Study 1

Study 2

Results of study 1

Replicating previous results (Diekhof, Wittmer & Reimers, 2014; Reimers, Büchel & Diekhof, 2017; Reimers & Diekhof, 2015) a clear tendency towards parochialism was found. A 2 (“group”: ingroup, outgroup) × 2 (“offer”: unfair, fair) repeated-measures ANOVA with the covariate “standardized testosterone” revealed a significant two-way interaction between “group” and “offer” (F_1,59 = 5.44; p = 0.023; partial eta² = 0.08). In addition, we also found a main effect of “group” (F_1,59 = 10.56; p = 0.002; partial eta² = 0.15) as well as of “offer” (F_1,59 = 194.07; p < 0.001; partial eta² = 0.77) (see also Table 1). Apart from that, there was a significant three-way interaction between “offer”, “group” and “standardized testosterone” (F_1,59 = 7.82; p = 0.007; partial eta² = 0.12). Post-hoc simple slope analysis of the estimated marginal means indicated that unfair offers were more likely to be rejected if they came from an outgroup proposer than from an ingroup member, however only if the respondents were men with relatively high testosterone levels (t₍₅₉₎ = 3.64, p = 0.001). For men with low testosterone the same comparison was not significant (t₍₅₉₎ = 0.93, p = 0.358) (Bonferroni-corrected threshold: p = 0.025). In case of rather fair offers, which also included the 40% offers, both participants with low and high testosterone level tended to be more likely to reject offers from outgroup than from ingroup members, yet these differences were only numerical as they missed the threshold for significance in a two-tailed t-test (high testosterone: t₍₅₉₎ = 1.81, p = 0.076; low testosterone: t₍₅₉₎ = 1.96, p = 0.055) (Bonferroni-corrected threshold: p = 0.025). The results from the simple slope analysis of the estimated marginal means were confirmed by an analysis based on the complete data that compared the subjects with a testosterone concentration above the standardized mean (0) as the high T group (n = 29) with those with a testosterone level below 0, i.e., the low T group (n = 32). Accordingly, only the high T group showed increased rejections of the unfair outgroup offers compared to unfair ingroup offers (t₍₂₈₎ = 2.97, p = 0.006), while this was not the case in the low T group (t₍₃₁₎ = 1.40, p = 0.173). Again both groups showed enhanced rejection rates when rather fair offers were made by an outgroup member, whereby only in the low T group this difference was significant (high T group: t₍₂₈₎ = 1.84, p = 0.077; low T group: t₍₃₁₎ = 2.13, p = 0.041) (see Fig. 1A). Yet, the latter effect did not survive the Bonferroni correction (p < 0.025), which needed to be applied since the same data were also used in the direct comparisons of rejection rates between the testosterone groups that yielded no significant differences. We only found a non-significant increase in the Delta of rejections of unfair offers made by the outgroup minus those from the ingroup (Delta_{(high T)}: mean = 17.93%; sem = 6.05%; Delta_{(low T)}: mean = 5.31%; sem = 3.81%; t₍₅₉₎ = 1.77, p = 0.084). Collectively, these results indicate a somewhat greater parochialistic tendency in men at the high end of the testosterone distribution, who differentiated more between unfair proposals from the out- and the ingroup.

In a second step, we also looked at the proposals made by the participants to anonymous ingroup and outgroup members after having completed the UG. A repeated measures ANOVA with the factor “group” and the covariate standardized testosterone only showed a significant main effect of “group” (F_1,59 = 19.48; p < 0.001; partial eta² = 0.25), while no significant interaction with nor a main effect of standardized testosterone could be observed. A post hoc t-test showed that participants offered significantly more points to an anonymous ingroup member (mean ± sem = 4.1 ± 0.1 points) compared to an outgroup member (mean ± sem = 3.4 ± 0.2 points) (t₍₆₀₎ = 4.45, p < 0.001), yet this effect was independent of testosterone.

Finally, being in the ‘fast’ or ‘slow group’ did not influence the rejection rates, which was tested by re-running the first ANOVA including the subject’s artificial group membership. Also when comparing rejection rates of the ‘fast’ and ‘slow group’ directly, no significant differences emerged (p > 0.347).

Results of study 2

Study 2 was intended to examine whether the results related to parochial altruism in study 1 could also be found in a more natural group context (supporters of political parties tested during the German election year 2017). First of all, we found that the political preferences of our student sample were clearly oriented towards the left wing parties (SPD, The Greens, The Left), which was also reflected by the mean ranking across participants (lower mean ranks indicate a more favorable rating (mean ± sem): SPD = 2.44 ± 0.23, The Left = 2.94 ± 0.29, The Greens = 2.97 ± 0.23, CDU = 3.06 ± 0.22, FDP = 4.03 ± 0.23, AfD = 5.56 ± 0.20). Moreover, in line with study 1 and our previous results (Diekhof, Wittmer & Reimers, 2014; Reimers, Büchel & Diekhof, 2017; Reimers & Diekhof, 2015) we found a significant two-way interaction (F_3.69,121.78 = 2.81; p = 0.032; partial eta² = 0.079) in the repeated-measures ANOVA with the factors rank of “political preference” (6 ranks) and “offer” (unfair, fair). This was also reflected by a significant linear trend with increasing social distance, as determined by an univariate ANOVA with the independent factor “rank” and the dependent variable “mean rejection rate for unfair offers” (F_1,203 = 9.94; p = 0.002). When comparing rejection rates between the most extreme cases, i.e., the first rank (one’s favorite political party) and the last rank (the 6th rank, which is associated with the least liked party with the most distant political values in the ranking) we found significantly enhanced rejection rates for offers by rank 6, both in the context of fair and unfair offers (p < 0.001). In the comparison of the rank 1 with the other ranks, we also documented another two significant differences in the treatment of unfair offers by supporters of distant political parties ranked as 4 and 5, whose rejection was also more frequent (p < 0.001) (see Fig. 2, which displays the rejection rates related to unfair offers). All these differences survived a Bonferroni corrected threshold for multiple comparisons of 0.01. This means, that the design of study 2 was also suitable to show parochialistic tendencies amongst the supporters of political parties. Also the subsequent subdivision of ranks into the “ingroup” (ranks 1–3) and the “outgroup” (ranks 4–6) below is in line with these observations of a differential treatment of ranks 4–6, but not ranks 2 and 3 compared to rank 1.

More importantly, the main question was whether these parochialistic tendencies were also related to endogenous testosterone as already documented in study 1. For this purpose and to make the data of study 2 (6 groups) most comparable to those of study 1 (2 groups) the rejection rates from ranks 1 to 3 were combined as the “ingroup”, while the remaining ranks represented the “outgroup”. These data were then subjected to a repeated-measures ANOVA with the factors “offer” and “group” and the covariate standardized testosterone. In contrast to study 1 this ANOVA only revealed two main effects of “group” and “offer”, while the three-way interaction of the factors merely approached statistical trend level (see Table 2 for details). Nevertheless, following the procedure of study 1 we subjected the estimated marginal means to an exploratory simple slope analysis. This exploratory analysis showed that, similar to study one, only men with very high testosterone showed a significantly increased rejection rate for unfair offers made by outgroup compared to ingroup members (t₍₃₂₎ = 4.46, p < 0.001), while this was not the case in men with low testosterone (t₍₃₂₎ = 1.78, p = 0.085) (Bonferroni-corrected threshold: p < 0.025). Further, fair offers of the outgroup were more often rejected than fair ingroup offers by both testosterone groups (high testosterone: t₍₃₂₎ = 2.68, p = 0.012; low testosterone: t₍₃₂₎ = 2.50, p = 0.018) (Bonferroni-corrected threshold: p < 0.025). If subdividing the sample by the standardized mean of testosterone (0) in a group with above (n = 16) and one below average testosterone (n = 18) direct comparisons showed no differential rejection pattern as originally demonstrated in study 1. Further paired t-tests comparing rejection rates for fair ingroup vs. outgroup offers and unfair ingroup vs. outgroup offers in the two testosterone groups all yielded significant differences, that were in line with the interaction effect between “group” and “offer” (all p < 0.020; Bonferroni-corrected threshold: p < 0.025; see Fig. 1B). This suggests that there was no clear association between testosterone and parochialism in this smaller sample of political supporters.

Similar to study 1, the ANOVA of the proposals the participants made to anonymous members of the first and the last rank after completion of the UG yielded no interaction with standardized testosterone, yet only revealed a main effect of “group” (F_1,32 = 26.40; p < 0.001; partial eta² = 0.45). A post hoc t-test showed that participants offered significantly more points to a member of the closest rank 1 (mean ± sem = 4.6 ± 0.1 points) compared to a member of most distant rank 6 (mean ± sem = 3.0 ± 0.3 points) (t₍₃₃₎ = 5.17, p < 0.001).

Finally, in study 2 subjects performed a surprise cued recall task that was similar to the one used by Hechler, Neyer & Kessler (2016). However, in our task version the group identity of the faces shown was not indicated during the memory test, but all faces were shown without any notion of group identity. It has been suggested that there may be a memory advantage for individuals of one’s own group, especially if they exhibited schema-incongruent behavior (e.g., behaved unfairly) (Hechler, Neyer & Kessler, 2016). We performed a repeated-measures ANOVA with the factors “offer” and “group” and the covariate standardized testosterone to assess their effects on memory performance. This yielded a significant two-way interaction between “offer” and testosterone (F_1,32 = 5.57; p = 0.025; partial eta² = 0.15), but no other significant effect emerged (Table 3). The simple slope analysis of the estimated marginal means revealed one significant difference that was restricted to subjects with low testosterone. These participants showed a better recall rate for fair than unfair proposers (t₍₃₂₎ = 2.53, p = 0.017), yet independent of group (Bonferroni-corrected threshold: p < 0.025). In the analysis of all subjects with a testosterone level below the standardized mean (n = 18) an increased hit rate for the identity of fair proposers per se (mean ± sem = 51.3% ± 3.8% correct) compared to unfair proposers could also be observed (mean ± sem = 44.0% ± 4.6% points), yet this numerical difference was not significant (t₍₁₇₎ = 1.47, p = 0.11). Similarly, the numerically difference in recall of unfair proposers between men with high testosterone (n = 16; mean ± sem = 52.6% ± 3.6% correct) and those with low testosterone (n = 18; mean ± sem = 44.0% ± 4.6% points) did not reach significance (t₍₃₂₎ = 1.450, p = 0.157). It thus remains to be determined in a bigger sample whether testosterone may promote memory for unfairness and whether better memory performance indeed depends on group affiliation as previously suggested.

Analysis of the combined data from study 1 and 2

The previous results suggested similar effects of testosterone on parochialistic behavior in the two studies. However, the observations made in study 2 were not significant, which was most likely a result of the smaller sample size. We therefore combined the ingroup and outgroup data from the UG in a meta-analysis of all cases (n = 95) and re-ran the repeated-measures ANOVA of the rejection rates described previously, this time with the additional between-subjects factor “study”. This analysis confirmed the significant three-way interaction of “offer”, “group” and standardized testosterone (F_1,92 = 10.65; p = 0.002; partial eta² = 0.10) as well as the two main effects of the factors “offer” (F_1,92 = 222.29; p < 0.001; partial eta² = 0.71) and “group” (F_1,92 = 29.46; p < 0.001; partial eta² = 0.24) (see also Table 4). Further, the simple slope analysis of the estimated marginal means showed that unfair offers were more often rejected if they came from an outgroup proposer than from an ingroup member, however only if the respondents were men with high testosterone (t₍₉₃₎ = 5.28, p < 0.001). For men with low testosterone the same comparison yielded no significant difference (t₍₉₃₎ = 1.71, p = 0.091). In case of the rather fair offers, all participants rejected offers from outgroup members more often than those from ingroup members (high testosterone: t₍₉₃₎ = 3.61, p = 0.001; low testosterone: t₍₉₃₎ = 3.58, p = 0.001). Apart from these replications, we also found a two-way interaction between “offer” and “study” (F_1,92 = 8.00, p = 0.006; partial eta² = 0.08). Post-hoc comparisons showed that the participants of study 1 rejected unfair offers more often (mean ± sem = 68.6% ± 4.2%) than the participants of study 2 (mean ± sem = 53.4% ± 5.5%) (t₍₉₃₎ = 2.2, p = 0.03) indicating a higher degree of inequity aversion in the first sample. We therefore also looked into the questionnaire data from the two studies as these might help to explain these differences in inequity aversion (please note that missing questionnaire data of some subjects explain the reduced degrees of freedom below). We found that the participants of study 1 were significantly more impulsive as indicated by a higher BIS score (BIS_{study 1}: mean ± sem = 65.7 ± 1.3) compared to the subjects of study 2 (BIS_{study 2}: mean ± sem = 60.4 ± 1.4) (t₍₈₈₎ = 2.60, p = 0.011). The participants of study 1 also showed a significantly enhanced negative reciprocity score in the trust and reciprocity questionnaire (negative reciprocity_{study 1}: mean ± sem = 8.65 ± 0.25; negative reciprocity_{study 2}: mean ± sem = 5.52 ± 0.29; t₍₉₁₎ = 7.94, p < 0.001). At the same time, both the trust and the positive reciprocity score were significantly reduced in study 1 (trust_{study 1}: mean ± sem = 7.20 ± 0.21; positive reciprocity_{study 1}: mean ± sem = 4.18 ± 0.13) compared to study 2 (trust_{study 2}: mean ± sem = 8.48 ± 0.27; positive reciprocity_{study 2}: mean ± sem = 10.58 ± 0.24) (trust_comparison: t₍₉₁₎ = 3.65, p < 0.001; positive reciprocity_comparison: t₍₉₁₎ = 23.57, p < 0.001). These results suggest that the participants from the two studies might have shown a general difference in the way to treat other humans in economic interactions, which might explain the increased rejection rate in relation to unfair offers made in study 1. Interestingly, when correlating of the scores from the trust and reciprocity questionnaire with the degree of political interest in subjects of study 2 (PIKS score), we found a negative relationship between negative reciprocity and PIKS score (r =  − 0.396; p = 0.021). Unfortunately, the p-value did not survive the Bonferroni-correcte threshold of p = 0.0167 (the PIKS score was used in three correlations with the trust, negative and positive reciprocity scores). So we can only speculate that people with a higher political interest, who also show a stronger need to get political information to make an informed voting decision, might be more conscientious, which is why they would show less retaliation, yet more positive reciprocity and trust.

In addition to that, we wanted to further ascertain to what extent the actual degree to which an individual shows a ‘parochialism bias’, i.e., the differential treatment of unfair offers made by the ingroup and the outgroup, may be related to testosterone. For this, we performed a correlation analysis with the data of all cases (n = 95). This analysis revealed a small, yet significant positive correlation between standardized testosterone and the delta of the rejection rate of unfair outgroup minus unfair ingroup offers (Delta_out-in: r = 0.245, p = 0.017; see Fig. 3).

Finally, in order to further explore this positive correlation we followed the procedure of our previous study (Reimers, Büchel & Diekhof, 2017), in which we subdivided the sample by the individual tendency to treat highly unfair proposals. It has previously been assumed that offers of 20% of the original share are assumed to be the threshold at which rejection rates start to increase tremendously, yet at the same time inter-individual variation of rejection rates increases (Güth, Schmittberger & Schwarze, 1982; Henrich et al., 2005). In the first study 80% of unfair proposals offered 20% or less of the initial endowment, while in study 2 50% of the unfair proposals fell within this range. Thus a rejection rate for unfair offers (regardless of group) that was exceeding these study-specific thresholds of 80% (study 1) and 50% (study 2) was assumed to reflect a high degree of inequity aversion that was marked as the first individual decision strategy (inequity averse subjects). In contrast, rejection rates well below the respective study-specific threshold were considered as reflecting both a high tolerance for unfairness and the motivation to collect as many points as possible for oneself (selfish strategy). Based on this classification, that was exclusively based on the experimental structure of the UG task (i.e., referred to the percentage of highly unfair trials as cut-off value), participants were divided into the two strategy groups (see also Reimers, Büchel & Diekhof, 2017 for a similar procedure). This sub-division showed that the positive correlation between standardized testosterone and the parochialism bias, i.e., delta of the rejection rate of unfair outgroup minus unfair ingroup offers, stemmed from the selfish subjects (n = 44; r = 0.394, p = 0.008) with their relatively few rejections overall (selfish group: mean rejection rate_unfair ± sem = 34.3% ± 3.6%), while it was not present in the more inequity averse subjects (n = 51; r =  − 0.016, p = 0.909). One might argue that the absence of a correlation could be explained by the generally high rejection rates in the latter group (mean rejection rate_unfair ± sem = 88.1% ± 1.9%) and the resulting ceiling effect. However, it may also be that, similar to Reimers, Büchel & Diekhof (2017), selfish and inequity averse subjects were differently affected by testosterone when deciding how to treat unfair offers made by different groups.

Parochial altruism in the intergroup UG emerged across different studies

As hypothesized and in line with previous studies (Diekhof, Wittmer & Reimers, 2014; Reimers, Büchel & Diekhof, 2017) a marked pattern of parochial altruism with higher rejection rates in outgroup than ingroup interactions was evident in both studies, which was also reflected by a linear trend for increased social discounting with increasing social distance in study 2 corresponding to the results of earlier studies (e.g., Strombach et al., 2015). Such rejections could indicate the willingness to forgo points in order to punish outgroup members, which would reflect a behavior consistent with the theory of parochial altruism (Choi & Bowles, 2007). An alternative view would interpret this behavior as spiteful rather than altruistic, with the major aim to minimize the other’s payoff in order to equalize relative gain between the interaction partners (Jensen, 2010). It has been assumed that spiteful individuals see others as competitors, whose gains negatively affect their own utility (Espín et al., 2015). Yet, it is not clear why spitefulness should necessarily follow a parochial pattern or should lead to increased punishment of one social group above the other. We observed that 41 of the 95 participants (n_study 1 = 20, n_study 2 = 21) showed a preference to minimize the payoff of unfair outgroup members more than that of unfair ingroup members (Mean Delta_out-in for unfair offers ± sem = 32.19% ± 4.19%). This behavior would fit with the theory of parochial altruism (Choi & Bowles, 2007). First, these rejections indirectly benefited the relative gain of the ingroup, as did the somewhat lowered fairness norm in ingroup interactions. Second, it also clearly harmed the payoff of the outgroup, yet this happened at the responder’s expense, since in the latter case the responder declined a share he could have otherwise acquired for himself. In contrast, another 14 subjects (n_study 1 = 11, n_study 2 = 3) preferentially rejected unfair ingroup offers compared to unfair outgroup offers (Mean Delta_out-in for unfair offers ± sem = −10.87% ± 1.09%). This decision behavior would rather fit with the theory of indirect reciprocity (Gintis et al., 2003; Yamagishi & Mifune, 2008), according to which rejections of unfair offers by ingroup members even in single-shot interactions promote the adherence to fairness norms and cooperation within one’s group. Altruistic punishment of ingroup unfairness is thereby assumed to be the glue of large anonymous societies that follow the principle of generalized exchange. In addition, harsher punishment of norm violations in the ingroup may not only increase the individual probability to get fairer shares in the future, but could benefit one’s social reputation and thus enhance individual social status (Fehr & Gächter, 2002; Yamagishi et al., 2012). Finally, the remaining 40 subjects (n_study 1 = 30, n_study 2 = 10) showed no difference in responding to unfair offers from the out- and the ingroup. This latter group also showed significantly higher rejection rates in response to unfair offers per se (mean ± sem = 73.01% ± 5.88%) than the subjects with a negative Delta_out-in (mean ± sem = 53.37% ± 4.00%) (t₍₇₉₎ = 2.78, p = 0.007), which might have indeed reflected a motive driven by increased spitelfulness (Jensen, 2010). Altogether, the present data do not conform to the assumption that the spitefulness motive was the only driving force of rejections in case of unfair offers, yet it could have motivated a subgroup of subjects, which needs to be determined by future studies.

When looking at the findings of study 1 we found that the minimal group formation task elicited parochial altruism in the same manner as did previous studies with natural groups (e.g., ethnic groups as in Fershtman & Gneezy, 2001, or the sample of political supporters tested in study 2). The minimal group formation task resembled previously applied methods of assigning subjects to different groups according to their performance on a meaningless task (like in a task that required the estimation of the number of presented dots). Given previous evidence indicating that even such minimal conditions for group assignment promote an intergroup bias (Brewer, 1979; Tajfel et al., 1971; Volz, Kessler & Cramon, 2009), we expected to find a similar pattern. But note that alternative methods to create group identity have also revealed conflicting findings indicating that norm violations committed by ingroup members are punished more often (e.g., McLeish & Oxoby, 2007; Mendoza, Lane & Amodio, 2014). Although costly punishment has been proposed to sustain group cooperation (e.g., Fehr & Gächter, 2002) and may thus be expected to occur more often in response to unfair ingroup members, we found previously that our version of the intergroup-UG particularly provoked increased punishment of unfair outgroup members by young healthy men, even when our participants played for themselves without a direct intergroup competition (Diekhof, Wittmer & Reimers, 2014; Reimers, Büchel & Diekhof, 2017).

Apart from similarities study 1 and study 2 nevertheless also exhibited a difference, namely in the percentage of rejections of unfair proposals per se (study 1 > study 2). This might have been related to differences in aspects of the personality of subjects, that were determined by two self-report questionnaires. For one thing, the participants of study 1 showed a higher degree of negative reciprocity, which comprised the increased tendency for retaliation or to harm someone who has previously harmed oneself, while the tendency to return a favor (positive reciprocity) and the trust in strangers was reduced in comparison of subjects from study 2. For another, the participants of study 1 were also more impulsive. Impatience as an aspect of impulsiveness has previously been shown to influence decisions in the UG. In particular, spitefulness in the UG may be driven by an increased impatience, which would result in increased rejections of unfair offers as well as reduced proposals in general (Espín et al., 2015). Interestingly, of the 40 subjects, who showed no difference in responding to unfair offers from the out- and the ingroup 30 were from study 1, which is approximately 50 percent of the sample of study 1 (see above). This would fit with the observed average personality profile of increased negative reciprocity in combination with heightened impulsivity of the men from study 1, and points in the direction of increased spitefulness as one motive for the exceedingly high rejection rates.

Testosterone may modulate parochial altruism in different social contexts

The major aim of this research project was to assess the relation between endogenous testosterone and parochial altruism in two related, yet distinct social settings. Taken together, we found only weak, yet consistent evidence for a positive connection between testosterone and the preferential punishment of unfairness in outgroup members in the two studies (see Fig. 3). Not surprisingly this correlation was most visible in subjects who exhibited a more selfish decision strategy (i.e., who did not show a general distaste for unfairness and thus overall high rejection rate, but in contrast exhibited a rather flexible rejection style that also differentiated more between the in- and the outgroup). This might indicate a modulatory role of testosterone on the behavioral expression of male parochial altruism as suggested previously (see also Reimers, Büchel & Diekhof, 2017). The effects of testosterone on altruistic punishment in the UG have often been subtle or were even contradictory at times (see ‘Introduction’). This currently precludes an unequivocal interpretation of the role of testosterone in altruistic punishment in the UG and its interaction with parochialism. For this reason, the present data need to be replicated in other intergroup contexts (e.g., in the comparison of different ethnic groups or in members of different universities). Further, the causal relationship between testosterone and parochial altruism has to be determined by pharmacological intervention studies that ideally test both men and women.

As it stands, the two employed intergroup paradigms are most comparable to the ones used by Diekhof, Wittmer & Reimers (2014) and by Reimers, Büchel & Diekhof (2017), who also assessed healthy young men. However, the present results do not correspond to our previous findings. In the behavioral study of Diekhof, Wittmer & Reimers (2014) the positive relationship between testosterone and parochial altruism was not observed in the unbiased context, but only occurred when soccer fans transitioned from the neutral session to the competitive part of the intergroup-UG, during which groups explicitly competed for an additional group bonus. This was reflected by the relative enhancement of rejections of rather fair outgroup offers (4:6), which was stronger in soccer fans with a high testosterone level. Moreover, the same subjects showed an increased parochialism bias, but only during the competition, whereas there was no such relationship in the unbiased session of the intergroup-UG. Finally, when being in the role of a proposer, soccer fans with high testosterone made more generous proposals to members of their ingroup, which could not be observed in the present studies. However, our previous and present behavioral studies differ in some important aspects. While in the previous study (Diekhof, Wittmer & Reimers, 2014) participants faced proposers of four different sports teams (i.e., the soccer ingroup, a neutral soccer outgroup, an unknown cricket outgroup and a disliked soccer outgroup), three of whom were directly competing in the German soccer league (Bundesliga), the present research either used artificially created group identities or employed a natural group of German voters with high political interest. We can only assume that the differences in the groups under research might have led to different degrees of group commitment, as demonstrated previously by Weisel & Böhm (2015). Despite the use of natural groups in both study 2 and the study by Diekhof, Wittmer & Reimers (2014), who were competing for desired resources, i.e., either for seats in the German Bundestag or a high Bundesliga ranking, we speculate that soccer fandom evokes a much stronger emotional group affiliation than being a supporter of a certain political party prior to an important election in Germany, even though this was not explicitly tested here. Compared to these natural social settings, even less emotional commitment should be assumed in a member of a minimal group that was based on the performance in a simple reaction time task like in study 1, although this assumption is again speculative since it was not tested here. As already discussed in detail by Weisel & Böhm (2015), the election campaigns of the different mainstream parties in Germany (SPD, Greens, CDU and Free Liberals) are less emotional than for example in the United States (US). As Weisel & Böhm (2015) put it, the political discourse in Germany is mild and the mainstream parties are not as polarized as for example the Democrats and Republicans in the US. Also, many voters may consider them to have more in common than in separation. Only parties on the extreme right or left wing may be considered as distinct, which was also demonstrated here by the ranking of the extreme right wing party, the AfD. Of the 34 participants of study 2, 28 categorized the extreme right as rank 6, and another three did so as rank 5. In contrast to interactions between political voters in real life, the context of soccer fandom is characterized by a high degree of enmity between teams and normally the affiliation with one’s own team is very strong and emotional, that it may often resemble a tribal identity (Van Vugt & Park, 2010). Our previous study only tested subjects who strongly agreed with statements like “soccer is my life” and who owned not only season tickets for matches of their favorite team, but also went to away matches and owned fan merchandise like bedclothes with a team logo (Diekhof, Wittmer & Reimers, 2014). Further, a soccer season comprises 34 weeks of a year with games every weekend, while an election for the German Bundestag happens only every 4 years and the hot phase of the election campaign comprises only a handful of weeks directly before the election. Soccer fandom thus requires a constant engagement with the success of one’s favorite team as well as real life interactions with other supporters of one’s team as well as those from rival teams (criteria that were all fulfilled by the participants of the study performed by Diekhof, Wittmer & Reimers, 2014). Weisel & Böhm (2015) found less outgroup hate in supporters of political parties than in soccer fans in an economic exchange task, which was also sensitive for the different aspects of parochialism. When comparing the parochialism bias in relation to unfair offers (i.e., the Delta of rejections of unfair offers made by the outgroup minus those from the ingroup) of the present two and our previous behavioral soccer study (Diekhof, Wittmer & Reimers, 2014), we see that the minimal group study 1 had the lowest value (mean ± standard deviation = 11.3% ± 27.8%), which was lower than the bias found in study 2 (mean ± standard deviation = 14.1% ± 19.1%). In contrast, the parochialism bias documented by Diekhof, Wittmer & Reimers (2014) when comparing the ingroup and the antagonistic outgroup was much higher (mean ± standard deviation = 24.8% ± 36.5%) as was the bias observed in the neuroimaging study by Reimers, Büchel & Diekhof (2017) (mean ± standard deviation = 21.7% ± 29.7%). Further, even the Delta based on the average of the three outgroups of the behavioral soccer study (i.e., the antagonistic soccer team, the neutral soccer team, and the unknown cricket team) minus the ingroup rejection rates remained the highest compared to the present two studies (mean ± standard deviation = 17.6% ± 28.3%). Based on this and the observations made by Weisel & Böhm (2015), we speculate that the emotional engagement might have been highest in the hardcore soccer fans, which could be the reason why we had been unable to observe the rather subtle effect of testosterone on behavior during the unbiased context of the UG there, during which soccer fans already exhibited a very high degree of parochialism per se. Nevertheless, the present result emerged across the two different social settings of study 1 and 2, which may point to a valid, but small effect of testosterone on latent behavioral parochialism during socio-economic interactions outside of the context of soccer fandom.

Finally, we also performed a cued memory recall task in study 2. We did not find a memory advantage for schema-incongruent information in the given social setting (e.g., for unfair proposals made by ingroup members) as suggested previously (Bell & Buchner, 2012; Hechler, Neyer & Kessler, 2016). However, in contrast to these prior studies group association was not reinstated in the cued recall task, but facial identities were shown without any reference to the group they belonged to. This may explain why the general recall performance ranged around 50% chance level for fair and unfair proposers across ranks (mean ± sem: hit rate for all fair proposers = 51.3% ± 2.7%; hit rate for all unfair proposers = 48.0% ± 3.0%), while the hit rate for new pictures was the highest (mean ± sem: hit rate for new pictures = 57.5% ± 2.0%). We also found a small effect of reduced testosterone on hit rates for fair as compared to unfair proposers (better recall of fair proposers) in the analysis of the estimated marginal means. A recent neuropsychological study with a focus on cue-induced forgetting and recall in relation to salivary testosterone found something similar, namely that low testosterone levels were associated with improved binding of the newly encoded memories to their context cue (Sterzer et al., 2015). However, since the present effect could not be ascertained in study 1, in which no cued recall test was performed, and did not occur when the complete sample was subdivided by the mean of standardized testosterone (below or above average testosterone), we cannot infer that this was not just a sporadic finding. Future studies have to more carefully assess the interaction of social distance by fairness in larger samples using different cued recall tasks, with or without indicators of group identity.

Limitations and future perspective

The present studies have several limitations that need to be addressed by future studies. First, sample size, especially that of study 2, was limited and therefore the power to identify the potentially subtle associations between testosterone and behavior was restricted. Nevertheless, we found similar effects in both studies that also held when combining both data sets.

Second, the present studies identified a single hormone as a correlate of parochial altruism in the intergroup-UG. However, testosterone may not be the only hormone involved in ingroup favoritism and outgroup hostility. There is recent evidence for other hormonal systems to play a role in shaping parochial altruism. For instance, the neuropeptide oxytocin has been shown to drive parochial altruism in men during economic interactions (De Dreu et al., 2010). The steroid hormones testosterone and also estrogen have been shown to modulate the expression of the neuropeptides oxytocin and vasopressin (Liening & Josephs, 2010; Soares et al., 2010), which are both involved in a variety of social behaviors and economic decision making (Bos et al., 2012). Further, another steroid hormone that might potentially influence testosterone’s effects on social behavior is cortisol. A growing number of studies provide evidence for the dual-hormone hypothesis, which states that the effects of testosterone on status-related behavior, such as dominance, depend on the levels of cortisol (Mehta & Josephs, 2010). In fact, a recent study has provided initial evidence for the dual-hormone hypothesis in the context of an UG showing that a rise in testosterone was associated with increased acceptance rates of unfair offers in individuals with decreased cortisol levels (Mehta et al., 2015). Yet, the dual-hormone hypothesis has not remained undebated and two recent meta-analyses demonstrate only weak effects (Dekkers et al., 2019; Grebe et al., 2019). These results call for much larger samples for hypothesis testing as well as pre-registered study protocols in the future.

Third, another aspect that currently remains unknown, as it was not in the focus of the present study, is the impact of genetic predisposition on individual differences in parochial altruism. For instance, a genetic polymorphism in the androgen receptor gene, the CAG tandem repeat length, is associated with the sensitivity for circulating androgens such as testosterone (Chamberlain, Driver & Miesfeld, 1994). Subjects with a more selfish decision strategy were reported to show a tendency towards shorter repeat lengths and thus supposedly increased androgen sensitivity (Reimers, Büchel & Diekhof, 2017), which would help to explain the present findings of a positive correlation between the parochialism bias and testosterone that was particularly evident in men who showed reduced inequity aversion across groups. Other studies also demonstrated nuanced relationships between personality, repeat length, testosterone and aggression (Geniole et al., 2019), suggesting that genetic predisposition may be an important moderator of the relation between fluctuating hormones and behavior as well as brain physiology.

Fourth, the present studies as well as our previous ones (e.g., Diekhof, Wittmer & Reimers, 2014) were restricted to men. It thus remains to be determined how women would react to intergroup manipulations in the UG and whether endogenous testosterone may be related to female parochial altruism. Evolutionary theories of parochial altruism like the one formulated by Choi & Bowles (2007) or the ‘male warrior hypothesis’ (Van Vugt & Park, 2010) do not make explicit claims about female behavior during group competition or conflict. However, the ‘steroid/peptide theory of social bonds’ (Van Anders, Goldey & Kuo, 2011) makes some suggestions regarding the role of testosterone in the achievement of social goals that may also apply for females. Accordingly, in females testosterone may trigger defensive aggression in a special case of social threat, namely when there is a need to protect her offspring against outside threat. One may therefore speculate that women may also show parochialistic behaviors that are related to testosterone, however these might become particularly evident in situations that involve the core ingroup (e.g., family, mother-infant interaction) rather than an abstract group (like in study 1) or a political view (like in study 2). So it remains to be determined by future studies to what extent female testosterone mediates parochialism in general and parochial altruism in particular.

Finally, the expression of parochial altruism may not only be driven by physiological factors, but may further be shaped and could even be intensified by cultural ramifications like gender or racial stereotypes, socially preferred or sanctioned behaviors and long-standing rivalries between groups. Yet, by thoroughly assessing the physiological basis of parochial altruism we might be better able to formulate hypotheses that address the potential interaction between physiological and cultural factors that may collectively lead to prevailing intergroup biases and may fuel racism across the globe.