I posted some maps the other day showing estimates of who would’ve won each state in the recent presidential election, based on a multilevel model of state and income categories, fit to data from the 2008 Pew pre-election polls.
I realized that the maps could be usefully augmented with scatterplots showing the estimated outcome in each income group by state: that way you can see which states were (estimated to be) landslides among different income groups, and which states were estimated to be close.
Here are the results (click on the image for a larger version):
As I noted before, the estimates are based on data from five income categories (for simplicity I display the highest, middle, and lowest groups here; the results for categories 2 and 4 fall pretty much in between) and, for each state, the poll-based estimate was shifted to be consistent with the actual election results there. Also, the low-income map changed slightly, with West Virginia and Utah very slightly edging over the 50% point for John McCain. These changes occurred because I changed the model slightly, adding region as a predictor in the model.
Any inference based on a survey sample will ultimately be somewhat model-based (I’m sure Nate Silver’s audience is familiar with this point); here, I’m doing my best with a quick analysis. One useful thing about the scatterplots is that it puts less of a burden on the sharp red/blue distinctions in the maps. (Yes, I know that I could use some purple to indicate close states, but that in turn would make the maps more difficult to read. So I compromise by making the maps sharp and then using the scatterplots to show more detail.)
Also, in response to some of the comments on the earlier post:
– Yes, we’re looking at other variables such as sex, ethnicity, age, and education. I’m pretty sure the patterns by education and sex will be much less dramatic than those by income, ethnicity, and age.
– The polls did not include Alaska and Hawaii, so any inference for them would be entirely model-based or else extrapolated from other poll information. In Hawaii, we can safely assume that Obama won among all income groups, and in Alaska, I don’t know that I’d trust the polls there anyway.
– Regarding Robert’s comment that “the map based on rich Americans only corresponds so closely to the image of the US presented by pundits TV talking heads”: Exactly. That’s a key point of Red State, Blue State.
– I’ll post the 2000 and 2004 maps too, at some point so you can see the changes over the three elections.
– Indeed, making $150,000 in New York does not make you so rich, after correcting for cost of living, as making $150,000 in Mississippi, and differences between states (or between counties within states) must be interpreted with that in mind. However, the data show that in almost every state, higher-income people are more Republican than lower-income people.
– The average income by state comes from the Census. I can’t quite remember if these are numbers from 2004 or even 2000, but it doesn’t really matter for the graphs, because the rankings of states by income haven’t really changed.
– Mary asked for a breakdown by religion. Some of that I posted last week.