In an earlier post, I noted that, presence of many Sanders primary voters in the Midwest did not lead to votes losses for Clinton in 2016 relative to Obama in 2012, except in Pennsylvania. Of course, Pennsylvania is not technically part of the Midwest, but the Northeast. Also, it is worth remembering that, even if states like New York, New Jersey, or Connecticut did not flip, Trump did draw far more voters in these states than Romney, or indeed, any other Republican did, while Sanders performed respectively in these states’ primaries. It seemed worth asking what the relationship between Sanders voters and Trump voters looked like in these states.
For reasons that are not obvious, however, the county level vote data that I had obtained from Kaggle are terrible, full of missing data, especially for Northeastern states. The only state that provides a reasonably complete set of county level data for both primary votes and presidential votes is New York, so I plotted the same numbers there as I did for Midwestern states.
The pattern here is exactly the opposite that in the Midwest: more Sanders votes there are, the bigger the drop off between Obama in 2012 and Clinton in 2016. So while the Sanders voters were, if anything, more loyal to Clinton than non-Sanders voters (certainly not responsible for Clinton losing Michigan, Wisconsin, Ohio), Sanders voters in New York (and presumably, also in New Jersey and Connecticut–but for some reason, these states were missing way too many observations in the Kaggle data I had on hand.) did abandon Clinton in large numbers.
An even more instructive plot is where the data is divided between fairly wealth counties (average median income greater than $50,000) vs. not so wealthy ones (below $50,000).
Red dots are the relatively wealthy counties and the blue dots are less wealthy ones. The dropoff is systematically greater for the less wealthy counties–at least as far as New York is concerned.
Hardly definitive proof, but certainly consistent with the argument put forward by Andy Gelman when he examined the linkage between wealth and partisanship in different regions of the country: in the poorer states i.e. the South, the state’s average population may be more Republican, but there is a clear relationship between wealth and partisanship–the poor are more Democratic and the rich are more Republican, even if everyone may be more Republican than their counterparts with similar incomes in the North; in the richer states, i.e. the North, the state’s average population may be more Democratic, but the wealth-partisanship relationship is much weaker. The poor tend to be relatively more Republican and the rich more Democratic–although nowhere so much that the rich are more Democratic than the poor. It seems that (although Gelman did specifically point to Connecticut as an example of a state where the rich-poor-Democrat-Republican linkages were blurred) Gelman’s argument applies more to the coastal Northeast than the Midwest, and Trump, not being a more stereotypical rich and/or Southern Republican, could draw off Republican-tending poor voters much more effectively than they could.
One has to imagine that the causal mechanism is exactly that pointed out by Thomas Frank in his books, What’s the Matter with Kansas and Listen, Liberal. The catch seems to be that Frank got his geography wrong.
interestingly, much the same pattern seems to hold in the other coast–the following graphs capture the same pattern in California.
Without differentiating between poorer and wealthier counties, more Sanders votes are indeed correlated with bigger dropoffs for Clinton.
When differentiated by income, the pattern replicates that of New York.
Of course, this needs to be taken with giant grains of salt: some counties in California are gigantic–Los Angeles, Orange, San Diego. Others are tiny (Inyo, Stanislaus, etc.) Still, we are looking at rough patterns at this point.