The 2016 Election: Analyzing a Survey of 64k Americans.

Summary: Why did Clinton lose? This analysis of CCES data suggests a combination of anti-immigration & racially conservative Democrats defecting to Trump, an inability to maintain 2012 turnout levels among black voters, and tepid support from (younger) Bernie voters left her with a coalition too small to win the electoral college.

Intro: The 2016 Cooperative Congressional Election Study is a national survey of ~64,600 adults. The following is broken into 1. a summary of a regression analysis on vote choice between Clinton and Trump and 2. plots visualizing differences among Clinton/Trump voters, Vote Switchers (e.g. Obama to Trump voters), and various subgroups like gender and age.

As for variable definitions: vote choice refers to stated vote choice and race is broken into hispanic, non-hispanic white, non-hispanic black etc.  Additionally, many of the issue questions ask similar questions on the same topic (e.g. abortion, racial attitudes etc.). For ease of comparisons I created a summary variable for these questions. In essence, I centered each question to mean 0 (i.e. converted to z-scores) and averaged the answers to get a single measure of the questions. For example, a score of 1 on the prochoice variable refers to being on average 1 standard deviation more prochoice on the set of abortion questions. Lastly, all responses have been appropriately weighted.

Regression Analysis

Analysis of the 2016 CCES is broadly consistent with the findings in from my county-level analysis as well as previous research. I obtain primary results by running logits on individual responses with state level fixed effects and survey weights, along with various robustness checks. I focus on the role racial attitudes and support for immigration has on vote choice which, for example, adds as much as 31% explained variance in vote choice to a demographics only model. Rendered code can be found here.

A notable distinction is that race and education are much more predictive at the county level than at the individual level. This suggests that living in areas with lots of college graduates or lots of racial and ethnic minorities is more important in predicting vote preferences than individual-level demographics — thus, consistent with the theory that right wing populism is driven in large share by those living in socially & culturally isolated or segregated areas. Furthermore, this interpenetration is reinforced by the expected behavior that education loses nearly all its predictive power when controlling for racial attitudes (see: DeSante & Smith 2017 for relevance of race attitude questions).


1. 2016 Stated Two-Way Vote: by 2012 vote choice, 2016 Obama approval, stated primary vote, ideology, age, and ideology*age.

Note “NA” includes those who stated they did not vote or voted for a third party. Also, net votes refers to how many more votes than Trump did Clinton receive among a particular group.
net_votes_ageideo5_by_age2. Those who voted for Obama in 2012 but voted for neither Clinton nor Trump in 2016. This group  predominately has a somewhat to strongly favorable opinion of Obama. Demographically they are disproportionately black and younger than 30. They are more likely to have not voted in the primary or to have voted for Bernie Sanders (In fact only ~75% of Bernie voters stated they voted for Clinton in the general election).  While ideologically they primarily identify as moderates and somewhat liberal, their stances on policy are virtually indistinguishable from other Democrats.

In short, their lack of regular voting and not their policy positions is what distinguishes this group from Obama-Clinton voters.o2na_by_o

3. Those who voted for Obama and then Trump. The main message is that while Trump won over many people who voted for Obama, these voters had already soured on Obama in terms of favorability and differ significantly on policy from other Obama voters. They are primarily centrist on issues while moderate to somewhat conservative in identity. The most distinguishing feature is that they are most conservative on race and immigration issues. In many ways they are the last of the Reagan Democrats or Blue Dogs: racial conservatives with more centrist positions on economics, but who are not themselves economically disadvantaged.

4.  Those who voted for Romney and then Clinton. This group is much smaller than the others. However, I’ve included two relevant plots suggesting Romney to Clinton votesr are moderate on ideological identity, center to center-left on policy Qs, and were disproportionately likely to have voted for a non-Trump or non-Cruz Republican during the primary.


6. Legislative Questions: df refers all respondents, df2 to Obama to Trump voters, df3 to Obama to Neither. 1 corresponds to 100% support for proposed legislation, 0 with 0% support. Interestingly, support for Highway funding is the only question which has virtually no association with vote choice among Clinton and Trump supporters.



Subgroup Plots:

  1. Marriage & Gender. One talking point which has emerged since the election is the observation that “more white women voted for Trump than Clinton”. However, this ignores splits by education and marriage. Here it’s evident that white women without a college degree and who’ve been married at least once are the only group of women that were more pro-Trump than pro-Clinton. For men, nonwhite men across the board and white men with a degree who have never been married are more supportive of Clinton than Trump.However at the individual level , marital status appears to be associated with vote preferences only indirectly as it’s indicative of age and living in a non-urban setting. Nonetheless, a county’s percentage of residents who’ve never been married is one of the best single variable predictors of either 2012 and 2016 results (correlation coefficient of .7) .
  2. New Voters. Here are respondents who stated they voted for one of the top two candidates in the 2016 election but not the 2012 election. Recall the first plot that showed Clinton and Trump pulling virtually the same number of voters from this group. It’s apparent that Clinton netted more new voters among college-aged whites and nonwhites regardless of age while Trump’s new voters were more likely to be middle age and were almost exclusively white. These differences support the narrative that Trump was particularly attractive to irregular “working class” whites concentrated in the Midwest.
This entry was posted in Uncategorized. Bookmark the permalink.

2 Responses to The 2016 Election: Analyzing a Survey of 64k Americans.

  1. Pingback: The 2016 Election: A County-Level Analysis | gauchnomica

  2. Pingback: Deindustrialization cannot explain Trump. | 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