patterns we do.

TECHNICAL NOTE: Using color to express numerical data accurately is quite hard. Color is a tricky thing, for example 50% red appears a lot redder than 50% blue is blue. To do this properly, you have to use a "perceptually linear color space" instead of RGB. In this case, I mapped the data to piecewise linear segments in the "CIE Lab" space. A whole lot more information on that fascinating subject is available here.