September 28, 2017 - admin

The Anger Games: Who Voted for Donald Trump in the 2016 Election, and Why?

David Norman Smith and Eric Allen Hanley have a pair of forthcoming articles in Critical Sociology that explore the Trump election in more detail.  What follows is a preview of the argument they develop, accompanied by the statistical tables that ground this argument in data drawn from the American National Election Study.

David Norman Smith

University of Kansas, United States

Eric Allen Hanley

University of Kansas, United States

 For many people the election of Donald Trump in 2016 is the proverbial enigma, wrapped in a riddle, shrouded in mystery.  How could a candidate who transgressed so often and so crudely against women, minorities, the truth and ordinary decency win 63 million votes?

Two theories are popular.  Both pivot around race, but in different ways.  The issue in question is whether racial prejudice, in a deep sense, lies at the heart of Trump’s election.  Some pundits answer this question affirmatively, saying that the electorate is sharply polarized over race and that Trump won by capturing the support of racially angry white voters.  Others argue that Trump won by appealing to the economic populism of  the white working class, which voted for him more out of desperation and hope than racial resentment.

At first glance these positions might seem to differ only slightly but, in reality, they are divided by a chasm.  Partisans of the “white working class” thesis tend to say that Trump’s election, however disturbing, was an aberration, and that, once white workers realize that the celebrity billionaire’s populism is phony, they will be open to authentic populist appeals – even of the centrist Clintonite variety.   Polarization theorists are far less optimistic.  They say that Trump’s white voters supported him because he pandered to their deep underlying racial resentment.  Overcoming that resentment is not the work of a day, or a slogan, or a program.  What is needed, rather, is a commitment to defeat intolerance in the here and now, accompanied by a push to overcome the inequities that divide us.

In “The Anger Games,” which Critical Sociology will publish in early 2018, we analyze data from the 2016 American National Election Study.  Multiple logistic regression of 17 key variables enables us to adjudicate between these rival  claims. What we see, in the regression tables (shown below), is a portrait of polarization.  White voters in every demographic category – whether college educated or less educated, older or younger, richer or poorer, married or single, male or female – voted in conformity with their attitudes.  White voters who harbored prejudices against women, minorities, immigrants or Muslims were far likelier to vote for Trump than voters in the same categories who did not share those prejudices.

Married voters, older voters, men and voters without four-year college degrees do figure prominently among Trump’s white supporters – but only when they share his biases.  Millions of voters in those same categories who rejected his biases rejected Trump as well.  We found, in fact, that once we take attitudes into account, every single demographic variable loses its power to explain the election.  In every sector of the white population, racially resentful voters flocked to Trump and racially tolerant voters did the opposite.

Pocketbook worries, which many people assume played a decisive role in the election, did not in fact distinguish Trump voters from others.  The attitudes that mattered most were racial, not financial; and racial resentment was strongly felt by pro-Trump voters across the spectrum.  Many analysts have speculated, or hoped, that an appreciable percentage of Trump’s support came from voters who winced when they cast their ballots, closing their eyes to his prejudices in the hope that he would heal the economy.  But the facts tell a different story.  Even Trump voters who did not classify themselves as “strong” partisans (just over a quarter of the total) scored much higher on racial resentment and a host of other prejudices than non-Trump voters.  In other words, the idea that Trump’s white electorate includes a significant percentage of dissatisfied but unbiased populists does not square with the facts.  Polarization is acute, and the center is shrinking.

Our 2016 findings shed fresh light on that polarization.  By constructing scales from items new to the Election Study, we are able to show, first, that strong Trump partisans protest what they perceive as reverse discrimination (in contrast to his less enthusiastic supporters, who score high only on more routine measures of anti-Black prejudice); and second, that Trump voters also show hostility towards women, especially non-traditional women.  Polarization, in other words, revolves around attitudes towards gender as well as race, and racial resentment comes in two forms:  strong, and stronger.

Another finding enables us to better grasp the complexity of this picture. Trump voters, it turns out, are uncommonly eager to agree that, in order to restore moral order in America, the authorities must “crush evil” and “get rid of the rotten apples who are ruining everything.”  These uninhibited statements appear in items that the Election Study included, at our suggestion, in 2012 and 2016.  In 2012, these items, which comprise what we call the “Domineering Leader” scale, proved to be powerfully predictive of racial resentment.  In 2016, they proved to be powerfully predictive of pro-Trump voting.  Our analysis suggests that these findings are not unrelated.  Support for domineering leaders who refuse to tolerate moral outcasts is, at the same time, a kind of bias against tolerant leadership.  Authorities who do not take sides against resented minorities and women are regarded as illegitimate, as usurpers, who favor “the undeserving” over the deserving, the unrighteous over the righteous, “takers” over “makers.”  The same principle applies to the media, and to other institutions (e.g., colleges and universities), which are decried as fraudulent when they stand up for tolerance.

One implication of this analysis is that Trumpism existed before Trump.  Why, critics have wondered, did Donald Trump triumph over 16 Republican rivals?  How did he defeat the formidable Hillary Clinton?  Part of the answer, it seems, is that Trump’s base was ready and waiting for him.  No one remotely like him had run for president before, and his campaign catalyzed pent-up demand for a new kind of leader.  Ordinary politicians thought they had to play by the rules.  But Trump sensed that tens of millions of voters wanted a domineering leader who would break the rules and scorn liberal pieties.  He arrived on the scene as if he had been chemically synthesized to play this role, and his base responded with fervor.  That fervor is unlikely to dissipate, even if Trump angers his base by cutting taxes for the rich or healthcare for the poor.  It poses an enduring challenge to everyone who fights for tolerance.

 Statistical Profile of the 2016 Election

The attitudes that crystallized in support of Donald Trump in 2016 will almost certainly outlast his presidency.  So the deepest imperative posed by his presidency is to better understand those attitudes.  Subterranean hostilities have come back into the light; sublimated prejudices are now television talking points and campaign slogan.  Trump himself is a larger-than-life icon of this return of the repressed, but he owes his success, above all, to the trends he detected and stirred in the electorate.  Hence, in two forthcoming articles in Critical Sociology (2018), we analyze data from the 2016 Election Study to better understand Trump’s base.  We do this, in “The Anger Games,” by examining and explaining the centrality of attitudes in the 2016 election; and then, in “Nativism, Populism, and the White Working Class,” we challenge the popular hypothesis that Trumpian hostilities owe their strength mainly to the toxic effects of globalization on less-educated white workers.  We agree that class counts; but ethnic and other biases are not reducible to class.  We posit a complex dialectic, in which attitudes and feelings are as much causes as effects.

In both papers, though we examine other data as well, our primary original contribution revolves around evidence about Trump’s base from the 2016 Election Study.  We present our findings from that study in the following five tables.

First, we look at the raw numbers, to see which categories of white voters were likeliest to vote for him.  In Table 1, we see that Trump won 52% of the white vote and that 73.1% of his voters called themselves his “strong” supporters.  Here and in Table 2, we see that Trump and non-Trump voters differ strikingly in many of their demographics and attitudes.  (Readers will notice that our tables explore the effects of 17 variables in all, only a few of which are summarized here; the full significance of these 17 variables will be explained in the “The Anger Games,” forthcoming.)

Next, in Table 3, we present a series of logistical regression analyses (which we call “models” here and “trials” in the main article) to explore support for Trump among white voters.  We find, when we look only at demographics (Model 1), that there are notable demographic differences between Trump and non-Trump voters.  But those differences lose their statistical significance when we look at attitudes as well as demographics in Models 2 and 3.  We start with a scale which asks voters to choose between desirable traits in children, which is often construed as a measure of “authoritarianism”; we then include our own Domineering Leader scale, which we drew from a different tradition of authoritarianism research.  The results show that Trump and non-Trump voters differ significantly on both measures, but that  the Domineering Leader scale, in particular, is powerfully associated with Trump voting. In Model 4, we find that eight attitude variables have sizeable significant associations with Trump voting, while demographic variables remain negligible.

In Table 4, we establish that mild Trump supporters more closely resemble non-Trump than strong Trump voters in education and income, but that, in their attitudes, they remain far closer to strong Trump voters.

In Table 5, we carry out a 17-variable regression of the kind we reported in Table 3, but with the goal, now, of explaining how strong and Trump voters differ.  We find, again, that demographic variables lose their significance when we include authoritarianism scales in Models 2 and 3, and that the Domineering Leader scale is particularly powerful.  When we examine 12 major attitude variables in Model 4, we learn that five of these attitudes (including the wish for a domineering leader, complaints of reverse discrimination, and anti-immigrant feeling) are held with significantly greater fervor by strong Trump voters than by mild Trump voters.

Table 1:  Sample means and standard deviations of variables included in our analyses of presidential choice and strength of support for Trump in 2016.

 white voters in 2016 white trump voters in 2016
outcomes mean sd mean sd
 % who voted for trump .520 .500  
 % who voted for trump enthusiastically   .731 .443
 demographic variables
 % male .470 .499 .502 .500
 average age 52.3 17.2 54.2 17.0
 education (some college or less) .518 .500 .623 .485
 income (tens of thousands USD) 9.13 7.38 8.41 6.83
 marital status (married=1) .648 .478 .668 .471
attitudes toward    
child traits 4.88 3.25 6.20 2.24
domineering leaders 5.50 3.20 7.37 2.24
African Americans 5.65 2.92 7.32 2.01
reverse discrimination 3.65 2.01 4.71 1.60
immigrants 5.05 2.39 6.41 1.88
Muslims 5.11 2.16 6.09 2.07
women 3.67 1.87 4.60 1.55
personal financial concerns 5.01 2.35 5.61 2.25
the health of the economy 5.01 2.31 6.25 2.04
liberalism vs. conservatism 5.34 2.61 6.97 1.86
general religiosity 5.00 3.94 6.32 3.57
fundamentalism 4.79 2.84 6.15 2.38
 number of cases 1883 979


For simplicity we have substituted the term “outcomes” here for “dependent variables” and the term “causal factors” for “independent variables.”

Table 2.  Mean values of independent variables included in analysis by presidential vote choice in 2016, white voters only. N=1883

voted for


another candidate difference
 demographic variables
gender: male 50.15 43.69   6.46
gender: female 49.85 56.31  -6.46**
education: some college or less 62.31 40.38  21.93***
education: ba or higher 37.69 59.62 -21.93***
marital status: not married 33.20 37.28  -4.08
marital status: married 66.80 62.72   4.08
age (years) 54.17 50.31   3.86***
Household Income ($10,000 USD) 8.41 9.91  -1.50***
 attitudes toward…
child traits 6.20 3.45   2.75***
domineering leaders 7.37 3.48   3.89***
African Americans 7.32 3.83   3.49***
reverse discrimination 4.72 2.50   2.22***
immigrants 6.41 3.57   2.84***
Muslims 6.10 4.05   2.05***
women 4.60 2.66   1.94***
personal finan ial concerns 5.61 4.36   1.25***
the health of the economy 6.26 3.66   2.60***
liberalism vs. conservatism 6.97 3.57   3.40***
general religiosity 6.32 3.57   2.75***
fundamentalism 4.92 2.41   2.51***


Table 3.  Logistic regression coefficients from four models predicting pro-Trump voter choice in 2016, white voters only (standard errors in parentheses).

model 1 model 2 model 3 model 4
demographic variables        
education (some college or less=1) .834***








marital status (married=1) .402**








age (years) .012***








gender (male=1) .097








income (in tens of thousands US$) -.013








attitudes toward…
child traits .277***






domineering leaders .487***




African Americans .177***


reverse discrimination .114


immigrants .197***


Muslims .137*


women .211**


personal finan ial concerns .002


the health of the economy .376***


liberalism vs. conservatism .486***


general religiosity .028


fundamentalism .107*


constant -1.175








pseudo r-squared .049 .149 .328 .588
number of cases 1883 1883 1883 1883

* p > .05                      ** p > .01        *** p > .001


Table 4.  Mean values of independent variables included in analysis by strength of support for Trump in 2016, white Trump voters only.

level of support for Trump strong mild difference (strong vs. mild)
gender: male 50.56 49.01   1.51
gender: female 49.44 50.95  -1.51
education: some college or less 66.34 51.33  15.01***
education: ba or higher 33.66 48.67 -15.01***
marital status: not married 33.38 32.70    .68
marital status: married 66.62 67.30   -.68
age (years) 54.41 53.48   .94
household income (in tens of thousands US$) 7.90 9.81 -1.91***
attitudes toward
child traits 6.45 5.52   .93***
domineering leaders 7.71 6.46  1.24***
african americans 7.60 6.57  1.03***
reverse discrimination 4.92 4.16   .75***
immigrants 6.74 5.53  1.21***
muslims 6.34 5.42   .93***
women 4.68 4.38   .30**
personal finan ial concerns 5.74 5.27   .47**
the health of the economy 6.54 5.48  1.05***
liberalism vs. conservatism 7.10 6.60   .50***
general religiosity 6.45 5.94   .51*
fundamentalism 5.22 4.11  1.11***


Table 5.  Logistic regression coefficients from four trials strength of support for Trump in 2016, white Trump voters only (standard errors in parentheses).

causal factors model 1 model 2 model 3 model 4
gender (male=1) .208








age (years) .002








education (some college or less=1) .528**








income (tens of thousands usd$) -.036**








marital status (married=1) .354








attitudes toward…
child traits .097**






domineering leaders .241***




African Americans .060


reverse discrimination .195**


immigrants .164**


Muslims .075


women -.069


personal finan ial concerns .025


the health of the economy .177***


liberalism vs. conservatism .064


general religiosity -.060


fundamentalism .156**


constant .559








pseudo r-squared .029 .040 .082 .173
number of cases 979 979 979 979

* p > .05                      ** p > .01        *** p > .001

Corresponding author:

Communications about this paper can be sent to David N. Smith, Department of Sociology, 716 Fraser Hall, 1415 Jayhawk Boulevard, University of Kansas,  Lawrence, KS 66045.  Email:




Leave a Reply

Your email address will not be published. Required fields are marked *