This is a guest post by Charles Stewart III:
There has been a lot of interest in how the 2012 Iowa Republican Caucus results compare with 2008, but in a non-systematic, or at least inefficient, way. A good example of a an inefficient comparison method was watching John King late Tuesday night on CNN rapidly flipping back-and-forth between the 2008 and 2012 Iowa maps that showed counties colored according to who won each county in each of these years. As Ed Tufte pointed out a long time ago, what John King really needed was some scatterplots.
Here they are.
The 2012 Iowa Republican caucus can be thought of as a competition between three wings of the party: the business wing, the social wing, and the libertarian wing. Two of these wings, the business and libertarian, were represented by one candidate each, Romney and Paul, respectively. The third wing was represented by a range of candidates, including Santorum, Bachmann, and Perry. Using precinct and county results, we can see the degree to which the caucus results mirrored 2008.
The easiest comparisons between 2012 and 2008 concern the results garnered by Romney and Paul. The following graphs depict the percentage of caucus-goers who supported each candidate in 2012 (y-axis) and 2008 (x-axis) at the county level. The area of the circles are proportionate to the number of total votes cast in 2012; the diagonal lines show where the 2012 percentage equals the 2008 percentage. You can click the figure to see a larger version.
Romney:
Paul:
Not surprisingly, Romney’s and Paul’s support in 2012, relatively speaking, came from roughly the same places as in 2008. At least the correlations across counties are high .68 in the case of Romney and .69 in the case of Paul. However, that’s where the similarities end with these two repeat candidates.
Although Romney’s overall statewide vote share was roughly the same in 2012 (24.6%) as in 2008 (25.2%), the scatterplot shows that there was a shift in the geographic distribution of that support. In particular, a big gain occurred in Polk County, the largest county in the state, which was offset by a reduction in vote share in 70 of Iowa’s 98 other counties.
Paul, on the other hand, did better everywhere, although he tended to do “more better” in the counties where he had been strongest in 2008. (Nonetheless, the surge in support for Paul in Polk County, where he did moderately well in 2008, is also evident in the scatterplot.)
What about the other candidates? Santorum, Perry, and Bachmann could fairly be said to be heirs to Mike Huckabee, drawing on his social conservative voters. Thus, it’s not surprising that their county-level vote shares are all highly correlated with Huckabee’s: .50 for Santorum, .53 for Perry, and .51 for Bachmann. Here are the scatterplots:
Santorum:
Perry:
Bachmann:
Because of the interest in comparing each of these three candidates directly to Huckabee, the magnitude of the correlation coefficients is masked by the scale of these graphs. What stands out in these graphs, in fact, is just how far each candidate came from matching Huckabee’s vote totals in any of Iowa’s counties.
However, the shared appeal of these candidates among Huckabee’s supporters (geographically speaking) is quite apparent if we sum the votes received by Santorum, Perry, and Bachmann, creating a composite candidate I’ll call “Santoperrymann”, and then graph that composite vote total against Huckabee’s vote in 2008:
Santoperrymann:
The Spearman correlation coefficient here is .70, the large coefficient described by any of these scatterplots.
This graph points out one thing that I haven’t seen commented on yet about the caucus results: the combined votes of the Sons and Daughter of Mike mostly outpaced what Huckabee received in 2008. If nothing else, there is evidence here that the Tea Party — or at least social conservatives — were a slightly larger share of the caucus-goers in 2012 than in 2008. Their problem was that they just couldn’t unite around a single candidate, as they did four years ago.
About Newt Gingrich: the county-level correlation in Gingrich’s 2012 vote with all candidates from 2008 is paltry in comparison to anything we’ve seen here: the largest correlation is a negative one, with Paul’s vote (-.09). Here is that scatterplot:
Nothing of interest here. Time to move on.
If journalism is the first draft of history, then graphs such as these are the first draft of social science. Clearly, nothing in these graphs establishes causality, nor do they describe the persistence of behavior at the individual level. Still, they establish a strong persistence in voting patterns between 2008 and 2012, despite the fact that when we color the counties according to who won a plurality of caucus-goers, the two maps look quite different. Unlike most commentary on the caucus, these patterns suggest that the Tea Party, or at least social conservatives, were actually more of a factor in 2012 than in 2008, at least to a small extent.
[A brief note about John McCain. Despite the fact he was the eventual nominee, McCain came in fourth in the 2008 Iowa caucus, with 13.0% of the vote. Thus, McCain has not been analyzed here. Interestingly enough, despite [or maybe because of] harsh words back then between Romney and McCain in 2008, Romney’s geographic support in 2012 is also highly correlated with McCain’s 2008 support. Consider the following set of regressions, performed using the precinct-level returns, predicting Romney’s 2012 vote share in terms of Romney’s and McCain’s 2008 vote share:
(Standard errors in parentheses. Observations are weighted by 2012 precinct turnout.)
No wonder McCain’s endorsement of Romney came on the heels of the caucus.
