Economic Anxiety or Cultural Backlash: Which Is Key to Trump’s Support? (Essay 4)
Economic Anxiety or Cultural Backlash: Which Is Key to Trump’s Support? (Essay 4)
Economic Anxiety or Cultural Backlash: Which Is Key to Trump’s Support? (Essay 4)
Morris P. Fiorina
2016 Was the Year White Liberals Realized How Unjust, Racist, and Sexist America Is.1
For anyone who voted for Donald Trump, bald-faced racism and sexism were not the deal-
breakers they should have been. Hatred of women was on the ballot in November, and it won.2
Donald Trump has won the presidency, despite an unprecedented level of unfitness and in
defiance of nearly every prediction and poll. And he’s done this not despite but because he
expressed unfiltered disdain toward racial and religious minorities in the country.3
You have to accept that millions of people who voted for Barack Obama, some of them once,
some of them twice, changed their minds this time. They’re not racist.4
The reason Donald Trump was elected was that we automated away four million manufacturing
jobs in Michigan, Ohio, Pennsylvania, and Wisconsin.6
Given that 85–90 percent of Trump’s (and Clinton’s) vote came from partisans—people who
nearly always vote Republican or Democrat—claims like these applied to the behavior of
a relatively small proportion of the electorate, although one residing disproportionately in
states critical for the outcome.7 In particular, among other factors, the 2016 outcome hinged
The COVID-19 pandemic relegated these competing narratives to a lower level in the 2020
election season, and Joe Biden’s vanquishing of Trump seemed to further decrease the atten-
tion to whether culture or economics better explained Trump’s appeal. Given another election
involving Trump in 2024, however, the debate seems worth revisiting so that analysts can be
better prepared to adjudicate the debate than when they first examined it in 2016.
This essay critically examines the findings of studies that consider economic versus “cultural”
explanations for Trump’s surprising victory. While recognizing the heterogeneity of the cate
gory, I use the term “cultural” for the collection of noneconomic reasons offered as expla-
nations for Trump support: racial prejudice, ethnic prejudice, religious prejudice, nativism,
misogyny, and various social-psychological conditions that incorporate or reflect (at least
partially) such motivations; for example, “social dominance orientation,” “status anxiety,”
“White consciousness,” “hostile sexism,” and “hegemonic masculinity.”10 Both economic
and cultural motivations surely played a role in voting for Trump, but the bottom line is that,
contrary to many premature conclusions in the literature, it is extremely difficult, perhaps
impossible, for even a disinterested and methodologically sophisticated analyst to identify
the relative contribution of each class of motivation. The relevant literature is large, so what
follows is not a comprehensive literature review. Rather, I select some prominent studies that
illustrate various problems of inference.11 For purposes of discussion, I divide the problems
into five categories, although they overlap in some cases. I begin with the most obvious
of the five categories.
PROBLEMATIC MEASURES
Economic concepts—the unemployment rate, real income per capita, household income,
and so forth—generally have more precise meanings and measures than do psychologi-
cal concepts, such as status anxiety, White consciousness, social dominance orienta-
tion, and racial resentment. The latter typically are measured by batteries of survey items
that purport to capture the theoretical concept. However, even if economic variables
have less measurement error than cultural measures, analysis after analysis concludes
that cultural variables show much stronger relationships to Trump’s vote than do more pre-
cisely measured economic v
ariables, consistent with the conclusion that culture trumps
economics.
One reason for the apparently greater importance of cultural variables may be that some
commonly used economic measures, although they may be more precisely defined, fail to
Some of the cultural measures may have the opposite problem: they capture too much—more
than the concept they are designed to measure. One of the long-standing examples in the
political science literature is the concept of “racial resentment,” which frequently appears
in the analyses under consideration.14 Four survey items comprise the measure:
1. How much R agrees or disagrees that Blacks should work their way up without special
favors, like the Irish, Italians, and Jews have
2. How much R agrees or disagrees that slavery and discrimination have created difficult
conditions for Blacks to work out of
3. How much R agrees or disagrees that Blacks have gotten less than they deserve over
the past few years
4. How much R agrees or disagrees that if Blacks would try harder they could be as well
off as Whites
Critics of the racial resentment measure charge that someone who strongly believes in tradi-
tional values like personal responsibility and hard work will necessarily appear to be racially
resentful: the scale measures American individualism as well as or perhaps even more than
racism.15 This objection is particularly pertinent to Trump’s working-class supporters inas-
much as many are descendants of immigrants who experienced relative deprivation without
the benefit of government aid or programs that are available to new immigrants today. The
scale’s creators have labored strenuously to meet such criticisms, but four decades after
its creation the debate continues.16 Thus, skeptics tend to discount the myriad analyses that
find a strong association between the racial resentment measure and support for Trump. The
empirical relationship is strong, but does measured “racial resentment” actually reflect only
racial prejudice or other things as well? 17
To take another prominent example, one woman’s cultural variable may be another man’s
economic variable. Consider the lengthy exchange between political scientist Diana Mutz
and sociologist Stephen Morgan about the former’s claim that “status threat, not economic
hardship, explains the 2016 presidential vote.” 18 Much of the exchange centered on statis-
tical questions, but for present purposes I focus on Mutz’s measures of status threat that
Please indicate whether you favor or oppose each of the following proposals addressing immigra-
tion: (i) provide a path to citizenship for some illegal aliens who agree to return to their home coun-
try for a period of time and pay substantial fines, (ii) increase border security by building a fence
along part of the US border with Mexico, (iii) return illegal immigrants to their native countries.
Do you favor or oppose the federal government in Washington negotiating more free-trade
agreements? Thinking about the increasing amount of trade between the United States and
other countries, do you think this has helped the US economy, hurt the US economy, or not
affected the US economy?
These days, there are different views about China. Some people see China as more of an oppor-
tunity for new markets and economic investment, while others see it as a threat to our jobs and
security. Still others are somewhere in between. Which view is closer to your own?
Mutz writes,
All three of these measures capture potential racial and global status threat. For example,
immigration captures the perceived threat of allowing those who are racially different into one’s
country. Trade opposition captures Americans’ fear of takeover by more dominant economic
powers as well as racial opposition based on resentment of “others,” including foreigners and
businesses in countries that are racially different. . . . Finally, China can be considered an out-
group threat both racially and with respect to threatening American global dominance.20
A general reader might reasonably wonder how attitudes toward immigration, trade, and
China reflect only threats to one’s racial or social status and not to one’s economic status.
Although studies are not unanimous, some economists argue that low-skilled immigrants neg-
atively affect the wages of low-skilled native workers.21 Certainly there are Democratic com-
mentators (not to mention union leaders) who believe in and decry the negative wage effects
on native workers of competition with low-skilled immigrants and foreign workers.22 More
generally, economic studies report that free trade has had significant negative impacts on
US labor markets.23 As Goldstein and Gulotty summarize, “Estimates of the effect of China’s
entry into the WTO [World Trade Organization] show that trade has produced a discontinu-
ous shift in global production patterns and is a major cause of subnational labor displace-
ment in the United States and elsewhere.”24 Morgan constructs a table in which he offers
different—mostly economic—interpretations of a number of the measures that Mutz considers
to be measures of status anxiety (table 1). Not surprisingly he concludes that economic factors
play a far more important role in support for Trump than Mutz does.
Few would deny that considerations of racial and social status are entirely absent from the
measures Mutz uses; rather, the concern is that attitudes toward immigration, trade, and
the rise of China surely reflect both status and economic considerations—and perhaps
other ones as well.
A fair critic’s
Variable Mutz (2018) alternative
Safety net: spend more taxes on safety net, cut taxes to Economic Material
eliminate government programs and services indicator interests
(2-item scale)
Current personal finances: better or worse than last year Economic Material
indicator interests
Social dominance orientation: consider all groups when Status threat Status threat
setting priorities, group equality should be our ideal,
should not push for group equality, superior groups
should dominate inferior ones (4-item scale)
Outgroup prejudice: other groups are hardworking/ Status threat Status threat
peaceful or lazy/violent (multiple-item scale; number
not provided by Mutz)
Worried about America: worried that the American way of Status threat Status threat
life is under threat
Support for free trade: support federal government nego- Status threat Material
tiating more free-trade agreements, past increases in free interests
trade have helped or hurt the US economy (2-item scale)
China is a threat to jobs: China provides new markets and Status threat Material
is an investment opportunity or is a threat to our jobs and interests
security
Support for inclusive immigration policy: support path Status threat Material inter-
to citizenship, border fence with Mexico, return of illegal ests and for-
immigrants to native countries (3-item scale) eign policy
Support for isolationism: active role in solving conflicts Status threat Material
around the world, take care of the well-being of Americans interests and
and not get involved with other nations, essential to work foreign policy
with other nations to solve problems, best for the future of
the country if we stay out of world affairs, have a responsi-
bility to fight violations of international law and aggression
wherever they occur (5-item scale)
Even assuming appropriate measures, a second problem arises from the fact that few ana-
lysts make any serious attempt to model how economic and cultural attitudes affect the vote.
Instead, they treat economic and cultural attitudes as logically independent and include
both in their model specifications. But economic and cultural variables may be causally
related and not independent as many analyses implicitly assume. The post–World War II
sociological literature argued that economic distress gives rise to increased xenophobia,
nativism, and other sentiments that are problematic for democracy.25 The rise of Hitler as the
German economy collapsed in the 1930s is the locus classicus.26 As Inglehart and Norris
observe, “Although the proximate cause of the populist vote is cultural backlash, its high
present level reflects the declining economic security and rising economic inequality that
many writers have emphasized.”27 Commentators who attribute the rise of Trump to American
racism and sexism often fail to consider the experience of other countries. Britain has seen
ethnic resentment directed against the Polish (White, Christian) plumber. In France, Algerians
are the target. In Italy, workers in the textile industry target the Chinese.28
Moreover, it is also likely that for some people economic and cultural variables are causally
related to the vote in the opposite direction from that which is usually assumed. Considerable
research suggests that candidate preference causes policy attitudes, as well as vice versa.
For example, during Trump’s campaign and after his election, surveys showed that Republican
voters became less interventionist in foreign policy, more protectionist on trade, and less
hostile to Putin and Russia.30 Enns and Jardina argue that in 2016 some voters first decided
whom to vote for and then adopted the economic and cultural positions of that candidate.31
In particular, they report that racial hostility did not only cause support for Trump; support
for Trump caused people to express more racially hostile attitudes. The possibility of such
endogeneity rarely is seriously considered when cultural issues are under examination, even
though it is widely recognized in other contexts.
When political scientists are confronted with some seemingly interesting factoid at a dinner
party, their professional training often leads them immediately to ask, “Is that higher or lower,
more or less than what we have seen before?” Simple facts have different interpretations
depending on their temporal context. Various analyses of the 2016 election report that some
variable had a statistically significant relationship to the vote and, by implication at least,
was a key component of the surprising outcome. Attitudes toward Blacks and other ethnic
minorities, women, immigrants, and Muslims; views on issues like abortion, climate change,
and trade; predispositions such as authoritarianism, nationalism, and narcissism; and vari-
ous economic measures all received attention. Insofar as we are interested in explaining the
surprising 2016 outcome, however, the critical question is whether any such measure had a
different—usually larger—impact on the vote than in past elections. No doubt, all US elections
(as well as elections in other democracies) have elements of racism, misogyny, and other
characteristics that Hillary Clinton’s supporters decried, but is it true that “Donald Trump’s
support in the 2016 campaign was clearly driven by racism, sexism, and xenophobia?”33
Did such factors appear to play a larger than usual role in the 2016 voting? Despite the
claims of some commentators, this simple question can be surprisingly difficult to answer.
Political scientists have addressed the question of temporal change using our standard tool,
regression analysis, comparing the coefficients of specific variables in 2016 with those esti-
mated in earlier elections. Williamson’s and Gelfand’s claim references (among others) work
by Schaffner, MacWilliams, and Nteta who estimate vote models for the 2012 and 2016 elec-
tions, finding that “the coefficients for hostile sexism and denial of racism are more strongly
associated with 2016 vote choice than they are in 2012.”34 Sides analyzes Voter Study Group
panel data and shows that “attitudes related to immigration, religion, and race were
more salient to voter decision-making in 2016 than in 2012.”35 Enders and Scott pool
American National Election Study (ANES) presidential surveys from 1988 to 2016 and
find that, although the relationship between racial resentment and the vote has not signifi-
cantly increased over the years, racial resentment’s association with candidate evaluations
and issue attitudes that presumably affect the vote has increased significantly.36
But the explanatory power of variables can vary across time for multiple reasons. Most obvi-
ously, we may observe a change in the distribution of a variable, which is commonly referred
to, naturally enough, as distributional change. So, for example, if “racial resentment” more
strongly relates to Trump voting than to Romney voting, a plausible hypothesis is that Trump
Strongly agree 57 50
Agree somewhat 28 30
Disagree somewhat 6 6
Disagree strongly 2 2
Strongly agree 5 6
Agree somewhat 23 24
Disagree somewhat 26 30
Strongly disagree 34 31
Strongly agree 1 3
Agree somewhat 7 10
Disagree somewhat 30 30
Strongly disagree 45 40
Strongly agree 31 30
Agree somewhat 29 34
Disagree somewhat 11 13
Strongly disagree 6 4
Source: ANES
Strongly agree 29 11
Agree somewhat 25 21
Disagree somewhat 19 24
Disagree strongly 11 28
Strongly agree 17 33
Agree somewhat 38 36
Disagree somewhat 19 15
Strongly disagree 17 8
Strongly agree 6 23
Agree somewhat 19 33
Disagree somewhat 29 15
Strongly disagree 19 10
Strongly agree 19 8
Agree somewhat 22 11
Disagree somewhat 22 26
Strongly disagree 18 40
Source: ANES
Most analyses that find that racial resentment is a more important correlate of voting in 2016
interpret the finding as behavioral change: the campaign raised the salience of race, leading
Trump voters to weight race more heavily relative to other factors in 2016. Engelhardt traces
the sorting of the parties on race between 1988 and 2016, particularly between 2012 and
2016.38 He concludes that changes in the regression coefficients reflect both distributional
and behavioral change: the significantly increased liberalism of Democrats and the cam-
paigns’ increased emphasis on race.
Figure 1 depicts the familiar spatial model of elections.40 The electorate is distributed across
a single economic dimension ranging from, say, total government control of the economy
on the extreme left to a completely laissez-faire economy on the extreme right. As on most
issues the electorate is centrist, with voters getting steadily scarcer as we move toward the
extremes (although the shape of the distribution is irrelevant for the following argument).41
Voters choose the candidate who is closer to them. If both candidates locate at the median,
the election is a tie—the standard median voter result. However, in the real world, candidates
represent parties, so they take off-center positions reflecting their party platforms. In figure 1,
the Democrat locates one standard deviation left of the median and the Republican one and a
half standard deviations right of the median. The result is a 60:40 landslide for the Democrat,
much like say, the 1936 election between Democrat Franklin Roosevelt and Republican
Alf Landon where economics was the principal, if not the only, issue.
Beginning in the 1950s and accelerating in the 1960s and 1970s, issues such as race, gender,
and family values joined economics in the space of electoral competition. Much public opin-
ion research finds that domestic issues now break into at least two dimensions: an economic/
social welfare dimension and a social/cultural dimension that is at least partly independent of
the economic dimension.42 Although these dimensions are highly correlated at the elite level,
they are more distinct at the voter level. Let’s assume that they are independent, although that
is not important for the conclusions that follow. So, now we use a hump, as in figure 2, to rep-
resent the two-dimensional voter distribution, and we change our perspective to look at the
hump from the top rather than the side, as in figure 3.43 The Democrat and Republican can-
didates retain their positions on the economic dimension, but they do not differ (yet) on the
social dimension. The heavy vertical line just to the right of the median separates the distribu-
tion into those voters closer to the Democrat on the left and those closer to the Republican
on the right.
60 40
σ 1.5σ
FDR Landon
1
0
3 −1
1 Cultural
0
−1
Economic −3
−3
FDR
Landon
Cultural
FDR Landon
Economic
Here is the surprising conclusion of this exercise. In each of the four contests during which
there is neither distributional nor behavioral (voter) change, the coefficients from regress-
ing the vote on the voter’s economic and cultural positions are different.45 Specifically, as
the Democratic and Republican candidates separate on the cultural dimension, the vote
coefficient associated with that dimension gets larger, and the coefficient associated with
the economic dimension gets smaller, as plotted in figure 5.
To reiterate, in this example there is no distributional change: the voter distribution stays the
same. Nor is there any behavioral change: voters were not “activated” by the campaigns but
simply vote for the candidate closer to them. What has changed is how close or far the candi-
dates locate themselves from voters: candidate behavior is what has changed.
Cultural
Romney
FDR Landon
Economic
Obama
0.05
0
E1 E2 E3 E4
–0.05
Regression coefficient
–0.1
–0.15
–0.2
–0.25
–0.3
–0.35
–0.4
–0.45
Election
Voters’ economic coefficient Voters’ cultural coefficient
There is every reason to believe that changes in candidate positions in some part underlie the
larger statistical relationship of cultural attitudes in 2016 relative to earlier elections. Sides,
Tesler, and Vavreck write, “Americans have rated the Democratic presidential candidate as
the more supportive of federal aid to blacks in every single survey since the question’s incep-
tion in 1972. . . . Then, in 2016, this disparity increased to record levels. . . . Whites saw Clinton
as more liberal than Obama in 2012 (a 0.13 shift on the scale) and Trump as significantly more
conservative than Romney (a 0.37 shift).”46 Similarly, after analyzing a panel survey, Hopkins
observes that respondents in 2016 viewed Clinton as slightly more liberal on immigration than
Obama in 2012, and Trump as far more conservative than Romney.47 My conjecture is that
other things being equal, as candidates move apart on an issue, the voter coefficients asso-
ciated with that issue grow larger. Unless analysts explicitly incorporate candidate positions
into their statistical estimations, we cannot identify whether an apparent increase or decrease
in importance of a variable reflects voter change, candidate change, or some mixture of the
two. Of the analyses I have located, only Mutz explicitly considers changes in candidate posi-
tions in her analysis, concluding—consistent with the preceding arguments—that changes in
candidate positions were more important than distributional change for explaining the differ-
ence between 2012 and 2016 voting.48
During the 2016 campaign and its aftermath, various surveys reported a finding that some
found surprising: Trump’s supporters were not particularly poor or suffering from joblessness.
In fact, Manza and Crowley reported that “Trump’s voters were, on the whole, significantly
more affluent and better educated than the average voters in primary states.”49 Given that one
explanation of the election focused on the prevalence of economic hardship, these findings
seemed counterintuitive to some people in the media.
Somewhat strangely, although that finding surely is well known to those scholars who
attempt to adjudicate between economic and cultural explanations of Trump’s appeal, their
analyses in many cases do not incorporate it. Pocketbook measures abound in their voting
models. Reny, Collingwood, and Valenzuela use change in household income over the pre-
ceding four years.52 So do Schaffner and coworkers.53 Mutz’s analysis includes household
income, whether the respondent is looking for work, changes in personal finance, and the
personal impacts of trade.54 Sides and colleagues use responses to survey questions about
the respondent’s current financial situation and whether they worry about losing their job or
missing a rent or health payment.55 Enns uses YouGov measures of family income and the
respondent’s employment status.56 Pocketbook measures like these typically show little or
no association with Trump’s vote once control variables and cultural variables are included
in the models. In the statistical horse race with cultural variables, the pocketbook variables
trail badly.
But why should anyone have expected otherwise? Consider Krystal Ball’s (a Bernie supporter)
colorful postelection rant:
One after another, the dispatches came back from the provinces. The coal mines are gone, the
steel mills are closed, the drugs are rampant, the towns are decimated. . . . And we offered a
fantastical non-solution. We will retrain you for good jobs! . . . And as a final insult, we lectured
a struggling people watching their kids die of drug overdoses about their white privilege. Can
you blame them for calling bullshit?57
Ball clearly refers to suffering communities, not suffering individuals. Similarly, Coontz reports,
“As a recent CNN poll shows, white working-class and rural voters without a college degree
are not the poorest of Americans, but they are the most pessimistic about their future pros-
pects. A full half expect their children’s lives will be worse than their own, and less than a
quarter expect their children to do better.”58
Political scientist Matthew Dickinson spent his sabbatical year talking to attendees at Trump
rallies. He writes, “It quickly became clear that two themes dominated the thinking of
Clearly, Ball, Coontz, Dickinson, and other commentators identify not just voters’ concerns
about their personal economic situations but also fear for the future of their progeny and their
communities and despair over the social dysfunctions produced by areawide economic dis-
tress. I may have a secure public sector job with good benefits, but that is small consolation
if my community is going to hell.
In sum, a fourth problem in the debate over the relative importance of economic versus cul-
tural considerations in the 2016 voting is more specific than the first two. Many analyses do
not include the appropriate economic concepts, substituting a pinched notion of personal
economic hardship for general economic distress, almost guaranteeing a finding of appar-
ently weak effects of economic considerations.60
An objection to the arguments in the preceding section is that some analyses do include
aggregate economic indicators that capture broader, more sociotropic notions of economic
distress. In addition to a survey item on job loss, Green and McElwee’s analysis includes
zip-code-level unemployment insurance receipts.61 Mutz adds zip-code-level civilian unem-
ployment, manufacturing employment, and median income measures to the survey measures
discussed earlier in this essay.62 Reny and colleagues include county-level manufacturing loss
and unemployment, as well as a survey item on household income. Nevertheless, their analy-
ses conclude that culture dominates economics.
Rarely in this literature do analysts recognize that including both aggregate- and individual-
level measures in a regression creates complications that require more sophisticated analy-
ses. Other things being equal, simply regressing indicators of individual behavior such as
voting on both individual and aggregate right-hand-side variables disadvantages the vari-
ables measured at the aggregate level. Again, I adapt an illustration from an earlier work
that encountered an analogous problem.63
In the 1970s and early 1980s, research on congressional elections generated a puzzle. In that
era of candidate-centered elections, members of Congress firmly believed that activities on
behalf of their constituents resulted in electoral payoffs, and therefore, they allocated signifi-
cant resources to such activities. For their part, surveys showed that constituents who felt
their members were attentive and helpful evaluated them positively and voted for them. The
puzzle was that attempts to relate what congressional offices reported doing with what con-
stituents reported experiencing or how they voted showed weak, inconsistent, or even nega-
tive relationships, as in Gary Jacobson’s classic contrarian finding that “the more they spend,
the worse they do.”64
Table 4 depicts six counties each containing five voters. The variables in the table are binary.
Column 1 is 2016 vote (Trump = 1). Column 2 contains the economic variable (hardship = 1). The
counties vary from wealthy ones that contain no economically distressed voters to severely
depressed units where everyone is distressed. By assumption, economic hardship perfectly
determines the vote: every person who is distressed votes for Trump, while every person not
distressed votes for Clinton. Every voter also has a cultural score (column 3; prejudice = 1).
These scores are very highly related to the vote but not perfectly so: county 1 contains a
Clinton voter who is prejudiced, and county 6 contains a Trump voter who is not prejudiced.
County 1
0 0 1 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
County 2
1 1 1 0.2
0 0 1 0.2
0 0 0 0.2
0 0 0 0.2
0 0 0 0.2
(continued)
County 3
1 1 1 0.4
1 1 1 0.4
0 0 1 0.4
0 0 0 0.4
0 0 0 0.4
County 4
1 1 1 0.6
1 1 1 0.6
1 1 0 0.6
0 0 0 0.6
0 0 0 0.6
County 5
1 1 1 0.8
1 1 1 0.8
1 1 1 0.8
1 1 0 0.8
0 0 0 0.8
County 6
1 1 1 1
1 1 1 1
1 1 1 1
1 1 1 1
1 1 0 1
The regressions of the Trump vote on the cultural and economic variables are shown here:
Trump Vote = 0.0 + 1.0 Hardship + 0.0 Prejudice R2 = 1.0 (standard errors = 0)
As is often the case, however, suppose the economic measure is available only at the county
level (column 4), so every voter is assigned the average figure for the county. The economic
regression is
The slope and intercept remain the same as in the individual-level regression, of course, but
the explained variance has been more than halved. This development reflects the fundamen-
tal identity of the analysis of variance: Total variance = within-unit variance plus between-unit
variance. When measured at the aggregate level the economic variable can only explain
between-unit variance, which from the aggregate economic regression is only 47 percent
of the total variance in this constructed dataset.
But when the regression includes the individual-level cultural variable and the aggregate-level
economic variable, the regression is as follows:
Although the cultural variable was previously irrelevant in the presence of the economic
variable, because it can access the within-unit variance, its coefficient now becomes larger
in magnitude than the economic coefficient. In sum, even though by assumption economic
distress perfectly determines the vote in this example, an analyst using an aggregate-level
economic variable would attribute greater importance to the cultural variable.
Other research subfields have long recognized problems like this one. Education researchers
for example, regress student achievement on individual student characteristics, classroom
characteristics, school characteristics, and sometimes even larger units. When the observa-
tions are “nested” within larger units, techniques more sophisticated than single regression
equations are required.66 Although these methods appear here and there in the political sci-
ence literature, they are rare or absent in this area of research where the nature of the data
clearly calls for them.
CONCLUSION
This selective review justifies no definitive conclusion, other than that no study of which I am
aware provides a conclusive answer to whether cultural or economic considerations were the
Unfortunately, announcing conclusions that are not well-rooted in the data may have negative
real-world consequences. As Enns notes, “If social scientists and journalists over-emphasize
the role of racist attitudes in the election, they risk inflaming political divisiveness.”67 Trump
supporters motivated by economic concerns felt rightly resentful when their concerns were
dismissed as disguised racism, sexism, or other social pathologies. And Trump opponents
did not realize that they may have had more common ground with Trump supporters than they
believed. The research conducted after the 2016 elections may inadvertently have contrib-
uted to the toxic quality of contemporary debate.
NOTES
1. L. V. Anderson, “2016 Was the Year White Liberals Realized How Unjust, Racist, and Sexist America Is,”
Slate, December 29, 2016, http://w ww.slate.com/ blogs/ x x_factor/2 016/ 12/2 9/_ 2 016_was_the_year_white
_liberals_learned_about_disillusionment.html.
2. Christina Cauterucci, “In 2016 America Was Forced to Face the Reality of Sexual Assault,” Slate,
December 28, 2016, http://w ww.s late.com/blogs/x x_f actor/2 016/1 2/2 8/_2 016_w as_the_year_america
_learned_what_sexual_assault_looks_like.html.
3. Jenee Desmond-Harris, “Trump’s Win Is a Reminder of the Incredible, Unbeatable Power of Racism,”
Vox, November 9, 2016, http://w ww.vox.com/policy-and-politics/2 016/11/9/13571676/trump-win
-racism-power.
4. James Barrett, “Michael Moore Slaps down Attempts to Smear Trump Voters as ‘Racist,’ ” Daily Wire,
November 12, 2016, accessed via Internet Archive, https://web. archive.org/ web/2 0161114043754/ http://
www.dailywire.com/news/10742 /michael-moore-slaps-down-attempts-smear-trump-james-barrett#. It
should be noted that Michael Moore was all over the map on this issue.
5. Eduardo Porter, “Where Were Trump’s Votes? Where the Jobs Weren’t,” New York Times, December 13,
2016, https://w ww.nytimes.com/2 016/12 /13/business/economy/jobs-economy-voters.html.
6. Andrew Yang, as quoted in Kevin Roose, “His 2020 Campaign Message: The Robots Are Coming,”
New York Times, February 10, 2018, https://w ww.nytimes.com/2 018/0 2 /10/technology/his-2 020
-c ampaign-message-the-robots-are-coming.html?module= inline.
7. Such high partisan loyalty figures are somewhat exaggerated because they treat leaning independents
as partisans. Considerable evidence indicates that some independents decide how to vote first and
then indicate how they lean based on how they intend to vote. This reverse causation artificially inflates
the relationship between partisanship and voting. See Morris Fiorina, Unstable Majorities (Stanford CA:
Hoover Press, 2017), chap. 6.
8. In the Economist/YouGov panel, about two-thirds of 2012 Obama White voters supported Clinton in
2016, whereas about three-quarters of Romney white voters supported Trump. Fiorina, Unstable Majorities,
197. The blue-ribbon post-election American Association for Public Opinion Research (AAPOR) report
concluded that the errors in 2016 state polls in part reflected polling samples that contained too many
The publisher has made this work available under a Creative Commons Attribution-NoDerivs
license 4.0. To view a copy of this license, visit https://creativecommons.org/licenses/by-nd/4.0.
Copyright © 2024 by the Board of Trustees of the Leland Stanford Junior University
The views expressed in this essay are entirely those of the author and do not necessarily reflect the
views of the staff, officers, or Board of Overseers of the Hoover Institution.
30 29 28 27 26 25 24 7 6 5 4 3 2 1
MORRIS P. FIORINA
Morris P. Fiorina is a professor of political science at Stanford University and a
senior fellow of the Hoover Institution. He has written or edited fourteen books,
most recently Who Governs? Emergency Powers in the Time of COVID. An
elected member of the National Academy of Sciences, Fiorina has received
career achievement awards from two sections of the American Political
Science Association.
A continuation of the Hoover Institution’s Unstable Majorities series from the 2016 election season, the first
half of this essay series leads up to the November 2024 elections with general discussions of the past and
present political situation, of particular interest to students and professionals in the fields of political sci-
ence and political journalism. The second half continues post-election with analyses focused specifically on
the 2024 elections, addressed to a wider audience. The series begins by looking back at the issues raised
in 2016 that continue today.