Fear and Loathing across Party Lines: New Evidence

on Group Polarization
Shanto Iyengar Stanford University
Sean J. Westwood Princeton University

When defined in terms of social identity and affect toward copartisans and opposing partisans, the polarization of the
American electorate has dramatically increased. We document the scope and consequences of affective polarization of
partisans using implicit, explicit, and behavioral indicators. Our evidence demonstrates that hostile feelings for the opposing
party are ingrained or automatic in voters’ minds, and that affective polarization based on party is just as strong as
polarization based on race. We further show that party cues exert powerful effects on nonpolitical judgments and behaviors.
Partisans discriminate against opposing partisans, doing so to a degree that exceeds discrimination based on race. We note
that the willingness of partisans to display open animus for opposing partisans can be attributed to the absence of norms
governing the expression of negative sentiment and that increased partisan affect provides an incentive for elites to engage
in confrontation rather than cooperation.

M
ore than 50 years after the publication of The social norms (Himmelfarb and Lickteig 1982; Maccoby
American Voter (Campbell et al. 1960), de- and Maccoby 1954; Sigall and Page 1971), there are no
bates over the nature of partisanship and the corresponding pressures to temper disapproval of po-
extent of party polarization continue (see Fiorina and litical opponents. If anything, the rhetoric and actions
Abrams 2008; Hetherington 2009). While early studies of political leaders demonstrate that hostility directed at
viewed partisanship as a manifestation of other group the opposition is acceptable, even appropriate. Partisans
affiliations (Berelson, Lazarsfeld, and McPhee 1954; therefore feel free to express animus and engage in dis-
Campbell et al. 1960), more recent work suggests that criminatory behavior toward opposing partisans.
party is an important form of social identity in its own Scholars have typically treated the sense of partisan
right (Green, Palmquist, and Schickler 2004; Greene 1999; identity as a major cue for political choices, most no-
Huddy, Mason, and Aarøe 2010; Iyengar, Sood, and Lelkes tably, voting behavior. We demonstrate that partisan cues
2012). As anticipated by social identity theorists (e.g., now also influence decisions outside of politics and that
Tajfel 1970; Tajfel and Turner 1979), under conditions partisanship is a political and social divide. Using novel
of group competition, the sense of group membership measurement techniques, we directly compare implicit,
inculcates positive evaluations of the ingroup and corre- explicit, and behavioral measures of partisan affect with
spondingly hostile evaluations of outgroups. In the case affect based on racial identity. We find that implicit affect
of partisanship, this divergence in affect toward the in and behavioral discrimination based on partisanship are
and out parties—affective polarization—has increased just as significant as affect and discrimination based on
substantially over the past four decades (Haidt and race.
Hetherington 2012; Iyengar, Sood, and Lelkes 2012). Our argument proceeds in two parts. We first scale
Unlike race, gender, and other social divides where the magnitude of the ingroup versus outgroup partisan
group-related attitudes and behaviors are constrained by divide against the comparable divide for race. To ensure a

Shanto Iyengar is Chandler Chair of Communication and Professor of Political Science, Stanford University, Building 120, Room 110, 450
Serra Mall, Stanford, CA 94305-2050 ([email protected]) Sean J. Westwood is Postdoctoral Researcher, Princeton University, 308
Robertson Hall, Princeton, NJ 08544 ([email protected]).
We are grateful for helpful comments and suggestions from the editors of AJPS, our anonymous reviewers, Alan Abramowitz, Lauren
Davenport, Jamie Druckman, Morris Fiorina, Justin Grimmer, Yphtach Lelkes, Jane Mansbridge, Solomon Messing, Erik Peterson, Paul
Sniderman, Sidney Verba, Rebecca Weiss, and Frances Zlotnick. Replication data and scripts are available in the AJPS Data Archive on
Dataverse (http://dvn.iq.harvard.edu/dvn/dv/ajps) and from the authors.

American Journal of Political Science, Vol. 59, No. 3, July 2015, Pp. 690–707

C 2014, Midwest Political Science Association DOI: 10.1111/ajps.12152

690
FEAR AND LOATHING ACROSS PARTY LINES 691

fair comparison, we use implicit measures. The use of im- In the contemporary American political environ-
plicit measures allows us to assess racial attitudes that are ment, there is evidence of increasing hostility across party
uncontaminated by social desirability biases (Asendorpf, lines, which has been attributed to a variety of factors, in-
Banse, and Mücke 2002; Boysen, Vogel, and Madon cluding candidates’ reliance on negative campaigning and
2006) and to capture unconscious attitudes (Greenwald, the availability of news sources with a clear partisan pref-
McGhee, and Schwartz 1998) that are difficult to manip- erence (see Iyengar, Sood, and Lelkes 2012; Lelkes and
ulate. By documenting significant implicit partisan affec- Iyengar 2012). Among Americans who say they identify
tive polarization, we demonstrate that hostile feelings for with a political party, negative views of the out party and
the opposing party are ingrained and automatic in vot- its supporters have risen sharply since the 1980s (Haidt
ers’ minds. The general agreement that race represents the and Hetherington 2012; Iyengar, Sood, and Lelkes 2012).
deepest divide in American society (Myrdal 1944; Pager Not only are group evaluations polarized along party
and Shepherd 2008; Schuman et al. 1997) makes racial af- lines, but the specific content of outgroup stereotypes has
fect a particularly robust benchmark for the assessment of also followed suit. While Republicans view fellow parti-
partisan affect. We show that the level of partisan animus sans as patriotic, well informed, and altruistic, Democrats
in the American public exceeds racial hostility. are judged to exhibit precisely the opposite traits (see
Second, through three experiments, we show that Iyengar, Sood, and Lelkes 2012). Citizens are also more
partisanship and partisan affect are strong cues for non- dubious of the motivations and ulterior motives of politi-
political judgments and behaviors. We continue to bench- cians from opposing parties than copartisans (Munro,
mark the effects of partisan bias against racial bias. In the Weih, and Tsai 2010).
absence of a social norm or sanction that discourages
partisan discrimination, we show in three independent
studies that partisans frequently choose to discriminate
against opposing partisans. We further document that Intrusion of Partisan Cues into
discrimination against the outgroup is based more on Everyday Life
outgroup animus than ingroup favoritism.
Unlike race and gender, which are physical traits, parti-
sanship is a less obvious affiliation. Nonetheless, there are
ample opportunities to discern partisan cues. Political dis-
cussion at the workplace provides one such opportunity
Affective Polarization by Party: The (Mutz and Mondak 2006). Exposure to social networks
State of the Evidence is another; on Facebook, 35 million Americans like Pres-
ident Barack Obama, 11 million like Mitt Romney, and
We define affective polarization as the tendency of people more than 25 million like various national politicians and
identifying as Republicans or Democrats to view opposing news correspondents for Fox News and MSNBC (Face-
partisans negatively and copartisans positively (Campbell book 2013). Partisan affiliation is also displayed openly
et al. 1960; Green, Palmquist, and Schickler 2004). This af- on cars and lawns. Two months before the end of the
fective separation is a result of classifying opposing parti- 2012 election, Obama for America sold—only counting
sans as members of an outgroup and copartisans as mem- official campaign sales—over $43 million in campaign
bers of an ingroup. The standard definition of an outgroup merchandise such as T-shirts and bumper stickers (Korte
is a group to which a person does not belong, whereas an 2012). Thus, for a large portion of the electorate, informa-
ingroup is a group to which a person does belong. Re- tion on individuals’ political affiliations is conveniently
search in psychology generally shows that members of accessible.
an ingroup frequently ascribe undesirable or inappropri- Even more striking than the availability of partisan
ate traits to members of outgroups (Tajfel 1970; Tajfel and cues is the gradual encroachment of party preference into
Turner 1979), though some high-status outgroups such as nonpolitical and hitherto personal domains. Residential
whites are viewed positively by outgroup members (Jost neighborhoods are increasingly politically homogeneous
and Banaji 1994). Classic studies by psychologists have (Bishop 2008), and geographic distance creates social
demonstrated that the mere act of identifying with a par- distance (Bogardus 1925). A standard measure of social
ticular group in competitive environments—no matter distance—parents’ displeasure over the prospects of their
how trivial the basis for group assignment—is often suf- offspring marrying into a family with a different party
ficient to trigger negative evaluations of outgroups (see affiliation—shows startling increases in the United States,
Billig and Tajfel 1973). but not in Britain (Iyengar, Sood, and Lelkes 2012).
692 SHANTO IYENGAR AND SEAN J. WESTWOOD

The stated preference for same-party marriage is but the Study 1: Anchoring Partisan Affect
tip of an evidentiary iceberg concerning the growing rel-
evance of partisan cues for interpersonal relations. Actual The growth of explicit partisan affective polarization is
marriage across party lines is rare; in a 2009 survey of mar- clear from analysis of survey data (Iyengar, Sood, and
ried couples, only 9% consisted of Democrat-Republican Lelkes 2012), but these data only show that people report
pairs (Rosenfeld, Reuben, and Falcon 2011; also see affective biases when evaluating partisan groups. Survey
Stoker and Jennings 1995). Moreover, marital selection data do not show the extent to which affective partisan
based on partisanship exceeds selection based on physical attitudes are ingrained in the minds of the public. Survey
(e.g., body shape) or personality attributes (Alford et al. data are also subject to cognitive manipulation, which
2011). Recent data from online dating sites are especially allows participants to filter or underreport actual partisan
revealing. Even though single men and women seeking affect. To address these limitations, we designed Study 1
companionship online behave strategically and exclude to measure implicit partisan affect.
political interests from their personal profiles (Klofstad, Implicit attitudes are the “traces of past experience
McDermott, and Hatemi 2012), partisan agreement that mediate favorable or unfavorable feeling, thought,
nevertheless predicts reciprocal communication between or action toward social objects” (Greenwald and Banaji
men and women seeking potential dates (Huber and 1995, 8). The detection of implicit partisan affect thus
Malhotra 2012). As the authors of one intermarriage shows that the sense of partisan identity is deeply em-
study put it, “the timeless character of political divisive- bedded in citizens’ minds. We also include an implicit
ness may emanate not just from the machinations of elites, assessment of racial affect to anchor the measure of im-
but also from the nuances of courtship” (Alford et al. plicit partisan affect. The general argument is that implicit
2011, 378). measures—not subject to cognitive processing—are more
All told, despite only mixed evidence of sharp ide- accurate since they do not permit active masking of feel-
ological or partisan divergence in their policy prefer- ings toward outgroups. Unobtrusive measures such as
ences, Americans increasingly dislike people and groups the Implicit Association Test (IAT) developed by Green-
on the other side of the political divide and face no social wald, McGhee, and Schwartz (1998) and the brief ver-
repercussions for the open expression of these attitudes. sion (BIAT) developed by Sriram and Greenwald (2009)
Heightened affective polarization has widened the reach are much harder to manipulate than explicit self-reports,
of partisan cues; party affiliation increasingly constrains producing more valid and less biased results (Asendorpf,
social and personal ties. Banse, and Mücke 2002; Boysen, Vogel, and Madon 2006).
The full IAT measures the reaction time necessary
to associate ingroups and outgroups (e.g., “Democrat”
Hypotheses and “Republican” or “African American” and “European
American”) with positive and negative attributes (e.g.,
“good” and “bad”). While completing the task, partici-
Our underlying research questions are (1) how does par-
pants are instructed to go as quickly as possible. Since
tisan affect compare with affect based on other social
people are able to respond faster to group-attribute pairs
divides, and (2) to what extent are partisans willing to
for which they have acquired automatic associations, the
discriminate against opposing partisans in nonpolitical
metric of the IAT compares the time taken to respond to
decisions? For the first research question, we hypothesize
pairings of ingroup + good with outgroup + good as well
(1) that partisan affect is sufficiently ingrained in citizen
as ingroup + bad and outgroup + bad. The differences
consciousness to manifest itself in implicit indicators of
in response times to the group pairings are used to gen-
partisan attitudes and (2) that the effect size of parti-
erate an indirect measure of group preference. Since the
san affect is larger than the effect size of affect for other
full version of the IAT requires more than 15 minutes to
social divides where social norms discourage discrimina-
administer, psychologists have developed (and validated)
tion. For the second research question, our hypotheses
a brief version (BIAT), which measures the same associ-
are (1) that partisanship and partisan affect motivate re-
ations, but with a reduced number of trials. We measure
spondents to make determinations and judgments that
implicit racial affect using the standard European Amer-
are biased in favor of copartisans and (2) that copartisan
ican/African American BIAT and implicit partisan affect
favoritism is a weaker influence than animosity toward
using a partisan BIAT that we created.
opposing partisans.
FEAR AND LOATHING ACROSS PARTY LINES 693

FIGURE 1 BIAT Design Worst, Horrible), and eight images (shown in the bottom
panel of Figure 1) referring to the Democratic and Re-
publican parties.1 We created a custom software tool to
implement theBIAT in a web browser.

Design
We recruited a sample of 2,000 adults from the Survey
Sampling International (SSI) panel.2 Respondents com-
pleted both the African American/European American
BIAT and our partisan BIAT at the end of a survey instru-
ment. To minimize possible order effects and to account
for reductions in implicit attitude extremity among those
who have completed one or more IATs (see Nosek, Ba-
naji, and Greenwald 1980), the order of the two BIATs was
randomized. We oversampled African Americans (500) in
order to capture racial affect among nonwhites.
Following Greenwald, Nosek, and Banaji (2003), we
utilize the “D-score” to interpret the BIAT results. The
score, which can range from −2 to 2, is calculated by
subtracting the mean response times for the round pair-
ing targets from Category A (Democrat/Black) with pos-
itive terms from the mean response times for the round
pairing targets from Category B (Republican/White) with
positive terms (for full details on the computation of
the D-score, see Greenwald, Nosek, and Banaji 2003).
This difference in response latency is then divided by
the pooled standard deviation over both rounds.3 Posi-
tive D-scores indicate that participants respond faster to
Republican-good than to Democrat-good pairings. Since
Note: The state of California and the “(D)” are colored blue, and people respond faster to group attribute pairs for which
the state of Texas and the “(R)” are colored red. they have acquired automatic associations, this pattern
would indicate greater positive affect for Republicans,
whereas the inverse response times would reflect greater
In a BIAT, participants complete four rounds of 20 positive affect for Democrats. The party stimuli used to
timed categorizations, with the first pair of rounds treated construct the D-score produced highly correlated laten-
as training and the last pair used for scoring the measure cies (r = .35) and a robust standardized Cronbach’s alpha
of implicit attitudes. The four blocks consist of two rep- of .89.4
etitions (randomly ordered) of the “ingroup + good”
block and the “outgroup + good” block. In each block,
the group not paired with “good” is grouped with nega- Results
tive words. The top panel of Figure 1 shows an example
The sign and magnitude of the obtained partisan D-
of a categorization round in the partisan BIAT. In this
score converged with traditional measures of partisan
example, the target stimulus is the Democratic mascot,
and the round pairs Democrats with “good.” Democratic 1
The study was fielded in July 2012.
respondents should more quickly categorize the mascot 2
After removing subjects with BIAT error rates above .35—the
as “good” since they have come to associate “good” with recommended filtering rate—and those who failed to complete the
Democrats. Conversely, Republican respondents should BIATs, we were left with a sample of 1,971.
take more time to associate the Democratic mascot with 3
The D-score is recommended for a variety of reasons, including
“good.” We constructed the partisan BIAT using the stan- only weak effects of cognitive ability on IAT results (Cai et al. 2004).
dard set of good stimuli (Wonderful, Best, Superb, Ex- 4
Cronbach’s alpha for the average latencies (pooled) within the two
cellent), the standard set of bad stimuli (Terrible, Awful, sets of party stimuli was .81.
694 SHANTO IYENGAR AND SEAN J. WESTWOOD

FIGURE 2 Implicit Partisan Affect among Partisan and Ideological

Group

Implicit Partisan Affect by Partisan Strength

Strong Republican ●
Weak Republican ●
Independent ●
Weak Democrat ●
Strong Democrat ●

−0.50 −0.25 0.00 0.25 0.50

Partisan D-Score

Implicit Partisan Affect by Ideology

Very Conservative ●
Conservative ●
Moderate ●
Liberal ●
Very Liberal ●

−0.50 −0.25 0.00 0.25 0.50

Partisan D-Score

Note: The dots are the means for each group, and the bars are 95% confidence intervals for the
mean.

and ideological affiliation. Figure 2 shows the average As a further validation test, and to compare the extent
partisan D-scores grouped by responses to self-reported of polarization across implicit and explicit measures, we
strength of partisan and ideological identity. The partisan examined the relationship between the partisan D-score
D-score corresponded closely with the conventional sur- and a conventional survey-based measure of affective po-
vey measure of party identification. “Strong Republicans” larization: the difference in feeling thermometer ratings
revealed the most bias in favor of Republicans (mean = of Democrats and Republicans (on a 0–100 scale). The
.35, s.e. = .03, n = 218), whereas “weak Democrats” were D-score correlated strongly with the net thermometer
the most biased in favor of Democrats (mean = –.26, rating (r = .418). Although our measure of implicit af-
s.e. = .02, n = 374). There were traces of intransitivity in fect closely parallels the explicit measure, there is also
the explicit indicator, as strong Democrats were slightly divergence, with the D-score explaining only 17.5% of
less biased against Republicans than weak Democrats the variance in the thermometer scores. The variation is
(mean = –.21, s.e. = .02, n = 466). substantively interesting insofar as it reveals differences
Turning to the measure of ideology, “very conserva- in the extent of affective polarization across the implicit
tive” individuals had the strongest implicit preference for and explicit indicators.5
Republicans (mean = .227, s.e. = .041, n = 171), fol- When respondents have less ability to control their
lowed by “conservatives,” who obtained the next highest affect, do they exhibit more or less polarized attitudes?
preference (mean = .170, s.e. = .026, n = 360). Once Figure 3 compares the distribution of both measures
again, there were minor inconsistencies in the pattern. for respondents who identified as Democrats and Re-
Thus, “liberals” were the most biased toward Democrats publicans. Clearly, the overlap between party responses
(mean = –.282, s.e. = .025, n = 338), followed closely by
respondents in the “very liberal” category (mean = –.267, 5
We document the predictive and convergent validity of the partisan
s.e. = .041, n = 149). D-score in the supporting information.
FEAR AND LOATHING ACROSS PARTY LINES 695

FIGURE 3 Comparing the Distribution of Implicit and Explicit

Partisan Affect

0.0100

0.0075
Density

Democrat
0.0050
Republican

0.0025

0.0000

−100 −50 0 50 100

Net Feeling Thermometer

0.8

0.6
Density

Democrat
0.4 Republican

0.2

0.0

−2 −1 0 1 2
Partisan D-score

Note: This figure shows the distributions of partisan affective polarization scores (explicit
in the top and implicit in the bottom) for Democrats (to the left) and Republicans (to the
right).

is small in both cases, suggesting strong polarization. Comparing Partisan and Racial Affect
To compare the magnitude of the differences, we use
Cohen’s d, a measure of effect size (Cohen 1988). Cohen’s How does implicit bias against racial outgroups com-
d is .95 for the implicit measure, suggesting considerable pare with implicit partisan bias? Since the partisan and
division between Democrats and Republicans, but it is race BIATs have the identical metric, we have comparable
even higher (1.72) in the case of the net thermometer estimates of the magnitude of partisan and race-based
score, suggesting that cognitive processing exacerbates polarization and can anchor partisan polarization with
rather than attenuates the level of affective polarization. racial polarization. Moreover, by using implicit measures,
696 SHANTO IYENGAR AND SEAN J. WESTWOOD

FIGURE 4 D-Scores for the Partisan and As shown in Figure 4 (graphs of D-scores for
African American/European Democrats and Republicans and blacks and whites, re-
American BIATs spectively), the separation of the D-Scores appears larger
between Republicans and Democrats than between whites
Implicit Partisan Affect by Party and African Americans. Party polarization exceeds polar-
ization based on race.
Partisan Affiliation

Independent ●
From our perspective, the difference in the magni-
Republican ●
tude of the partisan and racial divides in implicit affect is
especially telling. Racial identity is acquired at birth, and
Democrat ● racial attitudes are deeply ingrained (see Baron and Banaji
2006). For partisanship to approach (and surpass) race,
−0.50 −0.25 0.00 0.25 0.50 the underlying animosity must be more substantial than
Partisn D-Score previously thought. The data show that negative associ-
ations of opposing partisans are faster (more automatic)
Implicit Racial Affect by Race
than negative associations of African Americans.
Given the contrasting positions of the parties on is-
African American ●
sues relating to race and the clear racial divide in party
Race

affiliation, it is possible that the level of out-party ani-

European American ● mosity reflects a concatenation of racial and partisan or
ideological affect. At the level of explicit attitudes, there is
−0.50 −0.25 0.00 0.25 0.50
significant overlap between measures of antiblack senti-
Race D-Score ment and ideological sentiment (e.g., Carmines, Snider-
man, and Easter 2011). At the level of implicit attitudes,
Note: This figure shows the distributions of implicit partisan affect however, our evidence suggests that the overlap between
(top) and implicit racial affect (bottom) with 95% confidence partisan and racial affect is minimal. The correlation be-
intervals.
tween the partisan D-score and race D-score was .13
(p < .001), suggesting that implicit affect based on racial
identity does not simply mirror partisan leanings.
we circumvent the underreporting of racial bias caused
by normative pressures facing respondents asked explicit
questions about race relations (see, e.g., Crosby, Bromley, Partisan Affect among Independents and
and Saxe 1980; McConahay, Hardee, and Batts 1981). Leaners
Figure 4 shows the results of the partisan and African
American/European American BIATs. Negative D-scores Not surprisingly, self-identified partisans have the highest
indicate an affective preference for Democrats (and levels of polarization, but pure Independents and Inde-
African Americans), whereas positive scores indicate pendent leaners also show significant levels of partisan
an affective preference for Republicans and European affect. As shown in Figure 5, among Independents in
Americans, respectively. the SSI sample, the difference in Republican and Demo-
The spread between Democrats and Republicans on cratic feeling thermometer scores is slightly slanted to-
the partisan D-score was massive, t(824.66) = 17.68, p < ward Democrats (mean = 3.52, s.e. = 0.84, n = 697),
.001, with the Republicans averaging .27 (s.e. = .02), the t(696) = 4.19, p < .001. Among Democratic leaners,
Democrats –.23 (s.e. = .02), and Independents –.02 (s.e. = the net thermometer ratings reveal a stronger affective
.02). In the case of implicit racial bias, African Americans preference for Democrats (mean = –30.64, s.e. = 1.60,
showed a preference for African Americans (D-score = n = 297) that is smaller than the preference among
–.09, s.e. = .02), whereas whites displayed a somewhat Democrats (mean = –48.84, s.e. = 0.93, n = 1,340).
stronger ingroup preference (D-score = .16, s.e. = .01). Thus, Democratic leaners are significantly less affectively
Hispanics and Asians both revealed a slight preference polarized than self-identified Democrats, t(466.93)=
for whites over blacks. Consistent with previous research, –15.02, p < .001), but significantly more polarized
the black-white difference in implicit bias was substan- than Independents, t(516.52) = 9.98, p < .001. This
tial, t(740.10) = 11.04, p < .001, but the effect size for pattern is nearly identical on the Republican side,
race (Cohen’s d = .61) was not nearly as strong as the where leaners (mean = 28.18, s.e. = 1.85, n =
corresponding effect of party (Cohen’s d = .95). 222) are significantly less polarized than partisans
FEAR AND LOATHING ACROSS PARTY LINES 697

FIGURE 5 Explicit Partisan Affect Including Leaners and Independents

●
Partisan Affiliation
Democrat

Lean Democrat ●

Independent ●

Lean Republican ●

Republican ●

−100 −50 0 50 100

Net Feeling Thermometer

Note: The dots are the means for each group, and the bars are 95% confidence intervals for the mean.

(mean = 42.58, s.e. = 1.32, n = 624), t(461.55) = we randomly varied the academic achievements of each
6.33, p < .001), and significantly more polarized candidate (by assigning them either a 4.0 GPA or a 3.5
than Independents with no leanings, t(317.25) = GPA). We therefore have a four-cell design (candidate 1
–12.13, p < .001. Consistent with prior research showing more qualified; candidate 2 more qualified; both candi-
that leaners adopt a social identity of their preferred party dates equally qualified with a 4.0 GPA; both candidates
(Greene 1999), these results offer additional support for equally qualified with a 3.5 GPA). This design allows us
the contention that leaners behave like partisans. to measure the effects of partisan and racial bias when
the candidates are equally qualified and when one can-
didate is more qualified than the other. Participants were
Study 2: The Intrusion of Partisan randomly assigned to complete only one of the two se-
Affect into Nonpolitical Domains lection tasks. The order of the two candidates as well as
the order of their extracurricular activities was random-
ized.7 An example of the resumes presented is shown in
Our comparisons of implicit partisan and racial bias sug-
Table 1.
gest that partisan identity is a relatively strong source of
Depending on the task to which they were assigned,
group affect, but the behavioral implications remain un-
participants were exposed to candidates with either a par-
clear. We designed this follow-up study to compare the
tisan affiliation (cued through membership in a partisan
relative influence of partisan and racial affiliation in a
extracurricular group) or a racial identity (cued through
nonpolitical decision task and to examine whether affect
a stereotypical African American/European American
predicted decisions favoring the respondent’s own group
name and membership in an extracurricular group).
affiliation.
Table 2 shows the full set of names and extracurricular
activities for each task. Gender was not varied but fixed
Design as male; in the partisan task, race was fixed as European
American for both candidates; in the race task, partisan-
We randomly assigned 1,021 participants drawn from ship was not offered as an attribute of the student.
the SSI panel to one of two selection tasks. The first Given this design, comparisons across conditions
selection task required a selection between a Democrat shed light on the relative strength of ingroup preference
and a Republican, and the second required selection be- across partisan and racial lines. More specifically, we
tween a European American and an African American. In can compare the frequency with which respondents
each task, we asked participants to read the resumes for a demonstrate a discriminatory preference for the ingroup
pair of graduating high school seniors. We extend exist- candidate while varying an objective measure of merit
ing scholarship assessment designs (e.g., DeSante 2013; (GPA).
Munro, Weih, and Tsai 2010) to measure partisan and
racial bias.6 To increase the robustness of our design,

6 7
As a cover story, participants were instructed that an anonymous We use a paired instead of factorial design since crossing the vari-
donor had contributed $30,000 to a scholarship fund. The instruc- ous group attributes would have necessitated a sample beyond our
tions also stated that the selection committee had deadlocked over budget constraint and because many of the cells in a fully crossed
two finalists and commissioned a survey to decide the winner. design correspond to only small segments of the U.S. population.
698 SHANTO IYENGAR AND SEAN J. WESTWOOD

TABLE 1 Sample Instructions for Study 2

Applicant Highlights
Arthur Wolfe Jeremy O’Neill
Academic achievements Academic achievements
4.0 GPA 4.0 GPA
Community involvement Community involvement
Volunteer park ranger Volunteer middle school math tutor
Habitat for Humanity volunteer Red Cross volunteer
Extracurricular activities Extracurricular activities
Bowling team President of the Young Democrats
President of the Young Republicans Member of the marching band
Honor Society Art Club

TABLE 2 Full Set of Conditions and Treatments for Study 2

Task Identity Name Extracurricular Activity GPA
Partisan Republican Arthur Wolfe President of the Young Republicans 3.5 or 4.0
Democrat Jeremy O’Neill President of the Young Democrats 3.5 or 4.0
Racial European American Arthur Wolfe President of the Future Investment Banker Club 3.5 or 4.0
African American Jamal Washington President of the African American Student Association 3.5 or 4.0

TABLE 3 Favoritism in Candidate Selection by Group Membership

Partisan Selection Task

Participant’s Partisanship Democrat Winner (N) Republican Winner (N)

Democrat 79.2% (202) 20.8% (53)
Lean Democrat 80.4% (45) 19.6% (11)
Independent 57.9% (81) 42.1% (59)
Lean Republican 30.8% (12) 69.2% (27)
Republican 20.0% (24) 80.0% (96)
Racial Selection Task

Participant’s Race European American Winner (N) African American Winner (N)
European American 44.2% (129) 55.8% (163)
African American 26.9% (32) 73.1% (87)

Results party candidate. Democratic leaners showed a stronger

preference for the Democratic candidate than Republi-
Despite the explicitly apolitical nature of the task, it was can leaners showed for the Republican candidate, though
the party cue that exerted the strongest impact on se- both groups displayed the in-party preference (80.4% and
lection for the largest number of participants. Table 3 69.2%, respectively). Independents showed a slight pref-
shows the effects of participant partisanship and partici- erence for the Democratic candidate (57.9%).
pant race on winner selection, aggregating across the four Ingroup selection on the basis of race was con-
qualification manipulations. fined to African Americans (73.1% selecting the African
In the partisan task, approximately 80% of parti- American), with European Americans showing a small
sans (both Democrats and Republicans) selected their in- preference for the African American candidate (55.8% se-
lecting the African American). To clarify these results and
FEAR AND LOATHING ACROSS PARTY LINES 699

FIGURE 6 Predicted Probabilities for Partisan Winner Selection

Participant's Party Affiliation

Equally Qualified Republican More Qualified Democrat More Qualified

Republican ● ● ●

Lean Republican ● ● ●

Independent ● ● ●

Lean Democrat ● ● ●

Democrat ● ● ●

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
Predicted Probability of Selecting the Republican

Note: The horizontal values group the data by the partisanship of the participant, and the vertical facets group the data by the
qualifications of the scholarship candidates. The dots are the predicted probabilities of selection in each group, and the bars are 95%
confidence intervals.

account for the manipulation of academic qualifications, probability of a Republican selecting the Democrat was
we generated predicted probabilities (Figure 6) from lo- only .15 (95% confidence interval [.16, .35]); when the
gistic regression models.8 For the partisan task, the depen- two candidates were equally qualified, the probability of
dent variable was a dummy coded as 1 when the Republi- a Republican selecting the Democratic candidate was .21
can was selected; for the race task, the dependent variable (95% confidence interval [.13, .33]); and when the Repub-
was set as 1 when the African American was selected. lican was most qualified, the probability of Republicans
The independent variables were participant partisan af- selecting the Democratic candidate was .21 (95% confi-
filiation (including Independents and leaners) interacted dence interval [.10, .38]). The probability of a partisan
with academic qualification for the partisan task, and selecting an out-party candidate never rose above .3, and
participant race interacted with qualification for the race the coefficients for the various interaction terms between
task.9 participant partisan affiliation and candidate qualifica-
In the partisan task, candidate qualification had no tions were never significant; partisanship simply trumped
significant effect on winner selection. Compared to Inde- academic excellence in this task.
pendents, Democrats were more likely to select a fellow Consistent with the results from the party feeling
Democrat (b = –1.04, s.e. = .33, p < .01) and Republi- thermometers, leaners behaved like partisans and selected
cans were more likely to select a fellow Republican (b = the candidate from the party they prefer, but the effects
1.60, s.e. = .39, p < .001). Even when the candidate from were less robust than those of partisans due to smaller
the opposing party was more qualified, partisans opted numbers of leaners in the sample (top of Figure 8). Despite
to award the scholarship money to the copartisan candi- the larger variance in the behavior of leaners, compared
date. When the Republican was more qualified than the to Independents their tendency to select a winner from
Democrat, the probability of a Democrat selecting the Re- their preferred party was consistently significant for both
publican candidate was only .30 (95% confidence interval Democratic leaners (b = –1.35, s.e. = .33, p < .05) and
[.20, .44]); when both candidates were equally qualified, Republican leaners (b = .99, s.e. = .49, p < .05). Leaners
the probability of a Democrat selecting the Republican were also unresponsive to candidate qualifications, with
candidate fell to .21 (95% confidence interval [.14, .29]); all estimates of the interaction between leaner political
and when the Democrat was most qualified, the proba- identity and candidate qualification proving insignificant.
bility of a Democrat selecting the Republican candidate This result offers additional evidence that lea ners are in
was a meager .14 (95% confidence interval [.08, .24]). fact closet partisans.
Similarly, when the Democrat was more qualified, the Independents were nearly evenly split in all condi-
tions, even when one candidate was more qualified than
8
Full model results are included in the supporting information. another. When the candidates were equally qualified, the
9
In both tasks, there was no significant or substantive difference probability of an Independent selecting a Republican win-
in selection rates between the two conditions where candidates ner was .42 (95% confidence interval [.32, .54]); when
were equally qualified at different GPAs (two 3.5 GPAs and two the Republican was more qualified, the probability of
4.0 GPAs), so both conditions were combined to produce a singe
“equally qualified” condition. an Independent selecting the Republican candidate was
700 SHANTO IYENGAR AND SEAN J. WESTWOOD

FIGURE 7 Predicted Probabilities for Racial Winner Selection

Participant's Race

Equally Qualified African American More Qualified European American More Qualified
White ● ● ●
African American ● ● ●

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
Predicted Probability of Selecting the African American

Note: The horizontal values group the data by the race of the participant, and the vertical facets group the data by the qualifications
of the scholarship candidates. The dots are the predicted probabilities of selection in each group, and the bars are 95% confidence
intervals.

FIGURE 8 Effects of Race and Partisanship on Allocations to Player 2

Dictator Game
Different Race Same Race
Player 2

Opposing Partisan ● ●
Copartisan ● ●
$2.00 $2.50 $3.00 $3.50 $4.00 $4.50 $5.00 $2.00 $2.50 $3.00 $3.50 $4.00 $4.50 $5.00
Amount Allocated to Player 2

Trust Game
Different Race Same Race
Player 2

Opposing Partisan ● ●
Copartisan ● ●
$2.00 $2.50 $3.00 $3.50 $4.00 $4.50 $5.00 $2.00 $2.50 $3.00 $3.50 $4.00 $4.50 $5.00
Amount Allocated to Player 2

Note: The dots are the means for each group, and the bars are 95% confidence intervals.

.50 (95% confidence interval [.34, .66]); and when the tendency to select the European American as the winner
Democrat was more qualified, the probability of an Inde- when he was the more qualified candidate (b = –.93, s.e. =
pendent selecting the Republican candidate was .33 (95% .30, p < .01). There were no significant interactions
confidence interval [.19, .51]). Their unresponsiveness between participant race and candidate qualifications.
to academic qualification and the general preference for To illustrate these effects, we again present predicted
a Democratic winner corresponds with the small Demo- probabilities. Unlike partisanship where ingroup prefer-
cratic skew in Independents’ implicit and explicit partisan ences dominate selection, only African Americans showed
affect and suggests that Independents are also more re- a consistent preference for the ingroup candidate. Asked
sponsive to partisan cues than academic qualifications. to choose between two equally qualified candidates, the
Partisanship is important even for individuals who, in probability of an African American selecting an ingroup
terms of self-identified partisan affiliation, present them- winner was .78 (95% confidence interval [.66, .87]), which
selves as nonpartisan. Placed in the context of affective was no different than their support for the more qualified
preferences and behavior, the label “Independent” seems ingroup candidate—.76 (95% confidence interval [.59,
to function merely as an identity tag and not as a measure .87]). Compared to these conditions, the probability of
of the relevance of partisan cues to decision making. African Americans selecting an outgroup winner was at
The results of the race manipulation (Figure 7) its highest—.45—when the European American was most
showed generally weaker effects of outgroup bias. Most qualified (95% confidence interval [.26, .66]).
African American and European American participants The probability of a European American selecting an
selected the African American candidate. African ingroup winner was only .42 (95% confidence interval
Americans were significantly more likely than European [.34, .50]), and further decreased to .29 (95% confidence
Americans to select the African American candidate (b = interval [.20, .40]) when the ingroup candidate was less
.95, s.e. = .36, p < .01). However, there was an overall qualified. The only condition in which a majority of
FEAR AND LOATHING ACROSS PARTY LINES 701

European Americans selected their ingroup candidate triple the amount transferred to Player 2, who will have
was when the candidate was more qualified, with a a chance to transfer an amount back to Player 1 (though
probability of ingroup selection at .64 (95% confidence Player 2 is under no obligation to return any money). The
interval [.53, .74]). Overall, in contrast with the behavior dictator game is an abbreviated version in which there is
of partisans, the majority racial group consistently no opportunity for Player 2 to return funds to Player 1
selected the minority candidate unless the qualifications and where the amount transferred is not tripled by the
of the majority ingroup candidate exceeded the qualifi- researcher. Since there is no opportunity for Player 1 to
cations of the minority candidate. Ingroup selection was observe the strategy of Player 2, variation in the amount
stronger and the effects of academic qualifications much Player 1 allocates to different categories of Player 2 in the
weaker in the partisan cue conditions. dictator game is attributable only to group dislike and
Overall, the results from the candidate selection study prejudice. As Fershtman and Gneezy (2001, 354) put it,
suggest that evaluations of high school students’ academic “any transfer distribution differences in the dictator game
credentials are remarkably sensitive to small partisan cues. must be due to a taste for discrimination.”
The partisan cue consisted of a single activity (among a The behavioral economics literature suggests that
randomly ordered list of five); nevertheless, the effects of Player 1, contrary to the axioms of rationality, typically
the cue proved strong, even for leaners. Facing no so- allocates nontrivial amounts (Johnson and Mislin 2008;
cial pressures to make unbiased choices, partisans feel Wilson and Eckel 2011) and that the allocation varies
no compunction to discriminate against out-party can- depending on attributes of Player 1 and the group af-
didates. The data also show that discrimination based filiation of Player 2 (Fershtman and Gneezy 2001; Fong
on party affiliation exceeds discrimination based on race. and Luttmer 2011). Women, for instance, tend to allocate
Despite the obvious relevance of academic credentials to greater amounts and are less prone to discriminate on
scholarship eligibility, we found no evidence that parti- the basis of group attributes (Eckel and Grossman 1998).
sans took academic merit into account. Prior work by Fowler and Kam (2007) detected small but
significant traces of favoritism directed at copartisans,
but the study focused on the effects of biases in giving
on political participation and did not compare partisan-
Study 3: Behavioral Evidence of ship with other social divides as a basis for discrimination
Partisan Bias between recipients.
A sample of 814 adults, drawn from the SSI panel,
We have shown that partisans display strong implicit participated in the study. We oversampled Republicans
biases toward opposing partisans, but do these effects so that there were approximately equal numbers of
persist when discrimination has tangible, possibly ad- Democrats and Republicans in the sample.10 The sam-
verse, consequences for participants? The previous study ple was also stratified by race, age, region, and income
documented effects of partisan bias in a hypothetical so that the distribution of these background variables
decision-making situation where the decision itself did approximated census data.11
not affect the participant. In this study, we use trust Participants were randomly assigned to play four
and dictator games (Forsythe et al. 1994) to test the rounds of the dictator game or four rounds of the trust
robustness of our findings. The games provide a more game. In both games, participants were told that they
consequential test of bias, for they assess the extent to would receive $10 for each round that they could split
which participants are willing to donate or risk money with Player 2 in any way they wished. In the trust game,
they would otherwise receive themselves to copartisans participants were informed that the experimenters would
while simultaneously withholding money from opposing triple any amount given to Player 2 and that Player 2
partisans. Once again, we compare partisanship and race could then allocate some, all, or none of the funds back to
as bases for discriminatory behavior.
Behavioral games are used extensively to assess group 10
Independents were excluded from the sample, and leaners were
cooperation and conflict (Berg, Dickhaut, and McCabe grouped with partisans.
1995; Eckel and Grossman 1998; Fershtman and Gneezy 11
We fielded the study in September 2012. The permanent state of
2001; Habyarimana et al. 2007; Whitt and Wilson 2007). modern political campaigns (Ornstein and Mann 2000) and the
In the trust game, Player 1 is given an initial endowment persistence of partisan bias over time (Iyengar, Sood, and Lelkes
($10) and instructed that she is free to give some, all, or 2012) suggest that proximity to the campaign is a valid but likely
insignificant concern. Our results also replicate a pretest fielded in
none to Player 2 (said to be a member of a designated July before the start of the conventions. They were also replicated
group). She is further informed that the researcher will in Study 4, which was fielded in the winter of 2013.
702 SHANTO IYENGAR AND SEAN J. WESTWOOD

Player 1. Finally, to make clear the incentives, participants in the dictator game. In both versions of the game,
were told that the amount of money held at the end of players were more generous toward copartisans, but not
the study would determine their success. coethnics. The average amount allocated to copartisans in
For each round of the game, players were pro- the trust game was $4.58 (95% confidence interval [4.33,
vided a capsule description of the second player, includ- 4.83]), representing a “bonus” of some 10% over the aver-
ing information about the player’s age, gender, income, age allocation of $4.17. In the dictator game, copartisans
race/ethnicity, and party affiliation. Age was randomly as- were awarded 24% over the average allocation.13
signed to range between 32 and 38, income varied between Overall, Republicans tended to be less generous than
$39,000 and $42,300, and gender was fixed as male. Player Democrats (see Table 4) toward ingroup and outgroup
2’s partisanship was limited to Democrat or Republican, players in both forms of the game. They allocated about
so there are two pairings of partisan similarity (Democrats 35 cents less, on average, to copartisans and about 20
and Republicans playing with Democrats and Republi- cents less than Democrats to opposing partisans.14 The
cans). The race of Player 2 was limited to white or African additional amount allocated to copartisans in the dictator
American. Race and partisanship were crossed in a 2 × 2, game was exactly equal for Democrats and Republicans
within-subjects design totaling four rounds/Player 2s. The ($0.68), but three times as large for Democrats ($0.63)
order of each of the four rounds was randomized. than Republicans ($0.24) in the trust game.
To minimize demand effects, participants were given From Figure 8, it is clear that in comparison with
no indication they were participating in a study of race party, the effects of racial similarity proved negligible and
or partisanship. At the outset, participants were asked not significant—coethnics were treated more generously
to provide some basic demographic information and (by eight cents, 95% confidence interval [–.11, .27]) in the
were told that this information would be offered to other dictator game, but incurred a loss (seven cents, 95% confi-
participants who were simultaneously playing the game. dence interval [–.34, .20]) in the trust game. There was no
Before playing either version of the game, participants interaction between partisan and racial similarity; playing
were given detailed instructions, read examples, and with both a copartisan and coethnic did not elicit addi-
completed a short comprehension quiz. In both games, tional trust over and above the effects of copartisanship.
participants only took the role of Player 1. To minimize These results thus replicate Study 1 and Study 2. Despite
round-ordering concerns, there was no feedback offered lingering negative attitudes toward African Americans,
at the end of each round; participants were told all results social norms appear to suppress racial discrimination,
would be provided at the end of the study. but there is no such reluctance to discriminate based on
partisan affiliation.

Results
As shown in Figure 8, we assess the effects of par- Study 4: Separating Ingroup
tisanship and race on allocations through a within- Favoritism from Outgroup Animosity
subjects 2 × 2 analysis that estimates the independent and
joint effects of racial (white or African American) and Affect toward copartisans has remained relatively sta-
partisan (Republican or Democrat) similarity between ble in American National Election Studies (ANES) data,
Players 1 and 2. The within-subjects analysis (i.e., using a whereas affect toward opposing partisans has dramati-
multilevel model with random effects across individuals) cally decreased (Iyengar, Sood, and Lelkes 2012). Study
has the effect of controlling for individual-level variation 2 shows that when placed in a zero-sum decision task,
in generosity, sense of egalitarianism, and other relevant participants were more likely to select a copartisan. Study
predispositions (see Habyarimana et al. 2007; Whitt and 3 similarly shows that participants are more generous and
Wilson 2007). trusting toward copartisans. Neither study, however, al-
Consistent with prior research, players chose not lows us to disentangle the effects of outgroup prejudice
to profit maximize,12 but rather to allocate nontrivial from ingroup favoritism. This was the goal of Study 4.
amounts of their endowment—a mean of $4.17 (95%
confidence interval [3.91, 4.43]) in the trust game, and 13
The significantly larger allocation in the trust game suggests that
a mean of $2.88 (95% confidence interval [2.66, 3.10]) participants did in fact understand the incentives and were expect-
ing to receive some return from Player 2.
14
Republicans awarded 46 cents less to Democrats in the dictator
12
Player 1’s dominant strategy in both games is to share with or game, but were slightly more generous in the trust game (by a
entrust to Player 2 some amount of money. margin of 23 cents).
FEAR AND LOATHING ACROSS PARTY LINES 703

TABLE 4 Allocation Based on Party

Dictator Game
Participant Party Player 2 Party Amount Standard Error Allocating $0 to Player 2
Democrat Democrat $3.82 0.11 14%
Democrat Republican $3.14 0.11 23%
Republican Democrat $2.68 0.12 33%
Republican Republican $3.36 0.11 20%
Trust Game
Participant Party Player 2 Party Amount Standard Error Allocating $0 to Player 2
Democrat Democrat $4.71 0.12 7%
Democrat Republican $4.08 0.12 10%
Republican Democrat $4.21 0.16 15%
Republican Republican $4.45 0.15 11%

We drew a sample of 1,252 members of the SSI panel how likely it is that a given Player 2 will share some of the
in the winter of 2012. Participants were randomly as- returns offered by the researcher. If information on Player
signed to complete either a dictator game or a trust 2 is not useful in the assessment of the trustworthiness of
game. We used the same instructions and procedure as Player 2, the information should be ignored or given little
in Study 3, but we utilized a different set of Player 2 weight. However, if participants believe that a known trait
profiles. For both games, we deployed a four-condition, of Player 2 is an indicator of trustworthiness, they should
within-subjects design (Player 2 was described without be more likely to increase their allocation to Player 2 in
any partisan identity, as a Democrat, as a Republican, or anticipation of reaping greater financial rewards. Study
as an Independent). The inclusion of a true control lack- 3 clearly shows that partisanship is used to assess the
ing any reference to partisan affiliation provides a neutral trustworthiness of Player 2, a finding we replicate here,
baseline from which to assess preferences for ingroup and but the results from this study show that partisanship
outgroup members.15 is treated differently when Player 2 is a member of the
As shown in Figure 9, there was a significant copar- ingroup rather than the outgroup. Participants assume
tisan bonus of $0.67 (95% confidence interval [.50, .84]) that copartisans are more likely to return money and ad-
in the dictator game and $0.30 (95% confidence interval just their allocations upward to increase potential returns.
[.08, .52]) in the trust game. There was also a significant They remain, however, more punitive and untrusting of
penalty of $0.63 (95% confidence interval [–.80, –.46]) opposing partisans (with the opposing partisan penalty
for opposing partisans in the dictator game and a penalty amounting to almost twice the copartisan bonus). De-
of $0.63 (95% confidence interval [–.85, –.41]) for oppos- spite the stronger incentive to allocate funds to Player 2,
ing partisans in the trust game. Compared to the control participants are much less inclined to favor copartisans
(no information) condition, there was no effect when in the trust game compared to the dictator game, but
Player 2 was an Independent in the dictator game or the only slightly less inclined to discriminate against oppos-
trust game. ing partisans in the trust game compared to the dictator
The outgroup penalty remained approximately the game. Outgroup animosity is more consequential than
same across the two types of games (although the penalty favoritism for the ingroup
represented a smaller proportion of the overall allocation
to Player 2 in the trust game than the dictator game),
whereas the ingroup bonus was halved in the trust game Discussion
compared to the dictator game. Decisions in the trust
game require more complex analysis than the dictator Compared with the most salient social divide in
game. Not only must participants judge how much money American society—race—partisanship elicits more ex-
they wish to keep for themselves, but they must also assess treme evaluations and behavioral responses to ingroups
and outgroups. This remarkable patterns applies to both
15
Leaners were grouped with partisans. explicit and implicit measures of group affect and holds
704 SHANTO IYENGAR AND SEAN J. WESTWOOD

FIGURE 9 Effects of Ingroup and Outgroup Membership on Allocations to

Player 2

Dictator Game
Opposing Partisan ●
Player 2
Independent ●
Copartisan ●
Control (No PID) ●
$2.00 $2.50 $3.00 $3.50 $4.00 $4.50 $5.00
Amount Allocated to Player 2

Trust Game
Opposing Partisan ●
Player 2

Independent ●
Copartisan ●
Control (No PID) ●
$2.00 $2.50 $3.00 $3.50 $4.00 $4.50 $5.00
Amount Allocated to Player 2

Note: The dots are the means for each group, and the bars are 95% confidence intervals.

even when the tests of ingroup favoritism are unobtru- nonwhites (see Chong and Rogers 2005; Miller et al. 1981;
sive and completely nonpolitical, and partisans are incen- Schmitt et al. 2002). For men and whites, accordingly, the
tivized to treat copartisans no differently from opposing sense of gender/racial identity may be insufficiently salient
partisans. The most plausible explanation for the stronger to generate animus for the outgroup.
affective response generated by partisan cues is the non- Our comparisons across indicators of implicit and
applicability of egalitarian norms. These norms, which explicit party affect revealed no differences in the degree
are supported by large majorities, discourage the mani- of affective polarization; if anything, conventional survey
festation of behavior that may be constructed as discrim- measures may be somewhat inflated. Similarly, the analy-
inatory. In contemporary America, the strength of these sis of individuals’ responsiveness to party cues suggests
norms has made virtually any discussion of racial differ- that it is partisan affect—either implicit or explicit—
ences a taboo subject to the point that citizens suppress that underlies the broader political ramifications of
their true feelings (Greenwald, McGhee, and Schwartz polarization.
1998). No such constraints apply to evaluations of parti- More generally, our results provide further support
san groups. for the view that party identification in the United States
The larger animus associated with the party divide is more of an affective than instrumental or ideological
is further attributable to fundamental differences be- bond. This “primal” view of partisanship was first docu-
tween partisan and race-based identity. First, individ- mented in The American Voter (Campbell et al. 1960) and
uals choose rather than inherit their party affiliation. has since been reinforced by considerable work on the psy-
It is possible, therefore, that they are more likely to chology of partisan identity (see Green, Palmquist, and
be held responsible (i.e., blamed) for their partisanship Schickler 2004; Huddy, Mason, and Aarøe 2010) and by
rather than their ethnic affiliation. Second, Democrats corroborating evidence demonstrating that partisans are
and Republicans almost by definition stand in opposi- poorly informed about the policy positions advocated by
tion to each other, and Americans are regularly exposed party elites (Bennett 2003; Delli Carpini and Keeter 1996).
to rhetoric from their leaders conveying open hostility Current debates over the degree of ideological
toward political opponents. Racial identity, on the other polarization within the electorate and dismissals of
hand, is less transparently conflictual. Finally, there is polarization as a symptom of partisan sorting (e.g.,
considerable evidence that group identity is heightened Fiorina and Abrams 2008) do not come to grips with
among disadvantaged groups, that is, among women and this conception of partisan identity and the significant
FEAR AND LOATHING ACROSS PARTY LINES 705

role played by partisan affect in the psyche of ordinary position. Recent evidence on congressional “taunting”
Americans. The mass public may offer centrist prefer- fits precisely this pattern; representatives from safe seats
ences, but they certainly sense that “the other side” is are especially likely to taunt the opposition party. Con-
an outgroup. While Americans are inclined to “hedge” gressional press releases that fit the partisan taunting
expressions of overt animosity toward racial minorities, category—meaning that they utilize “exaggerated lan-
immigrants, gays, or other marginalized groups, they en- guage to put [the opposition] down or devalue their
thusiastically voice hostility for the opposing party and ideas”—make up more than one-quarter of all congres-
its supporters. sional press releases issued between 2005 and 2007 (Grim-
The extent of affective polarization—at least in these mer and King 2011, 2649).
studies—appears uniform across parties. Previous re- The level of animosity across party lines also im-
search suggests stronger bias against the outgroup among plies a reduced willingness to treat the actions of partisan
Republicans and conservatives (Jost et al. 2003; Jost, opponents as legitimate, resulting in more intense contes-
Hennes, and Lavine 2013; Stern et al. 2012), but our evi- tation of policy outcomes. The passage of the landmark
dence indicates that the polarization scores (both explicit 1964 and 1965 Civil Rights Acts were no doubt contro-
and implicit) for partisans on the left and right were gen- versial and opposed by large numbers of Americans, but
erally indistinguishable. However, when we limit the anal- they were not subject to persistent efforts at nullification.
ysis to people who identify as strongly partisan, outparty In contrast, two years after passage of the Affordable Care
animus is significantly higher among Republicans. Act, legislative efforts to repeal the law show no signs of
There are several limitations to this work. Racial cues weakening.
are easy to assess and hard to suppress, whereas parti- Finally, our evidence documents a significant shift
sanship must be disclosed or revealed. We can therefore in the relationship between American voters and their
document effects of partisanship on decision making, parties. Fifty years ago, comparative party researchers
and show that they are consistently more sizable than described American parties as relatively weak, at least
race, but we cannot show that partisanship is considered by the standards of European “mass membership” par-
as frequently as race. However, partisanship, as we note, ties (Committee on Political Parties 1950; Duverger 1963;
is a source of social identity that people embrace in their Kirchheimer 1966). The prototypical instance of the latter
social networks, their workplace discussions, and their category was a party “membership in which is bound up
place of residence. People may not wear their partisan- in all aspects of the individual’s life” (Katz and Mair 1995,
ship on their sleeves, but, for millions of Americans, it is 6). By this standard, American parties have undergone a
not difficult to discern. It is true that we run the risk of significant role reversal. Today, the sense of partisan iden-
artificially elevating the role of partisanship by revealing tification is all encompassing and affects behavior in both
it as an attribute of the target individuals in Studies 2–4. political and nonpolitical contexts.
However, these studies also provided participants with
information about the target person’s race, gender, in-
come, and academic achievements. If partisanship were
unimportant, we should expect participants to disregard
it when given a good reason to do so (e.g., the presence References
of a candidate with superior academic qualifications). It
is also possible that the political environment at the time Alford, John R., Peter K. Hatemi, John R. Hibbing, Nicholas
of the decision encouraged the use of partisanship in de- G. Martin, and Lindon J. Eaves. 2011. “The Politics of Mate
Choice.” Journal of Politics 73(2): 362–79.
cision tasks. However, our results replicate during times
Asendorpf, Jens B., Rainer Banse, and Daniel Mücke. 2002.
of relatively low political conflict (i.e., winter 2013 and
“Double Dissociation between Implicit and Explicit Person-
spring 2013) and times of stronger political conflict (i.e., ality Self-Concept: The Case of Shy Behavior.” Journal of
during the 2012 presidential election). Personality and Social Psychology 83(2): 380–93.
In closing, we note that the increased levels of par- Baron, Andrew S., and Mahzarin R. Banaji. 2006. “The Devel-
tisan affect have fairly clear implications for the polit- opment of Implicit Attitudes: Evidence of Race Evaluations
ical process. Hostility for the opposition party among from Ages 6 and 10 and Adulthood.” Psychological Science
17(1): 53–58.
rank-and-file partisans sends a clear signal to elected of-
Bennett, W. Lance. 2003. News: The Politics of Illusion. 5th ed.
ficials; representatives who appear willing to work across New York: Longman.
party lines run the risk of being perceived as “appeasers.” Berelson, Bernard, Paul F. Lazarsfeld, and William McPhee.
For the vast majority who represent uncompetitive 1954. Voting: A Study of Opinion Formation in a Presidential
districts, there are strong incentives to “bash” the op- Campaign. Chicago, IL: University of Chicago Press.
706 SHANTO IYENGAR AND SEAN J. WESTWOOD

Berg, Joyce, John Dickhaut, and Kevin McCabe. 1995. “Trust, Fong, Christina M., and Erzo F. Luttmer. 2011. “Do Fairness
Reciprocity, and Social History.” Games and Economic Be- and Race Matter in Generosity? Evidence from a Nation-
havior 10(1): 122–42. ally Representative Charity Experiment.” Journal of Public
Billig, Michael, and Henri Tajfel. 1973. “Social Categorization Economics 95(5): 372–94.
and Similarity in Intergroup Behaviour.” European Journal Forsythe, Robert, Joel L. Horowitz, Nathan E. Savin, and Martin
of Social Psychology 3(1): 27–51. Sefton. 1994. “Fairness in Simple Bargaining Experiments.”
Bishop, Bill. 2008. The Big Sort: Why the Clustering of Like- Games and Economic Behavior 6(3): 347–69.
Minded America Is Tearing Us Apart. New York: Houghton Fowler, John H., and Cindy D. Kam. 2007. “Beyond the Self: So-
Mifflin Harcourt. cial Identity, Altruism, and Political Participation.” Journal
Bogardus, Emory S. 1925. “Social Distance and Its Origins.” of Politics 69(3): 813–27.
Journal of Applied Sociology 9(2): 216–26. Green, Donald, Bradley Palmquist, and Eric Schickler. 2004.
Boysen, Guy A., David L. Vogel, and Stephanie Madon. 2006. Partisan Hearts and Minds: Political Parties and the Social
“A Public versus Private Administration of the Implicit As- Identities of Voters. New Haven, CT: Yale University Press.
sociation Test.” European Journal of Social Psychology 36(6): Greene, Steven. 1999. “Understanding Party Identification: A
845–56. Social Identity Approach.” Political Psychology 20(2): 393–
Cai, Huajian, N. Sriram, Anthony G. Greenwald, and Sam G. 403.
McFarland. 2004. “The Implicit Association Tests D Measure Greenwald, Anthony G., and Mahzarin R. Banaji. 1995. “Im-
Can Minimize a Cognitive Skill Confound: Comment on plicit Social Cognition: Attitudes, Self-Esteem, and Stereo-
McFarland and Crouch (2002).” Social Cognition 22(6): 673– types.” Psychological Review 102(1): 4–27.
84. Greenwald, Anthony G., Debbie E. McGhee, and Jordan L. K.
Campbell, Angus, Philip E. Converse, Warren E. Miller, and Schwartz. 1998. “Measuring Individual Differences in Im-
Donald Stokes. 1960. The American Voter. Unabridged edi- plicit Cognition: The Implicit Association Test” Journal of
tion. New York: Wiley. Personality and Social Psychology 74(6): 1464–80.
Carmines, Edward G., Paul M. Sniderman, and Beth C. Easter. Greenwald, Anthony G., Brian A. Nosek, and Mahzarin R. Ba-
2011. “On the Meaning, Measurement, and Implications naji. 2003. “Understanding and Using the Implicit Associ-
of Racial Resentment.” Annals of the American Academy of ation Test: I. An Improved Scoring Algorithm.” Journal of
Political and Social Science 634(1): 98–116. Personality and Social Psychology 85(2): 197–216.
Chong, Dennis, and Reuel Rogers. 2005. “Racial Solidar- Grimmer, Justin, and Gary King. 2011. “General Purpose
ity and Political Participation.” Political Behavior 27(4): Computer-Assisted Clustering and Conceptualization.” Pro-
347–74. ceedings of the National Academy of Sciences 108(7): 264–350.
Cohen, Jack. 1988. Statistical Power Analysis for the Behavioral Habyarimana, James, Macartan Humphreys, Daniel N. Posner,
Sciences. Hillsdale, NJ: L. Erlbaum Associates. and Jeremy M. Weinstein. 2007. “Why Does Ethnic Diver-
Committee on Political Parties. 1950. “Toward a More Respon- sity Undermine Public Goods Provision?” American Political
sible Two-Party System.” American Political Science Review Science Review 101(4): 709–25.
44(3). doi:10.1002/ncr.4110391113. Haidt, Jonathan, and Marc J. Hetherington. 2012. “Look
Crosby, Faye, Stephanie Bromley, and Leonard Saxe. 1980. “Re- How Far We’ve Come Apart.” Campaign Stops, The
cent Unobtrusive Studies of Black and White Discrimination New York Times, September 17. http://campaignstops.
and Prejudice: A Literature Review.” Psychological Bulletin blogs.nytimes.com/2012/09/17/look-how-far-weve-come-
87(3): 546–63. apart/.
Delli Carpini, Michael X., and Scott Keeter. 1996. What Amer- Hetherington, Marc J. 2009. “Putting Polarization in Perspec-
icans Know about Politics and Why It Matters. New Haven, tive.” British Journal of Political Science 39(2): 413–48.
CT: Yale University Press. Himmelfarb, Samuel, and Carl Lickteig. 1982. “Social Desir-
DeSante, Christopher D. 2013. “Working Twice as Hard to ability and the Randomized Response Technique.” Journal
Get Half as Far: Race, Work Ethic, and America’s Deserv- of Personality and Social Psychology 43(4): 710–17.
ing Poor.” American Journal of Political Science 57(2): 342– Huber, Gregory, and Neil Malhotra. 2012. “Political Sorting in
56. Social Relationships: Evidence from an Online Dating Com-
Duverger, Maurice. 1963. Political Parties: Their Organization munity.” http://xa.yimg.com/kq/groups/Relationships.pdf.
and Activity in the Modern State. New York: Wiley. Huddy, Leonie, Lilliana Mason, and Lene Aarøe. 2010. “Mea-
Eckel, Catherine C., and Philip J. Grossman. 1998. “Are Women suring Partisanship as a Social Identity, Predicting Political
Less Selfish Than Men? Evidence from Dictator Experi- Activism.” Paper presented at the annual meeting of the In-
ments.” Economic Journal 108(448): 726–35. ternational Society for Political Psychology, San Francisco.
Facebook. 2013. “Facebook.” http://www.facebook.com/press/ Iyengar, Shanto, Gaurav Sood, and Yphtach Lelkes. 2012. “Af-
info.php?statistics. fect, Not Ideology: A Social Identity Perspective on Polariza-
tion.” Public Opinion Quarterly 76(3): 405–31.
Fershtman, Chaim, and Uri Gneezy. 2001. “Discrimination in a
Segmented Society: An Experimental Approach.” Quarterly Johnson, Noel D., and Alexandra Mislin. 2008. “Cultures of
Journal of Economics 116(1): 351–77. Kindness: A Meta-Analysis of Trust Game Experiments.”
http://ssrncom/abstract1315325.
Fiorina, Morris P., and Samuel J. Abrams. 2008. “Political Polar-
ization in the American Public.” Annual Review of Political Jost, John T., and Mahzarin R. Banaji. 1994. “The Role of Stereo-
Science 11: 563–88. typing in System-Justification and the Production of False
FEAR AND LOATHING ACROSS PARTY LINES 707

Consciousness.” British Journal of Social Psychology 33(1): Pager, Devah, and Hana Shepherd. 2008. “The Sociology of Dis-
1–27. crimination: Racial Discrimination in Employment, Hous-
Jost, John T., Jack Glaser, Arie W. Kruglanski, and Frank J. ing, Credit, and Consumer Markets.” Annual Review of So-
Sulloway. 2003. “Political Conservatism as Motivated Social ciology 34: 181–209.
Cognition.” Psychological Bulletin 129(3): 3–39. Rosenfeld, Michael J., Thomas J. Reuben, and Maja Falcon.
Jost, John T., E. P. Hennes, and H. Lavine. 2013. “Hot Political 2011. How Couples Meet and Stay Together, Waves 1, 2, and
Cognition: Its Self-, Group-, and System-Serving Purposes.” 3: Public Version 3.04. Stanford, CA: Stanford University
In Oxford Handbook of Social Cognition, ed. Donal E. Carl- Libraries.
ston. Oxford: Oxford University Press. Schmitt, Michael T., Nyla R. Branscombe, Diane Kobrynowicz,
Katz, Richard S., and Peter Mair. 1995. “Changing Models of and Susan Owen. 2002. “Perceiving Discrimination Against
Party Organization and Party Democracy: The Emergence One’s Gender Group Has Different Implications for Well-
of the Cartel Party.” Party Politics 1(1): 5–28. Being in Women and Men.” Personality and Social Psychology
Bulletin 28(2): 197–210.
Kirchheimer, Otto. (1966). “The Transformation of the
Western European Party Systems.” In Political Par- Schuman, Howard, Charlotte Steeh, Lawrence Bobo, and Maria
ties and Political Development, ed. Joseph LaPalombara Krysan. 1997. Racial Attitudes in America. Cambridge, MA:
and Myron Weiner. Princeton, NJ: Princeton University Harvard University Press.
Press, 177–200. Sigall, Harold, and Richard Page. 1971. “Current Stereotypes:
Klofstad, Casey A., Rose McDermott, and Peter K. Hatemi. 2012. A Little Fading, a Little Faking.” Journal of Personality and
“The Dating Preferences of Liberals and Conservatives.” Po- Social Psychology 18(2): 247–55.
litical Behavior 35(3): 1–20. Sriram, N., and Anthony G. Greenwald. 2009. “The Brief Im-
Korte, Gregory. 2012. “Obama Has Edge in Political Merchan- plicit Association Test.” Experimental Psychology 56(4): 283–
dise.” USA Today, September 25. 94.
Lelkes, Yphtach, and Shanto Iyengar. 2012. “Fear and Loathing Stern, Chadly, Tessa V. West, John T. Jost, and Nicholas O.
across Party Lines? Cross-National Evidence on Party Polar- Rule. 2012. “The Politics of Gaydar: Ideological Differences
ization.” Working paper. in the Use of Gendered Cues in Categorizing Sexual Orien-
tation.” Journal of Personality and Social Psychology 104(3):
Maccoby, Eleanor E., and Nathan Maccoby. 1954. “The In- 520–41.
terview: A Tool of Social Science.” In Handbook of Social
Psychology, ed. Gardiner Lindzey. Cambridge, MA: Addison Stoker, Laura, and M. Kent Jennings. 1995. “Life-Cycle Tran-
Wesley, 449–87. sitions and Political Participation: The Case of Marriage.”
American Political Science Review 89(2): 421–33.
McConahay, John B., Betty B. Hardee, and Valerie Batts. 1981.
“Has Racism Declined in America? It Depends on Who Is Tajfel, Henri. 1970. “Experiments in Intergroup Discrimina-
Asking and What Is Asked.” Journal of Conflict Resolution tion.” Scientific American 223(5): 96–102.
25(4): 563–79. Tajfel, Henri, and John C. Turner. 1979. “An Integrative Theory
Miller, Arthur H., Patricia Gurin, Gerald Gurin, and Oksana of Intergroup Conflict.” In The Social Psychology of Inter-
Malanchuk. 1981. “Group Consciousness and Political Par- group Relations, ed. William G. Austin and Stephen Worchel.
ticipation.” American Journal of Political Science 25(3): 494– Monterey, CA: Brooks-Cole, 33–47.
511. Whitt, Sam, and Rick K. Wilson. 2007. “The Dictator Game,
Munro, Geoffrey D., Carrie Weih, and Jeffrey Tsai. 2010. “Moti- Fairness and Ethnicity in Postwar Bosnia.” American Journal
vated Suspicion: Asymmetrical Attributions of the Behavior of Political Science 51(3): 655–68.
of Political Ingroup and Outgroup Members.” Basic and Ap- Wilson, Rick K., and Catherine C. Eckel. 2011. “Trust and Social
plied Social Psychology 32(2): 173–84. Exchange.” In Cambridge Handbook of Experimental Politi-
Mutz, Diana C., and Jeffry Mondak. 2006. “The Workplace as cal Science, ed. James S. Druckman, Donald P. Green, John
a Context for Cross-Cutting Political Discourse.” Journal of H. Kuklinski, and Arthur Lupia. New York: Cambridge Uni-
Politics 68(1): 140–55. versity Press, 243–57.
Myrdal, Gunnar. 1944. An American Dilemma: The Negro Prob-
lem and Modern Democracy. Piscataway, NJ: Transaction.
Nosek, Brian A., Mahzarin R. Banaji, and Anthony G. Green-
Supporting Information
wald. 2002. “Harvesting Implicit Group Attitudes and Beliefs
from a Demonstration Web Site.” Group Dynamics: Theory, Additional Supporting Information may be found in the
Research, and Practice 6(1): 101–15. online version of this article at the publisher’s website:
Ornstein, Norman J., and Thomas E. Mann. 2000. The Perma-
nent Campaign and Its Future. Washington, DC: American Measure and sample information are included in the sup-
Enterprise Institute Press. porting information.