Cooperation during cultural group formation promotes trust towards members of out-groups

Abstract

People often cooperate with members of their own group, and discriminate against members of other groups. Previous research establishes that cultural groups can form endogenously, and that these groups demonstrate in-group favouritism. Given the presence of cultural groups, the previous literature argues that cultural evolution selects for groups that exhibit parochial altruism. The source of initial variation in these traits, however, remains uninformed. We show here that a group's economic production environment may substantially influence parochial tendencies, with groups formed around more cooperative production (CP) displaying less parochialism than groups formed around more independent production (IP) processes. Participants randomized into CP and IP production tasks formed cultural groups, and subsequently played hidden-action trust games with in-group and out-group trustees. We found CP to be associated with significantly greater sharing and exchanging behaviours than IP. In trust games, significant parochial altruism (in-group favouritism combined with out-group discrimination) was displayed by members of IP groups. By contrast, members of CP groups did not engage in either in-group favouritism or out-group discrimination. Further, we found the absence of out-group discrimination in CP to persist even following ‘betrayal’. Finally, belief data suggest that members of CP are not more intrinsically generous than IP members, but rather more likely to believe that out-group trustees will positively reciprocate. Our results have important implications for anyone interested in building cooperative teams, and shed new light on connections between culture and cooperation.

1. Introduction

In-group favouritism, or solidarity, is a well-documented aspect of human behaviour [1–8]. People offer preferable treatment towards members of their own group and discriminate against those belonging to other groups [4–10]. Various explanations have been offered for in-group favouritism. One is based on intergroup conflicts [5], and argues that winning groups usually exhibit above-average levels of in-group solidarity [11]. The result is evolutionary group selection for in-group favouritism. Another explanation is that, due to the high frequency of in-group interactions, it became evolutionarily advantageous for a reciprocal altruism heuristic to emerge [12].

Regardless of the selection mechanism, in-group favouritism may have coevolved with group culture [13], a process that may explain features of the contemporary socio-economic landscape [9]. For example, cultures that include a strong focus on lineage (e.g. Italy, Taiwan and Hong Kong) also have economies characterized by many small enterprises, which Fukuyama [9] argues may derive from family-oriented favouritism. While this is advantageous in many ways, in some cases it might also facilitate the emergence of out-group discrimination. The conditions under which parochial altruism (in-group favouritism combined with out-group discrimination) emerged remain a topic of active scholarly debate [5,7,13,14].

Most previous research explains parochial altruism by beginning with the assumption that multiple cultural groups coexist [4,7,8,10,15–17]. Recent research extends this by showing that cultural groups can emerge endogenously [13], and that these groups display parochial altruism. At the same time, there are widely documented cross-cultural differences in parochial altruism, in that people from different cultures behave differently towards members of in-groups and out-groups, particularly with respect to altruism, reciprocity and punishment decisions [18–20]. Left unexplained, however, are the sources of variation in parochial altruistic tendencies.

Research on small-scale societies suggests that the nature of the production process may have helped to plant the seeds of preference variation among cultural groups, particularly with respect to propensities for pro-sociality [14,21–23]. Henrich et al. [14] studied cross-cultural differences in cooperation, finding that group-level differences in pro-social activities can be largely explained by differences in pay-offs to cooperation and market integration. In particular, these pay-offs are positively correlated with pro-social behaviours within a group [14,21–24]. They also argue that individual-level economic and demographic variables seem not to explain behaviour either within or across groups. Our paper demonstrates that the extent of cooperation within the economic production environment can significantly impact the nature of parochial altruism displayed by a cultural group.

To investigate connections between cooperative culture and parochial altruism, we induce two distinct cultural groups. Our experimental approach draws from social identity theory [25]. Such studies typically either prime existing social identities (such as occurs, for example, in the stereotype threat literature [26]), or alternatively induce (artificial) group identities. Our study uses the latter, denoted by Tajfel and Turner [25] as the minimal group paradigm (MGP). This name draws attention to the idea that groups are created using trivial and almost meaningless tasks.

Our specific design is inspired by the MGP reported by Tajfel & Turner [27]. It involves randomly assigning subjects to groups and then asking each group to complete tasks. While similar, our procedure differs from Tajfel & Turner [27] in certain key respects. First, as for Chen & Li [4], our procedure facilitates social interaction within group members (but not between groups). Second, the nature of the social interaction is endogenous to each group's production task. This latter is the unique and crucial feature of our design. In particular, one of the production tasks turns out to generate relatively more social interaction among group members.

Our primary hypothesis is that the extent of cooperation within the economic production environment impacts the nature of parochial altruism displayed by a cultural group. The existing literature suggests two likely directions for this effect. First, some studies have argued that more cooperative societies are also more open to cooperation with outside groups [9,21–23]. By contrast, other studies suggest sharing and exchange among in-group members reinforce in-group favouritism, but also build boundaries outside the group, thus reducing cooperation with out-groups [10]. Our results offer support for the former: members of all cultural groups display in-group favouritism, but only those with a culture of independent production (IP) display out-group discrimination.

2. Experiment design

To form cultural groups, we used a production process similar to that reported by Eckel & Grossman [15]. We conducted 14 sessions where we induced cultural groups. Thirteen of these sessions consisted of 10 subjects randomized into two groups of five each, and one session included two groups of four subjects. In order to induce cultural groups, each group member was asked to complete one puzzle, and members of the same group were asked to complete the same puzzle in an open environment conducive to cooperation (see §4). Different groups, however, were separated into different rooms.

We used two different puzzles: one (a triangle puzzle) required relatively more cooperative production (CP), whereas the other (a square puzzle) required relatively more IP. There was competition between groups to finish the puzzles fastest, with each member of the fastest group earning an additional two dollars. The winning group was not announced until the conclusion of the experiment (see the electronic supplementary material for puzzle game instructions).

A total of 202 people participated in our experiment. Those in the culture condition participated first in a group production task (see §4), and then made trust decisions (as described below). Those in baseline did not participate in a group production task; thus, they made their trust decisions without the experience of a competitive group culture formation process. We did not include any pre-trust game task in baseline, due to the fact that any group activity could potentially generate group culture and correspondingly impact trust decisions. A total of 64 subjects participated in the baseline condition, with 32 subjects randomized into the role of investor and the other 32 subjects randomized into the role of trustee.

After completing the production task, participants made decisions in the modified trust game of Charness & Dufwenberg [28] (figure 1). This game includes the possibility that a trustee who tries to reciprocate an investor's trusting action will fail to be able to do so. Importantly, whether a trustee chooses to reciprocate is hidden from the investor, who sees only the outcome of the trustee's action. This feature is important for us as we are interested in knowing whether investors respond to a failure of reciprocity differently between in-group and out-group trustees. Such differences can emerge, for example, if investors are more likely to believe that failed reciprocity is due to bad luck with in-group trustees, but purposeful defection by out-group trustees.

Figure 1.

The modified trust game. Investor decides whether to choose IN or OUT when s/he is matched with an in-group or out-group trustee. Assuming the investor chooses IN, trustee decides whether to choose ROLL or DON'T ROLL depending on whether s/he is matched with an in-group or out-group investor. This game has a hidden action in the sense that when the trustee chooses ROLL, chance determines whether the investor receives $12 or $0 (with probabilities 5/6 or 1/6). Trustee decisions are unobservable, and trustees have a pecuniary incentive to choose DON'T ROLL. If investor foresees this, it will lead to the {OUT, DON'T ROLL} outcome, which is the unique subgame perfect Nash equilibrium for this game.

Each investor type (CP or IP) decided whether to choose IN or OUT according to whether they were matched with an in-group or out-group trustee. Trustees were asked to make their decisions assuming their investors chose IN. In particular, each trustee type (CP or IP) decides whether to choose ROLL or DON'T ROLL according to whether they were matched with an in- or out-group investor. The Nash equilibrium of this game is {OUT, DON'T ROLL}, meaning the investor does not trust and the trustee defects.

Participants played two periods of this game. In addition, we elicited investors' expectations regarding trustees' reciprocity towards in-group and out-group investors (see §4).

3. Results and discussion

We found CP and IP tasks to require different levels of cooperation. CP groups completed their tasks significantly more slowly (n = 12, time = 218 s; figure 2) than IP groups (n = 12, time = 102 s, p = 0.033, two-sided Mann–Whitney U-test¹). We also compared the puzzle pieces a participant initially received with the pieces ultimately included in his/her completed puzzle. Participants ending with a relatively greater number of new pieces evidently engaged in relatively more exchange and cooperation with other group members. Indeed, we observe a strong correlation between time taken to complete the puzzle and the frequency of sharing and exchange among group members. Those who completed their puzzle using relatively more new pieces also took longer to finish the task (n = 20, time = 215 s) than those who used fewer new pieces (n = 26, time = 103 s, p < 0.001). It follows that, overall, participants in CP groups spent more time with the task and experienced more cooperation than did participants in IP groups.

Figure 2.

Distribution of time to complete two puzzle tasks. Those who worked on the CP (triangle) puzzles (black bars, mean = 218 s, n = 12 groups) spent significantly more time than those who worked on IP (square) puzzles (white bars, mean = 102 s, n = 12 groups, p = 0.02). Further, we recorded the number of puzzle piece exchanges by comparing the specific pieces used in each person's finished puzzle to the specific pieces initially distributed to each person. The median percentage of new pieces included in a finished puzzle is 50%; and 14 observations exactly equal 50%. The result is robust to including these 14 in either group (p < 0.001). The number above each bar represents the mean fraction of new pieces in completed puzzles. People who spent more time finishing the task were more likely to be working on the triangle puzzle, and also engaged in more exchange of puzzle pieces (n = 27, mean = 72.2% new pieces), as compared with those who exchanged fewer pieces (n = 33, mean = 30.3% new pieces, p < 0.001).

Our manipulation successfully created cultural groups. In particular, our expectation elicitation (see §4) revealed systematically different expectations between CP and IP regarding the decisions of out-group trustees. CP investors expected out-group trustees to reciprocate more (mean = 0.47, n = 34) than IP investors (mean = 0.26, n = 35, p = 0.005). Consistent with these expectations, CP investors (mean = 0.59, n = 34) trusted out-group members significantly more than did IP investors (mean = 0.29, n = 35, p = 0.012; figure 3). On the other hand, our manipulation did not change the rate of in-group favouritism. CP (n = 34, mean = 0.65) and IP (n = 35, mean = 0.60) investors did not display differences in expectations regarding in-group trustees' reciprocation rates (p = 0.41); and CP (n = 34, mean = 0.74) and IP (n = 35, mean = 0.69) investors trusted in-group members at the same rate (p = 0.653).

Figure 3.

Trust among conditions. CP investors (n = 34) trust out-group members significantly more than IP investors (n = 35, p = 0.012), and as much as investors in baseline where no identity was created (n = 32, p = 0.834). IP investors display significant out-group discrimination, in that investors in baseline (n = 32) trust significantly more than their IP counterparts (n = 35, p = 0.02). IP investors present clear in-group favouritism: IP trusts in-group trustees (mean = 0.69) significantly more than out-group trustees (mean = 0.29, n = 35, p < 0.001); however, this favouritism disappears among the CP investors (in-group: mean = 0.74, n = 35; out-group: mean = 0.59, n = 34, p = 0.203). Note further that there is some evidence that CP (mean = 0.74, n = 34) investors trust in-group marginally more than their counterparts in baseline (mean = 0.56, n = 32, p = 0.072, one-sided Mann–Whitney U-test).

Differences between CP and IP investors were apparent in relation to investors in baseline. IP investors expected less reciprocation from out-group trustees (n = 35, mean = 0.26) than the overall rate of reciprocation expected by baseline investors (n = 32, mean = 0.51, p < 0.001). IP investors (n = 35, mean = 0.29) also trusted out-group members significantly less than the overall rate of trust displayed in baseline (n = 32, mean = 0.56, p = 0.023; figure 3). Thus, IP investors displayed out-group discrimination. No difference was observed between CP investors' (n = 34, mean = 0.47) expectations regarding per cent of out-group members who would reciprocate and rates of reciprocation expected in baseline (n = 32, mean = 0.51, p = 0.507). Consistent with this, CP investors (n = 34, mean = 0.47) displayed the same level of trust towards out-group members as investors in baseline (n = 32, mean = 0.51, p = 0.834). Hence, CP investors did not display out-group discrimination.

CP investors displayed the same rate of trust for both in-group (mean = 0.74) and out-group (mean = 0.59, n = 34, p = 0.203; figure 3), whereas IP investors presented strong in-group favouritism (mean = 0.69 for in-group; mean = 0.29 for out-group; n = 35, p = 0.001). When combined with the out-group discrimination indicated above, it follows that IP investors displayed parochial altruism, whereas CP investors did not.

Variation in expressions of parochial altruism between CP and IP persisted to the second round. CP investors (n = 25, mean = 0.72) trusted out-group members significantly more than their IP counterparts (n = 24, mean = 0.38), even after experiencing defection from in-group members (electronic supplementary material, figure S5; p = 0.016). In addition, CP investors continued not to present differences in rates of trust towards new in- or out-group trustees (n = 25, mean = 0.72 for both, p = 1.000). In-group favouritism was not observed among IP investors (n = 24, mean = 0.58 for in-group, mean = 0.38 for out-group, p = 0.15). This result is driven largely by reduced trust frequencies towards in-group members following betrayal. CP investors (n = 25, mean = 0.72) continued to trust out-group trustees significantly more than IP investors (n = 24, mean = 0.38, p = 0.016). In addition, we found that reciprocity towards out-group members in the second round was generally higher in CP than IP (see the electronic supplementary material, table S2).

Systematic differences in beliefs and behaviours between CP and IP groups stem from differences in their respective group formation processes. In relation to IP, participants in CP cooperated more during group formation as a consequence of their more challenging group production task. Our data suggest that this impacted expectations and beliefs. In particular, we observed that investors who experienced more CP were more likely to believe in the trustworthiness of out-group trustees than investors who experienced more IP. This was reflected in significantly more trusting decisions by CP towards out-group members. It is worth noting that connections between beliefs and behaviour can be formalized in our environment. In particular, when viewed through the lens of the Fehr–Schmidt [26] model of social preferences (see the electronic supplementary material), differences in expectations we elicit can well explain differences in frequencies of trust we observe.

We also found the group formation process to influence reciprocity, though to a lesser extent than trust. IP trustees reciprocated to the in-group more than to the out-group (n = 34, mean = 0.67 for in-group, mean = 0.38 for out-group, p = 0.016), whereas CP trustees reciprocated in-group and out-group investors at the same rate (n = 35, mean = 0.45 for in-group,  mean = 0.43 for out-group, p = 0.812).

The systematic influence of CP on both trust and reciprocity may be important for sustaining a cooperative culture. Previous research points to differences between trust and reciprocity [29], arguing that reciprocity is a social norm, while trust is not. Our results suggest that ongoing experiences with CP may be a powerful causal force helping to shield societies from disintegration of trust over time.

Our results shed light on the causal role of the group formation process in impacting cultural propensities to trust [30–32], and may have implications for which cultural differences are transmissible, as well as how this transmission takes place. One mechanism may perhaps involve transmitting economic production processes. Alternatively, transmission of a production process may involve an inherent transmission of cultural ideals, or a ‘cultural spillover’ associated with the adoption of new technologies. Such spillovers could be important in explaining historic examples of the coevolution of culture and group [32], and could also be valuable in determining new approaches for fostering economic growth and development.

Our findings raise a number of new questions. For example, we found that differences in helping behaviour impact willingness to trust members of out-groups, yet we did not explore whether helping behaviour itself, another pro-social activity with known adaptive value, might display patterns consistent with parochial altruism. The evolutionary importance and sensitivity of helping behaviour to subtle environmental variation has long been documented [33]. Exploring these behaviours within the context of in- and out-groups defines an agenda of substantial importance.

4. Methods and summary

A total of 202 subjects at George Mason University participated in our experiment (42% female and 58% male). A total of 138 subjects participated in the culture condition, with each session consisting of two groups of the same size. One group solved triangle puzzles (CP), whereas the other solved square puzzles (IP). Fourteen sessions were conducted (all with 10 subjects except for one session with eight subjects). A total of 64 subjects were included in baseline without-group cultural formation. On average, subjects earned about $15 for a 50 min session.

(a). Production tasks

Each participant received an envelope containing four cardboard pieces (figure 4). We designed the puzzle so that solving squares required a lower degree of cooperation, whereas solving triangles required a higher degree of cooperation. In particular, each of the pieces of the square puzzle included one of the square's corners, and there were no other right-angles included in the puzzle pieces (figure 4a). Then, we used a right isosceles triangle puzzle, cut so its pieces included only one right angle (one of the triangle's corners). The other pieces were shaped so that it was difficult to determine quickly how to assemble them to form the triangle (figure 4b). Successfully completing the task required each group member to make an identical square/triangle.

Figure 4.

(a) The cut of square puzzle. (i) The scattered pieces of the puzzle; (ii) a completed square puzzle; and (iii) the cuts. Each group member received a sealed envelope containing three of the four unique pieces necessary to complete the puzzle, in addition to one duplicate piece. The square puzzle could be completed quickly, because a corner of the square (a right angle) was present in each piece. (b) The cut of triangle puzzle. (i) The scattered pieces of the puzzle; (ii) a completed triangle puzzle; and (iii) the cuts. Each group member received a sealed envelope containing three pieces of the four unique pieces necessary to complete the puzzle, in addition to one duplicate. The square puzzle is easier as the four pieces determine its correct orientation, while this is not true of the triangle puzzle (because the vertex of any piece can be the vertex of a triangle, which is not true of the square). (Online version in colour.)

(b). The trust game

In each session, half of the subjects were randomly assigned as investor and the other half of the subjects as trustee in a two-period sequential trust game described by Charness & Dufwenberg [28] (figure 1). Subjects were randomly re-matched with a new partner in period 2. Subjects were paid based on a randomly chosen period. In period 2, they played the same game using all potential outcome scenarios that could have resulted from their period 1 decisions.

(c). Belief elicitation

After finishing the trust game, and prior to receiving feedback, investors (trustees) were asked to guess how many trustees (investors) reciprocated (invested) for both in-group and out-group matches. Participants earned $1 for each correct answer, and nothing otherwise.