2016 Election Analysis - Which Counties Moved Trump's Way and Why

Post election there has been no shortage of postmortems and analysis (So here is another one?). Many of these analyses have focused around the narrative of Hilary "ignoring" the white working class and economically depressed rural areas, while assuming that large urban areas would carry enough electoral weight to give her the presidency. 

To dig into this hypothesis I found election results by county (https://github.com/tonmcg/County_Level_Election_Results_12-16) for 2012 and 2016.

Here is the growth in GOP share by county compared with 2012 - 

You can see that major swings away from the GOP happened in Utah (Evan McMullin) and in large urban areas. The largest swings towards Trump relative to 2012 follow that midwest and rust belt line pretty well. You've probably already seen a map like this too. 

But which counties saw the most movement relative to 2012? 

Using the new built in K-Means clustering algorithm in Tableau 10, I was able to group counties into 7 groups based on their shift in voting to 2012. 

You can see the counties clustered in the gif below. The largest shifts to Trump came from parts of Midwest and the Rust Belt. 

Which states had the largest % of counties that fall into the Highest movement towards trump bucket?

Iowa, Ohio and West Virginia all saw 30% of their counties fall into the largest swing towards trump bucket. Montana, Illinois and Minnesota were not far behind.

Let's look at this data by county -

With the exception of a couple of Urban Counties (i.e. Cook County in Illinois, Hennepin County in Minnesota, Cuyahoga in Ohio), you can see this swing towards trump in these 10 states. Of these 10 states with the largest share of counties in the highest movement towards Trump bucket, only Illinois and Minnesota ended up blue. 

So why did we see this shift within the rust belt and in the Midwest in 2016?

Pundit Explanation 1 - Economic Anxiety and Economic Depression

Many pundits want to explain this away by claiming that the "economic anxiety" of the white working class created an environment where economic discontent lead to a desire to "blow up the system" and cast a vote for Trump, even if they didn't like him. 

See Michael Moore - 

"You live here in Ohio, you know what I'm talking about. Whether Trump means it or not, is kind of irrelevant because he's saying the things to people who are hurting, and that's why every beaten-down, nameless, forgotten working stiff who used to be part of what was called the middle class loves Trump. He is the human molotov cocktail that they've been waiting for. The human hand grande that they can legally throw into the system that stole their lives from them."

I wanted to see if this economic anxiety manifested itself in the data at all. I brought in county level employment data from the American Community Survey to see if there was any correlation between a shift towards trump and unemployment. While employment relative to population isn't a perfect metric I expected it to correlate with shifts towards Trump. 

Male employment and high school graduate employment, explains less than 5% of the variance in voting shifts from 2012 in the states with the highest number of highest movement trump counties.

White employment explains 16% of the variance. As the share of unemployed whites goes up, the shift to GOP increases in these states with the highest number of these highest movement Trump counties. 

Now, many of these states have large white majority populations, but it is surprising that the white employment rate would explain a higher portion of the shifts in voting by county than overall male employment.

That 16% number indicates that there is something in the economic anxiety argument. Looking at this data, this anxiety also may be more specific to whites in rural areas. The closing of factories, mines, and automation of manufacturing have been occurring for the last few decades. This gap between white employment and overall male employment may suggest that racial expectations are additive in this narrative. When a lack of opportunity for growth smacks against traditional expectations of what kind of work one deserves as a high school or college educated white male, anger at the system emerges, perhaps resulting in a vote for Trump. 

Counties in the Top 10 states with the largest shift towards Trump

Counties in the Top 10 states with the largest shift towards Trump

Pundit Explanation 2 - Victory of "Traditional" Values and Viewpoints

Many pundits want to explain Trump's victory as racially motivated and as a specific response to the racial progress symbolized (if not actualized) by electing a black president. Trump's campaign articulated hateful rhetoric against pretty much every minority group in America. He has ties to the White Nationalist movement and many argue that he deliberately did not distance himself from these hate groups.  

See Jemele Bouie articulate this better than I ever could -

Trump campaigned on state repression of disfavored minorities. He gives every sign that he plans to deliver that repression. This will mean disadvantage, immiseration, and violence for real people, people whose “inner pain and fear” were not reckoned worthy of many-thousand-word magazine feature stories. If you voted for Trump, you voted for this, regardless of what you believe about the groups in question. That you have black friends or Latino colleagues, that you think yourself to be tolerant and decent, doesn’t change the fact that you voted for racist policy that may affect, change, or harm their lives. And on that score, your frustration at being labeled a racist doesn’t justify or mitigate the moral weight of your political choice. 

The very slogan, "Make America Great Again" glorifies a time which was horrific for many Americans. While there are many polls which break down Trump supporter's racial attitudes, I wanted to see if there was American Community Survey data that could correlate with these voting shifts from 2012. I decide to look at county population (even if counties are segregated), and as suggested by sister-in-law Sarah, I brought in median age as well. 

Median age explains 18% of the variance in voting shifts from 2012 in the highest movement trump counties, and population size explains 20%. Older counties and counties with less population saw some of the highest shifts towards Trump compared with 2012. 

Counties in the Top 10 states with the largest shift towards Trump

Counties in the Top 10 states with the largest shift towards Trump

Does this mean that these counties were simply white-lashing against a black president? I'm not sure. Looking at this data it appears that nostalgia did drive some of the shift Towards trump, with age explaining a fair amount of voting behavior. That's not to say that all elderly people look back fondly on the past, but it is easy to imagine how an aging (largely white) rural population would see themselves looking towards the past with a fondness that a young person, minority, or relatively well off person living in an urban center, may not understand. 

What Happens When We Combine These Data Points, Does the Explanation Get Any Clearer?

First step was to bring all of our data points - voting shifts, age and employment by county in R studio and run a correlation analysis. Looking at the plot, it is clear that a higher age correlates with a greater shift in GOP vote share, less employment correlates with a greater shift in GOP vote share, and a higher population correlates with a lower shift in GOP vote share.

Now that we see that there are significant correlations, we can use the Leaps package to select a model. Leaps looks at all of your variables and then picks the factors that explain the greatest share of the variance. 

Ok! it's looking like that our inputs explain over 60% of the variance in voter shifts in Trump counties. It seems like we have some decent inputs in there. Now, the best model has a lot of redundancy (13 factors), so I reduced it to 3.

Here is the output:

Screenshot 2016-12-29 13.45.29.png

Putting it all together (as best I can)

 While this model is fairly simple, with only three inputs, it explains a lot of the variance in the 883 counties that shifted most significantly towards trump in 2016 (nearly 50%). So if 50% of the shift towards trump can be explained by age, economic anxiety and being in a less populous more isolated place, 50% is explained by something else. What is that something else? 

As we move forward, the question for our country becomes - can we provide the support for elderly, isolated, rural populations who are nostalgic for a (problematic) America of yesteryear while maintaining support for a younger and more diverse urban population impatient for for a more progressive and heterogenous future?

What kinds of systemic (or localized) change is needed to blend these viewpoints - Those who think America's best days are behind us and those who think our best days are yet to come?

Explore the Data for Yourself

Daniel Prager