A Deep Dive Into the Trump and Clinton Coalitions

A Deep Dive Into the Trump and Clinton Coalitions
AP Photo/Seth Perlman-File
X
Story Stream
recent articles

Who are Donald Trump’s supporters?

For the past week, a simple answer has been floating around in the media -- “working-class whites, in the South, the Great Plains and especially the Rust Belt.” No part of that answer is wrong, but it’s not a very detailed take. It fails to get at other central questions -- such as who voted for Hillary Clinton, where Trump outperformed Mitt Romney (and where he didn’t) and how Trump performed with racial minorities, to name a few.

Those details matter, so I retooled the Correlation Machine (an interactive I built during the GOP primary) to help readers explore what happened on Election Day. The basic idea is to pair Census Bureau data, past election results and other information with county-level 2016 election results in order to figure out where Trump and Clinton voters live and what that tells us about who they are.

The interactive is simple. Just pick a candidate variable (Trump’s share of the vote, Clinton’s share of the vote, the share that third party candidates earned or how much Trump outperformed Romney) and a state, and the Correlation Machine will look into 40 or more variables, figure out which are the best predictors of that candidate variable and display them.

Four quick but important mathematical notes:

First, the numbers you see in the interactive are correlations. Correlations are simple -- they’re numbers on a -1 to 1 scale that measure how closely two things track each other. So if Trump’s county-level vote share goes up as a county’s unemployment rate goes up, the correlation will be positive. The stronger that relationship is, the closer the value will be to 1. If one variable (e.g. Trump’s share of the vote) goes up as another one goes down (e.g. black share of the population), then the correlation will be highly negative. And if the variables have no effect on each other, the correlation will be closer to zero.

Second, these correlations are based on county-level rather than individual-level data. In other words, if Trump made gains in areas where fewer whites have college degrees, that doesn’t definitively prove that those voters cast their ballots for Trump. It suggests that they might have and it helps us characterize the political and economic climate those voters might live in -- but it doesn’t guarantee an individual-level relationship

Third, I used the difference between Trump's share of the two-party vote and Romney's share of the two-party vote to calculate how much Trump improved on Romney's performance. That means that if Trump outperformed Romney in every county in a given state, the interactive still tracks which variables led to a lower level of outperformance and displays that under the "Romney overperformance" column.

Finally, these numbers will change as more ballots are counted. Also, some states have yet to release county-level results, so not every state is included in this version of the interactive. So be cautious about drawing strong conclusions based on these numbers alone.

That being said, feel free to play with the interactive -- just pick a candidate variable, a state and press “go.” Please email or tweet at me (contact information below) if you find anything interesting (especially if you find a bug in the interactive). I’ve included a few observations below the interactive.

Trump’s gains among the white working class aren’t confined to the Midwest

Trump won the Electoral College by running up the margin with working-class white voters in Michigan, Ohio, Pennsylvania and Wisconsin. But his strength in that region can distract from the fact that he appeared to make gains with non-college-educated whites nationwide.

The Correlation Machine shows that for almost every state, Trump tended to do better than Romney (or at least underperform him by less) in counties where whites were less educated. Whites without a college degree are less numerous outside the Midwest, but the point here is that Trump’s appeal is not simply regional. He tended to outperform Romney wherever there were greater concentrations of working-class whites.  

That isn’t to say that Trump performed terribly among whites who have a college degree. Exit polls showed he won that group, 49 percent to 45 percent. But these correlations underline the fact that Trump’s increased national and regional strength came mainly from non-college-educated whites, a group that he won 67 percent-28 percent.

A rerun of the GOP primary?

It would be easy to look at these numbers and call Trump’s success with the white working class unprecedented -- but that’s not entirely true. Trump’s performance in the Republican primary foreshadowed the shape of his general election coalition.

Throughout the primary, Trump tended to perform better in areas with higher unemployment rates, lower levels of education or where a larger percentage of households made below $50,000 per year. The opposite was true of Ohio Gov. John Kasich and Florida Sen. Marco Rubio -- they ran better in more highly educated, wealthier, suburban areas.

In some ways, the general election was a rerun of that story. Trump tended to improve on Romney’s performance in counties that favored the president-elect in the primary. And the wealthy suburban areas that favored his establishment-friendly rivals were cooler to him in the general election. And although Trump may not have lost the people who voted for Kasich and Rubio in the primary – he won the overwhelming majority of Republicans – the correlation machine shows he may have turned off their neighbors and co-workers.

It’s important to note the class divides from the primary don’t map perfectly onto the general election. The general election electorate is more racially diverse than the GOP primary electorate, and Trump had trouble attracting racial minorities (whether they were working-class or not) to his candidacy. But Trump did run strongly in some of the primarily white, economically and culturally insecure areas that gave him the nomination.

The Latino backlash hasn’t showed up (yet)

During the general election, many analysts (myself included) considered the possibility that Trump might face a significant backlash from Hispanic voters over his immigration policies and some of his statements during the campaign. And while Clinton won this demographic by a large margin, it’s not clear that Trump underperformed Romney with that group.

In a number of states with large Hispanic concentrations (e.g. Texas, Florida, Nevada) the Correlation Machine shows a weak relationship between the percentage of Hispanics in a county and how much Trump underperformed or outperformed Romney there. It’s possible that data from California (not included because many ballots are still uncounted) or further data from other states will tell a different story, but current data suggest that Trump neither wildly outpaced nor fell far behind Romney with Latinos.

Third party candidates had unclear appeal

Finally, the Correlation Machine doesn’t show a clear demographic pattern for third party voters. That’s partially because at this point, those voters are all lumped together in one basket called “Other” (I intend to separate Libertarian Gary Johnson, Green Party candidate Jill Stein and Independent Evan McMullin in future versions of this tool), but it’s also partially because third party candidates didn’t perform especially well.

Johnson appears to have earned less than 5 percent of the popular vote after registering double-digit support early general election polls. And McMullin failed to win Utah (although high concentrations of Mormons did lead to higher vote shares for “other” in Utah and Idaho). When candidates fail to earn many votes, it can be difficult for a simple tool like correlation to pick up on a clear demographic pattern for their support. Other data sources like the voter list and exit polls may be more useful in figuring out if there are any clear demographic patterns in third party candidate support.

David Byler is an elections analyst for RealClearPolitics. He can be reached at dbyler@realclearpolitics.com. Follow him on Twitter @davidbylerRCP.

Comment
Show commentsHide Comments