Wendy Tam Cho and Jim Gimpel have a new piece at The Forum that analyzes county-level presidential voting patterns in 2008. They make two main contributions to our understanding of this election. First, they draw on aspects of economic hardship, including foreclosures and gas prices, that are rarely incorporated into these sorts of analyses. Second, their model — a geographically weighted regression — treats the effects of any factor as heterogeneous across counties. In other words, they can see where gas prices, or foreclosures, or unemployment, or whatever had a larger impact.
Here’s an example. The map below graphs the effect of John Kerry’s vote share on Obama’s. The darker areas indicate where there is a stronger relationship between the two — that is, where Kerry’s vote share better predicted Obama’s.
Cho and Gimpel write:
bq. Accounting for other influences on the county-level vote, we find that the support for John Kerry and Barack Obama was most closely related in the Mountain states, in very Republican areas of the Plains, and throughout most of the West. Correspondence between the two candidates’ vote percentages also ran high in a few other locations where presidential party support exhibits continuity, including the Wisconsin lakeshore, in New England, and in durable Republican pockets in South Carolina and Georgia.
bq. There was also a large territory centered on Arkansas, Louisiana, and Mississippi, shaded in white, where there was less of a relationship between the Kerry and Obama votes (and the Bush and McCain votes). Lighter turnout among ardent Republican populations was most likely responsible for the weak relationship between aggregate presidential preference in 2004 and 2008 in the South and Border States, as well as within some politically conservative pockets of Iowa, Indiana, and Ohio.
Here’s another map capturing the varying impact of unemployment:
They write:
bq. Several noteworthy dark patches in battleground states show locations of Obama strength as a result of the early onset of recession. These include large parts of states such as Pennsylvania and Florida, as well as Indiana, Ohio, Michigan, Wisconsin, Minnesota, and Iowa. Much of Northern New England, again, moves toward Obama as a consequence of rising jobless claims. The conspicuous dark shading through the entire Ohio River Valley contributed to significant victories in Indiana and Ohio, adding between 0.04% and 0.35% to the Democratic vote share for every single percentage point rise in the number of unemployed workers.
bq. Yet there were also locations where the rising level of unemployment claims may have marginally boosted the Republican ticket—largely less populated locations in the West and South, shaded in white in Figure 6. According to our estimates, McCain gained between 0.05% and 0.17% of the vote for every one point increase in jobless claims at these locations. To the extent the Republican candidate may have gained in a few places, however, he did not gain by as much as the opposition did elsewhere.
This is all very interesting. We need to identify heterogeneity — across states, counties, individuals — in the factors that drive voting behavior. But obviously these findings do raise many questions. There are, as yet, no real theories as to why certain factors should matter more in some places than others. These maps are primarily exploratory.
And I’m puzzled by how the well-established relationships between economic conditions and voting behavior are reversed in some places. For example, why in some counties is higher unemployment associated with a McCain vote? Typically, the incumbent party is supposed to suffer, not gain, when the economy is weak. This is especially puzzling where unemployment has opposite effects on vote share within the same state, as in southern and northern Michigan.
The upshot: I love these maps. Now I just want to understand them.