IDENTITY, BELIEFS, AND POLITICAL CONFLICT∗

GIAMPAOLO BONOMI
NICOLA GENNAIOLI
GUIDO TABELLINI

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

We present a theory of identity politics that builds on two ideas. First, when
policy conflict renders a certain social divide—economic or cultural—salient, a
voter identifies with her economic or cultural group. Second, the voter slants her
beliefs toward the stereotype of the group she identifies with. We obtain three
implications. First, voters’ beliefs are polarized along the distinctive features of
salient groups. Second, if the salience of cultural policies increases, cultural conflict
rises, redistributive conflict falls, and polarization becomes more correlated across
issues. Third, economic shocks hurting conservative voters may trigger a switch to
cultural identity, causing these voters to demand less redistribution. We discuss
U.S. survey evidence in light of these implications. JEL Codes: D03, D72, D74,
D83, Z13.

I. INTRODUCTION
Since 2010, the political systems of the United States
and other advanced democracies have undergone a momentous
change. Economic conflict over redistribution has weakened, while
conflict over cultural issues such as immigration, race, and abor-
tion has strengthened. Consider Figure I, constructed using data
from the American National Election Surveys (ANES). We create
an index of the demand for redistribution as the first principal

∗ We are grateful to Stefanie Stantcheva, four referees and Roland Bénabou,

Ben Enke, Giovanni Burro, Tommaso Crosta, Ed Glaeser, Elhanan Helpman,

Nathan Nunn, Raffaella Piccarreta, Giacomo Ponzetto, Jesse Shapiro, Andrei
Shleifer, Marco Tabellini, and Pierre Yared; to participants in workshops at AS-
REC, Bocconi, the Bundesbank, CEPR, CIFAR, the ESM, Harvard, the IMF, Hong
Kong Baptist University, National Taiwan University, National University of Sin-
gapore, CEIBS, the LSE ERINN conference, the NBER, Princeton, UPF, Univer-
sità di Padova, Università Statale di Milano, the University of Namur; to Viola
Corradini, Daniele d’Arienzo, Silvia Farina, Carlo Medici, Francesca Miserocchi,
and Giulia Travaglini for outstanding research assistance, and Silvia Tesauro for
editorial assistance. Tabellini thanks the ERC grant 741643 and Gennaioli the
ERC (GA 647782) and MIUR (FARE grant) for financial support.

C The Author(s) 2021. Published by Oxford University Press on behalf of President and
Fellows of Harvard College. This is an Open Access article distributed under the terms of the
Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which
permits unrestricted reuse, distribution, and reproduction in any medium, provided the orig-
inal work is properly cited.
The Quarterly Journal of Economics (2021), 2371–2411. https://doi.org/10.1093/qje/qjab034.
Advance Access publication on September 10, 2021.

2371
2372 THE QUARTERLY JOURNAL OF ECONOMICS

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

(A) (B)

FIGURE I
Population Moments
Panel A reports the variances of redistribution and culture. Panel B reports the
Pearson correlation coefficient for these two measures. Redistribution is the first
polychoric principal component of the following questions: (i) “Some people think
the government should provide fewer services, even in areas such as health and
education, in order to reduce spending. Other people feel that it is important for
the government to provide many more services even if it means an increase in
spending. Where would you place yourself on this scale?”; (ii) “Some people feel
that the government in Washington should see to it that every person has a job and
a good standard of living. Others think the government should just let each person
get ahead on their own. Where would you place yourself on this scale?” Answers
to these two questions are given on a 7-point scale. Culture is the first polychoric
principal component of desired immigration levels, attitudes toward race relations
and abortion policy. Below we report the corresponding questions. Immigration
levels: “Do you think the number of immigrants from foreign countries who are
permitted to come to the United States to live should be [1. increased a lot; 2.
increased a little; 3. left the same as it is now; 4. decreased a little; 5. decreased
a lot]?” Attitudes toward race are the difference between respondents’ feelings
toward Black and white people. Feelings toward Black (white) people: “How would
you rate the following group: Blacks (Whites),” on a 0–100 scale, from coldest to
warmest feelings. Abortion policy: “There has been some discussion about abortion
during recent years. Which one of the opinions on this page best agrees with your
view? [1. By law, abortion should never be permitted; 2. The law should permit
abortion only in case of rape, incest, or when the woman’s life is in danger; 3. The
law should permit abortion for reasons other than rape, incest, or danger to the
woman’s life, but only after the need for the abortion has been clearly established;
4. By law, a woman should always be able to obtain an abortion as a matter
of personal choice].” Both principal components (redistribution and culture) are
computed on the pooled sample from 1996 to 2016, based on polychoric correlations.
Higher values correspond to more liberal views. Before computing the moments in
Figure I, redistribution and culture are residualized on party identity, alone and
interacted with wave fixed effects (more details in the Online Appendix). Residuals
are standardized to have zero mean and unit variance across all waves. Estimates
are weighted with survey weights. Source: ANES.
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2373

component of two questions on public spending. We construct an

index of the demand for progressive cultural policies from opin-
ions on immigration, race relations, and abortion. To control for
shifting party positions, we estimate the residuals of both vari-
ables after conditioning on the respondent’s party affiliation, in-

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

teracted with wave fixed effects. Results are similar in the raw
data and when considering political independents only (see the
Online Appendix). Panel A reports the variance of these variables
between 1996 and 2016. After 2008, disagreement over cultural
policies has sharply increased, while disagreement over redistri-
bution has, if anything, declined.
Panel B shows that something else has changed: prefer-
ences over redistribution have become more correlated with those
on cultural policy. In 2008, a progressive cultural outlook was
positively correlated with more support for redistribution, but
this connection more than doubled in 2016. Wu (2020) docu-
ments similar patterns using a large data set on U.S. college
freshmen.
These trends cannot be explained by better sorting of ex-
tremists into parties (e.g., Klein 2020). Rather, they indicate
population-wide changes in voter demands: cultural conflict has
intensified and has become more correlated with conflict on re-
distribution. Another driver of increasing polarization may be
growing partisan divisions and hatred (e.g., Gentzkow 2016). This
mechanism likely plays a role, but it cannot easily explain why re-
distributive conflict—a historically partisan issue—has not itself
become more intense and why these changes are also observed
among political independents. Why have cultural divisions in-
creased? Why has redistributive conflict not risen despite growing
income inequality?
This article shows that these questions can be fruitfully ad-
dressed by studying how social identities influence voters’ behav-
ior. The basic idea is that when voters abandon their class identity
and redefine themselves in terms of their moral or religious val-
ues, the latter become more important to explain their beliefs in
several domains. Based on social psychology (Tajfel and Turner
1979; Turner et al. 1987), we build a model of this phenomenon
resting on two pillars.
First, voters can identify with their income group, upper ver-
sus lower class, or with their cultural group, social progressives
versus conservatives. At any point in time, voters identify with
the groups that are most salient, formalized as those having the
strongest policy conflict. When class conflict is stronger, voters
2374 THE QUARTERLY JOURNAL OF ECONOMICS

identify with their class. When cultural conflict is stronger, they

identify with their cultural group.
Second, identity causes voters to slant their beliefs toward the
stereotypical views of their group. In psychology, this phenomenon
is called belief polarization (Mackie 1986). It occurs because vot-

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

ers overweight the distinctive opinions of group members or those
of group-linked experts and media. Kahan (2015) shows that cul-
tural identities polarize risk perceptions about environmental is-
sues, abortion, and gun control. We formalize group-stereotyped
beliefs by adapting the model in Bordalo et al. (2016) and then
study its implications for politics.
Our setup features two policy instruments: a distortionary
income tax financing a public good and a “cultural policy” that
we interpret as civil rights or control of immigration. A voter is
described by two traits: expected future income and cultural pro-
gressiveness. Richer voters desire a lower tax, due to their higher
tax burden. More progressive voters want a more liberal cultural
policy and also a higher tax, because they like the public good
more. Culture reflects deep-seated values such as moral univer-
salism or religiosity, which shape beliefs across different domains
(Haidt 2012; Enke, Rodriguez-Padilla, and Zimmermann 2020).
Based on her personal traits, a voter belongs to an income class,
upper or lower, and to a cultural group, progressive or conserva-
tive, and she can identify with either.
We obtain three main insights. First, identity creates belief
distortions that polarize ingroup-outgroup conflict along the cur-
rently salient dimension. Under class identity, lower-class voters
are too pessimistic about their future income, and the reverse is
true for upper-class voters. These belief distortions boost redis-
tributive conflict. Under cultural identity, progressive voters be-
come even more progressive, and conservatives can become more
conservative. This boosts cultural conflict. Perceived polarization
is also larger than with rational voters: society is divided into “us
versus them.” Thus, identity can shed light on belief distortions
in politics and predicts that beliefs are shaped by the changing
salience of cultural, economic, or political groups (Alesina, Miano,
and Stantcheva 2018; Alesina, Stantcheva, and Teso 2018; Kahan
2015; Westfall et al. 2015).
Second, if the welfare relevance of cultural policy increases,
for instance because of a large inflow of immigrants, identity
switches from class to culture. As this happens, culture becomes
a stronger driver of policy views. In line with Figure I, then,
views over cultural policy become (i) more polarized and (ii) more
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2375

correlated with views over redistribution. Critically, by obfuscat-

ing class differences, cultural identity also dampens redistributive
conflict. This can explain why the latter is stable or declining
after 2008 despite growing income inequality. Using ANES
data, we show that—consistent with our theory—disagreement

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

between upper- and lower-class voters on redistribution has
sharply declined since about 2000.
Third, economic shocks that boost conflict among cultural
groups can also trigger a shift to cultural identity. We offer two
examples: skill-biased technical change and globalization. If these
shocks hurt less educated and hence more conservative voters, and
benefit more educated and hence more progressive voters, they
make cultural cleavages more salient and can induce a switch to
cultural identity. As a result, economic losers become more socially
and fiscally conservative. This is consistent with the evidence in
Autor et al. (2020) and Colantone and Stanig (2018), who show
that losses from international trade foster support for right-wing
parties.
We relate to recent work on the influence of moral universal-
ism and fairness on policy preferences (Enke, Rodriguez-Padilla,
and Zimmermann 2020; Stantcheva 2021). In our article, the im-
portance of cultural factors is time varying. Norris and Inglehart
(2019) stress the growing importance of cultural divisions. Frank
(2004) vividly describes the “cultural backlash” in Kansas, a state
that was Democratic in the past and then became culturally and
fiscally conservative. Relative to this work, we show that the cul-
tural divide can be amplified by specific economic shocks due to
trade or technology. We also connect to recent research explaining
the rise of populism as a reaction to economic distress (Guriev and
Papaioannou forthcoming).
Murphy and Shleifer (2004), Glaeser (2005), and Wu (2020)
explain changing voter beliefs based on political supply. Politi-
cians attract voters by catering to their cultural views, and
persuade them on issues they are less focused on, such as re-
distribution. One challenge for this approach is to explain why
persuasion should work with economic losers. Our model offers
a rationale: specific economic shocks have made cultural identity
more salient. More broadly, cultural conflict has become more im-
portant relative to redistribution in several countries with differ-
ent party systems but subject to similar globalization and technol-
ogy trends (Evans and Mellon 2016; Goodhart 2017; Guriev and
Papaioannou forthcoming). This phenomenon points to a role of
demand factors.
2376 THE QUARTERLY JOURNAL OF ECONOMICS

We also contribute to a growing literature on identity in eco-

nomics. Akerlof and Kranton (2000) develop the first economic
model where identity changes the payoffs of certain actions. They
do not consider beliefs. The seminal paper by Shayo (2009) in-
troduces identity in political economy. Grossman and Helpman

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

(2021) study how views over trade policy are affected by social
identities. In these papers, voters obtain utility from the welfare
of their group—which affects their policy demands—and iden-
tify either with their narrow class or the broader nation. In our
model identity affects beliefs, which allows us to relate to the evi-
dence on voters’ misperceptions (Achen and Bartels 2016; Flynn,
Nyhan, and Reifler 2017; Johnston, Lavine, and Federico 2017;
Stantcheva 2021). In addition, the groups we consider—class- and
culture-based—are heterogeneous. Thus, identity switches do not
cause social integration or disintegration but a realignment of
voters into different social partitions.1
A tradition in political science, started by Key (1955), stud-
ies electoral realignments (e.g., Sundquist 1983; Mayhew 2004).
This work focuses on the United States and seeks to explain last-
ing changes in party positions and in the composition of party
supporters. We endogenize these realignments, abstracting from
parties and their leaders.
The article proceeds as follows. Section II discusses the so-
cial psychology of identity and presents our model of beliefs.
Section III studies the causes of identity switches and how they
affect beliefs and equilibrium policy. Section IV derives testable
predictions on changes in policy preferences. Section V illustrates
the mechanism of our theory using U.S. survey data and assesses
alternative explanations for the same facts. Section VI concludes.
The Online Appendix contains mathematical proofs and details of
the data analysis.

II. THE SOCIAL PSYCHOLOGY OF IDENTITY AND STEREOTYPES

The social identity perspective, the leading theory of identity
and intergroup relations, combines social identity theory (SIT,
Tajfel and Turner 1979), and self-categorization theory (SCT,
Turner et al. 1987; Hogg and Abrams 1988). We discuss the

1. Bénabou and Tirole (2016) offer a different approach in which identity

reflects beliefs about oneself, and beliefs adjust to improve self-image and own
welfare, taking anticipatory utility into account.
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2377

psychology of identity, its connection to beliefs, and our formal-

ization of belief distortions.

II.A. Identity Formation

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

According to SCT, identity is a form of self-categorization that
leads people to perceive society through the lens of a group they
belong to. Each person belongs to many groups: income class, re-
ligion, nation, and so on, but they do not necessarily identify with
all of them at the same time. Identity depends on which social
partition is salient (e.g., Hogg and Abrams 1988). For instance,
at a football match, one’s own team is the salient group: it cap-
tures the cleavage of that moment, so it affects individual behavior
more than other groups do. When participating in a union strike,
though, the same person may identify with the lower class, and
assimilate their behavior to it. In these examples, identities are
flighty, but political and social identities can be highly persistent
when conflict-creating shocks are long-lasting or due to persistent
messaging by politicians and the media.
Which cleavage is salient depends on the so-called meta con-
trast ratio. A person identifies with a group of people who are (i)
similar to them, but also and crucially (ii) highly dissimilar from,
or in conflict with, the outgroup (Oakes 1987). For instance, at
a football match, the group of “team supporters” is more salient
than that of “football lovers” because it reflects the cleavage of
the moment. Similarly, during a strike the “lower class” is more
salient than the broader group of “economic producers.”
This idea naturally applies to politics. A shock, like a police
officer killing a Black person, renders a social partition salient, in
this case the one between those who believe that racial discrimi-
nation is significant and those who believe that the law has been
enforced. Many people identify as members of either group. How
does this affect beliefs and behavior?

II.B. Stereotyped Beliefs and Depersonalization

According to SIT and SCT, identity causes perceptions of self
and others to be tainted by group features (see also Sherif 1936;
Festinger 1950). Experiments on groups of objects by Tajfel and
Wilkes (1963) and subsequent work in social settings (Haslam
and Turner 1992) show that this occurs via two effects.
First, people stereotype groups by exaggerating differences
between ingroups and outgroups. In the previous example, those
2378 THE QUARTERLY JOURNAL OF ECONOMICS

who believe that the police are biased against minorities stereo-
type the other group as racist, while the skeptics stereotype the
others as radicals. Society is divided into “us versus them.” Second,
identity causes the individual to depersonalize, namely, to move
her beliefs closer to the stereotypical group member. McGarty

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

et al. (1992, 3) define the stereotype as the group’s most repre-
sentative type: “...the less a person differs from ingroup members
and the more he or she differs from outgroup members, the more
representative he or she is of the ingroup.” The stereotype is not
necessarily the modal or average group trait, it must be distinc-
tive relative to the outgroup. Thus, depersonalization creates the
phenomenon of group polarization, whereby group members hold
a more extreme position on an issue than their individually ex-
pressed positions (e.g., Mackie 1986). In the example of the police
officer, even moderate people slant their beliefs toward either left-
wing radicalism or racism, at least to some extent, enhancing
conflict.
Experiments using the “minimal group paradigm” found that
even arbitrary groups affect individual behavior when they are
made salient (Tajfel et al. 1971). SIT holds that beyond affecting
beliefs, identity can also yield positive self-esteem if one identifies
with a high-status group. Some experiments document ingroup
favoritism, which can be viewed as enhancing group status. Exist-
ing political economy models of identity are centrally built on this
ingredient (e.g., Shayo 2009). To highlight the new implications
of our mechanism, we abstract from this effect. Status cannot be
the sole driver of identity, because individuals often identify with
underdog groups.2

II.C. Formalizing Distorted Beliefs

We depart from a standard political economy setup because
voters’ beliefs and political preferences are malleable and change
based on the salient social identity. To see how this works, con-
sider a voter evaluating a particular policy, such as opening to
immigration or extending civil rights. The voter has a database of
information in her memory, captured by the pdf z(ψ̃|ψ), where
ψ denotes the mean of ψ̃ for that particular voter. A specific

2. There is debate on whether it is always the case that group identity leads
to ingroup bias (e.g., Hinkle and Brown 1990). Ingroup bias tends to arise when
discrimination is a group norm but not otherwise. A belief-based mechanism like
ours can arguably generate such an effect.
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2379

value ψ̃ captures an outcome associated with the policy (e.g., how

many crimes are committed by immigrants or benefits of diver-
sity) or a certain value judgment (e.g., altruistic attitudes toward
domestics or also foreigners). High values of ψ̃ favor a progres-
sive policy, low values favor a conservative policy. Parameter ψ

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

reflects the voter’s core beliefs, namely, her summary assessment
of the pros and cons of the policy. Voters with higher ψ are more
progressive.
There is a cdf H(ψ) of voters, who are partitioned into two
groups: the social conservatives, who have ψ < ψ , and the social
 
progressives, with ψ > ψ . The threshold ψ is historically given. A
voter identifies herself with one or the other group, depending on
her type. Here we take group identity as given. We endogenize it
in Section III.
In standard political economy models, where voters are ra-
tional, the voter aggregates all the information and evaluates the
policy at ψ. In our model, instead, the voter’s identity cues her to
retrieve facts or values that are stereotypical of her own group,
downplaying the rest.
We formalize this process following Bordalo et al. (2016,
BCGS): a person identified with group G overweights the trait
that is relatively more frequent in this group compared to out-
group G.3 As in McGarty et al. (1992), the stereotype is distinctive
of the group but not necessarily frequent.
Denote by zθ (ψ̃|ψ, G) the distorted beliefs of a voter of type ψ
when she identifies with group G. Following BCGS, this is given
by:
⎡  ⎤χ
zθ
( ψ̃ ψ G , G )
(1) zθ (ψ̃ |ψ, G ) ∝ z(ψ̃ |ψ ) ⎣  ⎦ ,

zθ (ψ̃ ψ G , G )

where ψ G ≡ E[ψ|G] is the core belief of the average member of G.

The likelihood ratio in equation (1) captures the overweighting of
facts or opinions ψ̃ that are more frequently held by the average

3. This model accounts for observed belief distortions in social domains, as

shown by BCGS, but also for beliefs in macro and finance (Gennaioli and Shleifer
2018). Closer to the current setting, Bordalo et al. (2019) show that gender stereo-
types influence self-confidence. When assessing own ability, people are too opti-
mistic/pessimistic in domains of knowledge where their gender has a competitive
advantage/disadvantage.
2380 THE QUARTERLY JOURNAL OF ECONOMICS

member of group G compared to G. Parameter χ  0 captures the

degree of overweighting.4
Relative to BCGS, equation (1) makes two innovations. First,
the stereotypical opinions of social groups also affect a voter’s be-
liefs about the world, not just her beliefs about others. Thus, group

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

beliefs emerge as a fixed point: the beliefs about the world of aver-
age group members are distorted by the stereotypical opinions of
groups; group stereotypes, in turn, reflect the distorted beliefs of
group members. Second, the group G through which social reality
is perceived depends on identity, which we make endogenous in
the next section.
The fixed point problem that jointly determines the beliefs of
average group members and group stereotypes proves tractable if
z(ψ̃|ψ) is Gaussian.
PROPOSITION 1. If χ < 12 the fixed point for average group beliefs
χ
exists, it is unique, and is stable. Let θ ≡ 1−2χ > 0. Identifi-
cation with group G distorts the beliefs of voter ψ as follows:

(2) ψGθ = ψ + θ (ψ G − ψ G ).

If θ = 0, beliefs are rational and identity plays no role. If θ >

0, identity distorts beliefs away from their core ψ. Identification
with the conservatives cues the voter to think about the risks of
progressive policies, so she becomes even more conservative. Iden-
tification with the progressives cues her to think about the benefits
of progressive policies, so she becomes even more progressive.
Equation (2) is consistent with the evidence in Kahan (2015),
showing that people with different religious or political orienta-
tions exhibit sharp factual disagreement over natural selection
and climate change, even if they are knowledgeable about science.
He proposes a theory of identity protective cognition, whereby in-
dividuals appraise information in a way that buttresses beliefs
associated with their ingroup.
Beliefs can also be distorted by identification with politi-
cal parties. In this case, party positions cue partisan voters to

4. In the BCGS model, the likelihood ratio is defined using the distributions of
 in groups G and G. To obtain these, one should aggregate a family of individual
ψ
|ψ, G), which is intractable. To simplify, we capture each group
distributions zθ (ψ
using the belief distribution for its average type. This assumption preserves the
idea that stereotypes magnify average group differences, and yields convenient
closed-form solutions.
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2381

retrieve party stereotypes. Our model is not inconsistent with this

possibility, but here we focus on social groups. This can explain
political realignments that weaken preexisting party affiliations
and occur for nonpartisan voters.
Equation (2) also sheds light on priming effects. A long-

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

standing tradition in studies of mass opinion holds that individual
beliefs change when a political or socioeconomic group is primed
(Zaller 1992; Han and Wackman 2017; Janky 2018). Equation (2)
disciplines these effects: if priming works through identity, it dis-
torts beliefs in the direction of observed disagreement between
θ θ
ingroups and outgroups (ψ G − ψ G ).5 Priming effects are then pre-
dictably heterogeneous because they cue voters to focus on the
different groups to which they belong.6
These priming effects are different from learning, whereby a
voter combines information with prior beliefs. When the voter is
primed, she is not provided with new information but recalls fac-
tual and value considerations associated with the primed group.
In particular, as we show in the next section, when the voter is
primed with a new social identity, she changes her beliefs in the
direction of overweighting the distinctive ingroup traits.
Recent research documents pervasive voter misperceptions.
Relative to right-wing respondents, left-wing respondents per-
ceive lower social mobility (Alesina, Stantcheva, and Teso 2018),
larger inequality (Gimpelson and Treisman 2018), a lower num-
ber of immigrants (Grigorieff, Roth, and Ubfal 2020), and per-
ceive immigrants as having better features than natives (Alesina,
Miano, and Stantcheva 2018). Future work may study whether
these misperceptions are determined by and change with prevail-
ing social or political identities.
One key implication of equation (2) concerns group polariza-
tion.

COROLLARY 1. When all voters identify with groups G and G,

disagreement among average group members is amplified

θ θ θ θ
5. Equation (2) can be rewritten as ψG = ψ + ( 1+θ )(ψ G − ψ G ).
6. In Alesina, Miano, and Stantcheva (2018), demand for redistribution in-
creases when subjects are provided with pessimistic information about mobility,
but this effect is only present for left-wing respondents. Kuziemko et al. (2015)
shows that informing poor people about their income rank in society increases
their demand for inheritance taxes. These findings are consistent with equation (2)
if one views the treatment as priming class identity.
2382 THE QUARTERLY JOURNAL OF ECONOMICS

relative to disagreement in rational beliefs. Formally,

θ θ
ψ G − ψ G = (ψ G − ψ G )(1 + 2θ )  ψ G − ψ G .

Identity boosts polarization by activating stereotypes. When

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

we endogenize identity, this result proves important to account
for growing cultural conflict in Figure I. We assume the selective
recall parameter θ to be exogenous. But if political leaders or
social media disseminate stereotypes, these more readily come to
mind—θ increases—causing more polarization.
Our model also implies that perceived polarization is greater
than actual polarization. Consider the beliefs held in society about
the distorted beliefs zθ (ψ̃|ψ G , G) of the average member of G,
namely, the second-order beliefs about ψ G . Due to stereotypes,
group-distinctive traits are overweighted as in equation (1), yield-
ing the result below.
θ and ψ
COROLLARY 2. Denote by ψ θ the perceived mean positions
G G
of the average group members. Group stereotypes imply:

1 + 4θ
θ = (ψ θG − ψ θG )
Gθ − ψ
ψ
θ θ
 ψ G − ψ G.
G 1 + 2θ

When thinking about the progressives, people overweight

their stereotypical members, who are very progressive. When
thinking about the conservatives, their stereotypical extreme
members are overweighted. Hence, perceived disagreement ex-
ceeds actual disagreement. This effect is not due to identity per
se: even if a voter is not identified with any social group, she will
have an exaggerated perception of polarization if she is subject to
stereotypical thinking.
There is a large literature measuring perceived and affective
polarization among U.S. parties. Westfall et al. (2015) and BCGS
show that U.S. voters exaggerate differences in the policy views of
Democrats and Republicans. These exaggerations are held by all
voters, including the nonpolitically affiliated ones, which is con-
sistent with Corollary 2. Bordalo, Tabellini, and Yang (2020) offer
intriguing new evidence: the gap between perceived and actual po-
larization is especially large in those policy issues that are more
salient and in which there is more actual disagreement. This is a
direct implication of our model when identity is endogenized.
Gentzkow (2016) shows that in the United States, per-
ceived polarization and distrust of political rivals has increased
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2383

more than actual divergence in policy views. As we show in

Proposition 4, our model can account for this effect as the
result of a shift from economic to cultural identity. Ahler
(2018) shows that correcting misperceptions about the out-
party also reduces affective polarization. Druckman and Leven-

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

dusky (2019) show that affective polarization is boosted by re-
trieval of extreme party stereotypes, again consistent with our
model.
In the current example, identity distorts beliefs and value
judgments regarding the general effects of a policy, say, immi-
gration. Identity can also distort a voter’s perception of her own
situation, such as her income rank in society or exposure to foreign
competition. When forming beliefs about these aspects, the voter
overweights the facts or opinions that are more frequently held in
her group compared to the outgroup, as in equation (1). A voter
identifying with the losers from trade retrieves examples of work-
ers similar to her who lost their job due to foreign competition, in-
creasing her demand for trade protection. Similarly, a voter identi-
fied with the lower class retrieves her group’s distinctive beliefs of
economic backwardness, so she becomes too pessimistic about her
future income. Consistent with this view, Stantcheva (2021) shows
that U.S. Republicans (Democrats) overestimate (underestimate)
their income rank in society. Of course, misperceptions of own
economic status may also be due to other factors. For instance,
Cruces, Perez-Truglia, and Tetaz (2013) document that Buenos
Aires residents misperceive their current income on the basis of
local conditions. Voters from rich neighborhoods underestimate
their income rank relative to voters from poor neighborhoods, as
if each voter uses the local income distribution as a reference. Our
point is that when class conflict becomes salient, many individu-
als identify with their class, so class-based misperceptions become
prevalent.

III. IDENTITY AND POLITICAL CONFLICT

We study a simple model of conflict over redistribution and
cultural policy, in which voters can identify with their economic
class or cultural group. We show that if cultural conflict be-
comes more salient, where salience is formalized in a precise
way, voters switch from class to cultural identity and their
beliefs change accordingly. This, in turn, changes equilibrium
policies.
2384 THE QUARTERLY JOURNAL OF ECONOMICS

III.A. The Model

There are two policies. The first is a proportional income tax
τ  0 that finances a public good. It entails quadratic distortions,
− ϕ2 τ 2 , ϕ > 0, that reduce aggregate income. The second is cultural

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

policy q. It captures domains mainly based on values, rather than
on material interests, such as abortion, civil rights, race relations,
and immigration. Larger q indicates a more liberal stance (e.g.,
more extensive civil rights or immigration). Our conclusions also
hold if we allow preferences over immigration to depend on in-
come, provided the dependence is weaker than that on culture,
consistent with the evidence in Card, Dustmann, and Ian (2012).
There is a set of measure one of voters who differ in their
expected relative income, ε, and their cultural traits, ψ. A voter
type is summarized by vector (ψ, ε), which is distributed in so-
ciety according to the normal cdf H(ψ, ε), with mean (0, 0), unit
standard deviations, and correlation coefficient ρ.
The policy evaluations of a voter depend on three sources of
uncertainty. The first one is about the effects of q. Preferences over
q follow the quadratic loss 12 (q − ψ̃)2 . ψ̃ is the voter’s preferred pol-
icy, which is Gaussian with voter-specific mean ψ. Higher ψ stands
for more socially progressive culture and hence higher preferred
q. We assume that income and social progressiveness are non-
negatively correlated, 0  ρ < 1. This is realistic because more
educated people are both richer and more culturally progressive.7
Second, voters are uncertain about their tax burden because
their income 1 + ε̃ is stochastic. ε̃ is Gaussian with voter-specific
mean ε. Thus, a rational voter of type ε expects to earn 1 + ε.
Beliefs about ε̃ capture beliefs over future relative income, and
higher ε implies a higher tax burden.

7. Consistent with this notion, in the ANES data used in Figure I and
Section V, if one regresses the demand for progressive policies defined in Fig-
ure I on wave fixed effects and income one obtains (omitting estimates for fixed
effects):
culture = −0.7672∗∗∗ + 0.0736∗∗∗ income.
(0.0447) (0.0119)

But when the same dependent variable is regressed on wave fixed effects, income
and education one obtains:

culture = −1.079∗∗∗ − 0.0021income + 0.1301∗∗∗ education.

(0.0486) (0.0132) (0.0090)

The Online Appendix provides details on variable definitions.

 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2385

Third, voters are uncertain about the benefit of the public

good g. Its marginal utility ν̃ is also Gaussian with expected
value ν + βψ. Because ψ has zero population mean, the aver-
age marginal value of g is ν > 1. Parameter β ∈ (0, 1) reflects
the influence of culture on preferences for the public good. The

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

distribution of ν differs from that of ψ̃ because these beliefs refer
to different policies.
The key point here is that the voter’s culture, embodied in
ψ, systematically affects her preferences over q and τ . One im-
portant cultural trait is moral universalism (Haidt 2012). More
universalistic people apply their value system to socially more
distant individuals (Tabellini 2008; Enke, Rodriguez-Padilla, and
Zimmermann forthcoming). Thus, they are more progressive on
civil rights or immigration, but they also trust the government
more and they are more favorable to redistribution (Haidt 2012;
Enke, Rodriguez-Padilla, and Zimmermann 2020). Another rele-
vant trait is religiosity. More religious people are more conserva-
tive on civil rights, featuring lower ψ̃, and they also trust the
government less (Enke, Rodriguez-Padilla, and Zimmermann
forthcoming), which reduces ν̃. Culture can also reflect fairness
norms, which affect preferences over tax policy (Stantcheva 2021)
and attitudes toward minorities. These traits do not need to be
independent from each other. Enke, Rodriguez-Padilla, and Zim-
mermann (forthcoming) show that religiosity is negatively corre-
lated with universalism. What matters for us is that they affect
voters’ views over q and τ .
Because ε has zero mean in the population, aggregate income
is 1. Thus, the quantity of g is equal to the tax rate τ . The expected
utility of the rational voter (ε, ψ) is, up to an additive constant:
ϕ 2 κ
(3) W εψ (τ, q) = (1 + ε) (1 − τ ) − τ + (ν + βψ)τ − (q − ψ )2 ,
2 2
where superscript εψ denotes the voter’s type and κ > 0 cap-
tures the weight attached to policy q. Neglecting nonnegativity
constraints, the rational bliss point of voter (ε, ψ) is:

(ν + βψ) − (1 + ε)
(4) τ εψ = , qεψ = ψ.
ϕ

More progressive voters (higher ψ) demand more redistribution

and a more liberal cultural policy. Richer voters (higher ε) demand
less redistribution because of their greater tax burden.
2386 THE QUARTERLY JOURNAL OF ECONOMICS

The socially optimal policy maximizes aggregate rational wel-

fare:

W εψ (τ, q) dH(ε, ψ)dεdψ.

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

Given linear private and public consumption, and given the as-
sumed normal cdf H(ε, ψ), the socially optimal policies are:

ν−1
(5) τ◦ = , q◦ = 0.
ϕ

Equation (4) shows that policy disagreement combines two

underlying conflicts. One is the rich versus poor conflict over taxes,
captured by expected relative income ε. The second is cultural
conflict, captured by ψ, which affects the evaluation of both re-
distributive and cultural policies. As emphasized by Sundquist
(1983), major realignments in U.S. politics have occurred when
the main parties shifted their positions along the economic and
cultural divides. We endogenize these realignments as the prod-
uct of voters’ changing identities across economic and cultural
groups.

III.B. Endogenous Identity

1. Groups. Social groups are defined based on income ε and
culture ψ. With respect to culture, a voter can either be socially
conservative, SC ≡ {ψ|ψ < ψ }, or progressive, SP ≡ {ψ|ψ  ψ },
where ψ  is historically given. With respect to income, voters
belong to the upper class U ≡ {ε|ε   ε} or to the lower class
L ≡ {ε|ε < ε}, where  ε is again historically given. Hence, there
are four groups, indexed by G = L, U, SC, SP. Each group is sum-
marized by the average traits of its members, (ε G , ψ G ), regardless
of their social identity. Thus, (ε L, ψ L) denote the average expected
income and progressiveness of the lower class, and the average
traits of the other groups are similarly defined. Note that, pro-
vided correlation is not full (ρ < 1), all cultural and economic
partitions overlap: each economic class contains both progressive
and conservative voters, and each cultural group contains both
upper- and lower-class voters. If ψ and ε are positively correlated,
these groups differ in both income and culture, irrespective of
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2387

the dimension that defines them. Bivariate normality of H(ε, ψ)

implies:

(6) ε SP − ε SC = ρ(ψ SP − ψ SC ),

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

(7) ψ U − ψ L = ρ (ε U − ε L ) .

If ρ > 0, the progressives are richer than the conservatives, and

the upper class is more progressive than the lower class. Of course,
since ρ < 1, cultural groups mostly differ along culture, ψ SP −
ψ SC > ε SP − ε SC and economic classes along income, εU − ε L >
ψ U − ψ L. We make the following reasonable assumption:

ψ SP − ψ SC 1
(A1) ρ< < ,
εU − ε L ρ

which implies that income differences are larger between classes

than between cultural groups, εU − ε L > ε SP − ε SC , and vice versa
for cultural differences, ψ SP − ψ SC > ψ U − ψ L.8

2. Metacontrast and Identity. We assume that there are only

two groups with which a voter can identify: her economic group
or her cultural group, but not both. For instance, a poor and con-
servative voter (ε < ε, ψ < ψ) can only identify either with the
lower class or with the conservatives, but not with the narrow
group of poor and conservative voters. In Online Appendix 3 we
allow for narrower social partitions, which allow a voter to identify
with her narrower two-dimensional group. The analysis is more
complicated, but our main results continue to hold, as we discuss
below.
How is a voter’s identity determined? As discussed in
Section II, identity maximizes (i) similarity between oneself and
the ingroups, and (ii) conflict between the ingroups and the out-
groups. This trade-off shapes which social partition, economic or
cultural, is more salient. To emphasize group conflict, which is

8. In what follows, we discuss the implications of whether ψ SP − ψ SC ≶ εU −

ε L. Given that H(ψ, ε) is normal with mean zero and unit variances, it can be shown
that if ε̂ and ψ̂ have the same sign, then

ψ SP − ψ SC  εU − ε L as |ψ̂|  |ε̂|.

See Schivardi, Sette, and Tabellini (2020) for a proof in a different context.
2388 THE QUARTERLY JOURNAL OF ECONOMICS

important in politics, we focus on criterion (ii), subject to the con-

straint that the voter can only identify with her income class or
cultural group because she is naturally more similar to them than
to the outgroups. As a result, all voters identify along the dimen-
sion, be it class or culture, that maximizes group conflict. Below we

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

discuss what happens if identity reflects a full trade-off between
criteria (i) and (ii).
Let G denote the complement of G. Conflict between groups
G and G is measured by the welfare loss that the average member
of G experiences when moving from her ideal policy (τ G , qG ) to the
bliss point of the average outgroup, (τ G , q G ):

(8) C(G, G) = W G (τ G , q G ) − W G (τ G , q G ).

The definition uses rational bliss points, but little changes if we

use stereotyped bliss points. Using equations (3) and (4), the On-
line Appendix proves that conflict is measured by:

C(G, G) = (ε G − ε G )2 + (β 2 + 
κ )(ψ G − ψ G )2
(9) − 2β(ε G − ε G )(ψ G − ψ G ).

The term (ε G − ε G ) captures economic conflict; the term (ψ G − ψ G )

captures cultural conflict. Thus, the conflict between two groups
(economic or cultural) is larger if their average types differ more
in income and culture. Positive correlation between culture and
income tends to dampen redistributive and thus overall conflict, as
captured by the last term. The weight attached to cultural conflict
increases with the importance of culture for the valuation of public
spending, β, and with the relative importance of cultural policy

κ ≡ κϕ. In turn,  κ increases in the welfare weight κ of cultural
policy and in tax distortions ϕ. If ϕ rises, everyone prefers lower
taxes, reducing the salience of redistributive conflict.
Equation (9) is symmetric, C(G, G) = C(G, G). Thus, if conflict
between economic classes is larger than conflict between cultural
groups, C(U, L) > C(SP, SC), all voters identify with their eco-
nomic class. Otherwise, they identify with their cultural group.
Define:
 2
2 ψ −ψ SC
(1 − βρ ) − (β − ρ )2 SP εU −ε L
(10) 
α≡  2 .
ψ SP −ψ SC
εU −ε L
− ρ 2
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2389

We then prove:
PROPOSITION 2. Everyone identifies with their cultural group if
2

κ > α and with their economic class if  α . If β  ρ 1+ρ
κ < 1+ρ 4
we
ψ SP −ψ SC
have that: (i) 
α > 0, (ii) 
α is strictly decreasing in ;

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

εU −ε L
ψ SP −ψ SC
(iii) 
α is strictly decreasing in ρ if εU −ε L
> 1.

If 
α < 0, identity is always cultural. If  α > 0, identity can be
either economic or cultural depending on parameters. To focus
on this case, which is more interesting, we impose the sufficient
2
condition β  ρ 1+ρ 1+ρ 4
.
Identity is shaped by three forces. First, cultural identity is
more likely to dominate when the importance of cultural policy
relative to taxes,  κ , increases. Higher welfare weight κ of cultural
policy, triggered for instance by a large inflow of immigrants or
by episodes of minority discrimination, brings cultural issues top
of mind relative to redistribution, promoting cultural identity. A
similar effect is created by higher tax distortions ϕ, caused for in-
stance by tax competition among countries, which reduce conflict
over taxes.
Second, cultural identity is more likely to dominate when dif-
ferences among cultural groups are large relative to those among
economic groups—namely when ψ SP −ψ SC
εU −ε L
is high. By Assumption
(A1), higher income inequality εU − ε L between classes increases
class conflict over all policies, making class identity more likely.
Stronger cultural divisions ψ SP − ψ SC increase cultural conflict
and reduce class cohesion, making cultural identity more likely. In
an 1893 letter, Friedrich Engels argued that class struggle proved
difficult in the United States because of deep ethnic cleavages in
the working class. This is consistent with the idea that strong
cultural divisions (large ψ SP − ψ SC ) hinder class identity. Simi-
larly, expansion of high education may enhance cultural conflict
between progressive elites and the rest of society (i.e., increase
ψ SP − ψ SC ), promoting cultural identity (Fukuyama 2018).
Third, the correlation ρ between income and progressiveness
also plays an important role. Higher ρ promotes cultural identity
if ψ SP −ψ SC
εU −ε L
> 1, otherwise it promotes class identity. The intuition
is simple. When ρ is higher, cultural groups disagree more over
redistribution and economic classes disagree more over cultural
policy. Thus, higher ρ increases policy conflict among both cultural
and economic groups. This effect, however, is more pronounced
2390 THE QUARTERLY JOURNAL OF ECONOMICS

for the groups that are more divided. Cultural groups are more
divided than economic classes if and only if ψ SP −ψ SC
εU −ε L
> 1, so in this
case higher ρ promotes cultural identity.
Skill-biased technical change is a shock that may increase
ρ, because it impoverishes less educated and hence more conser-

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

vative workers (Autor 2019). Online Appendix 2 formalizes this
argument by assuming that a voter’s skill endowment is positively
correlated with her progressiveness ψ. We show that if technical
change determines a higher remuneration of skilled labor, income
and culture become more positively correlated. This in turn may
induce voters to switch to cultural identity, even if the shock in-
creases income inequality.
Globalization is another possible driver of higher ρ if losers
from globalization are disproportionately conservative, which may
again be due to their lower education.9 Online Appendix 2 for-
malizes this argument by assuming that the labor endowment of
progressive voters is more tied to the export sector than that of
conservative voters. As globalization determines a higher remu-
neration of labor in the export sector and a lower one in the im-
port sector, the correlation between income and culture increases,
which can again promote cultural identity.10
We mentioned the possibility that voters may identify with
narrower groups. Online Appendix 3 shows that the main message
of Proposition 2 holds also if voters can identify with their finer
economic and cultural group. This means, for instance, that a voter
with ε < ε and ψ < ψ  can identify with the group of lower-class
and socially conservative people (L, SC). Two main results emerge
from this analysis. First, narrow identity is often not chosen, even
if available. The intuition is that it tends to reduce conflict relative
to broader cultural or economic groups. The narrower group is
closer to the voter, but less salient.11

9. As discussed by Autor, Dorn, and Hanson (2016), exposure to import com-

petition in the United States and other industrial countries has primarily hurt
low-skilled workers. As shown in note 7, education is positively correlated with
the demand for progressive cultural policies.
10. As shown in a previous version, trade shocks can induce a switch to cultural
identity by creating trade policy conflict. If conservatives are more exposed to trade
than progressives while economic classes are equally exposed, then trade shocks
can intensify trade policy conflict between cultural groups, favoring a switch to
cultural identity.
11. To see this, consider conservative and lower-class voters in (L, SC). Relative
to the lower class L, the narrower group (L, SC) features lower income conflict
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2391

The second result is that even if some voters may choose nar-
row identity under some conditions, the key qualitative implica-
tions of Proposition 2 still hold. In particular, it remains true that
an increase in  κ promotes cultural identity. This may occur be-
cause voters switch from their class to their narrow group, rather

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

than from class to culture. However, because the narrow group
is also defined along cultural values, it remains true that culture
becomes more important in shaping voters’ beliefs. This preserves
the gist of our results.
In our model, identity switches occur suddenly and for ev-
erybody at the same time. This is an artifact of our assump-
tions. In a more general model where identity reflects a trade-
off between group contrast and similarity between the individual
and the group average (as per criterion (i) above), the threshold
for identity switches differs across individuals. As a result, the
process of identity realignment triggered by growing  κ is more
gradual.12

III.C. Policy Preferences

How does identity affect policy preferences? As in Section II,
identity distorts beliefs through stereotypes, but beliefs now con-
cern income prospects, ε̃, and cultural views, ψ̃. This, in turn,
θ
affects preferences over τ and q. Let (εG , ψGθ ) denote the perceived
expected income and culture of voter (ε, ψ) if she identifies with
εψ εψ
group G. Let τG and qG denote her policy bliss points. Repeating

because its outgroup contains some lower-class voters, the socially progressive
ones. Relative to the conservatives SC, the narrower group (L, SC) features lower
cultural conflict because its outgroup contains some conservative voters, the upper-
class ones.
12. When identity also depends on a voter’s similarity to her groups, some
voters identify based on culture, others on class. It remains true, though, that
higher salience of culture relative to income (higher κ̂) favors cultural identity.
But in this case, extreme voters along income and culture are more likely to
change identity when a shock hits. Extreme voters feel intensely about both is-
sues, so they readily switch to the most fitting identity. The formal analysis of
that model can be found in a prior draft, Gennaioli and Tabellini (2019). We could
also allow for heterogeneity in the propensity to identify with any social group
(as if acquiring a social identity entailed a cost). In this case, identification with
a group would be triggered only if group conflict is sufficiently large. If unidenti-
fied individuals behave rationally, the results remain very similar, as long as at
least some voters are always identified in the dimension that maximizes group
conflict.
2392 THE QUARTERLY JOURNAL OF ECONOMICS

the steps that led to Proposition 1, and using equation (4), we

have:
PROPOSITION 3. A voter (ε, ψ) identified with group G = SC, SP, U,
L perceives her future expected income and her cultural trait

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

to be:
θ
(11) εG = ε + θ (ε G − ε G ),

(12) ψGθ = ψ + θ (ψ G − ψ G ).

The same voter’s ideal policies are given by:

εψ β(ψ G − ψ G ) − (ε G − ε G )
(13) τG = τ εψ + θ ,
ϕ

εψ
(14) qG = qεψ + θ (ψ G − ψ G ).

The voter distorts her perceived future income and her cul-
tural views by the belief difference between ingroups and out-
groups. This in turn distorts her perception of the ideal policy. If
the voter identifies with a group that is richer than the outgroup,
her perceived tax burden is enhanced, otherwise it is dampened. If
the voter identifies with a group that is more progressive than the
outgroup, her perceived benefit from liberal cultural policies and
universalistic redistribution is enhanced, otherwise it is damp-
ened.
Consider a lower-class and conservative voter, ε <  ε, ψ < ψ .
Using the above result and equations (6) and (7), if this voter
identifies with the lower class, her bliss points are:
(15)
εψ θ (1 − βρ)(εU − ε L) εψ
τ L = τ εψ + , qL = qεψ − ρθ (εU − ε L).
ϕ

Given that βρ < 1, the voter exaggerates her demand for redistri-
εψ
bution, τ L > τ εψ , especially if income inequality among classes
is strong. When thinking about herself, the poverty of her so-
cial group comes to mind. As evoked by Karl Marx, she feels
part of an economically oppressed class, and this increases her
demand for redistribution. Since ρ > 0, the lower class is on
average conservative, so the voter also demands a stricter cul-
εψ
tural policy qL < qεψ . Conservatism also dampens her demand
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2393

for redistribution but, as we already saw, this effect is dominated

because βρ < 1.
Now suppose that cultural policy becomes salient. If the voter
switches to a conservative identity, her bliss points become:
(16)

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

εψ θ (β − ρ)(ψ SP − ψ SC ) εψ
τ SC = τ εψ − , qSC = qεψ − θ (ψ SP − ψ SC ).
ϕ

The voter’s conservatism increases, due to Assumption (A1), so

εψ εψ
she is now even more opposed to liberal cultural policy, qSC < qL .
εψ εψ 13
She also demands less redistribution than before, τ SC < τ L . As
the voter abandons class identity, two forces work in this direc-
tion. First, the voter is more optimistic about her future income.
Second, she becomes more communitarian, which reduces her de-
mand for universalistic redistribution.
This change in preferences may be caused by a shock that
is barely related to redistribution, such as a salient conflict on
abortion or a large inflow of immigrants from a different culture.
Alesina, Miano, and Stantcheva (2018) find that making people
think about immigrants reduces their support for redistribution,
particularly if the respondent is less educated and right-wing.
They argue that respondents are unwilling to redistribute toward
foreigners. A different interpretation is that the treatment primes
cultural identity in conservative and anti-immigrant subjects. If
this is true, the treatment effect should be heterogeneous: con-
servative subjects should reduce their demand for redistribution
while progressives may even increase it, because they are willing
to redistribute in favor of immigrants.

III.D. Equilibrium Policy

We now study how identity affects equilibrium policy. As in
standard models of probabilistic voting, two candidates commit to
policy platforms ahead of the elections to maximize the probability
of winning (see Persson and Tabellini 2000). We assume that all
voters have the same degree of mobility across parties, so the
equilibrium policy maximizes perceived utilitarian welfare. Let
εψ
Wd (τ, q) denote the perceived expected utility of agent (ε, ψ) if
she identifies based on dimension d, for d = ε̃, ψ̃. The equilibrium

13. This is obviously true if β > ρ but it is also true if β < ρ because, by As-
sumption (A1), (1 − βρ)(εU − ε L) > (1 − βρ)ρ(ψ SP − ψ SC ) which is in turn larger
than (ρ − β)(ψ SP − ψ SC ) due to ρ < 1.
2394 THE QUARTERLY JOURNAL OF ECONOMICS

policy is defined by:

εψ
(17) (τ ∗ , q∗ ) = arg max Wd (τ, q) dH (ψ, ε) , for d = ε̃, ψ̃.
τ,q

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

Suppose that everyone identifies based on class. Exploiting
Proposition 3, the first-order conditions of the problem imply:

(1 − βρ) (εU − ε L) (π L − πU )
(18) τ∗ = τ◦ + θ ,
ϕ
(19) q∗ = q◦ − θρ (εU − ε L) (π L − πU ),

where τ ◦ and q◦ denote the socially optimal policies, and π L , π U

denote the size of the upper and the lower class, respectively.
If θ = 0, beliefs are rational and identity has no effect on
equilibrium policy. When instead θ > 0 identity matters. If  ε > 0,
the lower class is the larger economic group, π L > π U . Thus, taxes
are too high and, if ρ > 0, cultural policy is too conservative. These
distortions increase with class inequality (εU − ε L). In Marxist
theory, class consciousness is necessary for the poor majority to
succeed. In a similar way, class identity causes lower-class voters
to be more radical. Opportunistic politicians then accommodate
their demands because the lower class is larger. If the lower class
is more conservative, cultural policy is also too restrictive, for the
same reason.
Suppose now that everyone identifies based on culture. Then
we have:
(β − ρ)(ψ̄ SP − ψ̄ SC )(π SC − π SP )
(20) τ∗ = τ◦ − θ ,
ϕ
(21) q∗ = q◦ − θ (ψ̄ SP − ψ̄ SC )(π SC − π SP ).
Here, too, if θ > 0 the stereotypes of the larger group affect
the equilibrium. If ψ̂ > 0, the conservative group is larger, π SC >
π SP . Thus, cultural policy is too restrictive (q∗ < q◦ ). If the average
conservative demands lower spending (β > ρ), taxes are too low,
otherwise they are too high.
A switch from economic to cultural identity impacts (τ , q) in
a way that depends on (i) the relative severity of economic versus
cultural differences and (ii) the relative size of different groups.
If ε̂ = ψ̂ > 0 income inequality and cultural conflict are commen-
surate, ψ̄ SP − ψ̄ SC = εU − ε L and the dominant groups have the
same size π L = π SC > π U = π SP . In this case, a switch from class
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2395

to culture always causes cultural policy to become more conserva-

tive and taxes to fall (the latter due to the fact that 1 − βρ > β − ρ).
We summarize this discussion in the following:

PROPOSITION 4. Suppose that ε̂, ψ̂ > 0.

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

i. Under class identity, τ ∗ > τ ◦ and q∗  q◦ , with strict in-
equality if ρ > 0.
ii. Under cultural identity, q∗ < q◦ , and τ ∗  τ ◦ as β  ρ.
iii. If in addition ε̂ = ψ̂, a switch from class to cultural iden-
tity reduces both τ ∗ and q∗ .

In our model, candidates are opportunistic and follow chang-

ing voters’ preferences. Suppose instead that candidates are par-
tisan and have different policy platforms, as in Alesina (1987),
with the right-wing candidate being culturally and fiscally con-
servative, and the opposite for the left-wing candidate. Then, a
shift from class to culture reshuffles voters across parties. Con-
servative and lower-class voters, who voted for the left on the
basis of their class, are now attracted by the culturally conser-
vative right-wing party. Progressive and upper-class voters are
instead attracted by the culturally progressive left-wing party.
Gethin, Martinez-Toledano, and Piketty (forthcoming) shows that
after the 1960s, similar mobility patterns have occurred in several
Western democracies.

IV. EMPIRICAL PREDICTIONS

In this section we illustrate the key predictions of our model.
As shown in Figure I, after 2008 disagreement over cultural pol-
icy has increased, disagreement over redistribution has remained
constant or has declined, and views over cultural policy and redis-
tribution have become more correlated. An identity switch from
εψ εψ
class to culture produces these effects. Let τd , qd denote the bliss
points of voter (ε, ψ) if she identifies on dimension d = ε̃, ψ̃.

PROPOSITION 5. Suppose that ε̂ = ψ̂ > 0. Then an increase in κ trig-

gering an identity switch from class to culture causes the
following effects:
i. the variance of ideal cultural policies increases while the
εψ εψ
variance of ideal tax rates decreases: V ar(qψ̃ ) > V ar(qε̃ )
εψ εψ
and V ar(τψ̃ ) < V ar(τε̃ ).
εψ εψ
ii. the correlation coefficient between τd and qd in-
creases. If β > ρ, under cultural identity this correlation
2396 THE QUARTERLY JOURNAL OF ECONOMICS

coefficient is positive, but it may be negative under class

identity.

When identity switches to culture, conflict over q intensifies

because cultural beliefs polarize. Conflict over τ is subject to two

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

effects. On the one hand, increased cultural polarization implies
more disagreement over the perceived benefits of public spending.
On the other hand, dampening class conflicts reduces disagree-
ment over the perceived tax burden. The second effect is stronger
than the first if ε̂ = ψ̂. Thus, disagreement over τ falls.
The switch to cultural identity also causes the cultural factor
driving both q and τ to become more potent. Conservative vot-
ers increase their opposition to immigration and civil rights and
become less favorable to redistribution. Progressive voters move
in the opposite direction. As a result, the correlation between the
demand for redistribution and that for open cultural policies in-
creases.
Proposition 5 shows that an increase in the salience κ of cul-
tural policy can parsimoniously account for Figure I. A version of
our model with rational voters cannot do the same, because higher
κ plays no role when θ = 0. More generally, in the rational model,
no single shock can account for the facts, but a combination of
shocks is needed. We return to this point in Section V, where we
discuss other possible explanations of Figure I.
The preference changes of Proposition 5 are due to changing
conflict among underlying social groups. To see this in detail, let
τdG , qdG denote the average bliss points of members of group G if
they identify on dimension d = ε̃, ψ̃.
PROPOSITION 6. Suppose that ε̂ = ψ̂ > 0. Then, an increase in κ
causing identity to switch from class to culture affects the
policy preferences of different social groups as follows.
i. Relative to the conservatives, the progressives become
even more in favor of liberal cultural policies; if β > ρ,
they also become even more in favor of high taxes:

(22) qψ̃SP − qψ̃SC > qε̃SP − qε̃SC > 0

(23) τψ̃SP − τψ̃SC > τε̃SP − τε̃SC > 0 if β > ρ.
ii. Relative to the upper class, the lower class becomes even
more opposed to liberal cultural policies; it also becomes
less favorable to high taxes:
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2397

(24) qψ̃L − qψ̃U < qε̃L − qε̃U < 0

(25) τψ̃L − τψ̃U < τε̃L − τε̃U .

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

As voters switch to cultural identity, their cultural traits be-
come exaggerated. Across cultural groups, the progressives be-
come even more progressive, and the conservatives become even
more conservative. As a result, their conflict over cultural policy
intensifies. Conflict over τ between cultural groups, τdSP − τdSC ,
also increases, provided that the influence of culture on redis-
tributive preferences (i.e., β) is strong enough.
The opposite happens to perceived income differences be-
tween classes: lower-class voters feel less poor and upper-class
voters less rich. Hence, their assessment of the tax burden be-
comes less divergent. In addition, the conservative majority of
the lower class becomes more communitarian and demands less
government spending, and the reverse is true for the upper class.
For both reasons, the lower class demands lower taxes than be-
fore, while the upper class does the opposite. Finally, the lower
class becomes more supportive of conservative cultural policies,
compared with the upper class. As a result, conflict over cultural
policies between opposite income groups rises.
Going back to Proposition 5, the amplification of conflict
over cultural policy among economic and cultural groups explains
growing disagreement over q in the population as a whole. The re-
duction of redistributive conflict among income classes and its in-
crease among cultural groups explains why overall disagreement
over τ falls but preferences over τ and q become more correlated
over time.

V. COMPARING THE THEORY WITH THE DATA

We now present evidence from survey data consistent with
the predictions. The chain of causation in our model is:
salient issue =⇒ group identity =⇒ beliefs/policy preferences
=⇒ equilibrium policy.
The effect of an identity switch on the beliefs of voters de-
pends on their economic and cultural traits. In turn, beliefs af-
fect voters’ policy preferences. To test this mechanism, one should
ideally observe voters’ identities, beliefs, and policy preferences.
2398 THE QUARTERLY JOURNAL OF ECONOMICS

Unfortunately we do not observe identities or beliefs. As a result,

we cannot rigorously test our mechanism.
Nevertheless, in this section we show that several patterns in
the data are broadly consistent with our predictions and cannot
easily be accounted for by competing explanations. In the above

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

chain of causation, the trigger of the trends is an increasing con-
flict among cultural groups. This could be due to greater impor-
tance of immigration and civil rights, κ, or to an economic shock
hurting conservative voters (higher ρ). As identity switches from
class to culture, voters’ beliefs change and, by Propositions 5 and 6,
we should observe the following: (i) an increase in cultural conflict
(measured by the population variance of q and by disagreement
over q among cultural groups), (ii) a decrease in redistributive
conflict (measured by the population variance of τ and by dis-
agreement over τ among classes), and (iii) individual views on τ
and q becoming more positively correlated in the population.
To see whether Figure I could be attributable to a switch to
cultural identity, consider first whether the salience of cultural
conflict has increased in the past 20 years. Using data from a
repeated Pew survey between 2001 and 2018, Figure II plots the
share of respondents that report a certain issue as one of the three
most important problems facing the United States.
The importance of race and immigration has soared from 2012
onward, which coincides with the time in which the patterns of
Figure I are most pronounced.14 In our model, this phenomenon
corresponds to a higher welfare importance κ, and hence higher
salience of cultural policy.15
A second natural step is to see whether a growing importance
of cultural policies is associated with stronger cultural identity.

14. To highlight long-term trends, we omit the state of the economy and un-
employment, which follow the business cycle (especially the Great Recession) and
national security. Cultural issues were prominent also in the early 2000s. As em-
phasized by Abramowitz (2019), this is likely to reflect a continuing decline of the
share of white voters over the eligible population, due to immigration from Asia
and Latin America and a higher fertility of nonwhites. “According to ANES data,
(in 2016) nonwhites made up 39% of eligible voters under the age of 30, compared
with only 17% of eligible voters over seventy” (Abramowitz 2019, ch. 1).
15. In our model, there is only one cultural policy q, but the analysis is the
same if there are several dimensions of cultural policy q1 , ..., qk whose bliss points
are highly correlated because they are all driven by a voter’s socially progressive
or conservative stance ψ. If there is more than one cultural issue, parameter κ
captures the average importance of these issues. In this case, a growing importance
of several issues at the same time contibutes to a strong increase in κ.
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2399

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

FIGURE II
Most Important Problem Facing the Country
The graph shows the share of respondents mentioning the selected issues among
the top three most important problems facing the United States. The analysis of
the most important problem is based on the following question: “What do you
think is the most important problem facing the country today? [Record up to three
responses, in order of mention].” The question is open-ended, but in the public
release of the Pew data sets, answers have been classified in macro-categories.
Before constructing Figure II, categories “Abortion” and “Rights of Women under
Attack/Rolling Back” were aggregated in the macro-category “Abortion and Women
Rights.” To highlight long-term trends, we omit “the state of the economy” and
“unemployment,” which follow the business cycle (especially the Great Recession)
and national security. Estimates are weighted with survey weights. Source: Pew
Research Center public data.

We do not measure the strength of identification with cultural

groups or with income classes. In ANES, however, respondents
report “thermometer feelings” on how close they feel to certain
groups. We use this information. Respondents are assigned to
opposite economic classes based on a question on self-reported
social class. They are assigned to opposite cultural groups based
on their religiosity, a cultural trait that Enke, Rodriguez-Padilla,
and Zimmermann (forthcoming) have shown to be strongly (neg-
atively) correlated with moral universalism. Online Appendix 4
provides more details on the definition of social class, religiosity,
and all other variables used in the analysis.
As a proxy for affective class polarization, we take the dif-
ference between the thermometer of self-classified upper- and
2400 THE QUARTERLY JOURNAL OF ECONOMICS

lower-class respondents toward (i) unions and (ii) big businesses—

two symbols of class identity. As a proxy for affective cultural
polarization, we take the difference between the thermometer of
religious and secular respondents toward Christian fundamen-
talists and Catholics. To remove the confounding effect of iden-

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

tification with political parties, we use the estimated residuals
of these feelings, after conditioning on dummies for party affil-
iation interacted with wave fixed effects (patterns are stronger
or similar in the unconditional data and in the sample of politi-
cal independents, as shown in the Online Appendix). Figure III
plots these indices over time, together with their 95% confidence
intervals.
In Panel A, the upper class feels more distant from unions and
closer to businesses compared to the lower class, but these class
differences in feelings have shrunk and are not statistically sig-
nificant in recent years, suggesting a reduction in affective class
polarization. In Panel B, religious respondents feel closer to Chris-
tian fundamentalists and Catholics than secular respondents, and
these differences in feelings have diverged, suggesting a rise in
affective cultural polarization. This is consistent with conflict cool-
ing off among economic groups and heating up between religious
and secular groups.
The third step in the chain of causation is the change in beliefs
and policy preferences. We have already shown in Proposition 5
that our model can account for the trends of Figure I. Here we
ask whether these changes in policy preferences are tied to eco-
nomic and cultural groups, as predicted by Proposition 6. A switch
to cultural identity should dampen redistributive conflict among
classes and enhance it among cultural groups (if β > ρ). It should
also exacerbate cultural conflict among cultural groups as well as
between classes. Figure IV reports the position over redistribution
and cultural policies of respondents sorted into different income
classes and cultural groups. The latter are defined again based on
respondents’ religiosity. Here, too, we condition on party affilia-
tion interacted with wave fixed effects to control for shifting party
positions (the trends are stronger or similar if we do not condition
on party affiliation or if we focus on political independents, see
the Online Appendix).
The trends in group conflict are in line with our predictions.
In a robustness exercise, we split respondents into two equally
sized groups of conservatives and progressives based on the
first principal component of four questions on the importance of
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2401

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

(A) (B)

FIGURE III
Social Groups and Feeling Thermometer
Panel A plots the differences in the mean feelings of the upper-middle/upper
class versus lower/working class toward labor unions (solid line) and big business
(dashed line), with 95% confidence intervals. Panel B plots the differences in the
mean feelings of religious versus secular individuals toward Christian fundamen-
talists (solid line) and Catholics (dashed line), with 95% confidence intervals. Feel-
ings toward the four groups are measured with questions of this kind: “How would
you rate the following group: group X.” Answers are on a 0–100 scale, from colder
to warmer feelings. Before constructing Figure III, answers to the feeling ther-
mometer questions are residualized on party identity, alone and interacted with
wave fixed effects (more detail in the Online Appendix). Residuals are standard-
ized to have zero mean and unit variance across all waves. Class is a self-reported
variable with the following categories: Lower, Average Working, Working, Upper
Working, Average Middle, Middle, Upper Middle, and Upper. Lower/working class
(L/W class) is obtained aggregating Lower, Average Working, Working and Upper
Working (roughly 50% of the pooled sample); upper-middle/upper class (U-M/U
class) is obtained aggregating Upper Middle and Upper (roughly 15% of the pooled
sample). Religiosity is measured by the question “Do you consider Religion to be
an important part of your life? [Yes; No].” Respondents answering “Yes” (“No”)
are classified as Religious (Secular). Estimates are weighted with survey weights.
Source: ANES.

traditional values. The trends of Figure IV are confirmed (see

Online Appendix Figure A.7).
Another way to detect voters’ realignment into more acute
cultural conflict is to use cluster analysis. We use the K-means
algorithm to separate two clusters of voters in the bidimen-
sional policy space defined by the demand for progressive cul-
tural policies and redistribution. In the subperiod 2000–2008, the
two emerging clusters are almost exclusively defined by pro- and
2402 THE QUARTERLY JOURNAL OF ECONOMICS

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

(A)

(B)

FIGURE IV
Trends in Group Conflict
Panel A reports trends in the means of redistribution and culture for the
lower/working class (solid line) and the upper-middle/upper class (dashed line),
with 95% confidence intervals. Panel B reports trends in the means of redistri-
bution and culture for secular (solid line) and religious individuals (dashed line),
with 95% confidence intervals. Redistribution is the first polychoric principal com-
ponent of two questions on government spending and government’s role in seeing
to citizens’ jobs and living standards; culture is the first polychoric principal com-
ponent of desired immigration levels, attitudes toward race and abortion policy
(see the note of Figure I for the specific questions). For redistribution and culture,
higher values correspond to more liberal views. The two variables are residualized
on party identity, alone and interacted with wave fixed effects. Residuals are stan-
dardized, with zero mean and unit variance across all waves. Class and religiosity
are self-reported (see the note of Figure III). Estimates are weighted with survey
weights. Source: ANES.
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2403

anti-redistribution policy views. In the subperiod 2012–2016, the

clusters identified by the algorithm feature two-dimensional con-
flict: one cluster is pro-redistribution and culturally progressive,
the other is anti-redistribution and culturally conservative. In-
terestingly, during 2000–2008 cluster membership is predicted by

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

income class but not by religiosity, while during 2012–2016 clus-
ter membership is mostly predicted by religiosity (in the expected
directions). This additional evidence, reported in Online Appendix
Figure A.8, Table A.1, and Table A.2, is consistent with the notion
that the patterns in Figure I may be due to a growing importance
of the cultural divide, which has influenced policy views in the
cultural and economic domains.16
In sum, consistently with our model, a growing salience of
cultural conflict has gone hand in hand with a convergence of
feelings by social classes toward class-related groups, and with a
divergence of feelings by religious/secular groups toward religious
people. Moreover, and as predicted by Proposition 6, opposite eco-
nomic classes now disagree less over redistribution and more over
cultural policies, while opposite cultural groups disagree more in
both policy domains.
Consider alternative explanations of these facts, starting
from a rational version of our model. If θ = 0, the variance of bliss
points over q coincides with the variance of culture ψ, σψ2 . Thus, a
rational model must find an explanation for the growing disagree-
ment over cultural policy in Figure I. We explain growing cultural
disagreement through the growing salience of cultural policy κ,
as in Figure II. In a rational model, the salience of cultural policy
has no effect on polarization, although increased salience could be
a by-product of higher cultural polarization σ ψ .
Suppose that we take for granted an increase in cultural
disagreement σ ψ , due to mass education or changes in gender
roles (Fukuyama 2018). Is this consistent with the remaining
facts? In the rational model, a rise in σ ψ causes higher corre-
lation between the bliss points for q and τ , accounting for the
second fact of Figure I. The reason is that culture also shapes
preferences over the public good through parameter β > 0. If

16. The cluster detection algorithm identifies groups in the population that
feature the smallest within-group variance of policy views, which in the case of
two groups is equivalent to maximizing cross-cluster differences. This notion is
close, but not identical, to our conflict criterion in equation (8). Thus, this exercise
should not be interpreted as a test of the theory, even though the patterns are
consistent with the occurrence of a cultural realignment, as the theory predicts.
2404 THE QUARTERLY JOURNAL OF ECONOMICS

progressive voters become more extreme, they want a more lib-

eral cultural policy and more government spending; conservatives
want the opposite. Hence, views on these policies become more
tightly connected.
This intuition, though, suggests that growing cultural dis-

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

agreement σ ψ should also cause growing disagreement over redis-
tribution, contrary to what we observe. This is because culturally
extreme individuals also hold extreme views on τ . In particular,
if preferences over q and τ are positively correlated (like in the
ANES data from 2008 onward) and if θ = 0, then an increase in
σ ψ should cause the variance of bliss points over τ to increase.
Intuitively, a positive correlation between q and τ means that a
voter’s culture is an important determinant of her demand for
redistribution. In this case, a rise in cultural polarization should
also result in higher polarization over τ .17
Thus, an exogenous increase in cultural disagreement can
explain the increased coherence of views over τ and q, but not why
disagreement over τ has, if anything, decreased. In our model, a
switch from class to cultural identity produces these patterns via
two effects. First, it polarizes cultural preferences, which is akin
to increasing σ ψ . Second, it depolarizes class conflict, which is
akin to reducing the variance of expected income, σ ε . This second
effect reduces the variance of tax preferences in Proposition 5.
Thus, the rational model needs a combination of shocks (to σ ψ
and to σ ε ) to account for the above facts. Our model only needs
growing salience of cultural conflict.18
A second set of explanations for these facts emphasizes the
role of political supply. One version stresses partisan sorting.

17. Let R denote the correlation coefficent between the individual bliss points
of q and τ , let σ q and σ τ be the standard deviations of these bliss points. Using
equation (4), we have:
βσψ − ρσε
R=
ϕστ
β 2 σψ2 + σε2 − 2βρσψ σε
στ2 = ,
ϕ2
2
∂R
which in turn imply: ∂σ q
= β(1−R
ϕστ
) ∂στ
> 0 and ∂σ q
= βϕR > 0 if R > 0.
18. In the rational model, conflict over τ is also affected by other parameters.
An increase in tax distortions, ϕ, or in the correlation ρ between ε and ψ, reduces
the variance of ideal tax rates. But these changes also reduce the correlation of
views over taxes and cultural policy—recall that income and cultural progressive-
ness have opposite effects on τ .
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2405

The Democratic Party has become more progressive, the Re-

publican Party more conservative, causing voters with more ex-
treme cultural preferences to switch party allegiance (Gentzkow
2016; Klein 2020). This mechanism can explain the more co-
herent disagreement and increased cultural polarization be-

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

tween Republican and Democratic supporters, but it cannot
alone explain changing policy preferences in the population at
large.
Another version allows for persuasion by political leaders.
Parties have differentiated their platforms on immigration and
civil rights to attract voters who feel strongly about these issues.
Political leaders have then persuaded voters to follow their po-
sitions on issues that voters care less about, such as redistribu-
tion (Murphy and Shleifer 2004; Glaeser 2005). If messages by
political leaders increased the salience of cultural conflict, favor-
ing a switch to cultural identity, the endogenous transmission
to policy preferences is fully consistent with our model. It may
also be possible that political leaders are able to directly per-
suade voters to take more extreme cultural views while moder-
ating their demand for redistribution in a way that is unrelated
to identity. However, the mechanism for how this persuasion may
operate remains to be spelled out. Moreover, Figures I, II, and
IV hold after controlling for time-varying effects of party affil-
iation and after considering only independent voters, while di-
rect political persuasion should arguably mostly affect partisan
voters.
Another distinctive prediction of our model is that economic
shocks disproportionately hurting conservative voters should in-
duce a switch to cultural identity. Thus, economic losers should
become socially and fiscally conservative. In a rational model,
instead, voters hit by economic shocks would demand more redis-
tribution and would not change their views over cultural policy.
Autor et al. (2020) and Colantone and Stanig (2018) show that
in the United States and in Europe, losses from international
trade foster support for right-wing and conservative parties. Our
model can account for this effect if, as shown in Online Appendix
2, losers from trade are disproportionately conservative. In this
case, our model predicts that losers from trade also become more
socially and fiscally conservative in their policy opinions. A previ-
ous version of this article (Gennaioli and Tabellini 2019) reported
evidence from CCES survey data consistent with this prediction.
2406 THE QUARTERLY JOURNAL OF ECONOMICS

VI. CONCLUDING REMARKS

It is often argued that recent years have witnessed a rise
in identity politics, intended as the growing importance of con-
flict over civil rights and minorities (Fukuyama 2018). We take

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

a different perspective: current events underscore the role of
cultural identities, but periods of class conflict have their own
identity, too. As we suggest, at any given point in time, vot-
ers perceive their social and political reality from the vantage
point of the most salient group among the many latent ones
to which they belong. We considered groups defined along eco-
nomic and cultural traits, but regional, racial, or other groups
are also possible. As political cleavages change, voters switch
identity from their income class to their cultural, regional, or
racial group. These identity switches cause a realignment of
voters’ beliefs across all issues in which groups systematically
differ, including issues that are barely related to the identity
trigger.
We explored some key implications of this approach, but much
more remains to be done. First and foremost, survey data, possibly
combined with priming techniques, should be used to assess the
link between identity and beliefs, which we did not explore here.
This may shed light on the effects of fake or partial news, as well
as on the role of digital media, such as Twitter or Facebook, that
allow leaders to reach out to voters with emotional and symbolic
messages that appeal to their identities.
We have focused on the consequences of a shift from class
to cultural identity. But other important episodes in U.S. politi-
cal history can be interpreted as identity shifts in the opposite
direction: from culture to class. One prominent example is the po-
litical realignment that took place in the late 1930s and 1940s in
the United States that pushed the Democratic Party to abandon
its support for racial discrimination. As emphasized by Schickler
(2016), this was the result of a grassroots movement initiated by
the core groups that supported the New Deal. Industrial unions,
African Americans, and urban liberals in the Northeast pushed
for a fusion of class and race, in a joint defense of labor and civil
rights. They did so out of the ideological left-wing conviction that
racial division undermined class consciousness, but also because
of expediency: the inflow of Black immigrants from the South un-
dermined the threat of strikes as Black workers could be used as
replacements in the workplace. Incorporating Black people in the
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2407

working class would remove this threat and strengthen the la-
bor movement.19 In line with this, Calderon, Fouka, and Tabellini
(2021) study U.S. counties between 1940 and 1970 and show that
where the inflow of Black immigrants from the South was larger,
the Democratic Party gained more votes and grassroots activism

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

was strengthened.
We do not allow parties or politicians to shape identities. Yet
political and group leaders often play an important role in this,
for instance, by enhancing the salience of certain groups. Marx-
ist thinkers such as Gramsci stressed the role of the Commu-
nist Party and intellectuals in fostering class awareness. Nation
builders such as Bismarck used nationalism to mobilize support,
and the Catholic Church promoted identity politics on the basis of
religious values. Political leaders may also create new, party-based
identities that supersede traditional social groups. Glaeser (2005)
analyzes how a leader can mobilize voters by spreading messages
of hatred against a minority group. Glaeser, Ponzetto, and Shapiro
(2005) discuss how a party can energize its supporters by taking
more extreme positions. We think that our demand-side approach
may be useful to understand these and other supply-side aspects.
For instance, it may be easier to identify with politicians who im-
personate group stereotypes. This implies that when polarization
is strong, the most successful politicians come from the tails, not
the middle. Successful populists will then look similar to the un-
skilled and unexperienced labor market outsiders that voted them
in office (see in particular Dal Bó et al. 2019).
A related set of questions concerns the evolution of party
systems. When does a party represent a single identity group, and
when does it instead act as an ensemble of heterogeneous social
identities? The U.S. Republican Party seems to represent those on
the right that identify along the income dimension and the social
conservatives that identify on culture, whereas the Democrats
stand for the opposite groups in each dimension. But this has
changed at times of major political realignments. How do party
realignments interact with social identities, and how do political
and social identities influence each other?

19. In the words of union leader John Brophy, quoted by Schickler (2016),
“Behind every lynching is the figure of the labor exploiter, the man in the corpora-
tion who would deny labor its fundamental rights.” The vote of African Americans
was also pivotal in several key districts in the Northeast, and this too induced
the Democratic Party to recognize African Americans as full-right members of the
working class.
2408 THE QUARTERLY JOURNAL OF ECONOMICS

We believe that exploring these issues within the framework

of identity theory opens up a new and exciting research agenda.

DEPARTMENT OF ECONOMICS, UNIVERSITY OF CALIFORNIA, SAN DIEGO,

UNITED STATES

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

DEPARTMENT OF FINANCE AND IGIER, UNIVERSITÀ BOCCONI, ITALY
DEPARTMENT OF ECONOMICS AND IGIER, UNIVERSITÀ BOCCONI, ITALY

SUPPLEMENTARY MATERIAL
An Online Appendix for this article can be found at The Quar-
terly Journal of Economics online.

DATA AVAILABILITY
Data and code replicating the figures in this article can be
found in Bonomi, Gennaioli, and Tabellini (2021) in the Harvard
Dataverse, https://doi.org/10.7910/DVN/6COTQW.

REFERENCES
Abramowitz, Alan I., The Great Alignment: Race, Party Transformation, and the
Rise of Donald Trump (New Haven, CT: Yale University Press, 2018).
Achen, Christopher H., and Larry M. Bartels, Democracy for Realists: Why Elec-
tions Do Not Produce Responsive Government (Princeton, NJ: Princeton Uni-
versity Press, 2016).
Ahler, Douglas, “The Group Theory of Parties: Identity Politics, Party Stereotypes,
and Polarization in the 21st Century,” Forum, 16 (2018), 3–22.
Akerlof, George, and A. Rachel E. Kranton, “Economics and Identity,” Quarterly
Journal of Economics, 115 (2000), 715–753.
Alesina, Alberto, “Macroeconomic Policy in a Two Party System as a Repeated
Game,” Quarterly Journal of Economics, 102 (1987), 651–678.
Alesina, Alberto, Armando Miano, and Stefanie Stantcheva, “Immigration and
Redistribution,” NBER Working Paper no. w24733, 2018.
Alesina, Alberto, Stefanie Stantcheva, and Edoardo Teso, “Intergenerational Mo-
bility and Preferences for Redistribution,” American Economic Review, 108
(2018), 521–554.
Autor, David, “Work of the Past, Work of the Future,” American Economic Review
Papers and Proceedings, 109 (2019), 1–32.
Autor, David, and David Dorn and Gordon Hanson, “The China Shock: Learning
from Labor-Market Adjustment to Large Changes in Trade,” Annual Review
of Economics, 8 (2016), 205–240.
Autor, David, David Dorn, Gordon Hanson, and Kaveh Majlesi, “Importing Po-
litical Polarization? The Electoral Consequences of Rising Trade Exposure,”
American Economic Review, 110 (2020), 3139–3183.
Bénabou, Roland, and Jean Tirole, “Mindful Economics: The Production, Con-
sumption, and Value of Beliefs,” Journal of Economic Perspectives, 30 (2016),
141–164.
Bonomi, Giampaolo, Nicola Gennaioli, and Guido Tabellini, “Replication Data
for: ‘Identity, Beliefs, and Political Conflict’,” (2021), Harvard Dataverse,
https://doi.org/10.7910/DVN/6COTQW.
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2409

Bordalo, Pedro, Katherine Coffman, Nicola Gennaioli, and Andrei Shleifer,

“Stereotypes,” Quarterly Journal of Economics, 131 (2016), 1753–1794.
———, “Beliefs about Gender,” American Economic Review, 109 (2019), 739–773.
Bordalo, Pedro, Marco Tabellini, and David Yang, “Stereotypes and Politics,” NBER
Working Paper no. w27194, 2020.
Calderon, Alvaro, Vasiliki Fouka, and Marco Tabellini, “Racial Diversity, Electoral

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

Preferences, and the Supply of Policy: The Great Migration and Civil Rights,”
IZA Discussion Paper no. 14312, 2021.
Card, David, Christian Dustmann, and Preston Ian, “Immigration, Wages, and
Compositional Amenities,” Journal of the European Economic Association, 10
(2012), 78–119.
Colantone, Italo, and Piero Stanig, “The Trade Origins of Economic Nationalism:
Import Competition and Voting Behavior in Western Europe,” American Jour-
nal of Political Science, 62 (2018), 936–953.
Cruces, Guillermo, Ricardo Perez-Truglia, and Martin Tetaz, “Biased Perceptions
of Income Distribution and Preferences for Redistribution: Evidence from a
Survey Experiment,” Journal of Public Economics, 98 (2013), 100–112.
Dal Bó, Ernesto, Frederico Finan, Olle Folke, Torsten Persson, and Johanna
Rickne, “Economic Losers and Political Winners: Sweden’s Radical Right,”
Mimeo, UC Berkeley, 2019.
Druckman, James N., and Matthew S. Levendusky, “What Do We Measure When
We Measure Affective Polarization?,” Public Opinion Quarterly 83 (2019), 114–
122.
Enke, Ben, Ricardo Rodriguez-Padilla, and Florian Zimmermann, “Moral Uni-
versalism and the Structure of Ideology,” NBER Working Paper no. w27511,
2020.
———, “Moral Universalism: Measurement and Economic Relevance,” Manage-
ment Science (forthcoming).
Evans, Geoffrey, and Jonathan Mellon, “Social Class - Identity, Awareness and
Political Attitudes: Why Are We Still Working Class?,” British Social Attitudes,
33 (2016), 1–19.
Festinger, Leo, “Informal Social Communication,” Psychological Review, 57 (1950),
271–282.
Flynn, D. J., Brendan Nyhan, and Jason Reifler, “The Nature and Origins of Mis-
perceptions: Understanding False and Unsupported Beliefs about Politics,”
Advances in Political Psychology, 38 (2017), 127–150.
Frank, Thomas, What’s the Matter with Kansas?: How Conservatives Won the Heart
of America (New York: Metropolitan Books, 2004).
Fukuyama, Francis, Identity: The Demand for Dignity and the Politics of Resent-
ment (New York: Farrar, Strauss and Giroux, 2018).
Gennaioli, Nicola, and Andrei Shleifer, A Crisis of Beliefs: Investor Psychology and
Financial Fragility (Princeton, NJ: Princeton University Press, 2018).
Gennaioli, Nicola, and Guido Tabellini, “Identity Beliefs and Political Conflict,”
CEPR discussion paper 13390, 2019.
Gentzkow, Matthew, “Polarization in 2016,” Toulouse Network for Information
Technology White Paper, 2016.
Gethin, Amory, Clara Martinez-Toledano, and Thomas Piketty, “Brahmin Left
Versus Merchant Right: Changing Political Cleavages in 21 Western Democ-
racies, 1948–2020,” Quarterly Journal of Economics, forthcoming.
Gimpelson, Vladimir, and Daniel Treisman, “Misperceiving Inequality,” Economics
& Politics, 30 (2018), 27–54.
Glaeser, Edward, “The Political Economy of Hatred,” Quarterly Journal of Eco-
nomics, 120 (2005), 45–86.
Glaeser, Edward L., Giacomo A. M. Ponzetto, and Jesse M. Shapiro, “Strategic
Extremism: Why Republicans and Democrats Divide on Religious Values,”
Quarterly Journal of Economics, 120 (2005), 1283–330.
Goodhart, David, The Road to Somwhere: The New Tribes Shaping British Politics
(London: Penguin, 2017).
2410 THE QUARTERLY JOURNAL OF ECONOMICS

Grigorieff, Alexis, Christopher Roth, and Diego Ubfal, “Does Information Change
Attitudes towards Immigrants? Representative Evidence from Survey Exper-
iments,” Demography, 57 (2020), 1117–1143.
Grossman, Gene M., and Elhanan Helpman, “Identity Politics and Trade Policy,”
Review of Economic Studies, 88 (2021), 1101–1126.
Guriev, Sergei, and Elias Papaioannou, “The Political Economy of Populism,” Jour-

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

nal of Economic Literature (forthcoming).
Haidt, Jonathan, The Righteous Mind: Why Good People Are Divided by Politics
and Religion (New York: Pantheon Books, 2012).
Han, Jiyoung, and Daniel B. Wackman, “Partisan Self-Stereotyping: Testing the
Salience Hypothesis in a Prediction of Political Polarization,” International
Journal of Communication, 11 (2017), 603–625.
Haslam, Alexander S., and John C. Turner, “Context-Dependent Variation in
Social Sterotyping 2: the Relationship between Frame of Reference, Self-
Categorization and Accentuation,” European Journal of Social Psychology,
22 (1992), 251–277.
Hinkle, Steve, and Rupert Brown, “Intergroup Comparisons and Social Identity:
Some Links and Lacunae,” in Social Identity Theory: Constructive and Critical
Advances, Dominic Abrams and Michael A. Hogg, eds. (New York: Springer,
1990).
Hogg, Michael A., and Dominic Abrams, Social Identification: A Social Psychology
of Intergroup Relations and Group Processes (London: Routledge, 1988).
Janky, Béla, “Salience of Partisan Identity and Social Attitudes: A Sociological
Perspective,” Mimeo, Institute for Sociology, Hungarian Academy of Sciences,
2018.
Johnston, Christopher D., Howard G. Lavine, and Christopher M. Federico, Open
versus Closed: Personality, Identity, and the Politics of Redistribution (Cam-
bridge: Cambridge University Press, 2017).
Kahan, Dan M., “Climate-Science Communication and the Measurement Prob-
lem,” Advances in Political Psychology, 36 (2015), 1–43.
Key, V. O., “A Theory of Critical Elections,” Journal of Politics, 17 (1955), 3–18.
Klein, Ezra, Why We’re Polarized (New York: Simon and Schuster, 2020).
Kuziemko, Ilyana, Michael Norton, Emmanuel Saez, and Stefanie Stantcheva,
“How Elastic Are Preferences for Redistribution? Evidence from Randomized
Survey Experiments,” American Economic Review, 105 (2015), 1478–1508.
Mackie, Diane M., “Social Identification Effects in Group Polarization,” Journal of
Personality and Social Psychology, 84 (1986), 932–943.
Mayhew, David, Electoral Realignments: A Critique of an American Genre (New
Haven, CT: Yale University Press, 2004).
McGarty, Craig, John C. Turner, Micheal A. Hogg, Barbara David, and Margaret
S. Wetherell, “Group Polarization as Conformity to the Prototypical Group
Member,” British Journal of Social Pyschology, 31 (1992), 1–19.
Murphy, Kevin M., and Andrei Shleifer, “Persuasion in Politics,” American Eco-
nomic Review Papers and Proceedings, 94 (2004), 435–439.
Norris, Pippa, and Ronald Inglehart, Cultural Backlash: Trump, Brexit, and Au-
thoritarian Populism (Cambridge: Cambridge University Press, 2019).
Oakes, Penelope J., “The Salience of Social Categories,” in Rediscovering the Social
Group: A Self-Categorization Theory, John C. Turner, Michael A. Hogg, Pene-
lope J. Oakes, Stephen J. Reicher and Margaret J. Wetherell, eds. (Oxford:
Blackwell, 1987), 117–141.
Persson, Torsten, and Guido Tabellini, Political Economics: Explaining Economic
Policy (Cambridge, MA: MIT Press, 2000).
Shayo, Moses, “A Model of Social Identity with an Application to Political Economy:
Nation, Class, and Redistribution,” American Political Science Review, 103
(2009), 147–174.
Schickler, Eric, Racial Realignment: The Transformation of American Liberalism,
1932–1965 (Princeton, NJ: Princeton University Press, 2016).
 IDENTITY, BELIEFS, AND POLITICAL CONFLICT 2411

Schivardi, Fabiano, Enrico, Sette, and Guido, Tabellini, “Identifying the Real
Effects of Zombie Lending,” Review of Corporate Finance Studies, 9 (2020),
421–429.
Sherif, Muzafer, The Psychology of Social Norms (New York: Harper, 1936).
Stantcheva, Stefanie, “Understanding Tax Policy: How Do People Reason?,” Quar-
terly Journal of Economics, 136 (2021), 2309–2369.

Downloaded from https://academic.oup.com/qje/article/136/4/2371/6368349 by guest on 21 September 2023

Sundquist, James L., Dynamics of the Party System: Alignment and Realignment of
Political Parties in the United States (Washington, DC: Brookings Institution,
1983)
Tabellini, Guido, “The Scope of Cooperation: Values and Incentives,” Quarterly
Journal of Economics, 123 (2008), 905–950.
Tajfel, Henri, and John C. Turner, “An Integrative Theory of Intergroup Con-
flict,” in The Social Psychology of Intergroup Relations, William G. Austin and
Stephen Worcel, eds. (Monterey, CA: Brooks/Cole, 1979), 33–47.
Tajfel, Henri, and A. L. Wilkes, “Classification and Quantitative Judgment,”
British Journal of Psychology, 54 (1963), 101–114.
Tajfel, Henri, M. Billig, R. Bundy, and Claude Flament, “Social Categorization
and Intergroup Behavior,” European Journal of Social Psychology, 1 (1971),
149–178.
Turner, John C., Michael A. Hogg, Penelope Oakes, Steven D. Reicher, and Mar-
gareth S. Wetherell, Rediscovering the Social Group: A Self Categorization
Theory (New York: Blackwell, 1987).
Westfall, Jacob, Leaf Van Boven, John R. Chambers, and Charles M. Judd, “Per-
ceiving Political Polarization in the United States: Party Identity Strength
and Attitude Extremity Exacerbate the Perceived Partisan Divide,” Perspec-
tives on Psychological Science, 10 (2015), 145–158.
Wu, Alex, “The Changing Nature of Political Polarization in the Unites States,”
Mimeo, Harvard, 2020.
Zaller, John R., The Nature and Origins of Mass Opinion (Cambridge: Cambridge
University Press, 1992).