Tom Lee says that the “purple America” images I linked to the other day are actually misleading in their own way for a few reasons:
But also true: visualizing information by using a linear red/blue scale is about the worst way possible to make data legible to the human eye. First: our vision is logarithmic. When a photographer drags out his “50% gray” card for measuring lighting, it’s actually 18% gray. Judging by the triangular key in the corner of Vanderbei’s image, he’s just taking the percentage of vote totals and translating it flatly to 8 bit color — a 100% Republican district gets an RGB 24-bit value of (255,0,0).
The colors themselves are also a problem. As I’m sure you all remember keenly from this post I wrote in 2006, perceptual image codecs spend more bits on brightness than on color because the color-sensing cones in your eyes have a much lousier dynamic range than the light-sensing rods. We’re worse at distinguishing between levels of color than between levels of brightness. And since the percentage of the vote in any given spot on the map should always sum to 100, with negligible green (third party) contributions, the brightness will be relatively uniform (although admittedly not quite due to the perceptual differences between colors — monitor calibration and colorspace begins to enter the picture here, and is just as hideously complex as you might imagine).
He suggests instead a simple grayscale:

That shows the same rough similarity between the two elections, but highlights the geographical variation in a clearer way and lets you see where things did change: “Things are more black and white than they may seem, and certainly less purple.”
November 9th, 2008 at 12:03 pm
great grayscale and post!
http://culturedecoded.wordpress.com/2008/11/09/gopandthepress/
November 9th, 2008 at 12:16 pm
The biggest distortion is using a map which plots land area as one variable against percentage voting by party.
A more meaningful picture would plot population within a district against vote preference. There are several distorted maps around which aim to do this, but most just warp whole states.
Done this way, the big swath of red though the mid section would be seen to represent a much smaller fraction of the population than it appears.
The only place where the small state (in the sense of under populated) bias is important is in the senate, where 16% of the population controls 50% of the seats.
November 9th, 2008 at 12:22 pm
I’m with Robertdf. When NYC has fewer pixels than a rural Colorado county with fewer than 500 voters, there’s a problem. Additionally, I’d probably make 65% and up all black and 35% and less all white (contrast enhancement?) to make the changes a little more clear.
November 9th, 2008 at 12:30 pm
It looks like the map shows a shift towards the Republicans in places like OK, NE, and KS. What is the matter with Kansas?
November 9th, 2008 at 12:31 pm
Actually, the best representation would be to have a map with simply the differences. Presenting all the information just presents the eyes with a shit load of noise. The same map with just the differences between ‘04 and ‘08 would make everything crystal clear.
Probably still have to wait for that map, seeing as how the meme is “how similar”.
November 9th, 2008 at 12:39 pm
Hal,
Actually, that was the nature of the map from the NYT that Matt linked a while back, where shades of blue represented Obama increases versus Kerry and shades of red decreases versus Kerry. And, yes, I agree that captured the nature of the election the best, although a population-cartogram version would be even better.
November 9th, 2008 at 12:56 pm
I posted this yesterday too, but isn’t it easier to just look at a map of the difference between 2008 and 2004 rather than looking back and forth at the two maps? The good people at the New York Times have great “voting shift” maps that do just that:
http://elections.nytimes.com/2008/results/president/map.html
November 9th, 2008 at 1:16 pm
Actually the problem with the black white approach is that it while our eyes can detect levels of brightness over a very large range, computer moniters are unable to display at that range. This is why things like high-dynamic range lighting are used in pixar movies to create an illusion of greater brightness then can actually be displayed. I would argue that using blue and red helps create this extra contrast. More importantly red, white and blue are American colors and jackasses like John McCain who originally didn’t include these colors in his campaign signs deserve to lose. One thing adding Palin to the ticket did ad was improved red, white and blue signage, but Americans to their credit remembered his orignal signs and voted accordingly.
November 9th, 2008 at 1:32 pm
The grayscale image is definitely a vast improvement over the red-blue map perceptually, but there’s still a significant problem remaining. Our visual perception is tuned to differences in luminance – not just absolute brightness. A light gray patch on a white background will look darker than the same patch on a black background. So with this map, it’s difficult to tell where the red areas got redder versus where the surrounding bluer areas got bluer.
Of less perceptual importance, but still something that bugs me and may affect the perception, is the presence of the state outlines in 2004 and lack in 2008. It’s less glaring in the new grayscale map, but again it makes it difficult to compare the two on an equal basis.
As others have already noted, I find the maps showing the differences between 2004 and 2008 more illuminating.
November 9th, 2008 at 2:41 pm
To the extent that our eyes can perceive the color spread in the red-blue color legend that comes with the map, arguments about logarithmic vision and dynamic range are mostly irrelevant.
Much more relevant, by orders of magnitude, are the points made above:
- Maps that color geographical areas rather than population-weighted areas (or bubbles) are grossly misleading and do not accomplish what they attempt.
- Difference maps make more sense when what you want to highlight is differences.
The NYT map linked by AGT (above) addresses each of these points separately — but unfortunately stops short of doing the one correct thing: It shows (a) the percentages for each election as population-sized bubbles, or (b) the difference between the two elections as colored geographical areas, but not the optimal (c) differences as population bubbles. They probably didn’t do (c) to avoid the difficulty that population estimates changed between the two elections. But picking population in 2008, or average between the two elections, would have been the best chart.
By the way, when you go to that NYTimes map, you have to click “County bubbles” or Voting shifts” on the left to see the maps discussed here. If you display the bubbles, you can switch between ‘08 and ‘04 and thus get an animated effect that shows the difference in bubble format — but it’s not as effective as a static bubble map of the differences would have been.
November 9th, 2008 at 4:37 pm
How about this idea for graphically presenting the red – blue dynamic?
There are 435 congressional districts. Place a blue dot on the map for each (x) number of votes the Democrat receiveds in that district, and a red dot for each (x) number of votes the Republican receives. Do this so that the highest turnout district in the country gets five dots total. This will probably mean one dot for each 80,000 or so votes received per district.
For example, if Obama gets 190,000 votes in a district and McCain gets 140,000, the district is represented by two blue and one red dots. The map will literally show where each candidates are getting their votes.
You can also do this with percentage by district, assign a dot for each 20% received, this is less useful as it doesn’t show actual votes but will accentuate the Democratic advantages in the cities.
Either way you will see what looks like a pretty big solid blue dot in New York city, and a handful of red and a couple of blue dots spread out over the the Plains states.
November 9th, 2008 at 5:01 pm
Ed, did you check out the bubble charts we were talking about? Bubbles are dots of variable size. They avoid the granularity (and loss of precision) problem of equal-size dots, but they do need to overlap in densely populated regions.
November 9th, 2008 at 8:46 pm
So…we should start talking about black states vs white states?
And how the black guy in the white house comes turned several black states white on the way there?
November 10th, 2008 at 10:40 am
Looks to me as though the republican party is clearly functionally a regional party, per those pairs of maps. The thing is, that region looks NOT to be the deep south, but rather Texas on up through the breadbasket states.
November 15th, 2008 at 2:39 am
Welcome to http://www.powerleveling2000.com . We are a world class wow gold store online. We supply
cheap wow
gold to our loyal l2 power leveling and
reliable customers. you can buy really cheapwow gold here. We have mass
available stock of gold wow on most of the servers, so that we can do a really instant way of wow
gold
delivery. We can deliver your gold wow on the order in a short while. We have been an ebay power
seller and paypal confirmed sellerbuy WOW Gold
of wow gold for years.So it is securest and safest to buy wow
March 1st, 2009 at 5:54 am
viagra
It is the coolest site,keep so!
March 11th, 2009 at 4:48 am
It is the coolest site,keep so!
March 12th, 2009 at 11:40 pm
I bookmarked this site. Thank you for good job!
March 14th, 2009 at 5:30 am
I bookmarked this site. Thank you for good job!
xanax
March 22nd, 2009 at 6:21 am
tramadol
Incredible site!
April 2nd, 2009 at 5:27 am
Excellent site. It was pleasant to me.
buy cheap viagra
April 3rd, 2009 at 4:18 am
Very interesting site. Hope it will always be alive!
cheap brand pfizer viagra
April 9th, 2009 at 6:38 am
It is the coolest site,keep so! viagra