The co-occurrence of ingroup and outgroup prosociality across 121 societies

Significance

A common premise in the social and behavioral sciences is that ingroup love goes together with outgroup hate. To test this premise, we studied whether ingroup prosociality goes together with less or more outgroup prosociality. Across six datasets that span multiple domains of prosociality and group distinctions with 743,402 individuals in 121 societies, we find strong and robust evidence for a positive relationship between ingroup and outgroup prosociality. Although people are slightly more prosocial toward their ingroup than the outgroup, those who are prosocial toward the ingroup also tend to be prosocial toward the outgroup.

Abstract

In an increasingly interconnected world, prosociality across group boundaries becomes ever more important. Yet, a common premise in the behavioral and social sciences is that people are prosocial mainly toward their ingroup and not outgroups. So far, evidence for this premise comes mostly from studies in which people must choose between ingroup and outgroup prosociality, thereby precluding positive associations between the two types of prosociality. We study situations in which people make separate decisions regarding the ingroup and outgroup, allowing us to examine how ingroup and outgroup prosociality are related. Across six large-scale and cross-societal datasets spanning 743,402 individuals in 121 societies, we find a robust positive relationship between ingroup and outgroup prosociality. This relationship holds across societies, different group categorizations, and a diverse range of prosociality measures, including prisoner’s dilemmas, public good dilemmas, dictator games, and survey items on trust and tolerance. Although people are slightly more prosocial toward their ingroup than the outgroup, those who are prosocial toward the ingroup also tend to be prosocial toward the outgroup. The results offer an optimistic message for prosociality across group boundaries. When people are not made to choose between the ingroup and outgroup, they are generally prosocial toward both.

Humans possess a strong capacity for prosociality, i.e., to engage in behaviors that benefit others. At the same time, this prosociality is often suggested to be bound by group membership. A large body of research shows that people are more willing to behave prosocially with members from their own group (ingroup prosociality) than with members from other groups (outgroup prosociality) (1–9). Indeed, human behavior is often suggested to be characterized by high ingroup prosociality and low outgroup prosociality (10–12). Evolutionary models of parochial altruism argue that this occurs because competition and war between groups in our ancestral past led to the selection of a preference for ingroup prosociality and outgroup derogation (12). Indeed, because a favorable position for the ingroup can be achieved both by ingroup prosociality and by outgroup derogation, these two behaviors are sometimes said to go together (13). Others argue that the likelihood and intensity of ancestral war between groups have been overstated (14, 15), suggesting that ingroup prosociality is independent of outgroup derogation. Still others argue that evolution instead led to a universal “cooperative phenotype” (16, 17), suggesting that ingroup and outgroup prosociality are positively related. In this study, we empirically address the issue of how ingroup and outgroup prosociality are related using large-scale data from multiple societies and contexts.

Although it is well established that people tend to behave more prosocially with their ingroup than the outgroup (1–8), this alone does not tell us how ingroup and outgroup prosociality are related. Ingroup favoritism can arise when i) ingroup prosocials tend to harm outgroups, ii) ingroup prosocials act neither favorably nor unfavorably toward outgroups, and iii) ingroup prosocials also behave prosocially with outgroups but at lower levels. Thus, ingroup favoritism can coexist with a negative, absent, or positive relationship between ingroup and outgroup prosociality. Knowledge of the relationship between ingroup and outgroup prosociality is crucial for understanding the challenges and opportunities of contact across group boundaries. If ingroup prosocials tend to be outgroup defectors, they exacerbate intergroup conflict and hamper universal prosociality. If, on the other hand, ingroup prosocials tend to be outgroup prosocials, they can be instigators of positive intergroup relations and universal prosociality.

Similarly, whether ingroup prosociality is positively or negatively related to outgroup prosociality has important implications for interventions to establish prosociality across group boundaries. If ingroup prosociality is negatively associated with outgroup prosociality, an implication is that more prosociality is achieved by making the ingroup as large as possible. Such logic underlies efforts to expand inclusion into ingroups (18–20) and efforts to establish superordinate groups (21–23). While such efforts to expand the ingroup may see more people benefiting from prosociality, they also run the risk of reducing diversity and its associated benefits. Indeed, some have warned against “buying social harmony with dull sameness” (24). Diversity may be regarded as good in its own right, and it also brings important benefits in terms of innovation and solving complex problems (25–27). If ingroup prosociality is positively rather than negatively associated with outgroup prosociality, then prosocial behavior across group boundaries becomes possible without sacrificing diversity—for example, by introducing positive interdependencies between groups or removing negative interdependencies.

Previous studies that find support for a negative association between ingroup and outgroup prosociality typically focus on contexts with preimposed negative interdependencies between groups. That is, these studies mostly examine environments in which people can only behave prosocially with the ingroup at the expense of the outgroup or by harming the outgroup (28–30). These studies are often suggested to show a human tendency for ingroup “love” and outgroup “hate” (13, 31–33). Other studies let people decide whether they want to behave prosocially toward their ingroup with or without harming the outgroup. These studies find that most people choose the form of ingroup prosociality that does not hurt the outgroup, suggesting that ingroup “love” need not coincide with outgroup “hate” (32–34). However, by design, these studies cannot detect the potential for a positive association between ingroup “love” and outgroup “love.” For this, people need to be given the opportunity to behave prosocially with both the ingroup and the outgroup, without any preimposed negative interdependencies between them. The few studies that have done so provide tentative evidence that ingroup and outgroup prosociality can be positively related (35, 36). However, these studies are still scarce and have had to rely on small samples of university students in a limited set of societies.

We study the relationship between ingroup and outgroup prosociality in six large-scale and cross-societal datasets. The datasets measure different dimensions of prosociality, including trust, tolerance, cooperation in prisoner’s dilemmas, giving in dictator games, and contributions in public goods games. We selected datasets that 1) span multiple societies and 2) involve people in each society providing separate responses regarding ingroup and outgroup members. This allows us to examine how these responses are related and whether this differs across societies. These datasets have been used in multiple studies for various purposes—e.g., to study ingroup favoritism—but not yet to study the co-occurrence of ingroup and outgroup prosociality. In the World Values Survey (WVS), we examine the relationship between ingroup and outgroup trust for 253,681 individuals in 95 societies (37). In the European Social Survey (ESS), we examine the relationship between ingroup and outgroup tolerance for 467,634 individuals in 39 societies (38). In a large-scale dataset on prisoner dilemmas (PD), we examine the relationship between ingroup and outgroup cooperation for 18,411 individuals in 42 societies (6). In a dataset on dictator games and random allocation games (DGRAG), we examine the relationship between ingroup and outgroup allocations for 1,647 individuals in 11 socioecologically and religiously diverse societies (7). Using data from multilevel public goods experiments (MPG) in 6 societies (United States, Italy, Russia, Argentina, South Africa, and Iran), we examine the relationship between ingroup and outgroup contributions for 1,115 individuals (39). Finally, in a cross-national dataset from 5 societies in Latin and North America (LANA), we examine the relationship between ingroup and outgroup giving in dictator games for 915 individuals (8).

The included datasets are not only diverse in the dimensions of prosociality but also in the types of ingroup and outgroups they include. In the WVS, the ingroup consists of families, neighbors, and people one knows personally, whereas the outgroup consists of foreigners, people from other religions, and strangers. In the ESS, the ingroup consists of people of the same race/ethnic group, and the outgroup of people of a different race/ethnic group. In the PD dataset, the ingroup consists of people from the same country, and the outgroup of people from a different country. In the DGRAG dataset, the ingroup consists of people of the same religion, and the outgroup of people from a different religion. In the MPG dataset, the ingroup consists of people from one’s local area, and the outgroup of people from outside of the local area. In the LANA dataset, the ingroup consists again of people from the same country and the outgroup of people from a different country. Altogether, the six datasets offer the opportunity to test the relationship between ingroup and outgroup prosociality across a comprehensive set of societies, prosociality dimensions, and group distinctions.

Results

The Coexistence of Ingroup and Outgroup Prosociality Across Contexts.

Fig. 1 presents the relationship between ingroup and outgroup prosociality across the six datasets. Although these datasets differ in the type of ingroups and outgroups, the domains of prosociality, and the sets of societies, we see that there is a strong positive relationship between ingroup and outgroup prosociality across all datasets. Those who behave more prosocially with the ingroup also tend to behave more prosocially with the outgroup and vice versa. The positive relationship is mostly linear, suggesting that it holds regardless of people’s general tendency for prosociality (i.e., it holds for people with low as well as high levels of prosociality). The average correlation between ingroup and outgroup prosociality is 0.40 (P < 0.001) in the WVS dataset, 0.69 (P < 0.001) in the ESS dataset, 0.83 (P < 0.001) in the PD dataset, 0.59 (P < 0.001) in the LANA dataset, 0.56 (P < 0.001) in the MPG dataset, and 0.65 (P < 0.001) in the DGRAG dataset.

Fig. 1.

The coexistence of ingroup and outgroup prosociality across contexts. We find a positive relationship between ingroup and outgroup prosociality across all contexts and group distinctions. In each panel, markers show the levels of ingroup and outgroup prosociality for all individuals, and the black line shows the fractional-polynomial prediction for the relationship between ingroup and outgroup prosociality and its 95% CI (in the first three panels, the number of observations is so large that the CI is barely visible). Markers are weighted by the number of individuals per dataset, meaning that larger markers indicate more individuals within the dataset. (A) ingroup and outgroup trust among 253,681 individuals in the WVS dataset. (B) national ingroup and outgroup cooperation in prisoner dilemmas among 18,411 individuals in the PD dataset. (C) ethnic ingroup and outgroup tolerance among 467,634 individuals in the ESS dataset. (D) religious ingroup and outgroup allocations among 1,647 individuals in the DGRAG dataset. (E) national ingroup and outgroup giving among 915 individuals in the LANA dataset. (F) locality ingroup and outgroup contributions across 1,115 individuals in the MPG dataset.

We see that the strength of the relationship between ingroup and outgroup prosociality differs between datasets. It is strongest for prisoner’s dilemmas with national groupings (PD dataset) and tolerance with ethnic groupings (ESS dataset), whereas it is lowest for trust with the ingroup consisting of families, neighbors, and people one knows personally and the outgroup consisting of foreigners, people from other religions, and strangers (WVS dataset). Establishing the causes of these differences across contexts requires new research. In the current study, the differences across datasets cannot be pinpointed to specific prosocial domains or group distinctions because the datasets differ from each other in multiple aspects that cannot be isolated from each other (e.g., the measure of prosociality, included societies, group distinctions, and methodology). The main point is that despite all these differences between the datasets, we find a positive relationship between ingroup and outgroup prosociality in each of them.

In line with earlier studies (on the same datasets and others) (1–8), we also see that there is more ingroup prosociality than outgroup prosociality. The difference between ingroup and outgroup prosociality has a Cohen’s d effect size of 1.50 in the WVS (t(253680) = 754.47, P < 0.001), of 0.47 in the ESS (t(467633)= 322.64, P < 0.001), of 0.14 in the PD dataset (t(18410)= 19.55, P < 0.001), of 0.55 in the LANA dataset (t(914)= 16.74, P < 0.001), of 0.70 in the MPG dataset (t(1,114)= 23.39, P < 0.001), and of 0.06 in the DGRAG dataset (t(1,646)= 2.59, P = 0.01). Hence, we find evidence for both a positive relationship between ingroup and outgroup prosociality and for ingroup favoritism. This means that those who behave more prosocially with the ingroup also behave more prosocially with the outgroup, but at lower levels for the outgroup.

To assess simultaneously the association and difference between ingroup and outgroup prosociality, we conduct explanatory factor analyses on people’s decisions regarding ingroup and outgroup members. For each of the six datasets, we conduct an explanatory factor analysis on people’s decisions toward ingroup and outgroup members (the number of decisions differs per dataset, see Methods). If the factor analyses point to 1 factor, they suggest that ingroup and outgroup prosociality are best seen as part of one universal prosocial tendency. If, instead, the factor analyses point to 2 factors—one for ingroup prosociality and one for outgroup prosociality—they are best seen as separate behavioral tendencies. In line with conventional standards, we retain factors that have an eigenvalue above 1 (using different criteria, such as the scree plot, yields the same results) (40, 41). The only exception is the ESS dataset, where people made only 1 decision toward the ingroup and only 1 decision toward the outgroup. If people make only 2 decisions, it is not statistically possible to obtain 2 factors with an eigenvalue above 1. Hence, for the ESS dataset, we instead examine the difference in the eigenvalues of the two factors to decide whether a 1-factor or 2-factor solution best fits the data.

The results of the factor analyses are shown in Table 1. We see that for each dataset, there is only 1 factor with an eigenvalue above 1. The second-highest factor does not come close to an eigenvalue of 1 in any dataset. Indeed, the eigenvalues of the second-highest factor are only a small fraction of the eigenvalue of the first factor. These results show that a one-factor solution fits best for each of the six datasets, suggesting that ingroup and outgroup prosociality are best seen as part of one universal prosocial tendency. Note, however, that exploratory factor analysis is a data-driven approach that summarizes the average tendency in the data. In upcoming analyses, we take a more theoretically driven approach that distinguishes between predefined behavioral types.

Table 1.

Eigenvalues from exploratory factor analyses on ingroup and outgroup decisions

Factor	WVS	PD	ESS	DGRAG	LANA	MPG
1	2.112	5.154	1.174	1.413	2.732	1.600
2	0.496	0.200	−0.213	0.152	−0.025	−0.083
3	−0.003	0.169	–	−0.065	−0.075	−0.187
4	−0.136	−0.069	–	−0.253	−0.089	–
5	−0.175	−0.080	–	–	–	–
6	−0.202	−0.087	–	–	–	–
7	–	−0.097	–	–	–	–
8	–	–0.107	–	–	–	–

Note: Eigenvalues resulting from exploratory factor analysis per dataset on the different ingroup and outgroup decisions. The number of ingroup and outgroup decisions—and thereby the maximum number of factors that can be extracted—differs per dataset. The decisions are described per dataset in the Methods.

The Coexistence of Ingroup and Outgroup Prosociality Across Societies.

To assess how ubiquitous the positive association between ingroup and outgroup prosociality is, we can assess for how many societies the correlation between ingroup and outgroup prosociality is positive and statistically significant. To this end, Fig. 2 presents the correlation between ingroup and outgroup prosociality per society for each of the six datasets. We see that the correlation is positive and significant for all societies in all six datasets. The correlations range from a minimum of 0.17 in Pakistan to a maximum of 0.55 in the Netherlands in the WVS dataset, from 0.73 in Pakistan to 0.91 in Japan in the PD dataset, from 0.29 in Israel to 0.89 in Spain in the ESS dataset, from 0.29 in Inland Tanna to 0.84 in Huatasani in the DGRAG dataset, from 0.50 in Peru to 0.82 in the United States in the LANA dataset, and from 0.40 in Russia to 0.63 in Iran in the MPG dataset. Altogether, the correlation is positive and significant in all the 121 different societies included in this study (Fig. 2 shows more than 121 correlations because some societies are included in multiple datasets).

Fig. 2.

The coexistence of ingroup and outgroup prosociality across societies. We find a positive relationship between ingroup and outgroup prosociality across all societies in each dataset. In each panel, horizontal bars show the correlation between ingroup and outgroup prosociality with 95% CI. (A) correlations between ingroup and outgroup trust across 95 societies of the WVS dataset. (B) correlations between ingroup and outgroup cooperation in prisoner dilemmas across 42 societies in the PD dataset. (C) correlations between ethnic ingroup and outgroup tolerance across 39 societies in the ESS dataset. (D) correlations between religious ingroup and outgroup allocations across 10 societies in the DGRAG dataset. For one society (Samburu), we did not have enough observations to reliably estimate a correlation (N = 39). (E) correlations between national ingroup and outgroup giving across 5 societies in the LANA dataset. (F) correlations between locality ingroup and outgroup contributions across 6 societies in the MPG dataset.

As with the previously shown cross-context variation, we also find significant variation across societies in how strongly ingroup and outgroup prosociality are correlated. For example, the correlation is relatively high in countries such as the Netherlands (which appears in three datasets; with rank 1 out of 95 in the WVS dataset, rank 9 out of 42 in the PD dataset, and rank 4 out of 39 in the ESS dataset), whereas it is relatively low in countries such as Pakistan (which appears in two datasets; with rank 95 out of 95 in the WVS dataset and rank 42 out of 42 in the PD dataset) and Türkiye (which appears in two datasets; with rank 83 out of 95 in the WVS dataset and rank 41 out of 42 in the PD dataset). The societies included in this study differ from each other in numerous aspects, such as culture, geography, economic conditions, demographic composition, and so on. It is beyond the scope of this research to identify what aspects of these societies drive their differences in the correlation between ingroup and outgroup prosociality, but we provide some indications in upcoming analyses. Just as with the cross-context analyses, the point of the cross-societal analyses is that even with all these differences across societies, there is a positive and significant correlation between ingroup and outgroup prosociality in each society. This suggests that the association between ingroup and outgroup prosociality may be a universal feature of contemporary human societies.

Potential Mechanisms for the Co-Occurrence of Ingroup and Outgroup Prosociality.

To obtain some indication of potential mechanisms involved in the co-occurrence of ingroup and outgroup prosociality, we conducted three types of additional analyses. These analyses and their results are summarized here and given in full in the SI Appendix, sections S3–S5. First, we distinguish between different behavioral types in line with related literature (11, 42, 43)—universal prosocials, parochial prosocials, and self-interested actors. Universal prosocials behave prosocially toward both the ingroup and the outgroup, parochial prosocials behave prosocially toward the ingroup but not the outgroup, and self-interested actors behave prosocially toward neither the ingroup nor the outgroup. Although people’s prosociality toward ingroups and outgroups typically follows a continuous scale instead of a binary classification, it can be informative to classify them into distinct behavioral types. For current purposes, it allows us to examine whether a society’s correlation between ingroup and outgroup prosociality is related to its composition of behavioral types using regression analyses. The results show that a society’s correlation is higher if its share of universal prosocials is larger, if its share of parochial prosocials is lower, and the share of self-interested actors does not seem to matter (SI Appendix, section S3). These results suggest universal prosociality may play a role in the positive correlation between ingroup and outgroup prosociality.

Second, we examine whether the perceived closeness between the ingroup and outgroup matters for the correlation in prosociality between them. In the LANA and DGRAG datasets, we have measures on the perceived closeness to the ingroup and the perceived closeness to the outgroup. This allows us to calculate the difference score in the perceived closeness toward the ingroup and outgroup. We then examine whether the correlation between ingroup and outgroup prosociality is higher among participants who perceive the ingroup and outgroup to be closer, which indeed appears to be the case (SI Appendix, section S4). Finally, we examine whether the societal correlation between ingroup and outgroup prosociality is related to several societal features that have been theorized to play a role in intergroup behaviors, such as kinship intensity (44, 45), historical exposure to the Western Church (7, 46), cultural norms (e.g., individualism) (47, 48) and practices (e.g., nepotism and particularism) (49–51), economic conditions (e.g., average income) (52), parasite stress (53), and the prevalence of various religions (54, 55). The results of these cross-societal analyses suggest that a higher societal correlation between ingroup and outgroup prosociality is positively related to historical exposure to the Western Church, to individualism, and (weakly) to average income (Supplementary Material S5). In contrast, kinship intensity, nepotism, and particularism are related to a lower societal correlation between ingroup and outgroup prosociality. Parasite stress and which religions are prevalent do not seem to matter for this societal correlation. We also find that the societal correlation between ingroup and outgroup prosociality predicts the societal prevalence of a real-life form of prosociality beyond group boundaries, namely voluntary blood donations to nonfamily members.

Discussion

A common premise in the social and behavioral sciences is that ingroup love goes together with outgroup hate. While previous research indeed shows that people behave prosocially with their ingroup when this requires harming outgroups (28–30), less is known about their willingness to behave prosocially with both ingroups and outgroups without such preimposed negative interdependencies between the two. We studied whether ingroup prosociality goes together with less or more outgroup prosociality in situations where there are no preimposed negative interdependencies between the two. Across six datasets that span multiple domains of prosociality and group distinctions with 743,402 individuals in up to 121 societies, we find strong evidence for a positive relationship between ingroup and outgroup prosociality. Regardless of which prosocial domain, group distinction, or society is tested, the positive relationship appears. Thus, when people are not made to choose between the ingroup and outgroup, ingroup love and outgroup hate no longer go together. Rather, ingroup love goes together with outgroup love.

These results also speak to the academic debate about whether humans are best seen as parochial cooperators whose prosociality is mainly limited to the ingroup or as universal cooperators who behave prosocially across group boundaries (5, 6, 11, 12, 42, 56, 57). The results are not in line with theories of parochial altruism that posit that prosociality is ingroup-bounded and coevolved with outgroup hostility (12). Nor are they in line with theories that posit that prosociality is ingroup-bounded and essentially unrelated to the outgroup (58). The results are more in line with theories of a “cooperative phenotype” or social value orientation that posit that individuals vary in the extent to which they are prosocial to others in general (59–61), with this prosociality not being limited to the ingroup. Indeed, previous research finds that people with a prosocial orientation are also more likely to display prosocial behaviors toward outgroups, for example, in the form of helping refugees (62) or helping victims of a natural disaster (63).

However, the results do provide nuance to the view of universal prosociality. While people who behave prosocially with their ingroup also tend to behave prosocially toward outgroups, they do so at somewhat lower levels toward the outgroup. If we were to summarize the average prosocial tendency observed in our data, we could call it imperfect universal prosociality. This is similar to how some people are imperfect conditional cooperators (64, 65), whose cooperation increases with others’ cooperation but still falls somewhat below others’ cooperation. Imperfect universal prosociality then describes people who behave prosocially to both the ingroup and outgroup, but whose outgroup prosociality falls somewhat below their ingroup prosociality. There is of course strong variability in human behavior; next to summarizing average tendencies, it is also informative to investigate different behavioral types. When zooming in on these differences, we find a mix of self-interested actors, parochial prosocials, and universal prosocials across contexts and societies. However, our findings suggest that the average prosocial tendency is closer to the notion of imperfect universal prosociality than parochial prosociality.

While the co-occurrence of ingroup and outgroup prosociality is visible across different contexts and societies, its strength varies. Our analyses provide tentative indications for contextual and societal factors that may explain this variation. In line with suggestions in the literature (18, 66), we find that the co-occurrence of ingroup and outgroup prosociality is larger when the perceived differences between the ingroup and outgroup are smaller. The cross-societal variation in the co-occurrence of ingroup and outgroup prosociality may depend, among others, on the society’s culture (e.g., individualism vs. collectivism, norms of loyalty) (67, 68), institutions that facilitate cooperation across group boundaries (e.g., rule of law, antinepotism laws) (44), and economic conditions (e.g., standards of living, economic expansion versus recession) (57, 69). In line with some of these predictions, our results suggest that the co-occurrence of ingroup and outgroup prosociality is larger in societies with higher individualism, income, and historical exposure to the Western Church. In contrast, the co-occurrence is lower in societies with higher nepotism, particularism, and kinship intensity. However, these results are based on correlational, cross-sectional evidence. Future research is needed to provide more definitive conclusions on the contextual and societal factors that drive or moderate the co-occurrence of ingroup and outgroup prosociality, for example, by employing longitudinal designs or by exploiting exogenous variation (70, 71). It would also be interesting to examine whether this co-occurrence in the prosocial domain generalizes to other behavioral domains, such as competitive behavior. Prior research suggests that differences in people’s prosociality toward ingroups and outgroups are largely independent of differences in their competitive behavior toward ingroups and outgroups (72).

To be able to assess whether people who behave prosocially to the ingroup also behave prosocially to the outgroup, we relied on datasets in which decisions regarding the ingroup and outgroup are made by the same individual, at the same time, and with the same method. There are competing expectations on how such design features affect the association between different decisions of the same individual. On the one hand, common method bias in such designs may lead the decisions to be more similar than they would be otherwise (73), potentially leading to an overestimation of the correlation between ingroup and outgroup prosociality. On the other hand, these designs can lead to demand effects in which participants infer that the researcher expects them to make different decisions regarding the ingroup and outgroup and comply with this expectation (74, 75), leading the decisions to be more different than they would otherwise be. It is unclear which of these forces is stronger, but neither common method bias nor demand effects would be able to explain our observed heterogeneity in the behavioral types of universal prosocials, parochial prosocials, and self-interested actors. That the ingroup and outgroup measures are not temporally separated can also affect the correlation between them (76), for example, if participants who are in a good mood may be more prosocial in their decisions toward both the ingroup and outgroup. However, if the lack of temporal separation drives the results, we should not observe as we do that the societal co-occurrence of ingroup and outgroup prosociality predicts the societal prevalence of a real-life form of prosociality beyond group boundaries, namely voluntary blood donations to nonfamily members. Nevertheless, future research can improve in this regard by examining situations in which individuals make the ingroup and outgroup decisions further apart in time and/or in different behavioral domains.

Our findings have implications for intergroup relations within and across societies. It is known that intergroup conflicts are pervasive in situations where the goals of the ingroup and outgroup are negatively interdependent. In warfare, for example, territory gained by one country means territory lost by another country. Or in elections, votes gained by one party mean votes lost by other parties. These are zero-sum situations where ingroup love and outgroup hate can go together, but that does not tell us what happens without such structural constraints. Our results on the positive relationship between ingroup and outgroup prosociality without preimposed interdependencies suggest that transforming zero-sum situations into non-zero-sum situations would promote universal prosociality. It is not even necessary to transform the situations into those where the ingroup and outgroup are positively interdependent, they only should not be negatively interdependent.

Yet, altering the interdependence structure of a situation is no easy feat. For example, it can require involvement from an overarching third party (77, 78), e.g., supranational organizations applying sanctions for those involved in international conflicts, or national governments inducing shared goals among different local groups. However, altering social structures may not even be necessary. Research shows that people often perceive the goals of their ingroup to be more opposed to those of outgroups than they actually are (79, 80). This is also called zero-sum bias: judging a situation to be zero-sum when it is actually non-zero-sum. For example, some native-majority members have zero-sum competition beliefs toward immigrants that are not based in reality, and these perceptions are related to negative immigration attitudes (81). If such zero-sum beliefs can be changed—for example, by providing accurate information on ingroup and outgroup goals or by highlighting shared goals—perceived negative interdependencies can be reduced (82–85). Our results suggest that without such negative interdependencies, prosociality toward the ingroup goes together with prosociality toward outgroups.

Methods

First, we use the Word Values Survey (WVS), which consists of nationally representative samples from 95 countries (37). The WVS contains items regarding ingroup trust and outgroup trust. Following previous research (44), we measure ingroup trust by taking the average of people’s reported trust toward their families, neighbors, and people they know (“How much do you trust: [Your family/Your neighborhood/People you know personally]”). Answer categories range from 1 “Trust completely” to 4 “Do not trust at all,” which we reverse-scored such that higher scores indicate more trust. Similarly, we measure outgroup trust by averaging people’s reported trust toward foreigners, people from other religions, and people they have met for the first time (“How much do you trust: [People you meet for the first time/People of another religion/People of another nationality]”). Again, answer categories range from 1 “Trust completely” to 4 “Do not trust at all,” which we reverse-scored such that higher scores indicate more trust. The questions are asked in the order that we presented them. The WVS employs random probability samples that are representative of the adult population. The main method of data collection is face-to-face interviews at the respondent’s place of residence. We use all rounds of the WVS that include the ingroup and outgroup trust measures (which are rounds 5 to 7).

Second, we use the European Social Survey (ESS), which consists of nationally representative samples from 39 countries around Europe (38). We measure ingroup tolerance by whether respondents would “Allow many/few immigrants of same race/ethnic group as majority” and outgroup tolerance by whether respondents would “Allow many/few immigrants of different race/ethnic group from majority.” Response categories for both items were 1) Allow many to come and live here, 2) Allow some, 3) Allow a few, and 4) Allow none. We reverse-scored the items such that higher scores indicate more tolerance. These items are used in existing research to study tolerance and prosocial tendencies toward the ingroup and outgroup (86–88). The questions are asked in the order that we presented them. The ESS employs random probability samples that are representative of all persons aged 15 and over. The main method of data collection is face-to-face interviews. We use all 10 rounds of the ESS.

Third, we use data from a study on prisoner’s dilemmas (PD) in 42 countries, with each country sample being representative of the country’s age, gender, and income (6). In each country, participants had to make decisions in 12 prisoner’s dilemmas. In each prisoner’s dilemma, they were paired with a partner, and each person was given an endowment of 10 monetary units. Each person in a pair had to decide how many monetary units to keep for themselves and how many to give to their partner. Each monetary unit given to the partner was doubled, meaning that the pair was collectively better off if each person gives all monetary units to their partner, but each person is individually better off by giving nothing to their partner. In 4 of the 12 prisoner’s dilemmas, participants were paired with a person from their own country (the ingroup), in another 4 prisoner’s dilemmas they were paired with a person from another country (the outgroup), and in the other 4 prisoner’s dilemmas they were paired with a person without information about their country. We take the average of the number of monetary units given across the 4 prisoner’s dilemmas played with a person from the same nationality as the ingroup prosociality score. We take the average of the number of monetary units given across the 4 prisoner’s dilemmas played with a person from a different nationality as the outgroup prosociality score. The prisoner dilemmas played with a person without knowing their country are not relevant to the current study. In 3 of the 42 countries, participants were randomly assigned to have their decisions be monetarily incentivized or not, but this incentivization had no effect on behavior (6).

Fourth, we use data from Dictator Games (DG) and Random Allocation Games (RAG) employed in 15 socioecologically and religiously diverse societies (7). In the DG, participants are given 10 monetary units and have to decide how to allocate them between two recipients. What varies between conditions is who the recipients are. For our purposes, we need conditions where participants must allocate the monetary units between themselves and an ingroup member or between themselves and an outgroup member. In 10 of the 15 societies (1,201 participants), people played 1) a version of the DG where they had to allocate the monetary units either to themselves or an ingroup member (a distant coreligionist) and 2) a version of the DG where they had to allocate the monetary units either to themselves or an outgroup member (somebody from a different religion). These are respectively the ingroup and outgroup measures for the DG.

In the RAG, participants must roll two-colored dice and allocate a monetary coin to one of two cups depending on the dice roll (i.e., each cup is matched with a color). They do this 30 times, such that—on average—the money should be evenly distributed among the cups. The recipients of each of the two cups are varied between several conditions. For our purposes, we need conditions where the participant is the recipient of one cup and the recipient of the other cup is either an ingroup or outgroup member. Crucially, the allocations are made in private, such that the participants can cheat by giving monetary coins to the cup of one’s preference regardless of which color comes up in the dice roll. Hence, the number of coins allocated to the cups can be used to measure ingroup and outgroup preferences in giving. In 9 of the 15 societies (516 participants), people played 1) a version of the RAG where they had to allocate coins between themselves and an ingroup member (a distant coreligionist) and 2) a version of the RAG where they had to allocate coins between themselves and an outgroup member (someone from a different religion). These are respectively the ingroup and outgroup measures for the RAG. We divided the RAG allocations by three such that the DGs and RAGs have the same range (from 0 to 10). To obtain one overall measure for ingroup giving, we take the average of the ingroup DG and ingroup RAG (for participants who only played one of these two games, we take the values from that game). Similarly, to obtain one overall measure for outgroup giving, we take the average of the outgroup DG and outgroup RAG. Altogether, 1,647 participants played either the DG, the RAG, or both and could thus be included in our analyses. Both games were monetarily incentivized (7).

Fifth, we use data from a study on dictator game giving in five Latin-American and North American (LANA) countries (8). Each country sample is representative of the country’s age and sex distribution. Each participant played four dictator games; one with a coparticipant from the same country and three with coparticipants from countries different than one’s own. In each dictator game, a participant was given 100 monetary units and had to allocate these between oneself and the coparticipant. The number of monetary units allocated in the dictator game with the coparticipant from the same country is the measure of ingroup prosociality. The average number of monetary units allocated in the three dictator games with coparticipants from countries different than one’s own is the measure of outgroup prosociality. The dictator games were monetarily incentivized (8).

Sixth, we use data from single-level and multilevel public goods games (MPG) in six countries (39). In each game, participants received 10 monetary tokens and had to decide how many of these to keep for themselves and how many to contribute to various accounts that also benefit other people. In the single-level public goods games, participants had to decide how many tokens to keep for themselves or to contribute to one account that gave each of four members from one’s local area 0.5 tokens (including oneself) per contributed token. This means that each individual participant earns most money by not contributing anything (individual payoff is 10 tokens but others earn nothing), but that the local community earns most money if participants contribute all tokens (individual payoff is 5 tokens but the other three participants together earn 15 tokens). The number of tokens contributed to the local community in the single-group public good is our measure of ingroup prosociality.

In the multilevel public goods game, there are two accounts that participants can contribute to. The first is a local account that benefits the local community, the second is a global account that also benefits people outside the local community. Once again, the local account gives each of four members from one’s local community 0.5 tokens (including oneself). The global account gives 12 members 0.25 tokens each (including oneself, three other members from one’s local community, and eight members from outside the local community). These returns of the accounts are such that participants benefit themselves most by not contributing anything, benefit their local community most by contributing to the local account, and benefit the whole collective of 12 members most by contributing to the global account. The number of tokens contributed to the global community is our measure of outgroup prosociality.

Note that in the multilevel public goods game, we cannot use the number of tokens contributed to the local community to measure ingroup prosociality as a decision separate from outgroup prosociality. This is because in the multilevel public goods game, there is a preimposed negative interdependency between the local account and the global account. Any token contributed in the multilevel public goods game to the local account cannot be contributed to the global account and vice versa. However, there is no preimposed negative interdependency between contributions to the local account in the single-level public goods game (our measure of ingroup prosociality) and contributions to the global account in the multilevel public goods game (our measure of outgroup prosociality), as these games are played separately. We can thus correlate these two types of contributions to assess the association between ingroup and outgroup prosociality without preimposed negative interdependencies between the two. Participants played two versions of the multilevel public goods game, in both versions the local account benefits only people from one’s local area, and the global account all people regardless of whether they belong to the local community. In one version, the global account includes all people from the same nationality (but outside of their own local area), in the other version the global account also includes people from other nationalities. Because contributions to these two global accounts are highly correlated (0.68, p < .001), we take the average of these two contributions. The contribution decisions were monetarily incentivized (39).

All reported p-values are two-sided. We use the convention of P < 0.05 to determine whether a result is statistically significant. The conclusions remain the same when we apply much stricter significance thresholds such as P < 0.001. In the Supplementary Materials, we provide all data points regarding the decisions toward ingroup members and outgroup members per society in each dataset (SI Appendix, Figs. S1–S6). We also provide all correlations between ingroup and outgroup prosociality—and the number of participants on which these correlations are based—per society in each dataset (SI Appendix, Tables S1–S6).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

All data are openly available. The WVS data are openly available at https://www.worldvaluessurvey.org/WVSEVStrend.jsp (89). The ESS data are openly available at https://www.europeansocialsurvey.org/data-portal (90). The PD data are openly available at https://osf.io/68wds/ (91). The LANA data are openly available at https://osf.io/76bzq/ (92). The DGRAG data are openly available at https://osf.io/axz2b/ (93). The MPG data are openly available at https://osf.io/ecmkr (94). All code and materials needed to reproduce the results are available at https://doi.org/10.17605/OSF.IO/RFYSQ (95).