A student at another university writes in with some questions about Red State, Blue State:

1) The data used in the studies was obtained from election surveys and census interviews. Regarding the census data, how appropriate is it to use overall census income data in this study. Wouldn’t it be better to just determine the average income of voters in a state by averaging responses from the same surveys used to elicit voter choice? Is there a known relation between income and likelihood of voting in general? If so, is it constant, or segmented in such a way that a multiplier(s) can be used to normalize the study to generalize it to all Americans? Would that exacerbate the disparity (assuming the rich are more likely to vote)?

2) With regards to the multilevel logistic regression models, in you’re your paper “What’s The Matter With Connecticut,” what was the reasoning behind the fixed-sloped, varying-intercept model? How was the fixed-slope value determined?

3) Was the income based analysis done for the 2008 election at any levels besides the individual voters? Specifically how would the multilevel logistic regression (varying-slope, varying-intercept model) compare in the 2008 election against the 2000 and 2004 elections. Comparing 2000 with 2004 (Fig. 4- What’s… Connecticut?)* the correlation between income and voter choice seemed to be decreasing in the poor states and increasing in the rich states, possibly indicating that the discrepancy in the effect of income in different states was disappearing. Did this continue in 2008? Did the economic crisis further sensitize all Americans equally, making the slopes across the states converge (and possibly increase)?

On a second glance Figure 4-4 in the book, seems to show the opposite results- with Connecticut showing a slightly smaller correlation in 2004. But Figure 5 in the paper, shows the values of the slopes of all the states converging to some extent.

4) Obama clearly presented voters with a racial/minority bias. You mention that the Democratic Party as a whole saw similar gains (especially in the House elections). Is it possible that Obama inducted more voters into party voting or that the Democratic Party garnered new support from minorities as a response to the party supporting a minority candidate? Is there any way to test for that?

My reply:

1. We use survey responses on income. Since writing the book, we’ve been working with data from the Current Population Survey’s post-election supplement, which I think is considered the best source for demographic breakdowns of voters. I don’t think this would have a huge effect, so we didn’t redo our Red State, Blue State analysis. But, going forward, I’ll be using these numbers.

Yes, higher income people are more likely to vote. We’ve been making graphs of this recently ahd hope to complete our research article and post some graphs on the blog soon.

2. We fit the varying-intercept, constant slope model as a starting point, because it was simpler to fit and to understand than the varying-intercept, varying-slope model. But once we allowed the slopes to vary, we never looked back. At this point, the constant slope model is nothing more than a point of comparison.

3. We fit our model to 2008 and got similar results to 2000 and 2004. We posted some of this on the blog, for example here.

4. Yes, it is possible to look separately at turnout and vote choice, comparing 2008 to 2004. We’re in the midst of doing this now.