Here’s the graph that David made showing the Republican share of the two-party vote for president since 1940, for states in the upper third and lower third of per-capita income:
It used to be that rich states voted Republican, now they go for the Democrats (the famous red-blue map). The voting gap between rich and poor states has gradually widened since the early 1980s.
And here’s the plot comparing upper and lower income voters:
Rich people are much more Republican than poor people. Differences in voting by income have returned to 1940s levels.
Pulling out the South
We also did separate analyses for southern and non-southern states, since the South is poorer than average and has also moved steadily from the Democrats to the Republicans over the decades. First, a plot showing the difference between rich and poor states over time, overall and in southern and non-southern states:
And now the differences each year between rich and poor voters in the country as a whole and in south and non-south:
We used the Republican share of the two-party vote (for the state analysis in 1948, including Thurmond’s votes as part of the Democrats’). The state election data are public information and easy to find, for example from David Leip’s atlas.
For each election year, we defined rich and poor states in each election year as follows. We first sorted the states by per-capita income using data (from the Census, I think) that Justin Phillips gave us. We then aggregated by population from the top down and the bottom up, to construct a collection of states at the high end whose total population approximated 1/3 of the U.S. population in that year, and similarly for the low end. We rounded down to get rich state and poor state groupings that each had no more than 1/3 the population for that year.
When making the plots for states, we pooled the popular vote within each grouping (rich states and poor states) in each year. For individuals, we took the respondents from the National Election Study (since 1952) and data from Gallup polls prepared for us by Adam Berinsky and Tiffany Washburn (for 1940 and 1944). We don’t have individual level data for 1948, since our National Election Study data didn’t have state identifiers for respondents in that year.