The 2016 Election: A County-Level Analysis

What explains the 2016 presidential results at the county level? Running OLS and Spatial regressions on a dataset constructed from roughly 250 potential covariates, I find that support for Trump is strongly consistent with the burgeoning research on right-wing populism and is strongly inconsistent with the thesis that Trump’s electoral successes were driven by economic hardship. Furthermore, a follow-up analysis of the 2016 CCES (survey of 64k Americans) yields strongly corroborative results.

Right wing populism can be explained as cultural backlash to perceived out group threats (ex. contemporary demographic change in North American and Europe), especially among persons with authoritarian prone personalities and ethnocentric beliefs. Notably, Inglehart and Norris provide this summary:

[C]ultural values, combined with several social and demographic factors, provide the most consistent and parsimonious explanation for voting support for populist parties; their contemporary popularity in Europe is largely due to ideological appeals to traditional values which are concentrated among the older generation, men, the religious, ethnic majorities, and less educated sectors of society…. [In contrast,] evidence for the economic insecurity thesis, the results of the empirical analysis are mixed and inconsistent…. Populism, [a loose set of ideas that share three core features: anti-establishment, authoritarianism, and nativism] is a standard way of referring to this syndrome, emphasizing its allegedly broad roots in ordinary people; it might equally well be described as xenophobic authoritarianism.

This is to say, culture (specifically views on immigration, race, and gender) as opposed to economics appears to be the dominant cleavage in explaining Trump’s electoral successes. The following analysis is consistent with view. Left for further investigation is the role which Chinese trade competition and sociological isolation may have played in causing support for Trump.

Section I looks specifically at support during the general and primary for former Secretary of State Hillary Clinton and businessman Donald Trump. As described above, results are consistent with “cultural backlash” theories of support. Section II looks at third party support. Specifically, results are consistent with the concern that support for Senator Bernie Sanders led to greater third party defection. Section III looks briefly at turnout and ticket splitting. In particular, turnout among African American areas was decisively lower than 2012 while the sharp turnout decline in Wisconsin is suggestive of a negative effect from the 2016 voter identification law.

Regression tables and rendered code can be found: OLS and Spatial.

I Two Party Vote Margin & Primary Shares

Unsurprisingly, race and education are the most dominant explanatory variables. As is true with the 2012 and 2004 elections: after controlling for race and education, lower quality of life (poverty, drug mortality, unemployment, lack of health insurance, obesity, single parenthood, population stagnation) is associated with voting Democratic. While household income is negatively correlated with Clinton’s vote margin, it becomes insignificant as broader quality of life measures are included. As expected for candidates from incumbent parties, Clinton performed worse in counties with higher growth in unemployment or uninsurance rates. However, the effect sizes from these growth variables are too small to have affected the election outcome and may have even been negative when considering effects from the static QoL variables. In the base models, heavy alcohol consumption rates positively correlate with Clinton’s vote margin, but this is more likely an indicator of local culture and norms than economic hardship. Moreover, alcohol use is insignificant in the model which corrects for geographic dependence. Similarly, while injury mortality (a proxy for gun ownership and culture) is negatively correlated with Clinton, the effect size and significance are greatly mitigated by correcting for geographic dependence. Variables indicative of increased ethnic tension are predictive of Trump’s performance. Specifically, racial diversity, religious diversity, income inequality, foreign born share, and growth in a the percentage of non-whites are all associated with Trump’s electoral performance. Furthermore in the spatial model, the county proportion of African-Americans is associated with lower Clinton performance in bordering counties – tentatively suggestive of partisan self-sorting or “white flight”.

Primary election covariates for Trump are similar to the general election. In particular and more so then the general, Trump performed well in Catholic/Orthodox communities as well as older communities. Like the general election, Trump’s primary share is associated with smoking, lack of exercise, injury mortality and lower population density. These factors in conjunction with education greatly mitigate any negative relationship between Trump’s electoral successes with life expectancy. Curiously, unemployment is associated with Trump’s primary share (unlike the general) while other markers of hardship have either a zero or negative relationship (like the general). Finally, Trump does better in areas with a greater share of Caucasians as well as higher growth in the proportion of the non-white population.

II Third Party Voting.

Clinton’s primary share is negatively correlated with Stein’s vote share after controlling for observables. This is consistent with the concern that support for Bernie Sanders led to greater third party defection.

It is unlikely that Clinton’s performance in the primaries is acting as a proxy for county level ideology. Individual polling found only modest differences between Clinton’s and Sanders’s supporters. Additionally, at the county level, Clinton’s primary share is positively correlated with her two party margin in the general. Moreover, covariates such as exercise, alcohol consumption, trump’s primary share, and geography would mitigate the influence of ideology if it were a confounding variable. Lastly, outside the consideration of statistical models, the primary driving third party defection is consistent with the tens of thousands of write-in votes cast for Sanders.

III Turnout

Preliminary analysis of county level turnout is consistent with the general finding that black turnout subsided from 2012, and is suggestive that Asian areas saw increased turnout. Unsurprisingly, turnout was lower in Mormon areas. Turnout in counties with high alcohol use was also down, a variable which is highly geographically dependent and speculatively could be related to union membership rates or some other factor related to the Great Lakes and North Great Plains regions.

It’s been observed that Wisconsin’s turnout was abnormally low compared to earlier years. Even after controlling for state fixed effects, the counties of Racine and Milwaukee stood out in particular as outliers on both turnout and margin. While not definitive, it does strongly motivate the question how and why did WI’s strict voter identification law affect turnout. In the past, support for the effect of ID laws has been mixed. Thus, it is worth examining what makes WI’s ID law different and how much of any effect is temporary.

As for ticket splitting, Clinton performed better than her Senate counterparts in ethnic minority and Protestant areas, and counties with higher population density. She performed worse in areas with higher rates of intermediate educational attainment and higher rates of disability. These findings motivate the question on how best to leverage supporters both up and down ballot.

This entry was posted in Uncategorized. Bookmark the permalink.

One Response to The 2016 Election: A County-Level Analysis

  1. Pingback: The 2016 Election: Analyzing a Survey of 64k Americans. | gauchnomica

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s