Some interesting numbers coming out of the primary vs. general election data.
This is a very crude plot of voteshare gains/losses by Trump in 2016, relative to Romney in 2012 (red dots) vs. voteshare gains/losses by Clinton in 2016, relative to Obama in 2012 (blue dots). Not terribly useful since many primaries were not very competitive.
Since most Midwestern primaries were competitive, we might narrow the primaries to the states in the Midwest, which gives us the following (Not sure if I got all the Midwestern primaries):
The difference is a bit starker: Trump did gain votes relative to Romney consistently across the Midwest. Clinton did lose votes relative to Obama consistently, compared to the rest of the country. But, even though the slopes are quite shallow, the graphs suggest that, conditional on the voters (or geographies, more accurately) that changed their behavior in 2016 relative to 2012, the extent of the shift was smaller in the areas where support for Sanders was greater.
Now, we want to narrow the sample down even further, to whiter areas (80% or more) and poorer areas (county median income below national average).
Still the same pattern, even if starker. reflecting the fact that Trump did particularly well with the poorer, whiter demographics, while Clinton did especially poorly with them. But also persisting is the relationships with the Sanders performance in the primaries: the better Sanders did in the primaries, the smaller the gains by Trump and the losses by Clinton, even if the magnitude of the effect is relatively small.
The last point provides an interesting counterpoint to some speculations that rose in the immediate aftermath of the election, including by myself, that Obama votes migrated to Trump in the Midwest via Sanders. Nationally, there is a slight evidence that this may or may not be true: larger Sanders fractions ARE associated whatwhat with vote losses by Clinton, but national figures make for lousy data given the uneven nature of the primaries (a lot of primaries took place when they don’t matter much). Looking at the correlations state by state in the Midwest, the only state where more support for Sanders predicts greater dropoff in Clinton votes is Pennsylvania. All others show either insignificant relationship or slightly positive correlation, as seen above–more support for Sanders leads to smaller dropoff in Clinton voteshares.
So, in a sense, this is more interesting than not: Sanders voters, at least seen at the county level, seem to have been more loyal as Democrats than not: many of them may have looked like Trump voters (less educated, working class, and whiter), but they were actually less likely to be converted to Trump. This implies an interesting array of possibilities for the future of Democrats, including the possibility that Sandersism might be even better path forward for them than I had thought before.