RESEARCH ARTICLE | PSYCHOLOGICAL AND COGNITIVE SCIENCES OPEN ACCESS

Individuals prefer to harm their own group rather than help an

opposing group
Rachel Gershona,1,2, and Ariel Fridmana,1

Edited by Susan Fiske, Princeton University, Princeton, NJ; received September 14, 2022; accepted October 8, 2022

Group-based conflict enacts a severe toll on society, yet the psychological factors gov-
erning behavior in group conflicts remain unclear. Past work finds that group members Significance
seek to maximize relative differences between their in-group and out-group (“in-group
favoritism”) and are driven by a desire to benefit in-groups rather than harm out-groups Understanding the principles
(the “in-group love” hypothesis). This prior research studies how decision-makers guiding decisions in intergroup
approach trade-offs between two net-positive outcomes for their in-group. However, conflicts is essential to
in the real world, group members often face trade-offs between net-negative options, recognizing the psychological
entailing either losses to their group or gains for the opposition. Anecdotally, under barriers to compromise and
such conditions, individuals may avoid supporting their opponents even if this harms cooperation. We introduce a
their own group, seemingly inconsistent with “in-group love” or a harm minimizing novel paradigm for studying
strategy. Yet, to the best of our knowledge, these circumstances have not been investi-
group decision-making,
gated. In six pre-registered studies, we find consistent evidence that individuals prefer
demonstrating that individuals
to harm their own group rather than provide even minimal support to an opposing
group across polarized issues (abortion access, political party, gun rights). Strikingly, in are so averse to supporting
an incentive-compatible experiment, individuals preferred to subtract more than three opposing groups that they prefer
times as much from their own group rather than support an opposing group, despite equivalent or greater harm to
believing that their in-group is more effective with funds. We find that identity concerns their own group instead. While
drive preferences in group decision-making, and individuals believe that supporting an previous models of group
opposing group is less value-compatible than harming their own group. Our results decision-making claim that group
hold valuable insights for the psychology of decision-making in intergroup conflict as members are driven by a desire
well as potential interventions for conflict resolution. to benefit their in-group
Downloaded from https://www.pnas.org by 60.12.158.39 on February 28, 2023 from IP address 60.12.158.39.

(“in-group love”) rather than

intergroup conflict | identity | norms | decision-making | polarization
harm their out-group, our results
Group conflicts are a pervasive feature of society. Yet, despite the extensive literature on cannot be explained by in-group
the topic, a unified understanding of the psychology underlying decision-making in group love or by a harm minimizing
conflicts remains elusive. Prior work has documented two broad principles governing strategy. Instead, we propose
group-based decision-making. First, group members exhibit in-group favoritism (1, 2). that identity concerns drive this
That is, individuals prefer to create a favorable comparison between their in-group and behavior. Our theorizing speaks
the out-group, even leading to choices that prioritize relative gains compared to the out- to research in psychology,
group over greater absolute gains for their in-group (see Social Identity Theory; refs. political theory, and negotiations
3 and 4). Second, past work theorizes that group members are driven by a cooperative
by examining how group
motive to help the in-group (“in-group love”) rather than an aggressive motive to hurt
members navigate trade-offs
the out-group (“out-group hate”; 5, 6). Critically, these principles are derived from studies
in which participants chose between outcomes that are all ultimately favorable to the among competing priorities.
in-group. However, real-world decision-making often entails making choices where harm
is unavoidable (7, 8). Groups may have to choose between in-group losses and out-group
gains, a circumstance that has not been previously studied and reveals that individuals’
Author affiliations: aRady School of Management, UC San
decisions cannot be explained by existing theories. Diego, San Diego, CA 92093
Consider the example of Montgomery, Alabama, where only White residents were
allowed to use the publicly funded Oak Park Pool until, in 1959, a federal court deemed
the segregated pool unconstitutional. The White town council then faced two options Author contributions: R.G. and A.F. designed research;
that they considered unfavorable: give Black families access to the pool or close one of the performed research; contributed new reagents/analytic
tools; A.F. analyzed data; R.G. and A.F. wrote the paper.
town’s favorite gathering spots. Previous work on in-group favoritism and the dominant
The authors declare no competing interest.
role of “in-group love” would predict that white Montgomery residents would avoid
This article is a PNAS Direct Submission.
harming their own group even at the cost of extending pool use to Black citizens. Yet
Copyright © 2022 the Author(s). Published by PNAS.
famously, the white citizens of Montgomery closed the pool. Other public resources, such This open access article is distributed under Creative
as parks and zoos, were closed to all across the country to defy similar rulings (9). The Commons Attribution License 4.0 (CC BY).
Oak Park Pool is an exemplar of the fact that, anecdotally, when faced with two coun- 1
R.G. and A.F. contributed equally to this work.
ter-attitudinal choices—aid the out-group or harm the in-group —group members may 2
To whom correspondence may be addressed. Email:
[email protected].
avoid showing support for the out-group, even at the apparent expense of their own side
This article contains supporting information online at
(10–13). However, there has never been a rigorous investigation of how individuals nav- https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.​
igate unfavorable choices in intergroup conflicts and whether there is a broad preference 2215633119/-/DCSupplemental.
to harm one’s own group rather than support the opposition. Published November 29, 2022.

PNAS 2022 Vol. 119 No. 49 e2215633119 https://doi.org/10.1073/pnas.2215633119 1 of 12

 Table 1. Group decision-making theories.The ✓ indicates that the theory would predict a preference for the given
choice, whereas 50% indicates an indifference between the choices
Win-Win Lose-Lose
Add $1 to in-group Subtract $1 from Add $1 to opposing Subtract $1 from
opposing group group in-group
Identity-support model ✓ ✓
Zero-sum beliefs 50% 50% 50% 50%
Harm minimization ✓ ✓
(based on effectiveness
considerations)
Maximize in-group ✓ ✓
payoff
Maximize relative payoff 50% 50% 50% 50%
for in-group
Minimize payoff 50% 50% 50% 50%
difference
Maximize joint payoff, in ✓ 50% 50%
favor of in-group
Minimize opposing ✓ ✓
group payoff

Here, we develop a novel paradigm in which individuals must that individuals will choose the least value-expressive option,
either deduct funds from their in-group or add funds to an oppos- thereby best protecting their identity. Our results support these
ing group to examine how group members make trade-offs in predictions—individuals believe that helping an opposing group
lose-lose* intergroup conflicts. Our experiments were conducted is more harmful to their identity than inflicting equivalent harm
across multiple countries (United States and United Kingdom), to their in-group, even when this leads to a worse relative standing
several polarized issues (abortion access, political party, and gun for their in-group. Critically, we find that by ameliorating identity
control), and various experimental measures (financial donations concerns through shifting perceived in-group norms (20–24),
and incentive-compatible multiple price lists). Taken together, our individuals become more likely to support the opposing group,
Downloaded from https://www.pnas.org by 60.12.158.39 on February 28, 2023 from IP address 60.12.158.39.

results offer the first unambiguous evidence that individuals are providing a practical way to achieve more constructive outcomes
so averse to showing support for an opposing out-group that they in group conflict.
even prefer to do greater harm to their own group instead. Our While there are numerous existing models of group deci-
finding was symmetrically exhibited by individuals on both sides sion-making, the Identity-Support model can uniquely explain
of each issue we studied, and even among participants who iden- individual behavior for both favorable (win-win) and unfavorable
tified only weakly with their side. However, the degree to which (lose-lose) choices with opposing groups. We address other frame-
a participant identified with their side of the focal issue does play works below in Table 1. First, it is possible that individuals view
a moderating role—those with stronger attitudes in favor of their these decisions as a zero-sum trade-off (25, 26), wherein a gain
side (e.g., more strongly pro-choice) were more likely to choose for one group is perceived as an equivalent loss for the other group.
to harm their own side (and willing to cause greater overall harm In this case, group members should be indifferent between the
to their side) rather than help the opposing side. two options. It is also possible that group members consider which
Our results reveal the central role of identity in decision-making side is more effective at using funds to pursue their mission and
in polarized contexts. Identity often plays an important role in then choose based on harm minimization for their cause. Across
decision-making (14), such as when and to whom we offer support our studies, we find that group members typically believe their
(15, 16). We propose that individuals aim to protect their group- own group is more effective with funds, so those motivated by
based identity when facing intergroup conflict and therefore harm minimization should prefer to add to their own (more effec-
behave in ways that best express their values, especially those that tive) side in win-win scenarios and avoid subtracting from their
are central to their identity (17). Previous work finds that indi- own side in lose-lose scenarios.
viduals prefer expressions of support (e.g., “I support Democrats”) The social value orientation literature (e.g., ref. 27), which arbi-
to expressions of opposition (e.g., “I oppose Republicans”), trarily assigns in-groups and out-groups (i.e., a minimal group
because support is considered more “value-expressive” (18, 19). paradigm) to study group member decision-making, describes
We would therefore expect group members to choose actions of additional motives that may also guide decision-making for real-
support over actions of opposition when both options convey their world opposing groups. For example, individuals may simply
values (i.e., choosing to help their own side rather than harm their prefer allocations that maximize the payoff for the in-group (anal-
opponent). However, what would individuals choose when the ogous to being motivated by in-group love (5, 6)), maximize the
options are unfavorable, inconsistent with their values (i.e., a lose- relative difference in payoff between their in-group and opposing
lose choice between harming their own group or helping their group (consistent with findings of in-group favoritism (1, 2)), or,
opponent)? In such situations, since both options express values as demonstrated in Bornstein et al., (1983), either minimize the
that are counter-attitudinal, our Identity-Support model suggests difference in payoffs to each side or maximize joint profit in favor
of the in-group. Additionally, recent work on negative partisanship
*While the term “lose-lose” is often used to express a loss for each side, here we define
lose-lose (win-win) to describe situations where both options are unfavorable (favorable)
and increased out-group animosity (28–30) suggests that individ-
to the decision-maker. uals may be primarily motivated to minimize the payoff to the

2 of 12 https://doi.org/10.1073/pnas.2215633119 pnas.org
 opposing side. Table 1 summarizes how individuals guided by side or subtract $1 from the donation going to the organization
each of these motives would choose when faced with win-win and on the opposing side. In the lose-lose condition, both options
lose-lose scenarios. Notably, the Identity-Support model is the altered the donation in ways that were unfavorable given the
only model that can explain both findings across scenarios. participant’s stated attitude: either add $1 to the donation going
Finally, previous work on prospect theory (31, 32), finding that to the organization on the opposing side or subtract $1 from
people experience losses more strongly than equivalent gains, also the donation going to the organization on their side. Finally,
does not make clear predictions in this context. For example, when participants reported which side of each cause was more effective
faced with a lose-lose scenario, while participants technically at using funds to pursue their mission. Specifically, they were asked
choose between a loss or gain (of funding), they may encode both “Do you believe that [Pro-life/Pro-gun/Republican/Conservative]
adding to the opposing group and subtracting from the in-group or [Pro-choice/Anti-gun/Democratic/Labor] organizations are
as losses for their side. That is, any gains for the opposing group more effective at pursuing their mission? In other words, which
can feel like losses for the in-group and vice versa. However, if this one is able to do more with each dollar they receive?”
is not the case and individuals are indeed more affected by losses In the win-win condition, our results conceptually replicate and
to their side than gains to the opposing side, then our findings extend prior studies. We find that participants are more likely to
would be inconsistent with predictions based on prospect theory choose to support their side by adding $1 to the organization on
as well. their side of each cause (72.5%) rather than harm the opposition
by subtracting $1, t(594) = 13.87, P < 0.001, and 95% CI =
Open Science [69.7%, 75.2%]‡ (see Fig. 1). Thus, these results reproduce the
findings of the in-group love model in the context of a win-win
All study designs and analyses were pre-registered, and all data, choice in our sample, extending previous results to natural groups
analysis code, research materials, and pre-registrations are available (as opposed to minimal group paradigms).
at https://osf.io/gzxke/. Data were analyzed using R, version 4.1.2, The results of our lose-lose condition, however, contradict the
and the package ggplot2, version 3.3.5 (33). For all studies, we predictions of the in-group love model. While in-group love pre-
reported all manipulations and measures and recruited a minimum dicts that individuals will support their opposition to avoid harm-
of 100 participants per condition. All sample sizes and exclusion ing their in-group, participants in the lose-lose condition
criteria were determined in advance. predominantly chose not to help the opposition (helping the
opposing group was chosen by only 25.8% of participants), pre-
Results ferring instead to harm their own group almost three-quarters of
the time, t(594) = −13.85, P < 0.001, and 95% CI = [23.0%,
Study 1: Individuals Prefer to Harm Their Own Group Rather
28.7%] (see Fig. 1).
Than Support the Opposition. In Study 1, we develop a novel
The preference to harm the in-group rather than support the
paradigm to investigate how individuals behave in group conflicts
opposition is robust and consistent across a variety of subsamples.
in which they must choose between supporting an opposing
Downloaded from https://www.pnas.org by 60.12.158.39 on February 28, 2023 from IP address 60.12.158.39.

First, we find similar effects for both the nationally representative

group or harming their own group. In our paradigm, we asked
sample in the United States (24.6%; t(398) = −12.92, P < 0.001,
participants to indicate their position on several polarized issues
and 95% CI = [21.6%, 27.9%]) and the nationally representative
in two sub-studies conducted with participants from the United
sample in the United Kingdom (32.7%; t(195) = −4.74, P <
States and United Kingdom. Given the similarities in their designs
0.001, and 95% CI = [26.4%, 39.5%]). Second, the same pattern
and hypotheses, we report these studies together, noting only where
holds across every issue we tested (abortion: (27.8%; t(398) =
they differ. In the US study (Study 1A; N = 797, matched to US
−8.52, P < 0.001, and 95% CI = [23.6%, 32.4%]); gun control:
census data on age, sex, and ethnicity), issues included abortion
(25.3%; t(398) = −9.39, P < 0.001, and 95% CI = [21.3%,
access, gun control, and political party affiliation (Democratic or
29.8%]); party support: (24.7%; t(594) = −11.71, P < 0.001, and
Republican Party). In the UK study (Study 1B; N = 393, matched
95% CI = [21.4%, 28.3%]). Finally, the preference to harm the
to UK census data on age, sex, and ethnicity), participants were
in-group rather than support the opposition is present on both
asked about political party affiliation (Labor or Conservative
sides of the ideological spectrum§: (liberals: 26.4%; t(477) =
Party). After indicating their attitude toward the relevant issues (on
−11.46, P < 0.001, and 95% CI = [23.1%, 29.9%]; conservatives:
a six-point Likert scale†, which we subdivide into weak, medium,
24.1%; t(235) = −8.59, P < 0.001, and 95% CI = [19.6%,
and strong attitudes; see Methods section), participants learned
29.2%]). The results from the win-win condition were also con-
that real donations would be made to organizations supporting
sistent and significant across these robustness checks (all Ps <
each side of the partisan divide (e.g., both a pro-choice and pro-
0.001). See Table 2 for a summary of results across all studies.
life organization). Participants were then asked how they would
Although individuals prefer to deduct funding from their
choose to alter the donation (in Study 1A, they rated each of
in-group rather than support the opposing group in this study
the three issues separately, randomly ordered) and were informed
with real consequences, one possible explanation is that partici-
that for ten randomly chosen participants, their choices would
pants view this as the better outcome for their cause. Indeed, if
actually be implemented. Participants were randomly assigned
individuals believe that the opposing group is more effective at
to one of two between-subjects experimental conditions: (a) a
advancing their interests per dollar spent than their own group,
win-win condition or (b) a lose-lose condition. In the win-win
then reducing equivalent funding to the in-group maximizes the
condition, which serves as an experimental control to conceptually
relative difference in outcomes between groups in a lose-lose deci-
replicate prior findings (5, 6), both options altered the donation
sion. However, our results show the reverse: Participants indicated
in ways that were favorable given the participant’s stated attitude:
that they view organizations supporting their side of an issue to
either add $1 to the donation going to the organization on their
‡Statistical tests are from logistic regressions and account for repeated observations from
each participant across issues by clustering the standard errors by participant.
†We note that for this and all subsequent studies, we asked all participants to indicate their
position on each issue without allowing them to express indifference. While this introduces §69% of participants in the US sample held either liberal (pro-choice, anti-gun ownership,
noise by forcing truly indifferent participants to choose a side, it would not bias the results and Democrat) or conservative (pro-life, pro-gun ownership, and Republican) positions
in a particular direction and works against finding an overall effect. on all three issues.

PNAS 2022 Vol. 119 No. 49 e2215633119 https://doi.org/10.1073/pnas.2215633119 3 of 12

 Abortion Access
75%

50%

25%

75%

Gun Control
50%
Choice Share for Add

25%

U.S. Political Party

75%

50%

25%

UK Political Party
75%

50%

25%

Weak Medium Strong

Downloaded from https://www.pnas.org by 60.12.158.39 on February 28, 2023 from IP address 60.12.158.39.

Attitude Strength

Win−Win Lose−Lose

Fig. 1. Choice share in each condition for all four issues by attitude strength in Studies 1A (N = 797) and 1B (N = 393). The vertical axis shows the proportion
of participants choosing to add funds (in the lose-lose condition: add funds to opposing group vs. subtract from their in-group; in the win-win condition: add
funds to the in-group vs. subtract from the opposing group). Error bars represent 95% confidence intervals.

be more effective in spending donation money to achieve their 0.89, SE = 0.17, z = 5.21, P < 0.001, OR = 2.43, and 95% CI =
goal (SI Appendix, Results for Study 1). To verify that our results [1.74, 3.40]) or weak (β = 1.05, SE = 0.18, z = 6.01, P < 0.001,
are not driven primarily by individuals who believe the opposition OR = 2.87, and 95% CI = [2.04, 4.05]) attitudes. In the win-win
is more effective with funds, we conducted an ancillary analysis condition, we find that those with strong attitudes were more likely
in which we studied the subset of participants who believe organ- to choose to add $1 to their own side (vs. subtract $1 from the
izations on their side of an issue are strictly more effective with opposing side), compared to those with medium (β = −0.36, SE =
funds (43.8% of observations). These participants in the lose-lose 0.17, z = −2.13, P = 0.03, OR = 0.70, and 95% CI = [0.50, 0.97])
condition still preferred to “harm” their side (supporting the or weak (β = −0.83, SE = 0.16, z = −5.10, P < 0.001, OR = 0.44,
opposing side was chosen by only 22.8% of these participants; and 95% CI = [0.32, 0.60]) attitudes.
t(402) = −10.82, P < 0.001, and 95% CI = [19.2%, 27.0%]). Taken together, these results establish that not only do group
Finally, we conducted an initial test of our central theory that members prefer to deduct funds from their in-group rather than
individuals make decisions in group conflicts on the basis of pro- contribute an equivalent amount to their opposition, but they
tecting their identity (developed in-depth in Studies 3 and 4). As make this choice despite explicitly believing that this leaves their
previously outlined, individuals believe that acts of support are more group worse off than the alternative. While motives such as harm
value-expressive than acts of opposition. Individuals may choose minimization, in-group favoritism, and in-group love cannot
the least value-expressive option when offered two unfavorable explain the findings from this study, the results are consistent with
choices, thereby opposing their own group rather than supporting predictions from the Identity-Support model. Having established
the opposing group. We predicted that this desire to protect their that our findings in win-win scenarios are consistent with previous
identity (and therefore the choice not to support the opposition) work, we focus on decision-making in our novel lose-lose para-
should be stronger for participants with stronger group identities, digm in the remaining studies.
as assessed by stronger attitudes about the underlying issue. Indeed,
in the lose-lose condition, we find that those with strong attitudes Study 2: Quantifying the Aversion to Supporting an Opposing
were even less likely to choose to add $1 to the opposing side (vs. Group. Although Study 1 establishes that individuals prefer to
subtract $1 from their side), compared to those with medium (β = deduct a given amount of funds from their in-group rather than

4 of 12 https://doi.org/10.1073/pnas.2215633119 pnas.org
 Table 2. Summary table for studies 1–5 of the percentage of participants preferring to support the opposing group
(for lose-lose choices) and support the in-group (for win-win choices; study 1 only)
Lose-Lose (% choosing to Win-Win (% choosing to sup- Participant attitudes
support opposing group) port in-group)
Study 1
Abortion access 27.8% 81.9% Pro-choice: 76%; Pro-life: 24%
Gun control 25.3% 68.3% Anti-gun ownership: 75%;
Pro-gun ownership: 25%
US politics 20.8% 67.6% Democrats: 70%; Republi-
cans: 30%
UK politics 32.7% 72.1% Labor party: 64%; Conserva-
tive party: 36%
Study 2 Abortion access 35.1% Pro-choice: 74%; Pro-life: 26%
Study 3 US politics 36.4% Democrats: 65%; Republi-
cans: 35%
Study 4 US politics
Control 41.1% Democrats: 60%; Republi-
cans: 40%
Identity-strengthened 30.4% Democrats: 64%; Republi-
cans: 36%
Study 5 Abortion access
Control 39.2% Pro-choice: 67%; Pro-life: 33%
Norm-subtract 36.7% Pro-choice: 71%; Pro-life: 29%
Norm-add 57.7% Pro-choice: 72%; Pro-life: 28%
For study 2, the percentage captures the proportions of participants with an indifference amount of less than $1 (i.e., would rather add $1 to the opposing group than subtract $1 (or
less) from the in-group).

add the same amount to the opposition, Study 2 quantifies the the opposing organization. Strikingly, 28% of all participants
degree to which individuals prefer to harm their own group rather chose to entirely forgo the $10 donation to their side rather than
Downloaded from https://www.pnas.org by 60.12.158.39 on February 28, 2023 from IP address 60.12.158.39.

than support an opposing group. We approach this quantitative add $1 to the opposing organization. Lastly, we found that those
analysis by eliciting participants’ indifference amount between with greater attitude strength (i.e., more strongly pro-choice or
harming their own group and supporting the opposition using an pro-life) required more funds to be subtracted from their side to
incentive-compatible choice titration procedure (34; see Methods be indifferent toward adding $1 to the opposing side, and we also
section for more detail). For this choice titration analysis, we replicated the finding that participants believed that organization
recruited 300 US participants from Amazon’s Mechanical Turk on their side is more effective at spending their donation money
with a final sample of 268 following our pre-registered exclusions. to achieve their goal (for details on both results, SI Appendix,
Given that we found consistent evidence across issues in Study Results for Study 2).
1, here we focused on a single polarizing issue: abortion access.
We informed participants that the researchers would make two
$10 donations: one to a pro-life organization and another to a
pro-choice organization. Participants were asked to choose how 30%
to alter the donation amount in a series of 14 choices. For each
choice, they could select to either add $1 to the opposing organ-
ization’s donation or subtract an amount (sequentially from $0.10
to $10, order counterbalanced) from their own side’s organization
(similar to a price list; 35). To incentivize responses, we informed 20%
Proportion

participants that for one in ten participants, chosen at random,

we would actually make donations to both organizations and ran-
domly select one of their 14 choices to alter the donation amount
(for similar elicitation and bonus procedures, see refs. 36–38). As
in Study 1, participants also reported their beliefs about the rela- 10%
tive effectiveness of pro-life and pro-choice organizations, as well
as their attitude strength toward the issue.
Our results show that participants’ indifference amounts had a
mean value significantly greater than $1 (M = $3.85, median =
$1.50, max = $10, SD = $4.10, t(267) = 11.39, P < 0.001, d = 0%
0.70, and 95% CI = [0.56, 0.83]), and the majority of participants
(65%) had an indifference amount greater than $1 (χ2(1, N = 0 1 2 3 4 5 6 7 8 9 10
Indifference Amount ($)
268) = 23.29, P < 0.001; see Fig. 2). In other words, on average
participants required almost $4 to be subtracted from the donation Fig. 2. Results from Study 2 (N = 268): distribution of participants’ indifference
going to their organization, to be indifferent toward adding $1 to amounts.

PNAS 2022 Vol. 119 No. 49 e2215633119 https://doi.org/10.1073/pnas.2215633119 5 of 12

 In sum, Study 2 demonstrates the strength of participants’ pref- SI Appendix, Results for Study 3 for regression results and Fig. 3
erences when facing lose-lose choices. Despite believing that their for plots of both measures.
own side is more effective with funds, group members preferred In summary, Study 3 explicitly establishes that individuals feel
to subtract, on average, more than three times as much from their that their group identity suffers more when supporting the oppo-
own group rather than give a small amount of support to the sition than when harming the in-group. Moreover, in a direct test
opposing group. of the relative contributions of identity concerns and effectiveness
considerations, we find that whereas the identity measure is sig-
Study 3: Identity Concerns Trump Effectiveness Considerations nificantly correlated with the choice to harm the in-group, there
for Lose-Lose Decisions. Studies 1–2 provide evidence across is no significant association between the effectiveness measure and
contexts that individuals are so averse to supporting an opposing individual decision-making in lose-lose group conflicts.
group that they prefer to harm their own group instead. Moreover, These findings align with our Identity-Support model of group
the finding that individuals with stronger attitudes toward the decision-making, pointing to the important role identity plays in
focal issue (and therefore stronger identity-relevance; 39, 40) decisions involving intergroup conflict. In a pre-registered sup-
are more prone to harming their in-group rather than helping plemental study (SI Appendix, Study S1), we test another key
the opposing group offers preliminary evidence that identity element of the model, building on relevant research (18)—that
considerations govern decision-making in group conflicts. In acts of support are more value-expressive than acts of opposition.
Study 3, we directly test the hypothesis that identity concerns, as Participants read about and evaluated another participant from
opposed to effectiveness considerations, underlie the psychology their in-group (Republican or Democrat) who had to make either
of decision-making in group conflicts involving lose-lose choices. a lose-lose or win-win allocation (as in the US political party
To test the relative contributions of identity concerns and effec- choice from Study 1) and were told which option the participant
tiveness considerations, we recruited 400 US participants from chose. Based on this decision, they were asked to assess how
Amazon’s Mechanical Turk with a final sample of 393 following strongly they believed this target identified with their political
our pre-registered exclusions. After indicating whether they iden- party. In the win-win scenario, those who opted to support their
tified more strongly as Republican or Democrat (binary choice), in-group were perceived as identifying more strongly with their
participants were asked to make a lose-lose choice, identical to the party than those who subtracted money from the opposing group.
US political party choice from Study 1. Subsequently, participants In the lose-lose scenario, those who opted to support the opposing
responded to an effectiveness and an identity concern question, side were perceived as more weakly identifying with their party
both on a seven-point Likert scale, asked in a randomized order. than those who subtracted from their in-group. This suggests that
The effectiveness question was similar to the one used in the pre- participants do in fact believe that acts of support are more val-
vious studies, and the identity concern question asked participants ue-expressive than acts of opposition using our paradigm. In line
whether adding $1 to the opposing side or subtracting $1 from with the Identity-Support model, participants should prefer the
their side “would make you feel like a worse [Democrat/ choice that best promotes or protects their identity and therefore
Downloaded from https://www.pnas.org by 60.12.158.39 on February 28, 2023 from IP address 60.12.158.39.

Republican]? In other words, which option most undermines your choose the most value-expressive option (supporting their own
identity as a [Democrat/Republican]?”.
As in our previous studies, less than half of participants chose
to add $1 to the organization supporting the opposing political 80%

party (36.39%), χ2(1, N = 393) = 38.59, and P < 0.001, for both
Democrats and Republicans (both Ps < 0.001). Moreover, as in
the previous studies, the overall effectiveness measure was positive 60%
Choice Share for Add

(M = 0.61, SD = 1.35), t(392) = 8.95, P < 0.001, d = 0.45, and

95% CI = [0.35, 0.55]), indicating that, on average, participants
view the organization on their side to be more effective at using 40%
donated funds to pursue their mission.
Critically, the identity concern measure revealed that partici-
pants believed adding $1 to the opposition undermined their 20%
partisan identity more than deducting $1 from their own party
(M = −0.22, SD = 1.99, t(392) = −2.16, P = 0.03, d = −0.11, and
95% CI = [−0.21, −0.01]). The fact that individuals perceive their
0%
group-based identities to be at greater risk when supporting an
opposing group (vs. harming their own) offers a clear rationale
for why they choose in-group harm over supporting the opposing
−3 −2 −1 0 1 2 3
group. However, to explicitly test the contributions of identity Response
concerns and effectiveness considerations on choice, we regressed
participant choice (0 = subtract $1, 1 = add $1) on both our Measure Effectiveness Identity Concern
identity and effectiveness measures using a logistic regression. The
identity concern measure was positively associated with choice, Fig. 3. Results from Study 3 (N = 393): choice share by measure. The
such that participants were more likely to choose to add to the vertical axis shows the proportion of participants who chose to add funds
opposing side when they believed that subtracting $1 from their to the donation for the opposing organization (vs. subtract from their side).
The horizontal axis captures participants’ responses on the effectiveness
side undermined their identity more (β = 0.65, SE = 0.07, z = and identity concern measures (which were both centered at 0). For the
8.77, P < 0.001, OR = 1.91, and 95% CI = [1.66, 2.22]). The effectiveness measure, more positive values indicate the in-group is more
effectives measure was not significantly associated with choice (β effective at using donated funds (vs. the out-group; 0 indicates equal
= −0.09, SE = 0.09, z = −0.97, P = 0.33, OR = 0.91, and 95% CI effectiveness). For the identity concern measure, more positive values indicate
subtracting $1 from the in-group undermines identity more (vs. adding $1 to
= [0.76, 1.09]). A similar pattern of results was observed when the opposing group; 0 indicates both equally undermine identity). Error bars
choice was regressed on each measure in separate regressions. See represent 95% confidence intervals.

6 of 12 https://doi.org/10.1073/pnas.2215633119 pnas.org
 side) when offered pro-attitudinal choices and avoid the most (74% of participants; χ2(1, N = 369) = 5.36, P = 0.021, and ϕ =
value-expressive option (supporting the other side) when offered 0.13), Democratic participants (62% of participants; χ2(1, N =
counter-attitudinal choices. 308) = 4.24, P = 0.039, and ϕ = 0.12), and was directional but
non-significant among the (relatively smaller) subset of Republican
Study 4: Increasing the Salience of a Group-Based Identity participants (38% of participants; χ2(1, N = 189) = 1.20, P =
Decreases the Probability of Supporting the Opposing Group. 0.274, and ϕ = 0.09). As in prior studies, we also found that
Although Study 3 offers correlational evidence that identity participants believed that organizations on their side are more
considerations govern the preference to harm one’s in-group to effective at spending their donation money to achieve their goal
avoid supporting the opposition, in Study 4 we causally test the (SI Appendix, Results for Study 4). In sum, our results provide
relationship between identity concerns and choice in lose-lose additional evidence for the role of identity on decision-making in
group conflicts. Building on previous work showing that identities group conflict, demonstrating a causal effect of identity salience
are malleable and making certain identities more salient can affect on the decision to harm one’s in-group rather than support the
preferences and behaviors (15, 41, 42), we hypothesized that opposition.
strengthening group identity salience would lead to an increase Finally, we note that Study 4 was designed to also address the
in the probability of harming one’s own group to avoid supporting potential confound that participants may disproportionately focus
the opposing group. on the most negative ways in which the opposing group would
To test our hypothesis, participants were randomized into one use donated funds, but do not similarly consider the most positive
of two conditions: In the identity-strengthened condition, partic- ways in which the in-group would use donated funds (43). Such
ipants were asked to write about an event, story, or personal expe- a difference could explain the strong aversion to supporting the
rience where they strongly identified with their political party. In opposing side that we find across lose-lose choices (although would
the control condition, participants wrote about what they do on not explain the win-win preference to help one’s own side rather
a typical Monday evening. Participants then made the same choice than harm the opposition). Consequently, in this study, we also
as in Study 3—add $1 to the donation going to the organization specified that all donations would go to “administrative costs (e.g.,
supporting the opposing political party or subtract $1 from the maintaining the organization’s website)” to hold constant the use
donation going to the organization supporting their own political of donations, ensuring that participants would imagine similar
party. donation uses for each organization. We therefore conclude that
We recruited 500 participants from Amazon’s Mechanical Turk, the preference to harm one’s own group rather than support an
with a final sample of 497 following our pre-registered exclusions. opposing group is not explained by different imagined uses of the
Consistent with our previous studies, in the control condition, funds by the in-group and opposing group.
the proportion of participants choosing to add $1 to the organi-
zation supporting the opposing political party was less than 50% Study 5: Modulating Group Norms Alters Decision-Making in
(41.11%), χ2(1, N = 270) = 8.18, and P = 0.004. In the identi- Group Conflicts. In Study 5, we test a practical method for shifting
Downloaded from https://www.pnas.org by 60.12.158.39 on February 28, 2023 from IP address 60.12.158.39.

ty-strengthened condition, the proportion of participants choos- behavior in lose-lose group conflict, specifically testing whether
ing to add $1 was significantly lower (30.4%) than in the control shifting perceived in-group norms alters individual decision-
group, χ2(1, N = 497) = 5.67, P = 0.017, and ϕ = 0.11 (see Fig. making. Since our results suggest that individuals make choices
4). This result held among the subset of participants who believed to protect and promote their group identity, we hypothesized that
organizations on their side of a cause are more effective with funds decision-making will be sensitive to group norm information. In
the absence of clear norm information, we consistently find that
individuals avoid supporting the opposing group. In Study 5, we
50%
test whether providing participants with alternate group norm
information (i.e., others in your in-group chose to support the
opposing group) will increase the choice share supporting the
opposing group over harming their own side, as the norm serves as
a powerful guideline for making choices that maintain an identity
consistent with the in-group.
Choice Share for Add

40% To test whether modulating group norms alters decision-mak-

ing, participants in Study 5 were asked to report their position on
abortion access (“very much against abortion access” to “very
much in favor of abortion access”) and were then randomized into
one of three conditions: control, norm-add, or norm-subtract.
30% Participants in all conditions chose between adding $1 to a dona-
tion going to an organization supporting the opposing side or
subtracting $1 from a donation going to their side of the issue. In
the norm-add condition, participants were also informed that in
a previous study, 70% of participants who shared their views on
abortion access chose to add to the opposing side rather than
20% subtract from their own and that one of those participants had
Control Identity−Strengthened said the following: “I care way too much about my cause to take
money away from it.” In the norm-subtract condition, participants
Overall Democrats Republicans were instead told that 70% of previous participants on their side
of the cause had chosen to subtract from their in-group rather
Fig. 4. Results from Study 4 (N = 497): choice share by condition. The vertical than add to the opposing group. The statement from the previous
axis shows the proportion of participants who chose to add funds to the
donation to the opposing organization (vs. subtract from their side). Error participant was changed to: “I dislike the other side way too much
bars represent 95% confidence intervals. to give them money.” Finally, as in our prior studies, each

PNAS 2022 Vol. 119 No. 49 e2215633119 https://doi.org/10.1073/pnas.2215633119 7 of 12

 participant indicated which of the two sides of the cause they or support their opposition. We operationalize this choice by giv-
believe is more effective at pursuing its mission. We recruited 653 ing study participants the option to either deduct funds from
participants from Amazon’s Mechanical Turk, with a final sample organizations within their in-group or add the same amount of
of 635 following our pre-registered exclusions. funding to an opposing organization. Such choices help to separate
In the control condition, we replicated our fundamental find- various motives that could be driving decision-making, and
ing: The proportion of participants choosing to add $1 to the remarkably, we find that even though individuals report that
opposing side’s donation was significantly less than 50% (39.2%), organizations in their in-group (vs. opposing group) are more
χ2(1, N = 212) = 9.55, and P = 0.002 (see Fig. 5). effective with funds, they choose to deduct from their (more effec-
In the norm-add condition, the key test of the power of group tive) in-group rather than add an equivalent amount of funds to
norms, we found a significant increase in the proportion of par- the opposition. Indeed, individuals are so averse to providing any
ticipants choosing to add $1 in the norm-add condition (57.7%), support to the opposing group that they, on average, accepted
compared to participants in the control condition χ2(1, N = 420) triple the amount of financial loss to their in-group to avoid any
= 13.72, P < 0.001, and ϕ = 0.19. In fact, the proportion of par- gains for the other side (Study 2). We reproduce our main findings
ticipants in the norm-add condition choosing to add $1 to the across both sides of an array of group conflicts (abortion, gun
opposing side was significantly greater than 50% (χ2(1, N = 208) control, political party) and in multiple countries (United States
= 4.62, and P = 0.032). The results of the norm-add condition and United Kingdom; Study 1) to illustrate that the preference to
show that norms-based interventions about group identities can harm one’s in-group to avoid supporting the opposing group is a
powerfully shift decision-making away from harming the in-group robust, fundamental feature of individual decision-making in
and toward supporting the opposing group. group conflicts.
In the norm-subtract condition, the proportion of participants Moreover, we explored the role of identity concerns to under-
choosing to add $1 to the opposing side (36.7%) was not statis- stand the psychology underlying the preference to harm one’s
tically different from the control group, χ2(1, N = 427) = 0.17, in-group rather than support the opposition. We found that
P = 0.680, and ϕ = 0.02. The similarity between the norm-subtract whereas the strength of an individuals’ group identity strongly
and control conditions implies that in the absence of an explicit correlates with the decision to harm the in-group rather than
group norm (as in the control condition), the default norm is not support the opposition, individual assessments of group efficacy
to support the opposition even at the expense of harming one’s were uncorrelated with choice (Study 3). Manipulation of identity
own side. We further verified that the norm-add and norm-sub- salience modulated the choice to harm the in-group vs. support
tract conditions were statistically different, χ2(1, N = 423) = 17.79, the opposition (Study 4), further illustrating the central role of
P < 0.001, and ϕ = 0.21. In a supplementary analysis (SI Appendix, identity considerations in decision-making within group conflicts.
Results for Study 5), we found that the norms manipulation can Finally, we demonstrated a practical method to alter preferences
shift behavior even for those with strong attitudes toward an issue. in intergroup conflicts: Shifting perceptions of in-group norms
We also replicated the result that participants believed the organ- lead to corresponding changes in behavior—individuals who were
Downloaded from https://www.pnas.org by 60.12.158.39 on February 28, 2023 from IP address 60.12.158.39.

ization on their side is more effective at spending donation money told that other in-group members were willing to support the
to achieve their goal. opposing group became more likely to do the same (Study 5).

Discussion Identity Concerns as the Central Driver of Decision-Making in

Group Conflicts. Previous models of individual psychology in
In the present work, we investigate how individuals prefer to adju- groups, such as in-group favoritism (1, 3, 4, 44, 45) and in-group
dicate a lose-lose choice in intergroup conflict: harm their in-group love (5, 6), examined decision-making using win-win scenarios,
which cannot explain our findings in lose-lose scenarios. In a win-
win context, in which individuals choose between various gains for
the in-group and/or losses for the out-group, past work has found
60%
that individuals will seek the best relative outcome for their in-group
(“in-group favoritism”) while avoiding unnecessarily harming the
out-group (“in-group love” rather than “out-group hate”). While
this literature used “minimal” groups where trivial differences
Choice Share for Add

created in-group and out-group distinctions, we replicate the

50% preference to help one’s in-group rather than harm the out-group
using natural groups for preexisting polarizing issues. However, we
find that in a lose-lose context, individuals choose to financially
harm their in-group rather than support an opposing out-group.
This is a violation of both in-group love and in-group favoritism,
40%
as the alternative choice—supporting the opposition—maximizes
the relative position of the in-group, because organizations on
one’s side are generally viewed as more effective with funds than
opposing organizations. In fact, participants even chose to accept
triple the amount of financial losses to their own group to avoid
supporting the opposition, illustrating that group members were
30% not acting to establish the most favorable comparison between
Control Norm−Subtract Norm−Add their in-group and the opposing group. Rather than in-group
love, the results from lose-lose scenarios appear to be evidence
Fig. 5. Results from Study 5 (N = 635): choice share by condition. The vertical for the opposite—out-group hate—in line with recent work on
axis shows the proportion of participants who chose to add funds to the
donation to the opposing organization (vs. subtract from their side). Error negative partisanship, finding that partisans are demonstrating
bars represent 95% confidence intervals. increasingly negative affect toward the opposing party (28, 29).

8 of 12 https://doi.org/10.1073/pnas.2215633119 pnas.org
 Among political partisans in the United States, “out-party hate” from normative theory (56, 57), this prior work does not make
was recently found to be stronger than “in-party love” (30). clear predictions for which choice individuals will make in the
We synthesize prior work on support-framing (18, 19) and lose-lose scenarios in our studies. That is, individuals with deeply
propose the Identity-Support model, which can parsimoniously held beliefs tying themselves to their in-group would likely have
explain our findings across win-win and lose-lose scenarios. The a strong aversion to both harming the in-group and helping the
model suggests that individuals act in group conflicts to promote out-group, though it is unclear which would prevail based on this
their identity, and they do so primarily by providing support to literature. We also note that the preference to harm one’s in-group
causes they believe in (and avoid supporting causes they oppose; persists even among individuals for whom these are not strongly
see also SI Appendix, Study S1). Simply put, in win-win contexts, held beliefs (i.e., those who report weaker attitudes toward their
supporting the in-group is more expressive of one’s identity as a side of the issue). We would not expect the preference to harm
group member than harming the opposing group, thereby leading one’s own group rather than help an out-group to emerge for all
to a preference for in-group support. In lose-lose contexts, sup- out-groups, but rather for out-groups with which individuals do
porting the opposing group is more negatively expressive of one’s not want to align themselves (58, 59) or for groups that directly
identity as a group member than harming the in-group, resulting oppose the decision-maker’s beliefs.
in a preference for in-group harm. Therefore, the principle that
individuals make decisions in group conflicts to promote and Implications for Better Outcomes in Group Conflicts. One striking
protect their identity, primarily by allocating their support in ways facet of our work is that individuals resolve lose-lose decisions in
that most align with their values, offers a single framework that group conflicts in ways that leave their own in-group in a worse
predicts individual behavior in group conflicts in both win-win relative position than if they had simply supported the opposition.
and lose-lose contexts. When generalized across both sides of several issues, our work points
to the possibility that identity concerns may act as a barrier to better
Alternative Explanations and Related Literature. Although our outcomes for both groups. Therefore, groups engaged in conflict may
findings offer strong support for the role of identity considerations realize mutual gains if individuals are less averse to supporting the
in group conflict, our results do not address whether these identity opposition. Building on an extensive literature on in-group norms
concerns are driven by a motivation to maintain or boost one’s (60–63), we demonstrated that shifting group norms can modulate
self-image (46–48) or their reputation (i.e., for social approval; individuals’ aversion toward showing support for the opposing group.
49, 50). As an initial test of whether the aversion to supporting While many accounts suggest that the United States is becoming
the opposing side is driven primarily by reputation concerns, we more affectively and ideologically polarized (64–66), an emerging
ran a pre-registered supplemental study (SI Appendix, Study S2). literature on “false polarization” suggests that intergroup conflict is
As in previous studies, we offered participants a lose-lose trade-off, exacerbated by misperceptions about the magnitude and consistency
but here we also manipulated whether the choice was explicitly of out-group members’ beliefs (67, 68). In fact, recent work finds
anonymous or would be made public. If participants’ decision that Americans often tolerate and even show admiration for in-group
Downloaded from https://www.pnas.org by 60.12.158.39 on February 28, 2023 from IP address 60.12.158.39.

is based on public compliance or a desire for social approval members who seek to understand the out-group (69), indicating
(51, 52), we would expect the effect to be stronger when making both sides may have a desire for cooperation. Our findings offer a
their choice publicly (vs. privately). However, we found that practical approach that has the potential to increase cooperation:
individuals’ preferences did not differ when their choices were Providing information about in-group norms may reduce group
public vs. private and that they preferred to harm their group members’ identity concerns, thereby allowing for behaviors that
rather than help the opposing group in both conditions. Our support the out-group when advantageous. Future research might
results suggest that the motive to express one’s values by avoiding further examine the nature of this norm belief and test realistic and
out-group support is internalized. However, some work suggests effective methods for increasing the likelihood to work with the
that social influence may still be at play as individuals sometimes opposition, such as modeling cooperative behavior by high-status
act as though they are being observed by a third party even when in-group members.
they are not (as in the anonymous condition; 47, 53, 54, 55). Our findings add to a literature on how psychological barriers
Nevertheless, this supplemental study provides further evidence impede the advancement of important causes (13). In contexts in
for the robustness of the aversion to helping the opposing side which accommodating two groups’ desires is crucial for progress,
regardless of whether others would learn about their decision. how do we compromise when both sides would rather harm their
In the same supplemental study, we examined a possible alter- own cause than make concessions in which the opposition gains
native explanation for the pattern of preferences we observe. any benefits? For example, a congressperson wishing to cross the
Participants in our experiments may have chosen to subtract funds aisle to support legislation may be hindered by the assumption
from their side (rather than add to the opposing side) because it that it would be a sign of disloyalty to her constituents. In an era
feels easier to undo or reverse (e.g., by making an additional dona- of high perceived polarization, understanding how identity con-
tion to their side later). By contrast, participants may believe it is cerns and beliefs about group norms shape these decisions is crit-
more difficult to “undo” the addition of funding to the opposing ical. Otherwise, these psychological barriers are likely to impede
side. We therefore asked study participants to explain why they progress, not only for the causes we oppose, but also for those we
chose the option they selected. Of the 497 participants in the most strongly support.
study, only four mentioned reversibility as an explanation for their
choice to subtract funding, suggesting that this is not the primary Materials and Methods
driver for the preference to harm one’s own side in our studies.
Overview. All experiments were approved by the UC San Diego Institutional
This work studies decision-making for polarizing issues where
Review Board (IRB), and all participants gave their informed consent to participate.
individuals may have deeply held beliefs. We chose polarizing
contexts because of the importance of improved decision-making Studies 1A and 1B. Study 1A was conducted in January 2022. As outlined in our
around these contentious issues and our specific interest in inter- pre-registration (https://aspredicted.org/cz62t.pdf), we aimed to recruit a nation-
group conflict. While previous research finds that deeply held ally representative sample of 800 US participants through Prolific and ended
beliefs or sacred values lead to behavior that frequently departs up with a sample of 801 participants who completed the study (50.6% female,

PNAS 2022 Vol. 119 No. 49 e2215633119 https://doi.org/10.1073/pnas.2215633119 9 of 12

 mean age = 45.17 y). We excluded four participants who failed the reading check, researchers would make two $10 donations: one to a pro-life organization
leaving us with a final sample of 797 participants. and another to a pro-choice organization. Participants were then asked to
Participants were randomized into one of two conditions (lose-lose or win-win) choose how to alter the donation amount in a series of 14 choices, where for
to make a hypothetical choice. All participants first reported their position on three each choice they could select either a right-hand side or left-hand side option
issues, presented in a randomized order: abortion access (“very much against (similar to a price list; 35). The right-hand side option was always to add $1 to
abortion access” to “very much in favor of abortion access”), gun control (“very the donation going to the opposing organization. The left-hand side option
much against gun control” to “very much in favor of gun control”), and political was to subtract $X from the donation going to the organization on the partic-
party (“strongly Republican” to “strongly Democratic”). Responses were captured ipant’s side of the cause, where X took the values 0.10, 0.25, 0.50, 0.75, 1,
using a six-point Likert scale to prevent participants from expressing indifference, 2, 3, 4, 5, 6, 7, 8, 9, and 10 (see Fig. 6). We randomized whether participants
as alignment to a side of each issue was required for the scenario assignment. viewed X in ascending or descending order. To incentivize the responses, we
We used these responses to classify participants as having either weak (3, 4 on informed participants that for one in ten participants, chosen at random, we
the scale), medium (2, 5 on the scale), or strong (1, 6 on the scale) attitudes. would actually make donations to both organizations and randomly select
For each of the three issues, participants were told that, as part of the study, dona- one of their 14 choices to alter the donation amount.
tions would be made to organizations supporting each side, and that they would The outcome of interest was each participant’s indifference amount. We assumed
need to make a choice about how to alter the donation amount. We informed par- that the indifference amount is at the midpoint of the subtract amounts on either
ticipants that we would randomly select ten of them and adjust one of the donation side of the switch. For example (see Fig. 6), if a participant switches from preferring
amounts based on their choice and actually make the donations on their behalf. to subtract $3 from their side (instead of adding $1 to the other side), to preferring to
For each issue, brief descriptions of each organization’s mission were provided. For add $1 to the other side (instead of subtracting $4 from their side), the indifference
example, for the abortion access issue, participants read: “The mission of the Pro-life amount must be in the interval between $3 and $4 and was coded as the midpoint
organization is to reduce access to abortions. The mission of the Pro-choice organiza- ($3.50). However, our results are robust to coding the indifference amount as the
tion is to increase access to abortions.” No organizations were referred to by name to lower bound of each interval instead of the midpoint. Using the lower bound is a
avoid any associations a participant may have with a particular organization. All the highly conservative measure since it might underestimate each participant’s true
scenarios and corresponding binary choices were presented in a randomized order. indifference amount, which lies in between the end points of each interval. If a par-
For each cause, participants were asked to select one of two options. In the ticipant selected the left-hand option for every choice, we coded their indifference
win-win condition, both options altered the donation in ways that were favorable amount as $10. If a participant selected the right-hand option for every choice, we
given the participant’s stated attitude: either add $1 to the donation going to the interpreted their indifference amount to be between $0 and $0.10 and coded their
organization on their side or subtract $1 from the donation going to the organi- indifference amount as the midpoint ($0.05). There was no significant difference
zation on the opposing side. In the lose-lose condition, both options altered the in indifference amounts by price list order (P = 0.40), so we collapsed across the
donation in ways that were unfavorable given the participant’s stated attitude: ascending and descending conditions.
either add $1 to the donation going to the organization on the opposing side or After this series of choices, we assessed beliefs about the relative effectiveness
subtract $1 from the donation going to the organization on their side. of the organizations as in Study 1. Participants were asked “Do you believe that
After responding to all three scenarios, participants reported which side of each the Pro-Life or Pro-Choice organization is more effective at pursuing their mis-
issue had organizations that they believed to be more effective at pursuing their sion? In other words, which one spends their donation money more effectively?”.
Downloaded from https://www.pnas.org by 60.12.158.39 on February 28, 2023 from IP address 60.12.158.39.

mission. Participants were specifically asked “which one is able to do more with each Responses were collected on a seven-point Likert scale, allowing participants to
dollar they receive?”. Responses were collected on a seven-point Likert scale for all report equal effectiveness.
three issues, in a randomized order: abortion access (“pro-life organization are more
Study 3. Study 3 was conducted in February 2022. As outlined in our pre-regis-
effective” to “pro-choice organizations are more effective”), gun control (“pro-gun
organizations are more effective” to “anti-gun organizations are more effective”), and tration (https://aspredicted.org/ja6un.pdf), we aimed to recruit 400 U.S. MTurk
political party (“Republicans are more effective” to “Democrats are more effective”). participants and ended up with a sample of 401 participants who completed
Study 1B was conducted in February 2022 and was identical, except we col- the study (51% female, mean age = 39.57 y). We excluded 2.0% of participants
lected participants from the UK, and only focused on a single issue—political parti- who failed the reading check or had duplicate MTurk IDs, leaving us with a final
sanship. As outlined in our pre-registration (https://aspredicted.org/m4nq5.pdf), sample of 393 participants.
we recruited a nationally representative sample of 400 UK participants through Participants first indicated whether they identified more strongly as Republican or
Prolific (50.3% female, mean age = 44.68 y). We excluded seven participants Democrat (binary choice) and were subsequently asked to make a lose-lose choice,
who failed the reading check, leaving us with a final sample of 393 participants. identical to the US political party choice from Study 1. We told participants we would
All participants reported their political position on the following six-point Likert pick ten of them at random and make the donations according to their adjusted dona-
scale: “Strongly Conservative Party” to “Strongly Labor Party.” As with Study 1A, tion amounts. Subsequently, participants responded to an effectiveness question
we used these responses to classify participants as having either weak (3, 4 on and an identity concern question, in a randomized order. The effectiveness questions
the scale), medium (2, 5 on the scale), or strong (1, 6 on the scale) attitudes. asked participants, “Do you believe that the Republican or Democratic Party is more
For analyses across issues, we combined the datasets collected from Studies effective at using donated funds to pursue their mission? In other words, which
1A and 1B. To test whether attitude strength moderated the participants’ choices, one is able to do more with each dollar they receive?” (1 = Republicans are more
we regressed their choice (0 = subtract $1, 1 = add $1) on their condition and the effective, 7 = Democrats are more effective). The identity question asked, “Which of
interaction between condition and attitude strength (as a categorical variable), these two options would make you feel like a worse [Democrat/Republican]? In other
using a logistic regression. words, which option most undermines your identity as a [Democrat/Republican]?”
(1 = Definitely adding $1 to the [Republican/Democratic] organization, 4 = Both
Study 2. Study 2 was conducted in December 2020. We recruited 300 US par- choices equally undermine my identity as a [Democrat/Republican], 7 = Definitely
ticipants from Amazon’s Mechanical Turk (MTurk; 53% female, mean age = subtracting $1 from the [Democratic/Republican] organization). The first option in
36.71 y). As outlined in our pre-registration (https://aspredicted.org/cn7ry.pdf), square brackets was selected for participants who identified as Democrats, and the
we excluded participants who switched more than once between the left and second was selected for Republicans.
right-hand choices (10.7% of participants). This fraction of exclusion is within
the typical range observed in prior studies involving multiple price lists (70). All Study 4. Study 4 was conducted in June 2021. As outlined in our pre-registra-
remaining participants passed the pre-registered reading check, and there were tion (https://aspredicted.org/gu45s.pdf), we recruited 500 U.S. MTurk partic-
no duplicated MTurk IDs, so there were no additional exclusions, resulting in a ipants (50% female, mean age = 39.51 y). We excluded 0.6% of participants
final sample of 268 participants which was used for all analyses. who failed the reading check or had duplicate MTurk IDs, leaving us with a final
All participants reported the extent to which they are against or in favor of sample of 497 participants. All participants reported whether they identified
abortion access on a 12-point Likert scale (“very much against abortion access” more strongly as Republican or Democrat (binary choice). Participants were
to “very much in favor of abortion access”). We informed participants that the then randomized into one of two conditions: control or identity-strengthened.

10 of 12 https://doi.org/10.1073/pnas.2215633119 pnas.org
Downloaded from https://www.pnas.org by 60.12.158.39 on February 28, 2023 from IP address 60.12.158.39.

Fig. 6. Screenshot of the series of choices made by participants in Study 2. This example is for a pro-choice participant, with the choices listed in ascending order.

Participants in the identity-strengthened condition were then asked to write was identical, except we also informed participants that in a previous study, 70% of
about an event, story, or personal experience where they strongly identified MTurkers who shared their views on abortion access chose to add to the opposing
with their political party. In the control condition, participants were asked to side rather than subtract from their own and that one of those participants had said
write about what they do on a typical Monday evening. the following: “I care way too much about my cause to take money away from it.”
Subsequently, all participants were asked to make a lose-lose choice, similar In the norm-subtract condition, participants were instead told that 70% of previous
to Study 1—participants had to choose between adding $1 to the donation going participants on their side of the cause had chosen to subtract from their in-group
to the organization supporting the opposing political party or subtracting $1 from rather than add to the opposing group. The statement from the previous participant
the donation going to the organization supporting their political party. We also was changed to: “I dislike the other side way too much to give them money.”
specified that all donations would go to administrative costs (e.g., maintaining As with our previous studies, we also asked participants to indicate which of
the organization's website). the two sides of the cause they believe is more effective at pursuing its mission,
As with our previous studies, we also asked participants whether they believe using a six-point scale.
Republican or Democratic organizations are more effective at pursuing their mis-
Data, Materials, and Software Availability. All data, analysis code,
sion, using the same scale as Studies 1 and 2, except with six points.
research materials, and pre-registrations have been deposited in Open Science
Study 5. Study 5 was conducted in March 2020. As outlined in our pre-regis- Framework (Center for OpenScience) (https://osf.io/gzxke/?view_only=ea4d-
tration (https://aspredicted.org/cz2kf.pdf), we aimed to recruit 650 U.S. MTurk 7c32e4b3499487d1c661fd​4b5493).
participants and ended up with a sample of 653 participants who completed
ACKNOWLEDGMENTS. We would like to thank C. Dorison, E. VanEpps, D.
the study (55% female, mean age = 36.18 y). We excluded 2.8% of participants
Tannenbaum, A. Gneezy, O. Amir, W. Liu, and L. Ordóñez for providing feedback
who failed the reading check or had duplicate MTurk IDs, leaving us with a final
on the manuscript.
sample of 635 participants.
All participants reported the extent to which they are against or in favor of
abortion access on a 12-point Likert scale (“very much against abortion access”
1. H. Tajfel, M. G. Billig, R. P. Bundy, C. Flament, Social categorization and intergroup behaviour. Eur. J.
to “very much in favor of abortion access”). Participants were then randomized Soc. Psychol. 1, 149–178 (1971).
into one of three conditions: control, norm-add, or norm-subtract. 2. J. C. Turner, Social comparison and social identity: Some prospects for intergroup behaviour. Eur. J.
In the control condition, participants were informed that the experimenter would Soc. Psychol. 5, 1–34 (1975).
3. D. E. Abrams, M. A. Hogg, Social Identity Theory: Constructive and Critical Advances (Springer-Verlag
be making donations to a pro-life and a pro-choice organization, and that they would Publishing, 1990).
have to choose how to alter the amount—add $1 to the donation going to the oppos- 4. M. A. Hogg, "Social identity theory" in Understanding Peace and Conflict through Social Identity
ing side or subtract $1 from the donation going to their side. The norm-add condition Theory (Springer, 2016), pp. 3–17.

PNAS 2022 Vol. 119 No. 49 e2215633119 https://doi.org/10.1073/pnas.2215633119 11 of 12

 5. N. Halevy, G. Bornstein, L. Sagiv, “In-group love” and “out-group hate” as motives for individual 37. D. DeSteno, Y. Li, L. Dickens, J. S. Lerner, Gratitude: A tool for reducing economic impatience. Psychol.
participation in intergroup conflict: A new game paradigm. Psychol. Sci. 19, 405–411 (2008). Sci. 25, 1262–1267 (2014).
6. N. Halevy, O. Weisel, G. Bornstein, “In-group love” and “out-group hate” in repeated interaction 38. C. A. Dorison et al., Sadness, but not all negative emotions, heightens addictive substance use. Proc.
between groups. J. Behav. Decis. Mak. 25, 188–195 (2012). Natl. Acad. Sci. U.S.A.117, 943–949 (2020).
7. J. Z. Berman, D. Kupor, Moral choice when harming is unavoidable. Psychol. Sci. 31, 1294–1301 39. M. A. Hogg, J. R. Smith, Attitudes in social context: A social identity perspective. Eur. Rev. Soc.
(2020). Psychol. 18, 89–131 (2007).
8. L. J. Volz, B. L. Welborn, M. S. Gobel, M. S. Gazzaniga, S. T. Grafton, Harm to self outweighs benefit to 40. J. R. Smith, M. A. Hogg, "Social identity and attitudes" in Attitudes and Attitude Change, W. Crano, R.
others in moral decision making. Proc. Natl. Acad. Sci. U.S.A. 114, 7963–7968 (2017). Prislin, Eds. (Pyschology Press, New York, 2008), pp. 337–360.
9. H. C. McGhee, The Sum of Us: What Racism Costs Everyone and How We Can Prosper Together. 41. R. A. LeBoeuf, E. Shafir, J. B. Bayuk, The conflicting choices of alternating selves. Organ. Behav. Hum.
Racism Has a Cost for Everyone (One World/Ballantine, 2021). Decis. Process. 111, 48–61 (2010).
10. C. K. Morewedge, S. Tang, R. P. Larrick, Betting your favorite to win: Costly reluctance to hedge 42. E. Van der Werff, L. Steg, K. Keizer, Follow the signal: When past pro-environmental actions signal
desired outcomes. Manag. Sci. 64, 997–1014 (2018). who you are. J. Environ. Psychol. 40, 273–282 (2014).
11. C. Friedersdorf, The Conservatives who'd rather die than not own the Libs. The Atlantic, 24 43. P. Rozin, E. B. Royzman, Negativity bias, negativity dominance, and contagion. Pers. Soc. Psychol.
September 2021. https://www.theatlantic.com/ideas/archive/2021/09/breitbart-conservatives-john- Rev. 5, 296–320 (2001).
nolte-vaccine/620189/). Accessed 22 February 2022. 44. M. B. Brewer, In-group bias in the minimal intergroup situation: A cognitive-motivational analysis.
12. S. Benen, GOP Starts to Pay a Price for Opposing, then Promoting, Relief Bill. UpToDate (2014). Psychol. Bull. 86, 307 (1979).
https://www.msnbc.com/rachel-maddow-show/maddowblog/gop-starts-pay-price-opposing-then- 45. M. Hewstone, M. Rubin, H. Willis, Intergroup bias. Annu. Rev. Psychol. 53, 575–604 (2002).
promoting-relief-bill-n126. Accessed 22 February 2022. 46. M. Tonin, M. Vlassopoulos, Experimental evidence of self-image concerns as motivation for giving. J.
13. L. Van Boven, P. J. Ehret, D. K. Sherman, Psychological barriers to bipartisan public support for Econ. Behav. Organ. 90, 19–27 (2013).
climate policy. Perspect. Psychol. Sci. 13, 492–507 (2018). 47. R. Bodner, D. Prelec, Self-signaling and diagnostic utility in everyday decision making. Psychol. Econ.
14. G. A. Akerlof, R. E. Kranton, Economics and identity. Q. J. Econ. 115, 715–753 (2000). Decis. 1, 26 (2003).
15. J. B. Kessler, K. L. Milkman, Identity in charitable giving. Manag. Sci. 64, 845–859 (2018). 48. A. Gneezy, U. Gneezy, G. Riener, L. D. Nelson, Pay-what-you-want, identity, and self-signaling in
16. W. B. Swann Jr., Á. Gómez, J. F. Dovidio, S. Hart, J. Jetten, Dying and killing for one’s group: markets. Proc. Natl. Acad. Sci. U.S.A. 109, 7236–7240 (2012).
Identity fusion moderates responses to intergroup versions of the trolley problem. Psychol. Sci. 21, 49. E. Fehr, U. Fischbacher, Why social preferences matter–the impact of non-selfish motives on
1176–1183 (2010). competition, cooperation and incentives. Econ. J. 112, C1–C33 (2002).
17. S. Hitlin, Values as the core of personal identity: Drawing links between two theories of self. Soc. 50. D. Ariely, A. Bracha, S. Meier, Doing good or doing well? Image motivation and monetary incentives
Psychol. Q. 66, 118–137 (2003). in behaving prosocially. Am. Econ. Rev. 99, 544–555 (2009).
18. R. Catapano, Z. L. Tormala, Do I support that it’s good or oppose that it’s bad? The effect of support- 51. M. Deutsch, H. B. Gerard, A study of normative and informational social influences upon individual
oppose framing on attitude sharing. J. Pers. Soc. Psychol. 121, 23–42 (2021). judgment. J. Abnorm. Psychol. 51, 629 (1955).
19. C.-B. Zhong, K. W. Phillips, G. J. Leonardelli, A. D. Galinsky, Negational categorization and intergroup 52. J. Andreoni, R. Petrie, Public goods experiments without confidentiality: A glimpse into fund-raising.
behavior. Pers. Soc. Psychol. Bull. 34, 793–806 (2008). J. Public Econ. 88, 1605–1623 (2004).
20. R. B. Cialdini, Crafting normative messages to protect the environment. Curr. Dir. Psychol. Sci. 12, 53. S. J. Chaudhry, G. Loewenstein, Thanking, apologizing, bragging, and blaming: Responsibility
105–109 (2003). exchange theory and the currency of communication. Psychol. Rev. 126, 313–344 (2019).
21. H. Allcott, Social norms and energy conservation. J. Public Econ. 95, 1082–1095 (2011). 54. M. W. Baldwin, J. G. Holmes, Salient private audiences and awareness of the self. J. Pers. Soc.
22. B. Borsari, K. B. Carey, Descriptive and injunctive norms in college drinking: A meta-analytic Psychol. 52, 1087 (1987).
integration. J. Stud. Alcohol 64, 331–341 (2003). 55. H. A. Sackeim, R. C. Gur, Self-deception, other-deception, and self-reported psychopathology. J.
23. M. A. Hogg, S. A. Reid, Social identity, self-categorization, and the communication of group norms. Consult. Clin. Psychol. 47, 213 (1979).
Commun. Theory 16, 7–30 (2006). 56. P. E. Tetlock, Thinking the unthinkable: Sacred values and taboo cognitions. Trends Cogn. Sci. 7,
24. A. S. Gerber, T. Rogers, Descriptive social norms and motivation to vote: Everybody’s voting and so 320–324 (2003).
should you. J. Politics 71, 178–191 (2009). 57. P. E. Tetlock, O. V. Kristel, S. B. Elson, M. C. Green, J. S. Lerner, The psychology of the unthinkable: Taboo
25. S. Davidai, M. Ongis, The politics of zero-sum thinking: The relationship between political ideology trade-offs, forbidden base rates, and heretical counterfactuals. J. Pers. Soc. Psychol. 78, 853 (2000).
and the belief that life is a zero-sum game. Sci. Adv. 5, eaay3761 (2019). 58. K. White, D. W. Dahl, To be or not be? The influence of dissociative reference groups on consumer
26. C. L. Wilkins, J. D. Wellman, L. G. Babbitt, N. R. Toosi, K. D. Schad, You can win but I can’t lose: Bias preferences. J. Consum. Psychol. 16, 404–414 (2006).
against high-status groups increases their zero-sum beliefs about discrimination. J. Exp. Soc. 59. K. White, D. W. Dahl, Are all out-groups created equal? Consumer identity and dissociative influence.
Downloaded from https://www.pnas.org by 60.12.158.39 on February 28, 2023 from IP address 60.12.158.39.

Psychol. 57, 1–14 (2015). J. Consum. Res. 34, 525–536 (2007).

27. G. Bornstein et al., On the measurement of social orientations in the minimal group paradigm. Eur. 60. R. B. Cialdini, R. R. Reno, C. A. Kallgren, A focus theory of normative conduct: Recycling the concept
J. Soc. Psychol. 13, 321–350 (1983). of norms to reduce littering in public places. J. Pers. Soc. Psychol. 58, 1015 (1990).
28. A. I. Abramowitz, S. Webster, The rise of negative partisanship and the nationalization of US 61. N. J. Goldstein, R. B. Cialdini, V. Griskevicius, A room with a viewpoint: Using social norms to
elections in the 21st century. Elect. Stud. 41, 12–22 (2016). motivate environmental conservation in hotels. J. Consum. Res. 35, 472–482 (2008).
29. A. I. Abramowitz, S. W. Webster, Negative partisanship: Why Americans dislike parties but behave 62. D. T. Miller, D. A. Prentice, Changing norms to change behavior. Annu. Rev. Psychol. 67, 339–361
like rabid partisans. Political Psychol. 39, 119–135 (2018). (2016).
30. E. J. Finkel et al., Political sectarianism in America. Science 370, 533–536 (2020). 63. R. B. Cialdini et al., Managing social norms for persuasive impact. Soc. Influ. 1, 3–15 (2006).
31. A. Tversky, D. Kahneman, Advances in prospect theory: Cumulative representation of uncertainty. J. 64. Pew, Partisan antipathy: More intense, more personal (Pew, 2019).
Risk Uncertain. 5, 297–323 (1992). 65. M. Dimock, R. Wike, America is exceptional in its political divide (29 March 2021).
32. A. Tversky, D. Kahneman, Loss aversion in riskless choice: A reference-dependent model. Q. J. Econ. 66. E. Klein, Why We’re Polarized (Simon and Schuster, 2020).
106, 1039–1061 (1991). 67. P. M. Fernbach, L. Van Boven, False polarization: Cognitive mechanisms and potential solutions.
33. H. Wickham, W. Chang, M. H. Wickham, Package ‘ggplot2’, Create Elegant Data Visualisations Using Curr. Opin. Psychol. 43, 1–6 (2021).
the Grammar of Graphics (Version 2, Springer Cham, New York, 2016). 68. J. R. Chambers, R. S. Baron, M. L. Inman, Misperceptions in intergroup conflict: Disagreeing about
34. G. M. Becker, M. H. DeGroot, J. Marschak, Measuring utility by a single-response sequential method. what we disagree about. Psychol. Sci. 17, 38–45 (2006).
Behav. Sci. 9, 226–232 (1964). 69. G. Heltzel, Seek and ye Shall Be Fine: Attitudes towards Political Perspective-Seekers (University of
35. S. Andersen, G. W. Harrison, M. I. Lau, E. E. Rutström, Elicitation using multiple price list formats. British Columbia, 2019).
Exp. Econ. 9, 383–405 (2006). 70. D. M. Bruner, Multiple switching behaviour in multiple price lists. Appl. Econ. Lett. 18, 417–420
36. J. S. Lerner, Y. Li, E. U. Weber, The financial costs of sadness. Psychol. Sci. 24, 72–79 (2013). (2011).

12 of 12 https://doi.org/10.1073/pnas.2215633119 pnas.org