Simplistic county-level maps of election results have helped reinforce a mistaken sense of homogeneity. Inevitably, the non-metro counties of states such as Ohio are the same shade of red, while the counties containing Cincinnati, Cleveland, Columbus and Toledo look like small islands of blue.
Before the presidential election, I wrote an article pointing out that the homogeneity of “red” America is an illusion: Small and medium-size postindustrial U.S. towns routinely vote for Democrats — sometimes by very large margins. Few had noticed, because the largely rural counties in which these towns are located were often colored red on election-night maps. In fact, these counties are typically internally polarized, with a solidly Democratic downtown core around Main Street that is surrounded by Republican suburbs and rural areas. The Republican periphery has more voters who go to the polls at higher rates, and so the county is “red” overall.
This polarization only intensified in the most recent election. In this first of two posts, I take a closer look at the voting behavior of postindustrial towns in 2016.
Let’s begin with Ashtabula County, Ohio, which received a lot of media attention before the election. As anticipated, Donald Trump made very large gains and won the county. But in the city of Ashtabula, Hillary Clinton still received 58 percent of the two-party vote.
As you can see in the map below, Trump won majorities in virtually the rest of the county.
We can see a similar pattern in another area that got a good deal of media attention: the Scranton-Wilkes-Barre corridor in Pennsylvania. Clinton received 62 percent of the two-party vote in Scranton and 54 percent in Wilkes-Barre. Trump won large majorities in the surrounding higher-income suburbs and the rural periphery.
The story is quite similar elsewhere. Where the data are available, I have collected and geo-coded precinct-level data for each of the Pennsylvania and Indiana postindustrial towns mentioned in my pre-election article.
In each of the towns, Clinton won majorities, while Trump won the surrounding suburbs and rural periphery.
The idea that the United States is divided into large blue cities and a vast red middle is inaccurate. The postindustrial counties of the Midwest are extremely heterogeneous. They are polarized between town and country in a way that is similar to the polarization between “urban” and “rural” counties.
Sometimes maps do not provide the full picture. To show this division more clearly, I plot Clinton’s vote share against each precinct’s population density. You can see that relatively dense downtown areas — with rental units and smaller single-family homes — are on the right side of each graph; suburban neighborhoods are in the middle; and the rural periphery of the county is on the left. The size of the data markers is in proportion to the population of each precinct.
In each case, the more densely populated a precinct is — and the stronger its industrial history — the more heavily it votes Democratic.
In smaller towns such as Johnstown and Williamsport, Clinton won majorities in only a handful of precincts, but in larger industrial towns she won many precincts, often by large margins.
Has this relationship between population density and voting behavior changed over time?
A popular claim is that Trump’s populist anti-trade rhetoric resonated most in postindustrial towns with severe job losses. If so, we might expect that these towns suddenly started to vote more like their neighboring Republican precincts, with the graphs flattening in 2016.
To examine this possibility, I have collected and geo-coded data for recent presidential elections. For Pennsylvania’s counties, I am able to reach back to 2000, and in Ohio and Indiana, to 2008. Rather than presenting each scatter plot, I fit a curve to the precinct-level data using locally weighted regressions. We can see Clinton’s share of the two-party vote, captured with blue markers in the graphs above, as a blue curve. Al Gore’s 2000 share of the two-party vote is in green, and John F. Kerry’s in 2004 is in orange. Barack Obama’s 2008 vote share is in black, and his 2012 vote share is in red.
We can see that the blue line for the 2016 Democratic vote is no flatter than those for previous years. In fact, the relationship between population density and Democratic voting has only gotten steeper.
A big story of the 2016 election is that the blue line for Clinton’s vote is much lower than Obama’s red (2012) and black (2008) lines. But notice that in almost all of the graphs, the gap between Clinton’s vote share and Obama’s grew the most on the left side of the graphs — in the most rural places. Clinton lost fewer votes in the postindustrial towns than in the Republican-leaning surroundings.
In several counties, this decline in rural support for Democrats was foreshadowed by an asymmetric rural decline for Obama from 2008 to 2012.
Clinton lost crucial Midwestern swing states in large part because of a significant collapse of Democratic support outside of major city centers. But even in what look like “red” counties, Democrats maintain a significant base of support — despite Clinton’s historically poor performance — in the cities and towns that developed an industrial working class in the early part of the 2oth century.
The graphs also reveal something important about Obama’s victories in the Midwest. He was successful not only because he boosted turnout among young people and minorities in large cities. A closer look shows that he also won large majorities in low-income postindustrial towns, and almost half of the votes in the surrounding rural communities, where in 2016 Clinton won only 25 percent to 30 percent.
One of the clearest lessons from 2016 is that the Democrats cannot win the crucial swing states just by running up the score in the biggest cities. They will need to reinvest in understanding the heterogeneous areas beyond the city limits.
This is the first of two parts. Read the second part here.
Jonathan Rodden is a professor of political science and a senior fellow at the Hoover Institution at Stanford University.