Voters line up Thursday, Oct. 20, 2016, during early voting at Chavis Community Center in Raleigh, N.C. (AP Photo/Gerry Broome)

This is the second of two posts. Read the first one here

As they try to understand what happened in November 2016 in the industrial heartland, Democratic strategists are focusing on the “rural and small-town demographic.”

In my earlier post I suggested that voters in rural areas and small industrial towns are often two rather distinct demographic groups that should not be conflated. Yes, Democratic candidates lost votes in both postindustrial towns and their surrounding rural environs. But their losses were especially dramatic in the latter, while they still have pockets of support in the former. In larger towns with an industrial history or a university (or both), Democrats still win majorities.

If Democrats try to reacquaint themselves with rural America as part of a strategy for winning back the pivotal postindustrial states, they must also come to grips with another important difference between town and countryside: turnout.

In postindustrial town centers, where Democrats still win majorities, comparatively few people turn out to vote. By contrast, in the surrounding rural areas — where Republicans have gone from a slim to an overwhelming majority — a much higher percentage of the population votes.

In the graphs below, I plot turnout among registered voters in 2016 against the log of population density for several counties in Ohio, Indiana and Pennsylvania; the size of the dot shows the precinct’s comparative size. The blue curve plots the 2016 turnout; the red curve shows the 2012 turnout for comparison.



In each case, a far higher percentage of registered voters actually turned out in surrounding suburbs and rural areas than in the town centers. In many cases the difference is quite striking.

In the overwhelmingly Democratic downtown precincts of Terre Haute and Muncie, Ind., turnout was only around 30 percent of registered voters; in the rural Republican precincts a few miles away, more than 60 percent voted. In Pennsylvania, where the presidential campaigns focused more energy, turnout in the downtown Democratic precincts was around 50 percent, but around 75 percent in the most rural precincts.

By comparing the red lines and the blue lines, we can see that the pattern was quite similar in 2012, although the gap may have increased a bit in a handful of counties.

These graphs help explain Trump’s narrow victory in the electoral college. His campaign focused on rural whites who typically demonstrate relatively high levels of turnout in the crucial states. Conceivably, some habitual Democratic voters stayed home in these precincts while a set of new voters turned out for Trump. To show this would require individual-level data from voter files that are only now becoming available.

However, the lesson from these graphs, combined with those in my previous article, is that the Trump campaign successfully appealed to high-turnout rural areas that had been either evenly divided or delivered slim Republican majorities in the recent past.

The consistent high turnout of rural whites in postindustrial counties is not well understood. When examining exit polls or even post-election voter files, analysts tend to lump residents of low-turnout Muncie or Williamsport together with their high-turnout rural surroundings into a single, “rural white” category.

Worse, analysts might be tempted to draw inferences from aggregate county-level data, especially since they can be downloaded, merged with census data, and mapped in a matter of minutes. For instance, it is easy to download the Pennsylvania 2016 county-level presidential results, merge them with data on citizen voting-age population from the census, and then produce a graph like this one:


 

A look at this graph suggests a happy story for Democrats: Turnout is higher in relatively urban counties with higher overall population density (and higher overall support for Democrats). The graph looks similar if produced using data from Obama’s victories in 2008 and 2012, and it looks similar in many other postindustrial states.

However, it is quite misleading. Returning to the fine-grained precinct-level graphs above, you can see that the relationship within Pennsylvania counties is exactly the opposite: People in low-density Republican areas are showing up to vote while those in the high-density towns are staying home.

To be technical for a moment, this is a classic case of what statisticians refer to as the “modifiable aerial unit problem”: a statistical bias that can arise when data on spatial phenomena such as population density are aggregated to geographic units such as counties.

Within Pennsylvania counties, turnout is substantially lower in relatively dense clusters populated with renters and young people than in the surrounding areas populated with older homeowners.

It will take some time to collect and geo-code precinct-level data from a full set of counties in the 2016 election, but the pattern was already very clear in the precinct-level data from earlier elections, including the 2008 presidential election. This is true not only in Pennsylvania, but in a large cross-section of states. This includes yesterday’s battleground states, like Indiana and Missouri, as well as in the urbanizing battleground states of the future, including Georgia, Texas and Arizona.

As Democrats consider whether to completely give up on rural areas where they once brought in votes, they must keep in mind that rural voters might be able to make up for their dwindling raw numbers by voting at higher rates than growing urban populations.

Polarization between urban and rural areas is bad for the Democrats in several ways. Not only does it leave them underrepresented in the House and Senate, but it also magnifies a problem with turnout that faces left parties everywhere. Despite its gains among educated suburbanites, the Democratic Party still relies on a base of younger, lower-income renters who frequently change addresses. The Republican Party relies on a base of older, higher-income long-term homeowners who are more likely to vote.

This is just as true in towns such as Muncie and Williamsport as it is in cities such as Indianapolis and Philadelphia.

Jonathan Rodden is professor of political science and senior fellow at the Hoover Institution at Stanford University.