When the Underdog Apologizes: The Role of Intergroup Apologies in Historical Intergroup Conflicts

1.1. Intergroup apologies

Often, the first contemplated action is revenge when an individual or group is harmed by another (Rusbult et al. 1991). While seeking revenge may help the victim regain a sense of control and self-worth (Akhtar 2002), it often comes at a substantial cost to the relationship. Among others, it may serve as a pretext for subsequent retaliation, leading to a downward spiral of aggression and a progressive erosion of the relationship. One way to prevent retaliation following an unjust harm is through offering an apology. Intergroup apologies are public acknowledgments made by representatives of a group (Tavuchis 1991), indicating that the offending group recognizes its role in the conflict and is committed to addressing past wrongs. For the victim group, such apologies are important because they publicly affirm the injustice suffered and serve as a record by which the offending group can be held accountable for its future actions.

Some scholars view intergroup apologies as pivotal moments in the relationship between conflicting groups, as they can represent a meaningful turning point and the beginning of a renewed interaction (Wohl et al. 2011). Some of the most famous examples include the Canadian government’s apology to Aboriginal Canadians for the abusive residential school system or the Australian Prime Minister’s apology to the ‘Stolen Generations’ of Aboriginal people who were removed as children from their families (Wohl et al. 2011). In fact, apologies have been used so often in the past decades by governments, group leaders or corporations that some authors have begun speaking about an ‘age of apology’ (e.g., Brooks 1999).

It was extensively argued by theorists that intergroup apologies can be beneficial for intergroup relations (e.g., Lazare 2005; Tavuchis 1991), which led to a widely held presumption that intergroup apologies represent a key element for reconciliation (Wohl et al. 2011). The recipients of apologies themselves seem to share such views: For example, 93% of Aboriginal Australians rated the Australian government’s apology as very important for its abuses against the Aboriginal community, and 80% rated it as necessary for promoting amiable relations with Australians (Wohl et al. 2011). Apologies’ positive effects have been observed on outcomes such as improved cooperation with the offender (Halabi et al. 2013), decreased desire for retribution (Leonard et al. 2011), greater willingness for contact (Borinca et al. 2024) or more positive evaluations of the perpetrator (Blatz et al. 2014). However, when studied against more stringent criteria such as forgiveness or trust, the few existing experimental studies revealed conflicting results. For example, in Study 1, Brown et al. (2008) found that a U.S. apology for a friendly fire incident in Afghanistan, which resulted in the deaths of Canadian soldiers, led to increased forgiveness of the U.S. military among Canadian participants. Similarly, Leonard et al. (2011) found that an apology encouraged students to forgive an insult made by faculty members, while Reinders Folmer et al. (2021) showed that following a transgression in which a group of students kept most of the money in a Dictator game, offering an apology increased the victim group’s trust in the transgressors. A similar result was found by Ma et al. (2019), who showed that participants receiving an apology were more likely to trust their transgressors after a trust violation. However, other experimental studies have not supported these findings. In four experiments, Philpot and Hornsey (2008) observed that while an apology enhanced perceptions of remorse and increased victims’ satisfaction, it did not lead to greater forgiveness. Similarly, Steele and Blatz (2014), in five different apology conditions, found no effect of apologies on forgiveness or trust, while Kachanoff et al. (2017) found no significant effect of apologies on forgiveness.

One of the most widely used frameworks for understanding when apologies are effective is the needs-based model of reconciliation (Shnabel & Nadler 2008). This model proposes that reconciliation depends on the reciprocal fulfillment of the emotional needs of both victims and perpetrators. Specifically, victims are thought to experience a threat to their sense of power, whereas perpetrators are thought to experience a threat to their moral identity. Consequently, reconciliation is facilitated when perpetrators help restore victims’ sense of power and when victims affirm the perpetrators’ moral image.

While the needs-based model emphasizes emotional needs rooted in the roles of victim and perpetrator, other scholars have highlighted the importance of cognitive factors in shaping the effectiveness of apologies. For example, Wohl et al. (2013) argued that the perceived sincerity of the apology is crucial for its acceptance by the victim. If the perpetrator group is viewed as insincere, their apology may be perceived as a cynical tactic rather than a genuine expression of remorse (Wenzel et al. 2017). A sincere apology may also humanize the apologizing group by signaling to the victimized group that its members are capable of experiencing uniquely human emotions, such as remorse (Borinca et al. 2021; Wohl et al. 2012), thereby eliciting emotional responses, such as reduced anger or increased respect, that are conducive to positive intergroup outcomes (Leonard et al. 2011). Supporting this, Wohl et al. (2013) found that perceptions of sincerity in an official apology significantly predicted forgiveness among Chinese Canadians toward European Canadians for the head tax imposed on Chinese immigrants. Similarly, Wenzel et al. (2017) showed that higher perceptions of apology sincerity increased forgiveness toward the transgressor group, while Borinca et al. (2021) found that apologies perceived as genuine were successful in promoting positive intergroup outcomes.

Another important cognitive aspect concerns perceptions of the typicality of the apologizing representatives. Since the perpetrator group as a whole shares responsibility for the transgression (Wohl et al. 2006), those delivering the apology must be seen as legitimately representing the group on whose behalf they speak. Group members are generally conscious of the variability existing in outgroups, even though it is to a lesser extent than in their own (Quattrone & Jones 1980). Out-groups often consist of subgroups that vary across multiple dimensions, and these subgroups may also differ in their stance toward supporting an apology, which can significantly influence the apology’s effectiveness. For instance, perceiving the out-group as a coherent and unified entity (i.e., high entitativity) (Campbell 1958) resulted in viewing its apology as more sincere and less driven by ulterior motives, compared to apologies issued by out-groups perceived as low in entitativity (Leonard 2019). Similarly, Wenzel et al. (2017) demonstrated that apologies perceived as highly representative of the offending group elicited greater forgiveness than those perceived as less representative. In their study, victims were informed about the degree of support for the apology within the offender group, such as the proportion of group members endorsing it. In real-world contexts, however, victims rarely have access to such information and typically rely solely on the message conveyed by the apologizing representatives. Consequently, it is essential that these representatives are perceived as genuine members of and authorized spokespeople for the transgressor group. Even subtle discrepancies between the apologizers and the prototypical image of the offending group can activate subtyping processes (Taylor 2015), leading victims to interpret the apology as originating from an exceptional subgroup rather than from the group as a whole. In other words, apologizing representatives need to be perceived as typical members of the offender group. Although the mechanisms of perceived sincerity and typicality have traditionally been examined in contexts where apologies come from more powerful to less powerful groups, they are likely to operate similarly in reversed power dynamics, such as those investigated in the present study.

One of the objectives of the present study is to examine how an intergroup apology affects forgiveness and trust toward the apologizing group and whether this effect is influenced by the perceived sincerity of the apology and the perceived typicality of the apologizing representatives within the context of a real-world conflict between Romas (the apologizing group) and Romanians (the recipient group). A distinctive feature of this research lies in its reversed direction of apology—from the marginalized to the powerful group—allowing an investigation into whether members of less powerful victim groups can take an active role in reconciliation. Nadler and Liviatan (2006) investigated a similar dynamic in the context of a non-apologetic speech and found that expressions of empathy by a Palestinian authority (the lower-power group) increased Israelis’ (the higher-power group) willingness to reconcile but only among those who reported high levels of trust in Palestinians. To the best of my knowledge, however, the present study is the first empirical study to directly examine the effects of apologies issued by the less powerful group. These dynamics have been observed in real-world contexts as well. For example, following the end of apartheid, leaders of the African National Congress issued apologies to white South Africans—the former oppressors—for violent acts committed during the anti-apartheid struggle (Institute for the Study of Human Rights n.d.). Similarly, in the context of the Israeli–Palestinian conflict, Palestinian authorities have at various points apologized for attacks targeting Israeli civilians (e.g., Yang & Sagalyn 2023). However, despite these real-world examples, the extent to which such apologies are actually effective remains a question that needs to be empirically investigated.

1.2. Intergroup contact

Aside from intergroup apologies, intergroup contact was often used to improve intergroup relations. Allport (1954) was one of the first social psychologists to suggest that intergroup contact is conducive to improved intergroup outcomes. He specified four conditions under which contact leads to improved intergroup attitudes, namely, equal status, authority support, working towards a common goal, and cooperation towards reaching that goal. Research has generally supported Allport’s claims, although a later meta-analysis by Pettigrew and Tropp (2006) has provided evidence that intergroup contact is associated with more favorable intergroup attitudes irrespective if these conditions are met or not.

The positive effects of intergroup contact were observed for a number of intergroup outcomes, such as prejudicial attitudes (Hamberger & Hewstone 1997), anger reduction (Tam et al. 2007), as well as trust (Vezzali et al. 2012; Tausch et al. 2011) and forgiveness (Tam et al. 2007) and for diverse targets, such as ethnic minorities (Hamberger & Hewstone 1997), homosexuals (Herek & Capitanio 1996), or people with HIV/AIDS (Werth & Lord 1992), to name a few. This compelling evidence supports the use of intergroup contact as a primary intervention strategy when addressing problematic intergroup relations. However, caution is necessary when such contact occurs without meeting Allport’s optimal conditions, as research indicates that negative intergroup interactions can actually worsen intergroup attitudes (Techakesari et al. 2015).

As Hewstone et al. (2014) point out, with some exceptions (for a review, see Al Ramiah and Hewstone 2013), the positive effects of intergroup contact have largely been examined in relatively benign intergroup contexts. However, the situation may be quite different in contexts marked by historical conflict, where the conflict itself and negative perceptions of the opposing group have become embedded in the group’s collective ideology (Bar-Tal et al. 2012). Such ideologies can shape how individuals interpret intergroup interactions, potentially leading to more negative appraisals during and after contact and even reinforcing hostile attitudes toward the outgroup (Hayward et al. 2017). Nonetheless, several studies offer encouraging findings from conflictual contexts. In Northern Ireland, Tam et al. (2009) found that intergroup contact predicted more positive behavioral intentions toward the outgroup, while Tausch et al. (2011) showed that extended contact fostered outgroup trust. In Israel, Maoz and Ellis (2008) reported that Israeli Jews who engaged in structured dialogue with Palestinians expressed greater trust and stronger support for compromise-based solutions to the Israeli-Palestinian conflict. Similarly, in Bosnia and Herzegovina, Cehajic et al. (2008) found that intergroup contact was positively associated with Bosnian Muslims’ willingness to forgive Bosnian Serbs for wartime transgressions. Together, this compelling evidence suggests that, when the right conditions are met, intergroup contact can promote improved intergroup relations even in deeply divided contexts. To strengthen the validity of these findings, it is essential to further examine the positive effects of contact in real-world settings characterized by serious and ongoing conflict, such as the one addressed in the present study.

1.3. Interaction between apology and contact

Notably, the combined effects of intergroup contact and intergroup apologies have, to the best of my knowledge, not yet been examined. Beyond acknowledging wrongdoing, an apology may also signal a desire to restore damaged relations (Hareli & Eisikovits 2006). In a collectivistic context strongly focused on relationships such as the Romanian one (Gavreliuc 2011), living harmoniously with one’s neighbors and acquaintances is part of the cultural mandate and is highly valued. As such, Romanians who have frequent contact with Romas may have an added motivation to restore relations, which could further encourage acceptance of the Roma group’s apology.

Second, intergroup contact has been shown to impact perceptions of common humanity (Capozza et al. 2014), thus creating the psychological conditions under which an apology can be seen as a credible gesture from a humane group. Additionally, contact can enhance cognitive differentiation and perspective-taking, which are processes associated with integrative complexity, defined as the ability to differentiate between multiple viewpoints and to integrate them into a coherent whole (Tetlock 1985). Consequently, high-contact individuals should show less black-and-white moral reasoning regarding Roma behavior and rely less on the default distrust and suspicion that characterizes this intergroup context. Instead, they may be more inclined to attribute the apology to genuine remorse, rather than a strategic move from an adversarial group. Thus, intergroup contact and apology may interact, with the apology potentially having a stronger positive effect on those Romanians who have greater contact with Romas compared to those with limited contact.

1.4. The present study

In long-standing intergroup conflicts, both parties often assume the roles of both victims and perpetrators of violence. The present study focuses on the perspective of Romanians and examines the effects of apologies issued by the historically marginalized group (i.e., Romas), the effect of intergroup contact, and their interaction on forgiveness and trust toward the Romas. Additionally, it explores whether the perceived sincerity of the apology and the perceived typicality of the apologizing representatives influence the relationship between the apology and the outcome variables (forgiveness and trust). Drawing on the reviewed literature, the following hypotheses are proposed:

2.1. Ethical statement

The study was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki and was approved by the Ethics Committee of the West University of Timișoara. When requesting approval, a short description of the study, together with the method of data collection and expected results were provided.

2.2. Power analysis

Sample size was calibrated using G*Power, version 3.1.9.4 (Faul et al. 2007). To detect a small-to-medium effect size of d = 0.40, for α = 0.05 and a power of 0.80, the required sample size was N = 156. This value acted as a benchmark against which the study was calibrated.

2.3. Design

The study employed a 2 (contact) × 2 (apology) between-subjects quasi-experimental design, with contact treated as a measured variable and apology as a manipulated variable.

2.4. Participants and procedure

The study was conducted in the city of Timisoara (approximately 320,000 inhabitants), in western Romania. Participants were approached by the same person in parks, bars, festivals, churches, public universities, or the city center and asked to complete one of the two versions of a paper-and-pencil questionnaire. Such a variety of places was used so that the resulting sample is fairly representative of the general population of the city. All participants gave their informed consent for participation and were debriefed at the end of the study. Of the 339 persons approached, 196 agreed to complete the questionnaire, resulting in a participation rate of 57.8%. Of these, 18 participants were removed from further analyses on the following considerations: Twelve were not of Romanian ethnicity, two failed to follow the instructions, and four misidentified the article’s topic used in the manipulation. Therefore, the final sample consisted of 178 Romanian ethnics, which ranged in age from 16 to 67 years (M = 31.85, SD = 11.32). In terms of income, 13.5% declared below minimum income, 38.2% below average, 43.8% above average, and 4.5% did not state their income. Regarding education, 0.6% stated primary school, 1.7% secondary school, 24.2% high school, 6.2% post-high school studies, 42.1% bachelor, and 24.2% postgraduate studies as their highest educational achievement, while 1.1% did not declare their highest educational achievement.

2.5. Measures

Apology was manipulated using a printed online article embedded within the questionnaire, presented before the dependent variables were measured, and randomly assigned to approximately half of the participants (N = 87) (see Image 1 in Supporting Information). The article appeared to originate from a well-known Romanian news agency and was written in a style consistent with that of the publication. The apology was conceived to contain an expression of remorse (Nunney & Manstead 2021), an implied admission of responsibility (see Tavuchis 1991), as well as a stated desire to restore and strengthen the relationship with the victim group (see Hareli & Eisikovits 2006). Additionally, given that research suggests that costly apologies are perceived as more effective than less costly ones (Blatz & Philpot 2010), the article included a statement highlighting the symbolic costs for the Romas (see Supporting Information for the full text of the apology). Three items were used to determine if the apology manipulation was successful, each measured on a 5-point scale (1 = strongly disagree, 5 = strongly agree): ‘I think that Romas are sorry for their wrongdoings’ (remorse), ‘I believe that Romas took some responsibility for their wrongdoings’, (responsibility) and ‘I believe that Romas intend to strengthen their relationship with Romanians’ (restore relationship).

Contact with the apologizing group was measured following the approach of Voci and Hewstone (2003), assessing both the quality (‘In general, when you meet Romas, do you find the contact rather positive or negative?’; 1 = very negative, 5 = very positive) and quantity (‘How much contact have you had with Romas at home, work, school, or in social situations?’; 1 = not at all, 5 = a great deal) of contact. In line with their approach, a composite score was created by multiplying quality by quantity of contact.

Forgiveness was measured with three items adapted from Wohl et al. (2012), such as ‘It is important that Romanians never forgive the wrongs done by the Romas’ (1 = strongly disagree, 5 = strongly agree; α = .79).

Trust was assessed with three items adapted from Noor et al. (2008), including ‘Few Romas can be trusted’ (1 = strongly disagree, 5 = strongly agree; α = .75).

Perceived sincerity of the apology and perceived typicality of the apologizing representatives were each measured with a single item (1 = strongly disagree, 5 = strongly agree): ‘I believe that Romas’ apologies that I read in the article were sincere’ (sincerity) and ‘I believe that the Romas that apologized in the article are representative of Romas in general’ (typicality).

3.1. Manipulation checks

Success of randomization was checked by comparing the apology (N = 87) with the no-apology (N = 91) conditions on all demographics and on the contact variables. No significant differences between the two conditions emerged on all studied outcomes (all ps > 0.194), indicating that randomization was successful. Next, the apology and no-apology conditions were compared based on the three manipulation check measures. There were significant differences between the conditions regarding perceptions of remorse (mean difference = 0.47, t(176) = 3.21, p = 0.002, d = 0.48), perceptions of assumed responsibility by the Romas (mean difference = 0.46, t(176) = 2.90, p = 0.004, d = 0.43), and perceptions that the Romas intend to restore their relationship with Romanians (mean difference = 0.48, t(176) = 3.05, p = 0.003, d = 0.45). These results indicate that the manipulation was successful and produced effects of moderate magnitude.

3.2. Hypotheses testing

To test whether intergroup apologies lead to increased forgiveness (H1) and trust (H4) toward the less powerful group, two independent samples t-tests were conducted, in which the apology and no-apology conditions were compared on forgiveness and trust. Normality was examined for each group, and no substantial deviations from a normal distribution were detected. There was no significant difference between the apology (M = 3.85, SD = 0.76) and no-apology (M = 3.78, SD = 0.93) conditions regarding forgiveness, t(176) = 0.55, p = 0.291, d = 0.08, indicating that the apology had no impact on forgiveness of Romas. Therefore, H1 was not supported by the data. However, there was a significant difference between the apology (M = 2.85, SD = 0.77) and no-apology (M = 2.56, SD = 0.69) conditions regarding trust, t(176) = 2.62, p = 0.005, d = 0.40. This suggests that participants who received the apology expressed significantly higher trust in Romas, compared to those who did not, providing support for H4.

To examine whether perceived sincerity of the apology positively predicts forgiveness (H2) and trust (H5), and whether perceived typicality of the apologizing representatives positively predicts forgiveness (H3) and trust (H6), sincerity and typicality were entered simultaneously into two different regression models predicting forgiveness and trust. These analyses were conducted exclusively on participants exposed to the apology manipulation (N = 87), as sincerity and typicality were not measured in the no-apology condition. Analyses of linearity, normality of residuals, and homoscedasticity revealed no serious violations. There were also no multicolinearity issues between predictors (VIF < 4). In the model predicting forgiveness, sincerity emerged as a significant predictor (β = 0.54, p < 0.001), supporting H2, whereas typicality was non-significant (β = 0.13, p = 0.090), failing to support H3. Together, the predictors accounted for approximately 37% of the variance in forgiveness (R² = 0.365). In the model predicting trust, both sincerity (β = 0.29, p = 0.003) and typicality (β = 0.35, p < 0.001) were significant predictors, thereby supporting H5 and H6. The model explained about 29% of the variance in trust (R² = 0.289) (see Table 1).

To further examine how apology effectiveness varied with perceived sincerity and typicality, the no apology, low-sincerity/typicality, and high-sincerity/typicality groups were compared. For this purpose, sincerity and typicality were first dichotomized based on their respective mean scores. Participants scoring below the mean were assigned to the low group, while those scoring above the mean were assigned to the high group (see Table 2 for group means and standard deviations). Assumptions checks revealed no substantial deviations from normality and no violations of homoscedasticity. Four ANOVAs were then conducted to evaluate differences between the low, high, and no-apology groups for both sincerity and typicality. Post-hoc tests with LSD correction were applied. Regarding the effect of sincerity on forgiveness, participants in the high-sincerity group reported significantly greater forgiveness than those in the no-apology group, t(172) = 2.96, p = 0.004, d = 0.56, and the low-sincerity group, t(172) = 4.08, p < 0.001, d = 1.04. Interestingly, an almost significant boomerang effect emerged for insincere apologies, as participants in the low-sincerity group displayed lower forgiveness levels than those in the no-apology condition, t(172) = –1.79, p = 0.075, d = –0.33 (see Figure 1). Regarding the effect of sincerity on trust, participants in the high-sincerity group reported significantly greater trust than those in the no-apology condition, t(173) = 4.70, p < 0.001, d = 0.81, and the low-sincerity group, t(173) = 4.39, p < 0.001, d = 0.94. No significant difference was found between the no-apology and low-sincerity groups regarding trust, t(173) = 0.46, p = 0.644, d = 0.11.

Figure 1

Concerning the effect of typicality on forgiveness, there were no significant differences in forgiveness between the three groups (i.e., all ps > 0.05). However, significant differences in trust emerged between the three typicality groups: participants in the high-typicality group reported significantly greater trust than those in both the no-apology group, t(173) = 4.06, p < 0.001, d = 0.70, and the low-typicality group, t(173) = 3.40, p < 0.001, d = 0.72. No significant difference in trust was observed between the no-apology and low-typicality groups, t(173) = 0.11, p = 0.910, d = 0.01 (see Figure 2). These findings underscore the critical importance of recipients’ perceptions regarding the apology.

Figure 2

To test whether intergroup contact is positively associated with forgiveness (H7) and trust in the less powerful group (H8), contact was regressed on forgiveness and trust. Checks revealed no serious violations of model assumptions. Forgiveness was significantly predicted by contact (β = 0.24, p < 0.001), indicating that greater contact was associated with more forgiveness, thus supporting H7. Similarly, trust was significantly predicted by contact (β = 0.33, p < 0.001), indicating that greater contact was associated with higher trust, thereby supporting H8.

To examine whether intergroup contact moderates the relationship between apology and forgiveness (H9) and between apology and trust (H10), the interaction between contact and apology was investigated for both forgiveness and trust. For this purpose, two regression models were conducted in which apology, contact, and their interaction were introduced as predictors for forgiveness and trust, respectively. The interaction terms between contact and apology were non-significant for both forgiveness, β = 0.21, t(171) = 0.98, p = 0.329, and trust, β = 0.09, t(171) = 0.42, p = 0.678. Therefore, H9 and H10 were not supported by the data.

4.1. Limitations and recommendations for future studies

The present study has some important limitations that are worth mentioning. First, it might suffer from methodological limitations. Specifically, intergroup contact, sincerity, and typicality were measured rather than experimentally manipulated, meaning that conclusions about their impact on the studied outcomes are subject to the usual constraints of correlational research. In addition, sincerity and typicality were measured with only a single item, which may limit the reliability of their operationalization, while intergroup contact was self-reported. Consequently, participants may have been biased in their estimates of how much and how positive their contact with Romas was in the past. Nevertheless, such bias is likely limited, as evidence shows that self-reported measures are still valid measures of contact (see Hewstone et al. 2011). Furthermore, participants were assumed to possess a collectivistic orientation, although their degree of collectivism was not measured. Future studies could incorporate such measures to provide stronger support for the interpretations offered.

Second, the manipulation may suffer from ecological validity limitations. A written article does not necessarily carry the same impact as an apology delivered through official channels by high-status representatives. Additionally, broader social dynamics unfolding after an apology, such as media coverage or public reactions, may amplify or attenuate its effects. Future research could examine the potential influence of contextual realism on recipients’ reactions following an apology. The timing of measurement may also influence some conclusions, as we assessed the dependent variables immediately after the manipulation. Several authors (e.g., McCullough et al. 2003; Philpot & Hornsey 2008; Sells & Hargrave 1998) have argued that apologies may exert delayed rather than immediate effects on intergroup outcomes, potentially producing stronger impacts over time. However, the opposite may also occur; within the context of an ongoing intergroup conflict, the effect of an apology may fade quickly or be easily forgotten (see Philpot & Hornsey 2011). Future longitudinal research should therefore explore how the impact of intergroup apologies evolves and identify the mediating and moderating factors that shape these temporal dynamics.

Third, it is important to acknowledge that the control condition consisted of a simple no-apology text, which does not entirely rule out alternative explanations for the observed effects, such as demand characteristics or priming effects. Finally, as these findings were obtained in a specific socio-cultural context and based on a non-representative sample, they should be replicated in other intergroup conflict settings to determine whether the effects are robust and generalizable.

4.2. Conclusion

The results of this study suggest that apologies from low-power groups can produce positive intergroup effects, thereby showing that such groups can play an active role in reconciliation. Such apologies seem to be most effective when they are perceived as sincere and when the individuals issuing them are viewed as representatives of their group. In practical terms, this implies that reconciliation efforts should pay close attention not only to the content of the message but also to how apologies are communicated and by whom. At the same time, given the sensitive nature of such gestures, it is recommended that they be offered voluntarily, be supported by the group, and remain free from external pressure in order to avoid opposition within the group or reinforcing existing power inequalities.