Daily Kos Elections does a lot of good by compiling information about the politics of different places all over the country. But there's a real danger if we focus too much on all of those differences. The danger is that the considerable lack of difference between different places gets lost in all of that information. While there are definite regional trends and variable effects, most (really, nearly all) of the results of the 2012 election are explained by a very simple equation once you account for the 2008 election. Just look at most of the Huffington Post's election results pages and you'll see what I mean, but if you like, we can take a closer look.
Note: The state/county approach is not my preferred one, but it's what's most immediately available. Also, I know the scatterplots are small, but there's a lot to include. If you click through and make them their original sizes or download them (the files are all quite large), you should be able to read the labels.
And I tried to get 2012 election results from state sites, where possible, but sometimes I used the U.S. Election Atlas. Sometimes they're election night results, other times they've been updated. Click the specific links if you have any questions.
Introduction:
Lately, on DKE, I've been trying to push back against the idea that demographics and demographic change are the big story of 2012--because, in part, I don't think there's all that much to explain about 2012. Different people might disagree with me about that point, but hopefully everyone can find some of this information useful.
Let's begin with a national perspective. Here's a scatterplot of the Democratic share of the two-party vote, by state, for both 2008 and 2012, via Dave Wasserman (itself via Plain Blog):
As you can see, there's a very strong overall pattern. In technical terms, the r-squared rounds to 0.97, and the best-fit line is:
Obama2012 = 1.05*Obama2008 -4.86.
What does that mean?
More polarization: Obama's vote share dropped by a bit more in places where he did badly last time, but held up well in places where he did well last time.
On the other hand, uniform swing, which would be something like, say
Obama2012 = Obama2008-2.05
is very close to that line, especially in states that aren't extreme (they're equal at about Obama2008 = 56%). We'll stick with the first line, but a couple of the later graphs will demonstrate how similar the two cases are.
Here's a zoom in to around the 40%-60% range:
Perhaps you can see why I don't think Texas is "trending" Democratic to a particularly significant extent, or that Obama's advantages in the swing states were all that great relative to 2008.
The 2004-2008 fit was considerably worse:
The best-fit line was
Obama2008 = Kerry2004+4.87
but the fit was just 0.88--nearly 10 points worse than the 2008-2012 shift.
I think the dramatic 2004-2008 changes gave a lot of support to a kind of "demographism"--the (caricatured) idea that changing demographics would change the electoral map in a continuous and irreversible way. But 2012 didn't vary from 2008 the way 2008 varied from 2004, and we should consider the possibility that some extent of the 2004-2008 changes were about Barack Obama--raised here by John Sides, which I found originally via Plain Blog and again via Sean Trende. That would explain why there was a strong shift when Obama got onto the ballot, but not a strong shift as he remained there, even though the electorate presumably continued to diversify from 2008 to 2012.
States:
Now let's add some very different data: Obama's 2008 and 2012 performances in Kentucky, by county:
Most of Kentucky's county results fit right in around the same best-fit line as for the states, and the biggest exceptions (click through for a bigger image, or see my zoom in) are nearly all in eastern Kentucky, where Obama did considerably worse than the overall pattern. This almost certainly helped doom Ben Chandler, whose supposedly "shored-up" district included many of those outlier counties (although I'm not sure how populated they are). But it couldn't have helped.
Let's add Iowa from this time and last time, with the labels in blue:
The best-fit line for Iowa's counties?
IA2012 = 1.05*Obama2008 -5.25.
0.95 r-squared, and very similar to the state results, if a tiny bit more of a drop-off.
Let's throw on three more swing states: Florida, Ohio, and Colorado. Florida will be in green, Ohio will be in red, and Colorado will be in orange.
Florida results this time and last time from those links, Ohio results from last time and this time from those links, and Colorado results from this time and last time from those links.
Our graph is a bit crowded, so I'll include three new ones (scroll down for a cumulative zoom-in, though).
The best-fit lines for the three states are:
CO2012 = 1.01*CO2008 - 2.68.
OH2012 = 1.06*OH2008 - 4.5.
FL2012 = 1.01*FL2008-1.86.
With fits of--respectively--0.99, 0.96, 0.98.
We can see that things are a bit different for Colorado and Florida--the very best fit line is closer to uniform swing, suggesting those states had a bit less polarization and a bit more uniformity--but, as the graph makes apparent, the overall "2012 = 1.05*2008 - 5" equation still fits it pretty well.
To emphasize how small a difference this is, here are three graphs with just those swing states, plus the states, plus the best-fit line for each:
One last one: Texas, from 2008 and 2012. I include this because Texas is a perpetual topic of DKE discussion, and it also has one of Obama's largest over-performances, in tiny Reeves County:
As we can see, Obama really did over-perform in a few Texas counties, some of which--at least Webb County and Hidalgo County--are decently populated. As far as I can tell, they're all South Texas/border counties. So demographics really might be changing those regions. But otherwise, there doesn't seem to be much evidence of a trend.
(Also note that my decision to cut the graph off at 10% Obama is actually cutting off a few Texas counties!)
The Texas best-fit line is:
Texas2012 = 1.08*Texas2008 - 5.66.
Again, very similar, maybe a bit more polarization.
Conclusion:
Does that original best-fit line hold up? Let's plug it all into a single regression: States, Kentucky counties, Iowa counties, Florida counties, Ohio counties, Colorado counties, Texas counties. 743 data points, by my calculation.
We get:
Obama2012 = 1.04*Obama2008 -4.68.
The r-squared is 0.97. That's very high! And it's not very different from the original equation.
Here's the scatterplot for everything so far, with the original "50 states" best-fit line, the new best-fit line, and Texas in grey:
Looking across (the counties of) four of the biggest swing states and the very different Republican states of Texas and Kentucky added very little to what we knew from just looking at the 50 states. And what that says to me is:
There was rather little regional variation in the 2012 election once you account for the 2008 election.
So yes: Obama did well in South Texas, in Miami-Dade County and Osceola County, and in Alaska. And he did very poorly in eastern Kentucky counties, in Utah, in Monroe County OH, and in West Virginia. But beyond that, there's not much evidence of strong "trends"--from demographics or otherwise--that I can see in the 2012 Presidential election, at least in the counties I've looked at so far. Overall, either uniform national swing, or a slight variation to account for increased polarization (some of which might come from, yes, demographic changes) would seem to explain the vast majority of the 2012 election results.
That doesn't mean there's no story in the 2012 elections. I included "almost" in my diary title for a reason.
First of all: This county/state approach might be missing something. I'll hopefully look at it again once we have results by congressional district, or by state legislative district for a particular state, since those divisions are far more equally populated.
Although, I realized after I wrote most of this diary that there were some 2012 results available by Congressional District, thanks to David Nir and GradyDem: some 65 districts, actually (some of which have provisionals and mail-ins, some of which don't). With the 50 states plus D.C., the equation is:
Obama2012 = 1.04*Obama2008 - 4.08
with a fit of 0.98.
Second of all: I have found what might be some other statistically significant factors--yes, demographics, as well as campaign boosts and more--when looking at some of this information. There's plenty of work to be done in understanding those and in estimating their sizes and so on. And even in the above scatterplots, there really is variation left to be explained.
But we need to keep it in perspective.
What the strong overall fit does mean is that there really isn't all that much of a mystery left for other factors to explain. Not nothing--but not all that much. In my opinion, discussions of local trends, local campaign effects, and so on, should always keep the powerful and consistent national picture in mind, as well as the apparent relative sizes of their effects.
12:02 PM PT: I gave myself a deadline to publish this one by 3:00 today--so hopefully there aren't any mistakes (there always seem to be mistakes). Let me know.
12:10 PM PT: The emphasis on scatterplots, and on the outlier of Miami-Dade was also in this Hans Noel post, via Plain Blog. And the above John Sides post had this great summary line:
What is most remarkable about 2012 is not its radical change but instead enduring stability—very modest shifts in state outcomes relative to 2008, relative even to 2000.
