Abstract
Competition between groups is widely considered to foster cooperation within groups. Evidence from laboratory experiments hints at the existence of a proximate mechanism by which humans increase their level of cooperation with their ingroup when faced with an external threat. Further work suggests that ingroup cooperation should go along with aggressive behaviour towards the outgroup, although these theories are at odds with others that see high investments in outgroup relations as important means of stabilizing intergroup relations. Surprisingly, few of these arguments have been tested in the field, and existing studies are also limited by the lack of a direct measure of threat perception and aggressive behaviour. This study presents lab-in-the-field results from a rural context where exposure to an ethnic outgroup varies between villages. This context makes it possible to capture levels of threat perception, aggressive behaviour and cooperation without inducing intergroup competition artificially in the laboratory. All concepts are measured behaviourally. In- and outgroup cooperation was measured with a standard public goods game, and a novel experimental protocol was developed that measures perceived threat and aggressive behaviour: the threat game. The results show that levels of perceived threat, ingroup cooperation and aggressive behaviour are higher in regions more strongly exposed to ethnic outsiders. However, exposed regions also show high levels of outgroup cooperation and a concomitant lack of elevated ingroup bias. This pattern is explained by theorizing that communities show parochial altruism when faced with an ethnic outgroup, but balance aggressive behaviour with cooperative offers to diffuse tensions and to keep open channels of mutually beneficial exchange.
1. Introduction
Humans frequently engage in cooperative behaviour that benefits group members despite a cost to themselves [1–3]. Examples range from food sharing to mutual defence in warfare [4,5]. Although the causes for this behaviour remain debated, threat and competition between groups figure prominently among possible explanations [6–8]. Costly cooperation benefitting group members may ultimately have been selected for in a history of violent intergroup conflicts, in parallel with a tendency to disfavour the outgroup [9,10].
In terms of proximate mechanisms, scholars have described and tested a psychological response by which individuals raise their level of cooperation with ingroup members when faced with competition from an outgroup [11,12]. Numerous laboratory experiments have confirmed that members of groups contribute more to collaborative tasks when exposed to intergroup competition or comparison than when interacting in isolation. This is true when pay-offs directly depend on the behaviour of another group [13–19], or when the presence of another group is merely mentioned [20,21] (but see [22]).
Theoretic work and laboratory studies also demonstrate that ingroup cooperation can go along with exclusionary or aggressive behaviour towards outgroup members (i.e. result in ‘parochial altruism’) [23], although here the evidence is more mixed [24,25], with other studies showing no close link between ingroup favouritism and aggression against outgroup members [15,26]. More generally, it is not clear to what extent these findings reflect dynamics in the field, as the limited evidence here is inconclusive, too.
While evidence from post-war settings suggests that the experience of violent conflict may leave people more cooperatively inclined [27,28], politically engaged [29] or egalitarian [30], it remains unclear what aspect of the war experience may have caused these effects. Studies that more closely observed cooperative behaviour under intergroup competition either relied on artificially formed groups and intentionally created competition [31], or lacked indicators to rigorously measure levels of cooperation [11]. What is more, in the analytic games used in laboratory research and in real-world violent conflicts, it is often not clear whether it is the perception of threat—the fear of being harmed in the competition—or the prospect of winning the competition that motivates cooperative behaviour [13,32].
Existing work often conceptualizes contentious intergroup relations as residential proximity to an ethnic outgroup. These studies show that outgroup presence does go along with exclusionary attitudes, outgroup discrimination and increased participation in collective political processes [33–36]. Field studies that measure the effect of conflict and threat at the individual level have produced mixed findings, with some scholars identifying positive effects on forms of altruistic behaviour towards the ingroup [37,38], while others merely find lower levels of cooperation with in- and outgroup members [39–41].
The present study adds to existing research in three ways. First, the study presents data from a field context where exposure to ethnic outgroups varies between villages and thus did not have to be externally induced. Second, rather than assuming that threat perceptions positively covary with outgroup exposure, this assumption is explicitly tested. Third, all concepts of interest—perceived threat, cooperation and aggressive behaviour—are measured at the individual level and with behavioural measures. A standard public goods game is used to measure cooperation, and a new experimental protocol, the threat game, was developed to measure perceived threat and aggressive behaviour. I find that, in line with prevailing theories, outgroup exposure is associated with increased levels of perceived threat, and produce suggestive evidence for the presence of parochial altruism. At odds with current scholarship, I also find high levels of cooperation with the outgroup. To make sense of these findings, I theorize that this latter result is due to communities balancing parochialism with cooperative offers to the outgroup in an effort to diffuse tensions and keep open mutually beneficial channels of exchange.
2. Material and methods
(a) Study setting and design
In order to test for the impact of outgroup threat on cooperation levels, we would ideally like to compare communities that are identical in terms of history, geography, socio-economic opportunities and residential patterns, but vary in the degree to which they are exposed to an outgroup. The present study approaches these idealized conditions by exploiting variation in outgroup exposure within the same sub-national region. The experiments were conducted in six villages in the Kvemo Kartli region, Georgia. The Kvemo Kartli region is a compact (6528 km2), rural region in the vicinity of the country's capital, Tbilisi. The region is highly ethnically diverse, with ethnic Georgians constituting a numerical minority. In 2002, the year of the last census, 44.7% of the region's 497 530 inhabitants were ethnic Georgians and a similar share of the population (45.1%) were ethnic Azerbaijanis. Among the smaller minority groups, ethnic Armenians form the largest group, at 6.3% (see the electronic supplementary material for more information on the ethnic composition). Unlike other regions in Georgia, Kvemo Kartli has seen no sustained interethnic violence. This means that ethnic settlement patterns have remained relatively stable and have not been upset by large-scale displacement as elsewhere in the country [42]. However, intergroup relations in Kvemo Kartli are not free of tensions. Groups regularly compete over resources such as access to farmland, and in the past years, there have been violent clashes between ethnic Azerbaijanis and ethnic Georgians that left several people injured and one dead [43].
In the Kvemo Kartli region, ethnic diversity exists largely between villages, with individual village-communities tending to be ethnically homogenous. Moreover, there is within-regional variation in settlement patterns, as shown in figure 1a. This settlement pattern allows for close comparison of cooperation, threat perceptions and aggressive behaviour between communities exposed to a high or low share of ethnic outsiders in their surroundings.
(a) Villages in the Kvemo Kartli region, Georgia, by ethnic composition. Villages in which lab-in-the-field sessions were conducted marked with larger symbols. (b) Share of ethnic outsiders in the five villages closest to each village where a lab-in-the-field session was conducted. Villages in the low outgroup-exposure condition marked with blue-outlined bars, villages in the high outgroup-exposure condition marked with red bars.
Research was conducted in six ethnically Georgian villages, and outgroup exposure was measured as the share of non-Georgians in the population of the five villages closest to each village where a lab-in-the-field session was conducted. Based on these shares, villages were classified into a low outgroup-exposure condition (with greater than 50% ethnic Georgians in the surroundings) and a high outgroup-exposure condition (with less than or equal to 50% ethnic Georgians in the surroundings; figure 1B). In each of the six villages, between 20 and 28 participants took part in the experimental sessions, totalling 71 in the three villages in the low outgroup-exposure condition, and 69 in the three villages in the high outgroup-exposure condition. Villages in the two conditions were closely matched on a range of geographic and demographic indicators (see electronic supplementary material, table S1 for statistics and tests).
(b) Behavioural games
In order to measure threat perception and aggressive behaviour, a dedicated analytic game was developed: the threat game (more detailed information about the game is provided in the electronic supplementary material). In the threat game, participants are asked to make costly decisions that reveal how much they fear predatory behaviour from their interaction partner. The game captures the asymmetric nature of a typical threatener–threatened interaction where the more weakly positioned ‘threatened’ fears a dominant ‘threatener’. The threat game consists of two roles: P, the potential predator or threatener and T, the threatened. Both P and T are given an initial endowment of 20 monetary units (MUs). They are then asked to take decisions in parallel. P is asked to decide how many MUs she wants to take away from T's endowment in order to add to her endowment. P can claim anything between 0 MUs and 15 MUs. Each 1 MU that P manages to take away causes an additional loss of 1 MU to T, meaning that T can end up with a negative pay-off. The amount that P could claim was capped at 15 MUs rather than the theoretically possible 20 MUs in order to prevent participants from encountering excessive losses. This additional loss generates many of the advantageous properties of the threat game detailed in the electronic supplementary material. For explanatory purposes, while introducing the game the additional loss was likened to a burglary, where, in trying to steal valuables, the burglar also causes damage to the house. The amount claimed by a player in the role of P, therefore, is a behavioural measure of aggressive intent.
By contrast, the task of the threatened T is to estimate how much P might take away from his initial endowment, and then to decide how much of his initial endowment to spend on his protection. In his attempt to avoid losses or incurring negative pay-offs, T can spend parts or all of his endowment on his protection. Each 1 MU that T spends reduces the amount that P can take away by 1 MU, and also prevents the loss of the additional 1 MU that this would entail. However, the MUs T spends on his protection are non-recoverable, no matter the amount P actually claims. It is, therefore, optimal for T to spend exactly the amount on protection that P intends to take away. The number of MUs spent by T, therefore, is a behavioural measure of perceived threat. Unlike in the contest games often used to analyse cooperation under competition and threat, in the threat game, the threatened individual cannot gain, meaning that the wish to win over the threatener can be ruled out as a motivation.
Cooperation was measured with a standard public goods game (PGG). For the PGG, session participants were anonymously and at random divided into groups of 3. Each group member received an endowment of 10 MUs and was then asked to decide to invest parts, all or none of this endowment into a public account. Whatever amount was invested in the public account was doubled and evenly distributed among all group members. From this pay-off structure, it follows that while the group income is maximized when all group members contribute their whole endowment, for any given level of contributions by the other group members, an individual can always be better off by not contributing. The amount of MUs a participant invested in the public account is the behavioural measure of cooperation.
During the sessions, participants first played the PGG and then the threat game. This order was chosen so to avoid exposing them to the competitive set-up of the threat game before assessing their level of cooperation. In the PGG, participants made two decisions in randomized order, one when randomly matched with two co-villagers in the same room (i.e. coethnic co-villagers), and one when matched with two people from neighbouring villages, who could either be coethnics or not, depending on the exposure condition. In the threat-game, participants first decided twice in the role of P (in random order), again either when randomly matched with another participant in the room, or when paired with a person from a neighbouring village. They then made two decisions in the role of T, again in random order.
For all tasks, the interaction with members of neighbouring villages was simulated in the following way. Participants were told that they would be randomly matched with one (in the threat game) or two (in the PGG) inhabitants of the five closest neighbouring villages, whose decisions had been recorded during an earlier session. In fact, pay-offs for all participants were calculated by matching their decisions with a set of PGG and threat game decisions recorded during a trial session in a single Georgian village. Matching the participants' decisions with a stable set of decisions ensured that the average pay-offs were equal in all six sessions. In order to obtain measures free of priming effects, at no point was the ethnicity of the inhabitants of neighbouring villages mentioned. Instead, it was assumed that participants were aware of the ethnic composition of neighbouring villages and could base their behaviour on this knowledge. Information on outcomes was revealed only after all decisions had been taken. Detailed explanations and a comprehension test preceded the decision-making, and a survey was used to collect demographic information. A show-up fee ensured that participants would not incur overall negative pay-offs. Sessions lasted 90–120 min and on average participants earned 18 Georgian Lari (approx. USD 8/Euros 7), 80% of the average daily wage of an employee in agriculture or education.
3. Results
The goals of the analysis are (i) to assess whether outgroup exposure goes along with increased threat perceptions and (ii) to test for concomitant parochial altruism—increased levels of ingroup cooperation and bias in high-outgroup exposure villages, accompanied by aggressive behaviour towards outgroup members. The design of the study allows us to measure the concepts of interest—threat, cooperation and aggressive behaviour—in three ways. First, average absolute values for each game decision can be compared between the low- and the high-outgroup exposure condition. Of particular interest here are cooperation with co-villagers, threat perceived from neighbours and aggressive behaviour towards neighbours. Second, the data can be analysed in terms of differences in bias. Participants took all decisions when matched with co-villagers and neighbours. This allows us to compare the average differential between these two decisions between the low- and the high-outgroup exposure condition. To conduct the comparisons, regression models controlling for the unbalanced covariate, age and prior game decisions are used. To account for potential clustering at the village level and the small number of clusters/villages, standard errors are calculated with a wild cluster bootstrap procedure (5000 repetitions) [44]. Finally, the data allow us to analyse bias in terms of within-person differences in behaviour for participants interacting with co-villagers as opposed to neighbours, and to compare these within-person differences across the two exposure conditions. The within-person comparisons are evaluated using Wilcoxon's signed-ranks tests for clustered data (random permutation Datta–Satten method, two-tailed) [45,46].
(a) Outgroup exposure and threat
As illustrated in figure 2, participants in the high-outgroup exposure on average spent 59% of the theoretical maximum on their protection in the T-role of the threat game. This compares to the low-outgroup exposure condition, where they spent 52% (p = 0.07). Participants in the high-outgroup exposure condition also showed stronger bias in terms of threat against their non-coethnic neighbours, spending 8.8 percentage points (pp) more (1.17 times) when matched with neighbours than when matched with co-villagers, while participants in the low-outgroup exposure condition showed little bias, spending 3.3 pp more (1.06 times, 5.5 pp difference in bias, p = 0.08). Concordantly, within-person differences in spending on protection when matched with neighbours as opposed to co-villagers were no different in the low-outgroup exposure condition (p = 0.53), but clearly higher in the high-outgroup exposure condition (p < 0.001). As widely hypothesized in previous work, these tests demonstrate that exposure to a non-coethnic outgroup indeed went along with increased perceptions of threat.
Percentage of total possible MUs spent/claimed in the PGG and threat game in the low- (blue, triangle) as opposed to the high- (red, diamond) outgroup exposure condition, and difference in bias between the conditions. Depicted are the bias favouring co-villagers for cooperation, and bias against neighbours for threat and aggression. Marginal effects from wild cluster bootstrap regression, resampling at the village level, 5000 repetitions (electronic supplementary material, table S2).
(b) Parochial altruism
With regard to the first element of parochial altruism, ingroup cooperation and bias, we can see that overall levels of cooperation with co-villagers in the public goods game were higher in the high-outgroup exposure condition, where participants invested 73% of their endowment, as opposed to in the low-outgroup exposure condition, where they invested 63% of their endowment (p = 0.07). Compared to PGG decisions recorded in other contexts, where contributions between 40 and 60% of the endowment have typically been observed [3], these figures indicate remarkably high levels of cooperation, especially in the high-outgroup condition. However, there is no evidence for elevated bias in villages surrounded by non-coethnics. Participants favoured co-villagers over neighbours to virtually the same degree in both the high- and the low-outgroup exposure condition, giving 7.9 and 8.0 pp more (1.12 and 1.14 times) to co-villagers (0.1 pp difference in bias, p = 0.97). This is because in the high-outgroup exposure condition, higher cooperation rates with coethnic co-villagers went along with similarly elevated levels of cooperation with non-coethnic neighbours, with participants spending 65% of their endowment in the PGG, compared with 55% in the low-outgroup exposure condition (p = 0.25). This lack of elevated bias is discussed in more detail below.
With regard to the second aspect of parochial altruism, aggressive behaviour, levels appear to be higher in the high-outgroup exposure condition, although overall differences are not statistically significant. Participants in the high-outgroup exposure condition claimed 64% of their neighbour's endowment in the P-role of the threat game and made 10.6 pp (1.19 times) higher claims than when matched with co-villagers, while in the low-outgroup condition they claimed 57% (p = 0.22), and showed less (6.1 pp, 1.12 times) bias (4.8 pp difference in bias, p = 0.51). When considering within-person differences, however, clear differences between the exposure conditions are visible: while in the low outgroup exposure condition there were no differences in aggressive behaviour towards either co-villagers or neighbours (p = 0.36), in the high outgroup-exposure condition individuals behaved clearly more aggressively towards their non-coethnic neighbours (p = 0.001).
Theoretical models explain parochial altruism in terms of the signalling value of group coherence to deter attacks [12], and multilevel selection, whereby the negative individual fitness effects of heroic behaviour on the behalf of the group are overcompensated by the positive effects for the group [9]. According to the latter theory, intergroup conflict may have evolutionarily favoured parochial altruists that combine hostility to the outgroup with high levels of ingroup cooperation. We would, thus, expect to find a positive link between aggressive behaviour and cooperation at the individual level. The data at hand provide evidence for such a relationship. We can split the sample into aggressive individuals, who claim the median value of 10 MUs or more, and non-aggressive individuals, who spend less than this amount. When subdividing the sample in this way, 57% of individuals in the low outgroup exposure condition qualify as aggressive, compared with 67% in the high outgroup exposure condition. In line with the theoretical predictions, aggressive individuals overall invest somewhat more in the PGG with co-villagers, i.e. show a higher level of cooperation with ingroup members, although this difference does not reach conventional levels of significance (70%, compared with 64%, p = 0.13). They are, however, clearly more biased in favour of their co-villagers (13.3 pp bias, compared with 0.8 pp bias, p = 0.04). This result is confirmed when looking at within-person comparisons: while non-aggressive individuals do not distinguish between co-villagers and neighbours (p = 0.91), aggressive individuals are significantly more likely to favour their co-villagers (p = 0.003).
The data thus provide support for the hypothesis that communities exposed to a threatening outgroup show parochial altruism. This finding contrasts with those of other field studies, which find no support for parochial altruism in a context of real-world intergroup conflict in Belfast, Northern Ireland [39,40]. One possibility is that parochial altruism develops only in settings with high residential stability, as implied by the original model of Choi & Bowles [9]. Laboratory evidence shows that the option for free migration tends to suppress the cooperation-enhancing effect of intergroup conflict, potentially leading to smaller effects in urban than rural areas [47]. In this study, 61% of participants were born in the village where the experiment took place (35% of women and, due to a widely followed norm of patrilocality [48], 87% of men), which contrasts with the more fluid residential urban environment in which the cited studies were conducted. Residential stability may also explain the high overall cooperation rates observed [49].
(c) High outgroup cooperation/lack of elevated bias
As noted above, a surprising finding in the light of current scholarship is the high rates of cooperation with non-coethnic neighbours in the high-outgroup exposure condition, and the concomitant lack of elevated bias towards co-villagers. Here, two possible explanations for this novel finding are proposed. A first possibility is that outgroup cooperation serves to lower the probability of violent confrontation with neighbouring communities. In a context of weak institutional safeguards, it is possible for intergroup tensions to escalate into violence, which typically imposes high costs on all actors involved. One important reason for this, widely discussed in international relations theory, is the lack of information about the other groups' intentions, which can make escalation likely even if the actors involved would prefer to avoid violence [50,51]. Cooperative offers can help to communicate generally peaceful intentions, serve as signals of a group's intentions about future investment in a relationship, and hence lower the risk of violent confrontation [52,53]. Research in anthropology shows that groups, friendly and hostile, dedicate themselves to reciprocal gift-exchange relationships ostensibly to stabilize inter-group relations, and confidence-building measures between states serve a similar purpose [54,55]. It follows that in a context of tense intergroup relations, threat may induce cooperation with an outgroup rather than deter it. The relationship between spending on protection in the threat game and cooperation with neighbours in the PGG provides some support for this interpretation. In the low outgroup-exposure condition, where threat emanates from coethnics, there is a monotonic negative relationship between perceived threat (measured in terms of absolute spending on protection or bias) and above-average contributions in the PGG. This is the relationship we would expect to see based on the previous research showing that fearful and distrustful individuals are typically also less cooperative [56]. By contrast, in the high-outgroup exposure condition, where threat emanates from non-coethnics, a more complex pattern arises, with participants who perceive high levels of threat from their neighbours also cooperating at high levels with them (figure 3).
Probability of above-average contributions to neighbours in the PGG, depending on perceived threat as measured by T-role spending in the threat game in the low- (blue, triangle) as opposed to the high- (red, diamond) outgroup exposure condition. Marginal effects from logistic regression on the first- and third-order (best-fit) polynomial of T-role spending and bias in T-role spending. Shaded areas are 95% confidence intervals, standard errors clustered at the village level.
A second possibility that could explain the simultaneous presence of parochial altruism and lack of elevated bias is that individuals, even if parochially inclined, do not want to forfeit the benefits of exchange with their neighbours. This idea parallels findings in international relations where countries have been observed to continue trading even while at war [57]. Trading is particularly profitable across group lines, as groups are often economically differently specialized [58–60]. In the case of the Kvemo Kartli region, ethnic Azerbaijanians often have better access to sought-after trading products from Turkey, whereas ethnic Georgians find it easier to access markets elsewhere in Georgia, offering opportunities for mutually beneficial exchange. Tentative evidence for the interpretation that participants likened cooperation in the PGG to trade when interacting with outgroup members is present in the data. In the post-game questionnaire, participants were asked whether they approved of their co-villagers doing business with individuals from neighbouring villages. In the ethnically homogenous sub-region, there was no association between approval for business exchanges and cooperation (b = −0.08, p = 0.49), whereas in the high-outgroup exposure the association was substantial although not statistically significant (b = 1.28, p = 0.24).
While reciprocal exchange at first glance may seem at odds with parochial altruism, it may, in fact, be beneficial for individuals and communities to, at the same time, showcase aggressive behaviour to deter attacks, while lowering tensions and keeping channels of exchange open by means of inter-group cooperation. The strategy of combining parochial altruism with high levels of intergroup cooperation implies that the redistributive effects of the aggressive behaviour are strongly reduced. In the games presented here, the losses for non-coethnic neighbours due to the parochial strategy adopted in the threat game were almost exactly compensated by the additional gains from the more cooperative stance in the PGG ( 1.8 MUs versus 1.9MUs, p = 0.59), so that coethnic and non-coethnic neighbours received virtually identical average pay-offs.
4. Conclusion
The study presents evidence for parochial altruism in reaction to outgroup threat from a real-world setting. Parochial altruism is accompanied by high levels of outgroup cooperation/low levels of bias in a cooperation task. The latter result is novel and should be studied in more detail. The study is unique in that threat and aggressive behaviour were measured using dedicated behavioural games in a real-world setting. Owing to the complex data collection process, the sample sizes used are small (but comparable to other field studies of intergroup relations [38,40]). The results obtained must, therefore, be interpreted with due care and should be replicated elsewhere. It should also be reiterated that this study was conducted in a context where intergroup relations are strained, but overt violence is rare. On the one hand, it is interesting that even in this context, clear evidence for parochial altruism could be obtained. In a more hostile context, we would arguably expect to find stronger results in this regard. It remains uncertain, however, whether the other core result, the lack of elevated ingroup bias, would hold up in such a context. The comparison to other studies further suggests that the results may be specific to contexts with high residential stability. Such contexts are arguably similar to the conditions under which human cooperative behaviour evolved and which has inspired theoretical models, but may become scarce in today's world, marked by migration and increasing urbanization [61].
Ethics
The study design was approved by the European University Institute's Research Ethics Committee.
Data accessibility
Data are available at: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi%3A10.7910%2FDVN%2FP7Z8FH.
Competing interests
I have no competing interests.
Funding
Funding for this study was provided by the Department of Political and Social Sciences of the European University Institute (EUI) and through the EUI President's Fund.
Acknowledgments
I thank Delia Baldassarri, Stefania Bortolotti, James Fearon, Diego Gambetta, Benedikt Herrmann, Susanne Veit and the participants of trial sessions in Florence and Tbilisi for their helpful comments; Londa Beria and Tamuna Qeburia for their assistance in organizing and conducting the experimental sessions; and Lela Chakhaia and Nino Rcheulishvili for introducing me to most of my contacts in Georgia.
Footnotes
Electronic supplementary material is available online at https://dx.doi.org/10.6084/m9.figshare.c.3904936.
- Received July 11, 2017.
- Accepted September 26, 2017.
- © 2017 The Author(s)
Published by the Royal Society. All rights reserved.
References
Volume 284, issue 1865
