Since Trump was elected U.S. President in 2016, researchers have sought to explain his support, with some focusing on structural factors (e.g., economics) and others focusing on psychological factors (e.g., negative emotions). We integrate these perspectives in a regional analysis of 18+ structural variables capturing economic, demographic, and health factors as well as the aggregated neuroticism scores of 3+ million individuals. Results revealed that regions that voted for Trump in 2016 and 2020 had high levels of neuroticism and economic deprivation. Regions that voted for Trump also had high anti-Black implicit bias and low ethnic diversity, though Trump made gains in ethnically diverse regions in 2020. Trump’s voter base differed from the voter base of more traditional Republican candidates and Democrat Bernie Sanders. In sum, structural and psychological factors both explain Trump’s unique authoritarian appeal.
Regions differ on many important psychological, economic, demographic, and health-related dimensions. Investigating these dimensions can inform our understanding of the nature of regional differences and their implications for consequential political behaviors like voting.
We analyzed characteristics of more than 2,080 counties to see whether support for Donald Trump in the 2016 and 2020 U.S. presidential elections was concentrated in certain kinds of regions. Importantly, we considered a wide array of psychological and structural factors, which have typically been examined independently in the voting literature. In doing so, we provide a multi-faceted portrait of the roots of regional Trump support and how his voter base differs from that of other Republican and Democratic candidates.
Our analyses uncover four central findings. The first is that regions high on neuroticism—whose populations are more prone to experiencing negative emotions, like fear and anxiety—were more likely to vote for Trump. The robustness of this finding reveals a strong emotional basis for Trump support and suggests that enduring aspects of personality explain voting behavior beyond the effects of other factors.
The second major finding is that economically deprived regions were more likely to vote for Trump. These economically deprived regions were characterized by low college attainment, low income, high levels of manufacturing and agriculture, etc. However, the economically deprived regions voting for Trump were not the very poorest regions in the U.S. The very poorest regions tended to be ethnically diverse and voted for the Democratic presidential candidates (i.e., Clinton in 2016 and Biden in 2020).
Our third major finding is that Trump-supporting regions had low levels of ethnic diversity and high levels of anti-Black implicit bias. This finding shows that purely economic accounts of Trump support that do not take race and racism into account are incomplete. Interestingly, however, Trump gained more support in ethnically diverse regions over time. He performed better in ethnically diverse regions in 2020 than he did in 2016, even though ethnically diverse regions still preferred Democratic candidates overall. Trump’s performance also improved in other kinds of regions, like regions with poor health. He made gains in these regions in 2016 and 2020, even though regions with poor health still preferred Democratic candidates overall.
Our last major set of findings concerns important differences between Trump’s voter base and the voter base of other Republican and Democratic candidates. Trump made gains in 2016 over Republican presidential candidate Mitt Romney’s performance in 2012 in regions high on neuroticism and economic deprivation, and low on ethnic diversity, showing Trump’s unique appeal in these regions. There was some overlap in the characteristics of regions that voted for Trump and those that voted for rival Republican primary candidate Ted Cruz and Democratic primary candidate Bernie Sanders. However, there was no overlap between the characteristics of regions that voted for Trump and those that voted for more traditional Republican primary candidates like John Kasich and Marco Rubio. Thus, Trump seemed to have a unique authoritarian appeal unshared by most of his rivals, a view that is bolstered by additional evidence that regions that voted for Trump tended to score high on measures of right-wing authoritarianism.
In sum, the tendency to experience negative emotionality (i.e., neuroticism) and the objective economic, demographic, and health conditions of one’s environment are jointly associated with voting behavior in nuanced ways. The more general picture, however, is that fear and deprivation characterize Trump’s America.
We shed light on the psychological and structural roots of Donald Trump’s unique authoritarian appeal in a large-scale regional analysis of voting behavior in 2016 and 2020, highlighting the importance of personality, economics, ethnic diversity, and health.
More neurotic regions voted for Trump in 2016 and 2020. Economically deprived regions voted for Trump in 2016 and 2020. Low ethnic diversity and high implicit bias predicted Trump voting. Trump’s voter base differed from that of other candidates. Psychological and structural factors are both related to Trump voting.
Since Donald Trump’s election as U.S. President in 2016, researchers have sought to understand the roots of his support. This interest stems in part from perceptions that Trump is both a symptom and cause of democratic decline. Since Trump’s election, the most respected reports on democratic performance have all documented signs of democratic erosion in the U.S. In 2016, The Economist Group demoted the United States from the category of “full democracy” to “flawed democracy” ( This report also shows that Trump supporters and detractors make different evaluations of American democratic performance. Moreover, the public’s views of American democratic performance diverge when it comes to which democratic principles are under threat. For instance, a majority of experts believe that American elections are free of pervasive fraud, but this view is not shared by the majority of the public (
If the appeal of authoritarian leaders is on the rise, then it makes sense to turn to theories of authoritarianism for an explanation. Foundational and modern theories of authoritarianism contend that leaders like Trump win support from people experiencing feelings of threat (
Many studies have found evidence, including causal evidence, for the role of economic threat on authoritarian support. Regions that experienced more automation and trade competition with China were more likely than other regions to vote for Trump in 2016, to vote for far-right parties in Europe, and to have authoritarian values (
These findings suggest that Trump’s popularity is not just a matter of economics; his popularity is caused by perceived threats to white people’s status in an increasingly ethnically diverse United States (
Whites’ declining status is also evident in their declining health (
Nevertheless, psychologists have shown for decades that people’s behavior—which includes their voting behavior—cannot be explained solely by their environment (
Neuroticism is the personality trait most closely associated with the experience of negative emotions (
The link between neuroticism and support for authoritarian leaders has been theorized for over half a century, but robust empirical evidence for this link has been lacking until recently.
The recency and inconsistency of evidence for the link between neuroticism and right-wing voting behavior suggests that this link may be contingent on structural factors, such as the economic, demographic, and health-related conditions of one’s environment. Neuroticism may predict voting for leaders like Trump only in the presence of threatening structural factors, which trigger a neurotic cascade. Similarly, threatening structural factors may predict voting for Trump more strongly in regions high on neuroticism, because neuroticism may make threatening conditions particularly burdensome. Both possibilities are consistent with interactionist theories advanced in social and personality psychology, which argue that behavior is the product of psychological and environmental factors (
Despite this empirical and theoretical precedent, research taking an interactionist approach to accounting for political behavior remains rare. A related but distinct tradition of research examines how threatening situations (usually manipulated in the lab) interact with other individual differences (e.g., authoritarianism, not basic traits) to predict political behavior and attitudes (e.g.,
In exploratory and confirmatory analyses of the 2016 and 2020 U.S. presidential elections, we test the independent effects of psychological and structural factors, as well as their additive and interactive effects, on voting behavior at the regional level. We probe the generality of our findings across two regional levels of analysis (counties and Core-Based Statistical Areas) and address problems unique to spatial analyses (by accounting for differences in population density, variance shared between regions in the same state, and similarity between neighboring regions). We capture the general structural conditions of a region by factor analyzing an array of publicly available variables, which yielded three factors per region reflecting economic deprivation, ethnic diversity, and health disadvantage. We capture the prevalence of negative emotionality in a region by aggregating the neuroticism scores of more than three million individuals according to the regions in which they live.
In all our analyses, we compare the effects of regional neuroticism to the effects of other personality traits, which have also been found to predict political attitudes. In addition, we benchmark the effects of our focal psychological and structural variables against other implicit and explicit variables that reflect a desire for group-based dominance (i.e., social-dominance orientation, right-wing authoritarianism, implicit racial and gender bias), which have been shown to predict status threat and voting in past research (e.g.,
The method used to examine the 2016 and 2020 elections were identical, except that the structural and voting data were collected in different years. All data sources are described in We report the following deviations from our pre-registration. First, our pre-registration mistakenly stated that all variables would be standardized prior to being used in analyses. Only the independent variables were standardized. Second, we used Bureau of Economic Analysis population density data and the Federal Communications Commission internet data instead of the American Community Survey’s data because we wanted to use 1-year estimates where possible (rather than ACS’s 5-year estimates). Third, the income ratio described in the pre-registration is the same as that included in the main manuscript, but we describe the 90th income percentile as “the bottom 10% of earners.” Fourth, we did not mention weighting personality traits for representativeness as an additional test of robustness. Fifth, we did not anticipate the voting data exclusions due to lack of data availability in the pre-registration. Finally, the analyses of group-based dominance variables (i.e., SDO, RWA, race IAT, gender-career IAT) as well as analyses of Kasich, Rubio, and Cruz primary votes were conducted in response to reviewer comments and thus were not pre-registered.
Variable | Description | Data Source |
---|---|---|
Voting | Trump votes 2016: Republican two-party vote share in 2016 general election |
2012 general election data: OpenDataSoft |
Big 5 Personality Traits | Neuroticism, Openness to Experience, Conscientiousness, |
Gosling-Potter Internet Personality Project |
Group-Based Dominance | Race IAT (implicit anti-Black bias) |
Project Implicit |
Income | Real per-capita income in US dollars in 2016 and 2019 | Bureau of Economic Analysis |
Unemployment | Unemployment rate in 2016 and 2020 | Bureau of Labor Statistics |
Manufacturing | Employment share in manufacturing/mining in 2016 and 2019 | Bureau of Economic Analysis |
Agriculture | Employment share in agriculture in 2016 and 2019 | Bureau of Economic Analysis |
Birthweight | Percentage of live births with birthweight under 2500g (~5.5 lbs.) in 2016 and 2020 | County Health Rankings |
Smoking | Percentage of adults who smoke in 2016 and 2020 | County Health Rankings |
Obesity | Percentage of adults who report a BMI ≥ 30 in 2016 and 2020 | County Health Rankings |
Teen births | Percentage of births with mothers aged 15-19 in 2016 and 2020 | County Health Rankings |
Physical health | Number of physically unhealthy days reported per month in 2016 and 2020 | County Health Rankings |
Uninsured | Percentage of people under age 65 without health insurance in 2016 and 2020 | County Health Rankings |
Ethnic diversity | Inverse of Hirschman-Herfindahl-Index (i.e., inverse of the probability that two randomly selected people will be of the same ethnicity) such that higher numbers reflect more diversity in 2012–2016 and 2015–2019 | American Community Survey |
Income inequality | Ratio of the average income of the top 10% of earners divided by the average income of the bottom 10% of earners in 2015 | |
Poverty | Percentage of the population in poverty in 2016 and 2019 | Census Bureau |
Crime | Crime rate (arrests for murder, assault, theft, and burglary) per 100,000 individuals in 2016 and 2019 | FBI Uniform Crime Reporting |
Migration | Percentage of the population that moved to the region from abroad within the last year from 2012-2016 and 2015-2019 | American Community Survey |
Education | Percentage of people age 25+ with a bachelor degree from 2012–2016 and 2015-2019 |
American Community Survey |
Internet access | Percentage of households with high-speed internet connection in 2016 and 2018 | Federal Communications Commission |
Population density | Population per square mile in 2016 and 2020 | Bureau of Economic Analysis |
The main unit of analysis is the county level, which we supplement with analyses at the Core-Based Statistical Areas (CBSA) level for reasons we describe below. Each CBSA is a metropolitan area composed of one or more counties: an urban core and its surrounding commuting territory.
Spatial analyses require unique analytical considerations. First, these analyses require appropriate geographic control variables. We control for population density because voters in rural regions with low population density tend to vote for conservative candidates. We also control for state-fixed effects by including state dummies that account for variance shared between regions in the same state, including any omitted confounding variables at the state level. It is particularly important to account for state-fixed effects in election research because election laws and procedures that differ from state to state could influence the results. Second, we must anticipate potential violation of the statistical assumption that error terms will be uncorrelated because neighboring regions are non-independent ( We conducted this robustness check out of an abundance of caution. We actually did not observe problematic spatial autocorrelation as indicated by non-significant levels of Moran’s I (2016 Trump votes = 0.22, 2016 Trump gains = 0.61, 2020 Trump votes = 0.44, 2020 Trump gains = 0.84).
Personality data come from the Gosling-Potter Internet Personality Project’s most recent dataset, collected from 2003 to 2015 from participants who voluntarily completed personality surveys on the website
We aggregated the individual-level personality scores (
The 2016 U.S. general election data come from open data sources (
For our dependent variable, we focus on three measures of voting behavior. The first measure is Trump’s simple two-party vote share in 2016 and 2020. Vote share was calculated by taking the raw votes for Trump in each region as a proportion of the combined votes for Trump and the Democratic candidate in each region during the 2016 and 2020 U.S. presidential elections. This measure ignores third-party votes. The second measure is Trump’s gains beyond the Republican presidential candidate in the previous election. This measure captures the degree to which regions
The third category of voting behavior captures the vote share of other 2016 Democratic and Republican primary candidates’, which we compare to Trump’s 2016 primary vote share. Analyzing primary election results allows us to investigate the alleged common appeal of Trump and left-wing populist Bernie Sanders as well as whether Trump’s voter base was similar to that of more traditional Republican candidates. We focused on the primary candidates who were arguably Trump’s biggest competition: Bernie Sanders on the Democratic side and John Kasich, Ted Cruz, and Marco Rubio on the Republican side. Although there were other Republican primary candidates in 2016, none received more than 3% of the popular vote.
Sanders’s primary vote share was calculated as the percentage of votes cast for him out of all primary votes cast for candidates of his party in a given region. Put differently, Sanders’s vote share was calculated as the percentage of votes cast for Sanders out of all votes cast for both Bernie Sanders and Hillary Clinton in the Democratic primary because all other 2016 Democratic primary candidates withdrew before or shortly after the primary season began. The vote share for each primary Republican candidates was calculated as the percentage of votes cast for that candidate out of the total number of votes for all four candidates combined (Trump, Kasich, Rubio, and Cruz). We did not consider votes for Bernie Sanders in the 2020 Democratic primary elections because the political landscape had shifted from 2016 such that there were multiple Democratic candidates in 2020 (e.g., Elizabeth Warren) who were running on what might be considered left-wing populist platforms. We did not consider votes for other Republican primary candidates in 2020 because, as sitting president, Donald Trump was the de facto Republican nominee.
As is evident from
Using factor scores as indicators of economic, demographic, and health conditions allows us to address several concerns. Every variable has strengths and weaknesses when it comes to representing a construct of interest. Per capita income may be a good indicator of people’s absolute buying power in a given region, whereas unemployment may be a good indicator of the health of the region’s job market. Both variables provide important information about a region’s economic conditions. Relying on a diversity of sources and measures also helps diminish the influence of limitations, errors, or biases in any individual source’s data. To avoid arbitrarily choosing one variable over another, one could include
Therefore, we conducted exploratory factor analyses for 2016 and 2020 at both regional levels. Before conducting each factor analysis, we consulted a scree plot which suggested that three factors had an eigen value > 1 (
Variable | 2016 |
2020 |
||||
---|---|---|---|---|---|---|
Economic |
Ethnic |
Health |
Economic |
Ethnic |
Health |
|
College degree | -.16 | -.05 | -.16 | -.13 | ||
Agriculture | .12 | -.22 | .12 | -.19 | ||
Income | -.09 | -.26 | -.02 | -.32 | ||
Income inequality | .35 | .08 | .19 | .23 | ||
Migration | .22 | .00 | .20 | -.01 | ||
Internet access | -.15 | -.20 | -.11 | -.22 | ||
Obesity | -.12 | -.01 | .29 | |||
Manufacturing | -.14 | .02 | -.05 | -.06 | ||
Uninsured | .10 | -.01 | .13 | -.06 | ||
No high school | .35 | .13 | .32 | .21 | ||
Ethnic diversity | -.37 | -.08 | -.01 | |||
Teen births | .32 | .24 | .34 | .38 | ||
Smoking | .06 | -.15 | .14 | -.21 | ||
Physical health | .01 | .29 | .01 | .04 | ||
Poverty | .01 | -.02 | .28 | |||
Birthweight | -.14 | .36 | -.17 | .26 | ||
Unemployment | .17 | .39 | .26 | -.18 | .10 | .37 |
Crime | -.10 | .26 | .23 | .13 | .20 | .20 |
Regions that scored high on the first factor, which we call
Regions that scored high on the second factor, which we call
Regions that scored high on the third factor, which we call
Indeed, correlations between the factors show that regions that were economically deprived also tended to have greater ethnic diversity and worse health. In 2016, the economic deprivation and ethnic diversity factors were positively correlated (
The ethnic diversity factor provides a good sense of the objective demographic conditions of a region, but it does not reflect how people perceive or feel about minorities and other historically marginalized groups. Negative subjective attitudes towards minorities were considered core to the original conceptualization of the authoritarian personality (
Therefore, we benchmark the main effects of our focal psychological and structural variables against variables that capture regions’ tendency to experience several forms of group-based dominance: social dominance orientation (SDO, We provide a range of alpha reliabilities for SDO and RWA because different years used different response scales.
Moreover, personality traits that were not included in our previous regression models—extraversion and agreeableness—may also play a role in people’s tendency to aggress against others. People who are willing to denigrate women, minorities, and other groups are likely to be assertive (a facet of extraversion) and lacking in compassion (a facet of agreeableness). Indeed, recent research has linked both agreeableness and extraversion to support for Trump and other populists at the individual level (
We aggregated individual scores on the group-based dominance and personality variables to the county level. We dropped counties that included less than 50 individual observations rather than those with less than 100 individual observations (as was the case in our previous analyses) to increase our total sample of counties. Even so, our final sample size for regressions included group-based dominance variables was about half (
Variable | Min | Max | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | |||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1 | Trump votes 2016 | 63.43 | 15.62 | 4.25 | 92.26 | 1.00 | |||||||||||||||||||||||||
2 | Trump gains 2016 | 5.21 | 5.29 | -17.27 | 21.6 | .43 | 1.00 | ||||||||||||||||||||||||
3 | Trump votes 2020 | 62.35 | 15.79 | 5.53 | 91.28 | .99 | .47 | 1.00 | |||||||||||||||||||||||
4 | Trump gains 2020 | -1.08 | 2.52 | -8.12 | 28.13 | -.01 | .29 | .15 | 1.00 | ||||||||||||||||||||||
5 | Sanders primary votes 2016 | 44.99 | 14.99 | 3.93 | 99.75 | .08 | .22 | .03 | -.27 | 1.00 | |||||||||||||||||||||
6 | Trump primary votes 2016 | 45.82 | 14.80 | 0.34 | 87.99 | -.05 | .11 | -.06 | -.09 | .04 | 1.00 | ||||||||||||||||||||
7 | Kasich primary votes 2016 | 10.77 | 9.85 | 0.00 | 63.78 | -.28 | .06 | -.31 | -.16 | .28 | .00 | 1.00 | |||||||||||||||||||
8 | Cruz primary votes 2016 | 26.75 | 12.15 | 2.07 | 70.39 | .29 | .05 | .30 | .11 | .03 | -.58 | -.31 | 1.00 | ||||||||||||||||||
9 | Rubio primary votes 2016 | 11.70 | 9.22 | 0.14 | 62.66 | -.17 | -.44 | -.18 | -.07 | -.23 | -.51 | -.32 | -.19 | 1.00 | |||||||||||||||||
10 | Neuroticism | 2.93 | 0.08 | 2.56 | 3.24 | .36 | .44 | .39 | .17 | .21 | .08 | .05 | .02 | -.25 | 1.00 | ||||||||||||||||
11 | Openness | 3.61 | 0.09 | 3.31 | 3.93 | -.48 | -.48 | -.51 | -.22 | .03 | .12 | .15 | -.27 | .24 | -.19 | 1.00 | |||||||||||||||
12 | Conscientiousness | 3.59 | 0.08 | 3.35 | 3.94 | -.05 | -.08 | -.04 | .05 | -.37 | -.03 | -.13 | .02 | .08 | -.38 | -.06 | 1.00 | ||||||||||||||
13 | Extraversion | 3.30 | 0.07 | 2.95 | 3.62 | -.03 | -.02 | -.02 | .00 | -.07 | -.16 | .02 | .11 | .01 | -.41 | -.18 | .13 | 1.00 | |||||||||||||
14 | Agreeableness | 3.78 | 0.08 | 3.51 | 4.10 | -.08 | -.09 | -.06 | .14 | -.44 | -.05 | -.18 | .03 | .08 | -.38 | -.18 | .69 | .24 | 1.00 | ||||||||||||
15 | Economic dep. factor 2016 | 0.00 | 1.00 | -5.91 | 1.87 | .61 | .67 | .67 | .46 | -.14 | .02 | -.32 | .27 | -.31 | .38 | -.56 | .08 | -.07 | .12 | 1.00 | |||||||||||
16 | Economic dep. factor 2020 | 0.00 | 1.00 | -6.10 | 1.86 | .63 | .69 | .69 | .43 | -.10 | .01 | -.31 | .27 | -.32 | .38 | -.58 | .07 | -.07 | .11 | .99 | 1.00 | ||||||||||
17 | Ethnic diversity factor 2016 | 0.00 | 1.00 | -1.98 | 5.02 | -.11 | -.29 | -.05 | .37 | -.49 | .08 | -.38 | .00 | .21 | -.09 | .15 | .22 | -.12 | .28 | .18 | .12 | 1.00 | |||||||||
18 | Ethnic diversity factor 2020 | 0.00 | 1.00 | -1.84 | 4.65 | -.11 | -.32 | -.05 | .35 | -.52 | -.01 | -.39 | .07 | .26 | -.12 | .13 | .21 | -.09 | .26 | .16 | .10 | .95 | 1.00 | ||||||||
19 | Health dis. factor 2016 | 0.00 | 1.00 | -2.91 | 4.31 | .16 | .24 | .22 | .39 | -.29 | -.02 | -.25 | .08 | -.07 | .21 | -.19 | .24 | -.08 | .32 | .48 | .46 | .44 | .32 | 1.00 | |||||||
20 | Health dis. factor 2020 | 0.00 | 1.00 | -2.29 | 4.16 | .14 | .21 | .21 | .41 | -.32 | .04 | -.24 | .06 | -.10 | .21 | -.14 | .24 | -.11 | .30 | .49 | .45 | .56 | .42 | .95 | 1.00 | ||||||
21 | Race IAT | 0.33 | 0.05 | -0.04 | 0.45 | .48 | .34 | .47 | -.05 | .41 | .04 | .17 | -.01 | -.22 | .37 | -.34 | -.31 | .06 | -.26 | .17 | .20 | -.46 | -.48 | -.13 | -.14 | 1.00 | |||||
22 | Gender-career IAT | 0.38 | 0.03 | 0.26 | 0.5 | .13 | .04 | .13 | -.02 | -.08 | -.07 | -.08 | .08 | .03 | -.04 | -.11 | .13 | .02 | .17 | .06 | .06 | -.01 | -.02 | .12 | .11 | .02 | 1.00 | ||||
23 | RWA | 3.62 | 0.33 | 2.60 | 4.79 | .52 | .19 | .54 | .11 | -.39 | -.15 | -.32 | .32 | -.05 | .05 | -.43 | .28 | .09 | .27 | .53 | .51 | .19 | .20 | .38 | .37 | -.01 | .14 | 1.00 | |||
24 | SDO | 2.31 | 0.19 | 1.65 | 3.24 | .22 | .08 | .21 | -.05 | -.08 | -.05 | -.04 | .07 | -.03 | .08 | -.21 | -.02 | .11 | .05 | .09 | .08 | -.07 | -.04 | .03 | .02 | .19 | .02 | .22 | 1.00 | ||
25 | Population density 2016 | 845.75 | 13817.59 | 1.49 | 590508 | -.13 | -.12 | -.13 | .00 | -.04 | -.01 | .12 | -.12 | .23 | -.01 | .08 | -.05 | -.02 | -.04 | -.17 | -.17 | -.01 | .00 | -.04 | -.05 | -.04 | -.07 | -.13 | -.09 | 1.00 | |
26 | Population density 2020 | 852.26 | 13877.33 | 1.49 | 593084 | -.13 | -.12 | -.13 | .00 | -.04 | -.01 | .12 | -.12 | .23 | -.01 | .08 | -.05 | -.02 | -.04 | -.17 | -.17 | -.01 | .00 | -.04 | -.05 | -.04 | -.07 | -.13 | -.09 | 1.00 | 1.00 |
Next, we conducted OLS regressions to test the potential direct effect of neuroticism and its interactive effects with structural factors on voting behavior. For each voting outcome (i.e., Trump votes in 2016 and 2020, Trump gains in 2016 and 2020, and primary election votes in 2016), we tested six models. The first model included only the population density covariate and the state-fixed effects. The second model added main effects of the economic, health, and ethnic diversity factors. The third model substituted these three factors for the main effects of neuroticism, openness, and conscientiousness. The fourth model included all personality and structural main effects. The fifth model added our focal interactions between neuroticism and the structural factors (economic deprivation, ethnic diversity, and health disadvantage). The sixth model added interactions between all covariates (openness, conscientiousness, population density) and all three structural factors. We included these additional terms in the sixth model because including main effects of covariates alone does not properly control for the covariate in models that test interactive effects (
All independent variables were
Models 1, 2, and 3 evaluated the extent to which regional differences in the regional covariates, personality traits, and structural factors explained voting behavior, respectively.
Variable | 2016 Trump Votes |
2016 Trump Gains |
||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
(1) | (2) | (3) | (4) | (5) | (6) | (1) | (2) | (3) | (4) | (5) | (6) | |
Neuroticism | 4.27*** | 2.22*** | 1.58*** | 1.41** | 1.75*** | 0.55*** | 0.46*** | 0.49*** | ||||
(3.34 – 5.21) | (1.38 – 3.05) | (0.72 – 2.44) | (0.55 – 2.26) | (1.36 – 2.13) | (0.31 – 0.80) | (0.21 – 0.71) | (0.29 – 0.69) | |||||
Openness | -6.06*** | -2.47*** | -2.44*** | -2.58*** | -1.62*** | 0.00 | 0.02 | 0.10 | ||||
(-6.96 – -5.16) | (-3.30 – -1.64) | (-3.25 – -1.62) | (-3.33 – -1.84) | (-1.86 – -1.38) | (-0.17 – 0.18) | (-0.16 – 0.21) | (-0.12 – 0.32) | |||||
Conscientiousness | -0.90 | -0.88* | -0.62* | -0.38 | 0.55*** | 0.09 | 0.07 | 0.16 | ||||
(-2.18 – 0.38) | (-1.60 – -0.16) | (-1.21 – -0.02) | (-0.84 – 0.08) | (0.32 – 0.78) | (-0.09 – 0.27) | (-0.12 – 0.26) | (-0.04 – 0.36) | |||||
Economic deprivation 2016 | 11.36*** | 9.08*** | 9.21*** | 8.42*** | 3.69*** | 3.47*** | 3.66*** | 3.85*** | ||||
(10.45 – 12.27) | (8.03 – 10.12) | (8.03 – 10.38) | (7.34 – 9.50) | (3.33 – 4.05) | (3.08 – 3.86) | (3.25 – 4.08) | (3.43 – 4.27) | |||||
Ethnic diversity 2016 | -5.67*** | -5.44*** | -5.41*** | -4.47*** | -1.12*** | -1.10*** | -1.17*** | -1.37*** | ||||
(-7.94 – -3.39) | (-7.46 – -3.43) | (-7.42 – -3.40) | (-6.53 – -2.42) | (-1.52 – -0.72) | (-1.50 – -0.69) | (-1.56 – -0.77) | (-1.79 – -0.94) | |||||
Health disadvantage 2016 | -5.74*** | -5.16*** | -4.95*** | -5.36*** | 0.75** | 0.76** | 0.77** | 0.98*** | ||||
(-8.30 – -3.19) | (-7.44 – -2.88) | (-6.77 – -3.14) | (-7.10 – -3.62) | (0.24 – 1.27) | (0.26 – 1.26) | (0.25 – 1.28) | (0.47 – 1.49) | |||||
Population density 2016 | -1.56** | -0.04 | -1.27*** | -0.27** | -0.25* | -10.12** | -0.52*** | 0.05 | -0.40*** | 0.03 | 0.05 | -2.36*** |
(-2.64 – -0.49) | (-0.19 – 0.10) | (-1.95 – -0.59) | (-0.44 – -0.11) | (-0.45 – -0.04) | (-16.51 – -3.74) | (-0.77 – -0.26) | (-0.01 – 0.12) | (-0.54 – -0.26) | (-0.03 – 0.09) | (-0.02 – 0.12) | (-3.22 – -1.50) | |
Neuroticism x Ec. dep. 2016 | -0.70* | 0.39 | 0.38*** | 0.46*** | ||||||||
(-1.37 – -0.04) | (-0.31 – 1.09) | (0.18 – 0.59) | (0.23 – 0.70) | |||||||||
Openness x Ec. dep. 2016 | 1.16*** | -0.36*** | ||||||||||
(0.57 – 1.76) | (-0.56 – -0.16) | |||||||||||
Conscientiousness x Ec. dep. 2016 | -0.33 | -0.15 | ||||||||||
(-0.83 – 0.18) | (-0.41 – 0.12) | |||||||||||
Pop density x Ec. dep. 2016 | -1.97*** | -0.27*** | ||||||||||
(-2.76 – -1.18) | (-0.41 – -0.13) | |||||||||||
Neuroticism x Ethnic div. 2016 | 1.21*** | 0.33 | -0.19 | -0.20 | ||||||||
(0.54 – 1.88) | (-0.33 – 0.99) | (-0.38 – 0.01) | (-0.40 – 0.01) | |||||||||
Openness x Ethnic div. 2016 | -1.70*** | 0.29** | ||||||||||
(-2.63 – -0.78) | (0.10 – 0.48) | |||||||||||
Conscientiousness x Ethnic div. 2016 | -0.62 | 0.03 | ||||||||||
(-1.28 – 0.04) | (-0.20 – 0.27) | |||||||||||
Pop density x Ethnic div. 2016 | 2.57*** | 0.08 | ||||||||||
(1.60 – 3.55) | (-0.15 – 0.30) | |||||||||||
Neuroticism x Health dis. 2016 | 1.35*** | 0.95** | -0.09 | -0.19 | ||||||||
(0.77 – 1.92) | (0.32 – 1.58) | (-0.34 – 0.17) | (-0.45 – 0.07) | |||||||||
Openness x Health dis. 2016 | 0.08 | -0.15 | ||||||||||
(-0.77 – 0.92) | (-0.34 – 0.03) | |||||||||||
Conscientiousness x Health dis. 2016 | -0.86** | -0.10 | ||||||||||
(-1.44 – -0.27) | (-0.32 – 0.13) | |||||||||||
Pop density x Health dis. 2016 | -3.30* | -1.04*** | ||||||||||
(-6.03 – -0.57) | (-1.40 – -0.68) | |||||||||||
Neuroticism x Openness | 0.89** | 0.34*** | ||||||||||
(0.24 – 1.55) | (0.19 – 0.48) | |||||||||||
Neuroticism x Conscientiousness | 0.23 | 0.13 | ||||||||||
(-0.47 – 0.93) | (-0.04 – 0.30) | |||||||||||
Constant | 69.25*** | 74.62*** | 67.92*** | 74.33*** | 74.34*** | 73.95*** | 4.38*** | 2.84*** | 3.91*** | 2.91*** | 2.79*** | -2.00*** |
(69.20 – 69.31) | (71.82 – 77.42) | (67.52 – 68.31) | (71.84 – 76.83) | (72.40 – 76.28) | (71.90 – 76.00) | (4.36 – 4.39) | (2.39 – 3.30) | (3.83 – 3.99) | (2.47 – 3.35) | (2.33 – 3.25) | (-2.41 – -1.59) | |
Observations | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 |
Adjusted |
0.28 | 0.65 | 0.49 | 0.68 | 0.70 | 0.72 | 0.39 | 0.78 | 0.57 | 0.79 | 0.79 | 0.51 |
*
Variable | 2020 Trump Votes |
2020 Trump Gains |
||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
(1) | (2) | (3) | (4) | (5) | (6) | (1) | (2) | (3) | (4) | (5) | (6) | |
Neuroticism | 4.73*** | 2.34*** | 1.65*** | 1.49*** | 0.46*** | 0.11 | 0.03 | 0.03 | ||||
(3.82 – 5.64) | (1.61 – 3.08) | (0.80 – 2.49) | (0.70 – 2.27) | (0.34 – 0.58) | (-0.04 – 0.27) | (-0.13 – 0.20) | (-0.13 – 0.20) | |||||
Openness | -6.50*** | -2.33*** | -2.34*** | -2.47*** | -0.44*** | -0.05 | -0.04 | 0.03 | ||||
(-7.23 – -5.76) | (-3.03 – -1.64) | (-3.03 – -1.64) | (-3.12 – -1.82) | (-0.65 – -0.22) | (-0.28 – 0.17) | (-0.25 – 0.18) | (-0.18 – 0.23) | |||||
Conscientiousness | -0.70 | -0.84* | -0.57 | -0.40 | 0.20* | -0.01 | -0.03 | -0.05 | ||||
(-1.92 – 0.53) | (-1.56 – -0.12) | (-1.16 – 0.02) | (-0.88 – 0.07) | (0.04 – 0.37) | (-0.12 – 0.10) | (-0.13 – 0.08) | (-0.16 – 0.07) | |||||
Economic deprivation 2020 | 12.42*** | 10.14*** | 10.34*** | 9.89*** | 0.69*** | 0.61*** | 0.76*** | 1.11*** | ||||
(11.43 – 13.41) | (9.19 – 11.08) | (9.25 – 11.44) | (8.93 – 10.85) | (0.43 – 0.94) | (0.33 – 0.90) | (0.43 – 1.09) | (0.89 – 1.32) | |||||
Ethnic diversity 2020 | -2.79*** | -2.83*** | -2.89*** | -2.59*** | 1.16*** | 1.16*** | 1.14*** | 0.87** | ||||
(-4.06 – -1.51) | (-3.99 – -1.67) | (-4.01 – -1.76) | (-3.86 – -1.32) | (0.54 – 1.79) | (0.54 – 1.79) | (0.50 – 1.78) | (0.28 – 1.47) | |||||
Health disadvantage 2020 | -6.16*** | -5.51*** | -5.29*** | -5.13*** | 0.42** | 0.43** | 0.42** | 0.55*** | ||||
(-8.94 – -3.37) | (-7.90 – -3.12) | (-7.08 – -3.49) | (-6.95 – -3.31) | (0.16 – 0.67) | (0.17 – 0.69) | (0.16 – 0.68) | (0.29 – 0.80) | |||||
Population density 2020 | -1.52** | 0.11* | -1.20*** | -0.12 | -0.08 | -12.47** | 0.04 | 0.15** | 0.07 | 0.15** | 0.16** | 0.60 |
(-2.49 – -0.56) | (0.01 – 0.21) | (-1.74 – -0.65) | (-0.24 – 0.00) | (-0.21 – 0.05) | (-19.90 – -5.04) | (-0.07 – 0.15) | (0.05 – 0.25) | (-0.06 – 0.21) | (0.05 – 0.24) | (0.06 – 0.27) | (-1.91 – 3.10) | |
Neuroticism x Ec. dep. 2020 | -0.57 | 0.45 | 0.28** | 0.24** | ||||||||
(-1.33 – 0.20) | (-0.23 – 1.14) | (0.10 – 0.45) | (0.07 – 0.42) | |||||||||
Openness x Ec. dep. 2020 | 0.71* | -0.35*** | ||||||||||
(0.11 – 1.31) | (-0.43 – -0.26) | |||||||||||
Conscientiousness x Ec. dep. 2020 | 0.03 | 0.19* | ||||||||||
(-0.54 – 0.60) | (0.03 – 0.34) | |||||||||||
Pop density x Ec. dep. 2020 | -2.21*** | 0.03 | ||||||||||
(-2.96 – -1.46) | (-0.28 – 0.34) | |||||||||||
Neuroticism x Ethnic diversity 2020 | 0.75* | 0.16 | -0.10 | -0.01 | ||||||||
(0.17 – 1.33) | (-0.35 – 0.68) | (-0.22 – 0.02) | (-0.12 – 0.09) | |||||||||
Openness x Ethnic div. 2020 | -0.84* | 0.47*** | ||||||||||
(-1.51 – -0.18) | (0.29 – 0.65) | |||||||||||
Conscientiousness x Ethnic div. 2020 | -0.71* | -0.01 | ||||||||||
(-1.25 – -0.17) | (-0.10 – 0.08) | |||||||||||
Pop density x Ethnic div. 2020 | 3.19*** | -0.21 | ||||||||||
(1.78 – 4.59) | (-0.53 – 0.11) | |||||||||||
Neuroticism x Health dis. 2020 | 1.71*** | 1.08** | -0.06 | -0.08 | ||||||||
(1.10 – 2.31) | (0.41 – 1.75) | (-0.16 – 0.05) | (-0.21 – 0.05) | |||||||||
Openness x Health dis. 2020 | -0.51 | -0.17 | ||||||||||
(-1.28 – 0.26) | (-0.37 – 0.02) | |||||||||||
Conscientiousness x Health dis. 2020 | -0.85** | -0.01 | ||||||||||
(-1.46 – -0.25) | (-0.15 – 0.14) | |||||||||||
Pop density x Health dis. 2020 | -3.14 | 0.34 | ||||||||||
(-7.02 – 0.73) | (-0.60 – 1.28) | |||||||||||
Neuroticism x Openness | 0.95** | 0.09 | ||||||||||
(0.36 – 1.54) | (-0.04 – 0.22) | |||||||||||
Neuroticism x Conscientiousness | 0.28 | -0.02 | ||||||||||
(-0.35 – 0.91) | (-0.14 – 0.10) | |||||||||||
Constant | 68.75*** | 72.23*** | 67.28*** | 71.99*** | 72.00*** | 71.12*** | -0.50*** | -1.80*** | -0.64*** | -1.79*** | -1.89*** | -2.00*** |
(68.70 – 68.80) | (69.61 – 74.86) | (66.94 – 67.63) | (69.69 – 74.29) | (70.27 – 73.74) | (69.22 – 73.02) | (-0.51 – -0.49) | (-2.17 – -1.42) | (-0.71 – -0.56) | (-2.17 – -1.42) | (-2.28 – -1.51) | (-2.41 – -1.59) | |
Observations | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 | 2,083 |
Adjusted |
0.31 | 0.69 | 0.54 | 0.73 | 0.74 | 0.76 | 0.19 | 0.45 | 0.24 | 0.45 | 0.46 | 0.51 |
*
Model 1 revealed that state-fixed effects and population density explain roughly 20–40% of total variance in votes for Trump in 2016 and 2020. Model 2 showed that contextual factors explain an additional 30–40% of variance in Trump’s vote share and gains. More specifically, the economic deprivation factor (2016 votes:
In Model 3, we substituted the three structural factors with three personality traits: neuroticism, conscientiousness, and openness. Compared to the baseline model 1, the traits explained an additional 10–20% of variance in votes for Trump. Our focal trait—regional neuroticism—was positively associated with Trump votes and gains in both elections (2016 votes:
In Model 4, we jointly include the personality traits and structural factors. Most of the results from the Models 2 and 3 stay unchanged except that the neuroticism trait no longer significantly predicts Trump gains in 2020. Additionally, openness and conscientiousness do not consistently predict Trump gains in 2016 and 2020 in this model.
Models 5 and 6 investigated interactive effects of the personality traits with each of the three structural factors: economic deprivation, ethnic diversity, and health disadvantage. Model 5 included only the neuroticism interactions with the three structural factors, whereas Model 6 included interaction between all three traits (neuroticism, openness, and conscientiousness) and structural factors (economic deprivation, ethnic diversity, and health disadvantage). Therefore, from the latter model we only discuss the openness and conscientiousness interactions.
The interaction between neuroticism and the economic deprivation factor negatively predicted Trump’s vote share in 2016 but positively predicted gains in both elections (2016 votes:
The interaction between neuroticism and the ethnic diversity factor positively predicted Trump votes in 2016 and 2020 (2016 votes:
Similarly, the interactions between neuroticism and the health disadvantage factor positively predicted Trump votes in 2016 and 2020 (2016 votes:
Taken together, the interactions show the expected pattern: neuroticism tended to exacerbate the strength of the relationship between structural factors and Trump voting. However, size of the interaction coefficients demonstrate that the interaction effects were weak, not explaining much variance in voting beyond main effects of personality and structural factors. Moreover, the interaction effects were not particularly robust across dependent variables (vote share and gains) or after accounting for robustness checks (as summarized in
Therefore, we focus our interpretation of the results on the large and robust main effects of neuroticism and structural factors on voting. Openness and conscientiousness also interacted with structural factors on votes for Trump, consistent with interactionist theories. We do not interpret these interactions because they also were not particularly robust or consistent, and we did not have specific predictions regarding interactions with other personality traits.
Variable | 2016 Kasich Primary Votes |
2016 Cruz Primary Votes |
2016 Rubio Primary Votes |
||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
(1) | (2) | (3) | (4) | (5) | (1) | (2) | (3) | (4) | (5) | (1) | (2) | (3) | (4) | (5) | |
Neuroticism | -1.08*** | -0.16 | -0.18 | -0.46 | -0.74*** | -0.71** | -0.93** | -0.12 | -0.49* | ||||||
-1.56–-0.60 | -0.53–0.20 | -0.54–0.18 | -0.94–0.02 | -1.11– -0.38 | -1.18–-0.25 | -1.46–-0.40 | -0.39–0.14 | -0.91–-0.07 | |||||||
Openness | 1.26*** | 0.02 | 0.14 | -1.03*** | -0.55** | -0.54** | 1.16*** | 0.09 | 0.07 | ||||||
0.92–1.59 | -0.19–0.24 | -0.08–0.36 | -1.45–-0.61 | -0.88–-0.23 | -0.94–-0.14 | 0.72–1.61 | -0.24–0.43 | -0.28–0.41 | |||||||
Conscientiousness | -0.43** | -0.05 | -0.11 | -0.28 | -0.21 | -0.22 | -0.40* | -0.09 | -0.03 | ||||||
-0.70–-0.16 | -0.25–0.15 | -0.35–0.13 | -0.72–0.16 | -0.55–0.13 | -0.60–0.17 | -0.71–-0.08 | -0.35–0.17 | -0.26–0.21 | |||||||
Economic deprivation. 2016 | -2.86*** | -2.78*** | -2.46*** | 1.53*** | 1.55*** | 1.41** | -2.63*** | -2.53*** | -2.24*** | ||||||
-3.31–-2.41 | -3.25–-2.31 | -2.99–-1.93 | 0.80–2.27 | 0.79–2.31 | 0.59–2.23 | -3.71–-1.55 | -3.59–-1.47 | -3.15–-1.33 | |||||||
Ethnic diversity 2016 | 0.49 | 0.48 | 0.33 | -2.06*** | -2.07*** | -1.77*** | 0.29 | 0.27 | -0.04 | ||||||
-0.05–1.03 | -0.05–1.01 | -0.15–0.81 | -2.86–-1.26 | -2.86–-1.28 | -2.42–-1.12 | -0.44–1.01 | -0.44–0.98 | -0.58–0.49 | |||||||
Health disadvantage. 2016 | -0.70 | -0.70 | -0.78* | -0.42 | -0.38 | -0.30 | -0.46 | -0.46* | -0.19 | ||||||
-1.43–0.03 | -1.41–0.00 | -1.47–-0.08 | -1.45–0.61 | -1.38–0.62 | -1.20–0.60 | -0.94–0.01 | -0.91–-0.00 | -0.52–0.13 | |||||||
Population density 2016 | 3.07 | -0.68 | 1.56 | -0.66 | 0.71 | -1.67* | 1.52* | -0.94 | 1.79* | -1.27 | 7.56*** | 3.12*** | 5.47*** | 3.07*** | 0.18 |
-0.17–6.31 | -2.54–1.17 | -0.94–4.05 | -2.47–1.16 | -2.20–3.63 | -2.97–-0.37 | 0.34–2.71 | -2.12–0.25 | 0.41–3.17 | -3.15–0.60 | 3.36–11.76 | 2.07–4.17 | 3.15–7.80 | 2.10–4.05 | -2.44–2.80 | |
Neuroticism x Ec. dep. 2016 | 0.26 | -0.09 | 0.45* | ||||||||||||
-0.13–0.64 | -0.51–0.34 | 0.11–0.80 | |||||||||||||
Openness x Ec. dep. 2016 | -0.20 | -0.16 | -0.07 | ||||||||||||
-0.49–0.08 | -0.62–0.30 | -0.55–0.41 | |||||||||||||
Conscientiousness x Ec. dep. 2016 | -0.05 | 0.05 | 0.05 | ||||||||||||
-0.35–0.24 | -0.49–0.60 | -0.36–0.46 | |||||||||||||
Pop density x Ec. dep. 2016 | -1.64** | 0.52 | -3.05** | ||||||||||||
-2.61–-0.66 | -0.29–1.33 | -4.86–-1.24 | |||||||||||||
Neuroticism x Ethnic div. 2016 | 0.05 | 0.36 | -0.69*** | ||||||||||||
-0.36–0.46 | -0.02–0.74 | -1.01–-0.38 | |||||||||||||
Openness x Ethnic diversity 2016 | 0.07 | -0.38 | 0.37** | ||||||||||||
-0.14–0.29 | -0.89–0.13 | 0.13–0.61 | |||||||||||||
Conscientiousness x Ethnic div. 2016 | 0.21 | -0.29 | -0.17 | ||||||||||||
-0.02–0.44 | -0.57–0.00 | -0.44–0.10 | |||||||||||||
Pop density x Ethnic div. 2016 | -3.34*** | 2.55*** | -1.52 | ||||||||||||
-4.87–-1.81 | 1.15–3.95 | -5.08–2.03 | |||||||||||||
Neuroticism x Health dis. 2016 | -0.24 | -0.43* | 0.20 | ||||||||||||
-0.57–0.08 | -0.83–-0.03 | -0.14–0.54 | |||||||||||||
Openness x Health dis. 2016 | 0.03 | 0.59*** | 0.01 | ||||||||||||
-0.20–0.26 | 0.26–0.91 | -0.27–0.28 | |||||||||||||
Conscientiousness x Health dis. 2016 | 0.04 | 0.13 | -0.23 | ||||||||||||
-0.21–0.29 | -0.25–0.51 | -0.56–0.10 | |||||||||||||
Pop density x Health dis. 2016 | 2.23 | -0.76 | 1.06 | ||||||||||||
-0.21–4.68 | -2.38–0.87 | -0.33–2.44 | |||||||||||||
Neuroticism x Openness | 0.05 | -0.17 | -0.20 | ||||||||||||
-0.17–0.26 | -0.41–0.07 | -0.43–0.03 | |||||||||||||
Neuroticism x Conscientiousness | 0.07 | 0.25 | 0.11 | ||||||||||||
-0.04–0.18 | -0.02–0.52 | -0.02–0.24 | |||||||||||||
Constant | 3.69*** | 5.09*** | 3.99*** | 5.08*** | 5.06*** | 20.32*** | 21.71*** | 20.11*** | 21.55*** | 21.40*** | 16.10*** | 17.25*** | 16.34*** | 17.24*** | 17.00*** |
3.52–3.86 | 4.41–5.77 | 3.79–4.18 | 4.41–5.74 | 4.30–5.82 | 20.25–20.39 | 20.87–22.56 | 19.95–20.28 | 20.71–22.39 | 20.48–22.32 | 15.87–16.32 | 16.87–17.63 | 16.13–16.55 | 16.87–17.62 | 16.47–17.52 | |
Observations | 1,901 | 1,901 | 1,901 | 1,901 | 1,901 | 1,901 | 1,901 | 1,901 | 1,901 | 1,901 | 1,660 | 1,660 | 1,660 | 1,660 | 1,660 |
Adjusted |
0.85 | 0.92 | 0.88 | 0.92 | 0.92 | 0.77 | 0.78 | 0.77 | 0.78 | 0.79 | 0.80 | 0.86 | 0.82 | 0.86 | 0.87 |
*
As previously mentioned, variables that capture a tendency towards group-based dominance like right-wing authoritarianism (RWA), social dominance orientation (SDO) as well as implicit race and gender bias have been shown to be strong predictors of voting for Trump. Similarly, people high in extraversion and low in agreeableness may be particularly likely to engage in aggressive behavior. To test these alternative explanations, we regressed Trump vote shares and gains on these variables in
Variable | 2016 Trump votes |
2016 Trump gains |
2020 Trump votes |
2020 Trump gains |
||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
(1) | (2) | (3) | (1) | (2) | (3) | (1) | (2) | (3) | (1) | (2) | (3) | |
Race IAT | 7.97*** | 8.04*** | 4.63*** | 0.72*** | 0.80*** | 0.15 | 7.63*** | 7.72*** | 4.39*** | -0.34* | -0.31* | -0.01 |
(6.78 – 9.16) | (6.73 – 9.34) | (3.38 – 5.88) | (0.34 – 1.11) | (0.40 – 1.19) | (-0.18 – 0.48) | (6.56 – 8.70) | (6.55 – 8.90) | (3.29 – 5.49) | (-0.62 – -0.07) | (-0.61 – -0.02) | (-0.24 – 0.23) | |
Gender-Career IAT | 0.57 | 0.56 | 0.44 | 0.25* | 0.23 | 0.20 | 0.53 | 0.51 | 0.45 | -0.04 | -0.05 | -0.00 |
(-0.15 – 1.29) | (-0.20 – 1.32) | (-0.19 – 1.08) | (0.03 – 0.47) | (-0.00 – 0.47) | (-0.01 – 0.40) | (-0.16 – 1.22) | (-0.23 – 1.25) | (-0.14 – 1.05) | (-0.19 – 0.11) | (-0.21 – 0.10) | (-0.11 – 0.11) | |
RWA | 6.19*** | 6.15*** | 1.98*** | 1.89*** | 1.85*** | -0.34** | 6.41*** | 6.35*** | 1.79** | 0.22* | 0.20* | -0.32*** |
(5.11 – 7.27) | (5.07 – 7.24) | (0.98 – 2.98) | (1.51 – 2.26) | (1.51 – 2.19) | (-0.59 – -0.10) | (5.34 – 7.47) | (5.29 – 7.42) | (0.77 – 2.81) | (0.03 – 0.40) | (0.02 – 0.38) | (-0.50 – -0.15) | |
SDO | 0.18 | 0.20 | 0.37 | -0.19 | -0.17 | 0.04 | 0.04 | 0.06 | 0.33 | -0.14 | -0.14 | -0.10 |
(-0.50 – 0.86) | (-0.47 – 0.86) | (-0.14 – 0.87) | (-0.46 – 0.09) | (-0.43 – 0.09) | (-0.13 – 0.20) | (-0.58 – 0.66) | (-0.54 – 0.67) | (-0.11 – 0.76) | (-0.33 – 0.04) | (-0.32 – 0.05) | (-0.25 – 0.06) | |
Agreeableness | 0.07 | -0.96 | 0.16 | -0.42* | 0.21 | -0.82 | 0.14 | 0.31 | ||||
(-1.26 – 1.40) | (-2.00 – 0.08) | (-0.42 – 0.74) | (-0.81 – -0.02) | (-1.10 – 1.51) | (-1.73 – 0.10) | (-0.17 – 0.44) | (-0.05 – 0.68) | |||||
Extraversion | -1.12* | 0.29 | -1.02*** | 0.25 | -1.36** | 0.44 | -0.24* | 0.15 | ||||
(-2.06 – -0.18) | (-0.62 – 1.20) | (-1.49 – -0.54) | (-0.05 – 0.55) | (-2.30 – -0.43) | (-0.43 – 1.30) | (-0.43 – -0.05) | (-0.08 – 0.38) | |||||
Neuroticism | 2.00** | 1.03*** | 2.38*** | 0.38** | ||||||||
(0.76 – 3.24) | (0.58 – 1.48) | (1.28 – 3.48) | (0.15 – 0.60) | |||||||||
Openness | -3.14*** | -0.19 | -3.02*** | 0.11 | ||||||||
(-4.29 – -1.99) | (-0.50 – 0.13) | (-3.95 – -2.09) | (-0.22 – 0.44) | |||||||||
Conscientiousness | 1.05* | 0.55** | 0.83 | -0.28 | ||||||||
(0.01 – 2.09) | (0.23 – 0.86) | (-0.04 – 1.69) | (-0.56 – 0.01) | |||||||||
Economic deprivation | 6.65*** | 3.20*** | 7.32*** | 0.46** | ||||||||
(5.43 – 7.87) | (2.75 – 3.65) | (6.16 – 8.49) | (0.07 – 0.84) | |||||||||
Ethnic diversity | -4.93*** | -0.95** | -2.33*** | 1.33*** | ||||||||
(-6.98 – -2.88) | (-1.52 – -0.38) | (-3.77 – -0.89) | (0.76 – 1.89) | |||||||||
Health disadvantage | -2.96** | 1.05** | -3.38** | 0.54 | ||||||||
(-4.72 – -1.20) | (0.37 – 1.73) | (-5.12 – -1.64) | (-0.19 – 0.88) | |||||||||
Population density | -0.63* | -0.65* | -0.10 | -0.26** | -0.27*** | 0.02 | -0.56* | -0.58** | 0.00 | 0.07 | 0.07 | 0.10** |
(-1.18 – -0.09) | (-1.18 – -0.12) | (-0.23–0.04) | (-0.42 – -0.10) | (-0.43 – -0.12) | (-0.03 – 0.07) | (-1.00 – -0.12) | (-1.00 – -0.15) | (-0.13 – 0.14) | (-0.04 – 0.19) | (-0.04 – 0.19) | (0.02 – 0.18) | |
Constant | 62.93*** | 63.26*** | 72.92*** | 0.82*** | 1.09*** | 3.79*** | 60.24*** | 60.60*** | 70.28*** | -2.69*** | -2.66*** | -2.34*** |
(61.42 – 64.43) | (61.76 – 64.75) | (70.60 – 75.24) | (0.41 – 1.23) | (0.63 – 1.55) | (3.15 – 4.43) | (58.79 – 61.68) | (59.13 – 62.07) | (68.10 – 72.46) | (-2.98 – -2.40) | (-2.93 – -2.38) | (-2.66 – -2.02) | |
Observations | 1,096 | 1,096 | 1,096 | 1,096 | 1,096 | 1,096 | 1,096 | 1,096 | 1,096 | 1,096 | 1,096 | 1,096 |
Adjusted |
0.59 | 0.60 | 0.79 | 0.50 | 0.51 | 0.82 | 0.61 | 0.61 | 0.81 | 0.21 | 0.22 | 0.46 |
*
Across the three models, implicit anti-Black bias positively predicted Trump vote share in 2016, his gains in 2016, and his vote share in 2020. However, this relationship did not hold for Trump’s gains in 2020, when anti-Black bias was either unrelated or negatively related to Trump performance. This result is consistent with the previous finding showing that Trump made gains in 2020 over his performance in 2016 in more ethnically diverse regions. Right-wing authoritarianism also consistently predicted Trump’s vote share in 2016 and 2020. Unexpectedly, however, the relationship between right-wing authoritarianism and Trump’s vote gains in 2016 and 2020 became negative in Model 3, which includes neuroticism and our structural factors. Contrary to past research, SDO and implicit gender bias did not consistently predict Trump voting.
Perhaps most importantly, the main effects of neuroticism, economic deprivation, ethnic diversity, and health disadvantage largely held when accounting for variables capturing group-based dominance. Moreover, the effect size of our focal variables were comparable to or larger than the effect size of the group-based dominance variables.
To investigate whether Trump’s voter base is similar to that of other primary candidates, we analyzed Trump’s vote share in the 2016 primary elections, comparing the characteristics of regions that voted for Trump to those that voted for Bernie Sanders, John Kasich, Ted Cruz, and Marco Rubio. Results of these analyses are depicted in Note that we included five models in
Beginning with comparisons between Sanders and Trump, more neurotic counties were indeed more likely to vote in 2016 for both Sanders (
Variable | 2016 Sanders Primary Votes |
2016 Trump Primary Votes |
||||||||
---|---|---|---|---|---|---|---|---|---|---|
(1) | (2) | (3) | (4) | (5) | (1) | (2) | (3) | (4) | (5) | |
Neuroticism | 0.62 | 1.04** | 0.52 | 2.30*** | 1.11*** | 1.38*** | ||||
(-0.06 – 1.31) | (0.43 – 1.64) | (-0.05 – 1.09) | (1.65 – 2.94) | (0.49 – 1.73) | (0.73 – 2.02) | |||||
Openness | 0.90** | 0.46 | 0.43 | -1.15*** | 0.42 | 0.48* | ||||
(0.24 – 1.56) | (-0.37 – 1.29) | (-0.42 – 1.27) | (-1.64 – -0.65) | (-0.02 – 0.85) | (0.08 – 0.88) | |||||
Conscientiousness | -1.76*** | -1.40*** | -1.22*** | 1.06*** | 0.51* | 0.47* | ||||
(-2.53 – -1.00) | (-1.96 – -0.85) | (-1.70 – -0.75) | (0.58 – 1.54) | (0.10 – 0.92) | (0.11 – 0.83) | |||||
Economic deprivation 2020 | -0.14 | -0.45 | -0.22 | 3.59*** | 3.36*** | 3.08*** | ||||
(-1.33 – 1.05) | (-1.94 – 1.04) | (-1.95 – 1.51) | (2.97 – 4.20) | (2.56 – 4.16) | (2.43 – 3.74) | |||||
Ethnic diversity 2016 | -3.96*** | -3.94*** | -3.85*** | 0.61 | 0.65 | 0.60 | ||||
(-5.20 – -2.71) | (-5.22 – -2.66) | (-4.93 – -2.78) | (-0.99 – 2.21) | (-0.89 – 2.18) | (-0.64 – 1.84) | |||||
Health disadvantage 2016 | 0.05 | 0.47 | 0.52 | 0.72 | 0.65 | 0.57 | ||||
(-1.80 – 1.89) | (-1.23 – 2.17) | (-0.78 – 1.82) | (-0.59 – 2.03) | (-0.65 – 1.94) | (-0.61 – 1.75) | |||||
Population density 2016 | -3.33 | -0.97 | -4.26 | -1.71 | 2.00 | -3.80 | 0.12 | -1.92 | -0.14 | -2.46 |
(-9.35 – 2.69) | (-6.29 – 4.34) | (-10.23 – 1.71) | (-6.91 – 3.48) | (-9.07 – 13.08) | (-9.17 – 1.58) | (-3.59 – 3.83) | (-6.20 – 2.37) | (-3.61 – 3.34) | (-6.59 – 1.68) | |
Neuroticism x Ec. dep. 2016 | 0.28 | -0.58 | ||||||||
(-0.58 – 1.14) | (-1.18 – 0.02) | |||||||||
Openness x Ec. dep. 2016 | 0.05 | 0.12 | ||||||||
(-0.54 – 0.64) | (-0.43 – 0.66) | |||||||||
Conscientiousness x Ec. dep. 2016 | 0.45 | -0.07 | ||||||||
(-0.15 – 1.05) | (-0.54 – 0.40) | |||||||||
Pop density x Ec. dep. 2016 | -1.46 | 2.08*** | ||||||||
(-3.25 – 0.32) | (1.04 – 3.11) | |||||||||
Neuroticism x Ethnic div. 2016 | 0.01 | 0.14 | ||||||||
(-0.94 – 0.97) | (-0.33 – 0.61) | |||||||||
Openness x Ethnic div. 2016 | -0.48 | -0.16 | ||||||||
(-0.98 – 0.03) | (-0.56 – 0.25) | |||||||||
Conscientiousness x Ethnic div. 2016 | -0.44* | 0.22 | ||||||||
(-0.80 – -0.08) | (-0.17 – 0.62) | |||||||||
Pop density x Ethnic div. 2016 | -3.62 | 4.42** | ||||||||
(-9.36 – 2.11) | (1.88 – 6.96) | |||||||||
Neuroticism x Health dis. 2016 | 0.25 | 0.48 | ||||||||
(-0.78 – 1.27) | (-0.42 – 1.38) | |||||||||
Openness x Health dis. 2016 | 0.44 | -0.69*** | ||||||||
(-0.35 – 1.22) | (-1.02 – -0.37) | |||||||||
Conscientiousness x Health dis. 2016 | -0.53* | 0.24 | ||||||||
(-1.02 – -0.03) | (-0.12 – 0.60) | |||||||||
Pop density x Health dis. 2016 | 2.89 | -3.08** | ||||||||
(-1.62 – 7.39) | (-4.89 – -1.27) | |||||||||
Neuroticism x Openness | -0.03 | 0.42* | ||||||||
(-0.65 – 0.58) | (0.08 – 0.76) | |||||||||
Neuroticism x Conscientiousness | 0.15 | -0.30 | ||||||||
(-0.35 – 0.65) | (-0.76 – 0.16) | |||||||||
Constant | 21.26*** | 24.12*** | 21.78*** | 24.14*** | 24.50*** | 47.86*** | 45.39*** | 47.55*** | 45.59*** | 45.38*** |
(20.94 – 21.58) | (22.37 – 25.87) | (21.32 – 22.23) | (22.64 – 25.64) | (22.97 – 26.02) | (47.58 – 48.15) | (44.01 – 46.77) | (47.23 – 47.87) | (44.25 – 46.92) | (44.11 – 46.66) | |
Observations | 1,932 | 1,932 | 1,932 | 1,932 | 1,932 | 1,901 | 1,901 | 1,901 | 1,901 | 1,901 |
Adjusted |
0.67 | 0.70 | 0.69 | 0.71 | 0.72 | 0.78 | 0.83 | 0.80 | 0.83 | 0.84 |
*
Counties that voted for Sanders were also lower on conscientiousness (
It is also interesting to note the major differences between the Republican candidates in the primaries. Not a single independent variable consistently predicted voting for Republican primary candidates. Indeed, there is no overlap between the profile of regions that voted for Trump and those that voted for Rubio and Kasich. In fact, counties that voted for Kasich and Rubio tended to have
Interestingly, there are some differences between the determinants of Trump’s 2016 general election voting share and his primary voting share. While neuroticism and economic deprivation are significant predictors in both elections, neither ethnic diversity, health disadvantage, nor openness predicted Trump’s primary results. One reason for the discrepancies between primary and general election results could be that Trump’s campaign changed somewhat over time to appeal to additional voting groups. A related interpretation is that the regions that contributed to Trump’s nomination are not exactly the same as those that contributed to his general election win.
As noted in the Methods, spatial analyses require unique considerations such as including appropriate controls like state fixed effects, as we included in the models reported above. In addition, we conducted robustness checks to account for spatial autocorrelation (by running analyses with spatial lags on the independent variables using the normalized inverse of distances between regions), the potential MAUP problem (by conducting analyses at the CBSA level), and non-representativeness of the personality data (by weighting respondents by age and gender). Full results of these analyses are reported in the
Variable | Trump 2016 Votes |
Trump 2016 Gains |
Trump 2020 Votes |
Trump 2020 Gains |
||||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
(1) | (2) | (3) | (4) | (5) | (1) | (2) | (3) | (4) | (5) | (1) | (2) | (3) | (4) | (5) | (1) | (2) | (3) | (4) | (5) | |
Personality Main Effects | ||||||||||||||||||||
Neuroticism | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + | |||
Openness | - | - | - | - | - | + | - | - | - | - | - | |||||||||
Conscientiousness | - | - | + | + | + | - | - | |||||||||||||
Structural Factor Main Effects | ||||||||||||||||||||
Economic deprivation | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + |
Ethnic diversity | - | - | - | - | - | - | - | - | - | - | - | - | - | + | + | + | + | + | ||
Health disadvantage | - | - | - | - | - | + | + | + | + | + | - | - | - | - | - | + | + | + | + | + |
Interactions | ||||||||||||||||||||
N x Economic deprivation | - | - | - | / | + | + | + | / | - | / | + | + | + | / | ||||||
N x Ethnic diversity | + | + | + | / | - | - | / | + | + | + | / | - | / | |||||||
N x Health disadvantage | + | + | + | + | / | / | + | + | + | + | / | / |
As is evident from the table, the main effects of neuroticism and the structural factors on voting for Trump were consistent across nearly all robustness checks and dependent variables. Higher neuroticism, more economic disadvantage, and lower ethnic diversity predicted Trump voting. Worse health predicted more Trump gains but fewer Trump votes in 2016 and 2020. Of the 80 main effects of neuroticism and structural factors depicted in
To many observers, Donald Trump’s rise was surprising, but the roots of his support corroborate foundational and modern theories of authoritarianism. Regions that were more prone to negative emotionality (as captured by higher neuroticism) and that were experiencing threatening structural conditions (as captured by 18 variables reflecting economic deprivation, lack of ethnic diversity, and health disadvantage) were those most likely to vote for Trump. In fact, regional neuroticism, economic deprivation, and lack of ethnic diversity predicted voting for Trump even after accounting for geographic confounds (e.g., population density, spatial autocorrelation), other personality traits (e.g., conscientiousness, openness), and measures of group-based dominance (e.g., social dominance orientation). Moreover, these findings generalized in exploratory and confirmatory pre-registered analyses, across regional levels (i.e., counties and CBSAs), to the population as a whole (i.e., when weights were used to make the sample more representative), and across election years (i.e., 2016 and 2020). The fact that the same kinds of regions were likely to vote for Trump in 2016 and 2020 demonstrates that these regions’ preference was not merely for an anti-incumbent candidate. In short, both threat from internal psychological dispositions
Our findings with regard to neuroticism speak to a large and contradictory literature on the role of negativity bias and negative emotions in political behavior (
Neuroticism and threatening structural factors were not only robustly related to Trump voting, but they were also
Our findings suggest that Trump voting can be conceptualized as authoritarian voting and is explained by an array of regional factors that vary psychologically and structurally (
The work of scholars like Fromm, who sought to explain the rise of Nazism in the 20th century, has striking parallels in the modern-day context. Trump’s base may be different from the voter base of his contemporary political rivals. Still, his voter base is uncannily similar in some respects to the voter base of the historical figure who instigated the modern study of authoritarianism—Adolf Hitler. In particular, regions that voted for Trump were not those experiencing the greatest economic hardships, just as people who voted for Hitler were not those hardest hit by the Great Depression (
Our factor analytic method painted a broad portrait of the threatening economic conditions that explain Trump’s vote share and his performance gains over time. But one variable from the economic deprivation factor stood out: college degree attainment. An examination of zero-order correlations showed that lack of college attainment correlated more strongly than any other variable with Trump’s vote share in 2016 and 2020, Trump’s gains in 2016 over Romney’s performance in 2012, and Trump’s gains in 2020 over his previous performance in 2016. Therefore, boosting rates of college attainment by increasing college enrollment or retention (e.g.,
As might be evident from the historical example described in the previous section, economic deprivation in Trump-supporting regions must be understood in the context of white people’s perceptions that minorities’ status is increasing and the antipathy that can result from such perceptions ( In addition, the economic deprivation factor (which was positively associated with Trump voting) was characterized by lower, not higher, income inequality. This may seem inconsistent with evidence that economic inequality undermines democracy and enhances support for strong leaders (e.g.,
Importantly, our method was not designed to determine the relative influence economic deprivation
Our analyses show that more ethnically diverse regions preferred Democratic presidential candidates Clinton in 2016 and Biden in 2020, and that ethnically diverse regions were less likely to vote for Trump than they were to vote for Mitt Romney in 2012. However, Trump performance improved in ethnically diverse regions, and anti-Black implicit bias did not positively predict Trump’s gains in 2020 the way it did in 2016. Further work is needed to understand why Trump’s performance improved in ethnically diverse regions. One possible explanation is that the Republican effort in 2020 to paint Democratic rival Joe Biden as a socialist might have been effective in provoking fear among some Hispanic community members who may have had experience with socialist regimes in their countries of origin (e.g., Cuba, Venezuela). Another interpretation is more mundane – Trump may have already been so popular in ethnically homogenous regions that the only places where he could improve his performance were ethnically diverse.
Regarding regions high on health disadvantage, Democratic presidential candidates in 2016 and 2020 outperformed Trump in these regions overall. This finding is consistent with theories arguing that people vote for the leaders that they believe can best address the given threat at hand (
Prior research shows an inconsistent relationship between neuroticism and right-wing voting behavior. Only recently has this relationship been robustly empirically documented (
One promising avenue for future research would be to examine interactions at different levels of analysis. For instance, regional neuroticism may interact with threatening
When it comes to other personality traits, openness and conscientiousness have, in past research, been most strongly linked to political ideology at the individual and regional levels (e.g.,
Although we attempted to control for a wide variety of potential confounds, the most obvious limitation of this work is the lack of evidence for a causal role for psychological and structural factors on Trump voting. Indeed, recent research suggests that personality traits and political preferences can influence each other bidirectionally (
Moreover, the factors we studied are highly stable over time at the regional level, suggesting that the roots of Trump support have been long in the making (as was also the case for the roots of Nazi support;
Neuroticism and threatening economic, demographic, and health conditions are not the only psychological and structural variables that may explain Trump support. With regard to psychological variables, we emphasized the role of “hot” emotional factors, but “cold” cognitive factors like cognitive ability and rigidity have also played a role in recent voting behavior (
One limitation of our analytic approach is that it ignores third party votes, which might have an influential effect on election outcomes. Furthermore, our focus was on predicting voting for Trump, but vote choice is just one component of what determines election
Around the world voters in democratic regimes are electing leaders who “routinely ignore constitutional limits on their power” (
The data collection was declared exempt from informed consent by the approval of the Institutional Review Board at the University of Texas at Austin because there were no significant risks to participants (IRB #2004–10-0073).
A pre-print of this manuscript was uploaded to
Samuel D. Gosling is a senior consultant member of Personality Science. We have no other known competing interests.
We thank Diana Jordan and Faiza Quadri for research assistance.
For this article, data is freely available (
The Supplementary Materials include pre-registration, data, code, additional methodological details (e.g., regarding data quality, exclusions, robustness checks, data source details), and additional results (e.g., regarding factor analyses, CBSA results, weighted results, spatial autocorrelation controls, and additional visualizations). For access see
The authors have no funding to report.