Triadic intergroup relations: Studying situations with an observer, an actor, and a recipient of behavior
Abstract
The question of how onlookers make sense of acts they witness between an actor and a recipient of the action, and how the group memberships of the three inform outcomes, is of clear relevance to intergroup research on contact, collective action, and beyond. However, most intergroup research to date has focused on dyadic relationships rather than triadic relationships with an observer, an actor, and a recipient of behavior. Maybe because of the dominance of the dyadic approach, the design implications of focusing on triadic (rather than dyadic) relationships have not always been fully appreciated, leading to the introduction of confounds. This contribution will refer to some previously published studies to illustrate this, and outline recommendations for the study of triadic intergroup relationships in the future.
1 INTRODUCTION
Situations which involve an observer, an actor, and a recipient of a certain behavior displayed by the actor arise frequently. People often observe the behavior of someone else (the actor) toward a third party (the recipient). For example, a random 5 min of watching the evening news, or browsing the internet, might provide information to a British consumer of information about how the Germans behave toward Syrian refugees; how the USA is amending its Middle East policies toward Israel; about France suffering attacks by IS; and so on. How do group memberships between the observer, actor, and recipient of the behavior influence those reactions?
Situations with three parties are also relevant to research on intergroup contact and collective action. In a nutshell, intergroup contact theory (Allport, 1954, see also McKeown & Dixon, 2017) proposes that positive intergroup contact is a powerful mechanism to improve intergroup attitudes. The large body of work on collective action highlights those conditions which will cause members of disadvantaged groups to rebel against the status quo (Van Zomeren, Postmes, & Spears, 2008). Recent debates in the field have focused on the question to which extent intergroup contact should be seen not as a positive force for intergroup harmony, but as a negative force which adds to the sedation and repression of disadvantaged groups who have just cause to rebel (Cakal, Hewstone, Schwär, & Heath, 2011; Dixon, Levine, Reicher, & Durrheim, 2011).
Importantly for the present contribution, originally both intergroup contact research and collective action research was very much focused on dyadic intergroup relations (e.g., contact between group A and group B, collective action of disadvantaged group C against group D). However, more recently both literatures have branched out to consider situations with not two but three actors. In the case of contact research, scholars have looked at the effects of vicarious and extended contact (e.g., Vezzali, Hewstone, Capozza, Giovannini, & Wölfer, 2014). Such studies show that simply knowing that ingroup members have outgroup friends, or observing these friendships vicariously, can enhance intergroup attitudes. In this situation, then, the three relevant entities in the triad are the self, other ingroup members, and outgroup members. In the case of collective action research, scholars (e.g., Glasford & Caraballo, 2016; Mallett, Huntsinger, Sinclair, & Swim, 2008) have recently focused on situations where members of a group might be motivated to engage in collective action to better the situation not of their own group, but of a second group who is being oppressed by a third group. In such situations, clearly three or more groups are psychologically relevant (the observer's ingroup A, a privileged group B, and an underprivileged group C). A few scholars (e.g., Caricati, 2018; Kerr, Durrheim, & Dixon, 2017) have reflected on the theoretical implications of having more than two entities present in a psychological situation. However, the design implications of studying triadic intergroup relations have to date not received focused attention; and the goal of the present contribution is specifically to highlight how such situations can be studied (or not) using experimental design.
The present contribution will aim to outline some future pathways to better understand such triadic situations. In the following, I will (a) first present a brief overview of intergroup research to illustrate the point that the study of triadic intergroup relations takes the back seat compared to the study of dyadic intergroup relations. I will then (b) briefly review how triadic relations have been studied in the interpersonal (as opposed to intergroup) literature, as those insights will be of heuristic value. I will then (c) discuss how triadic intergroup situations can be systemized, and outline, with reference to published research, how neglecting a systematic approach to triadic intergroup relations has led to the introduction of confounds. I will (d) conclude with some design recommendations and avenues for future work.
2 PREVALENCE OF DYADS IN INTERGROUP RESEARCH
A casual glance at any text book chapter on intergroup relations/processes (e.g., Crisp & Turner, 2010; Hogg, & Vaughan, 1995) demonstrates that the vast majority of seminal research in this field has been focused on dyads. Brown's (2000) volume on group processes, entirely dedicated to intra- and intergroup processes, illustrates the point. The two chapters which focus on intergroup processes present an in-depth discussion of four classic approaches to intergroup relations, namely realistic group conflict theory, relative deprivation theory, social identity theory (SIT), and intergroup contact theory.
Realistic group conflict theory emerged out of the famous summer camp studies (Sherif, 1966) which investigated the intergroup dynamics between two groups of boys. Of focal interest in those studies were the effects of the presence or absence of superordinate goals, that is, goals shared commonly between both groups. Unsurprisingly, given the origin of the theory, it is concerned with dyadic dynamics between groups.
Relative deprivation theory (e.g., Walker & Smith, 2002) considers, aside from temporal deprivation whereby individuals compare their current situation with their past, collective deprivation. Collective deprivation is thought to arise when one relatively worse-off group compares their lot with a relatively more well off group (Vanneman & Pettigrew, 1972). Some important studies in this realm focus on groups which are, in lay persons’ eyes, typically conceived of as “naturally” dyadic, for example, men and women (Crosby, 1982). Moreover, sometimes the relations between certain dyads are so heavily politicized that juxtaposing them is the obvious choice (e.g., Black–White relations during the civil rights movement). In such contexts, the research focus on prominent dyads enables social psychology to address those issues which are most pertinent and pressing in people's minds, and those issues which have the most practical and applied relevance. Other deprivation studies, however, concern themselves with perceived deprivation between two groups which do not at all form obvious dyads. For example, it is far from obvious that London residents of Asian descent will turn to comparisons with “White people” when evaluating whether they should feel deprived. Comparisons with a Black next-door neighbor might be much more pertinent, psychologically speaking (Zagefka & Brown, 2005). In sum, then, while in some instances the focus on dyads can be solidly justified; it might be misguided in other instances. Nonetheless, the bulk of work on relative deprivation considers dyadic intergroup relations only, without acknowledging the greater complexities of the social world. Perceived deprivation will often result from comparisons whose structures are considerably more complex than the dyadic approach typically taken in relative deprivation research can do justice to.
Social Identity Theory’s (SIT, Tajfel & Turner, 1986) primary focus is dyadic—it theorizes the conditions which motivate a suppressed group to rebel against the suppressing group. Nonetheless, the complexities of the social world are acknowledged in certain parts of the theory. In its original formulation, SIT stated that if members of suppressed groups are unable to address the status quo and improve either their own personal or their group's position, they will engage in social creativity strategies, which are effectively psychological coping mechanisms to deal with an unpleasant situation which cannot be changed. One such strategy is to engage in downward comparisons, that is, to focus on even less fortunate groups—this will inevitably result in one's conundrum appearing less severe. Implicit in this prediction is, of course, the assumption that often there are more than two groups relevant and salient in a given intergroup context (the ingroup, the superior outgroup, and one or more inferior outgroup(s)). Nonetheless, despite the theory acknowledging that there are different social groups within the social strata, the main focus of SIT is on dyadic relations.
Intergroup Contact Theory (Allport, 1954) suggests that relations between two groups can be improved through intergroup contact, as long as this contact occurs under favorable conditions. Of course, some theoretical refinements of the theory acknowledge that the social world is a complex place, for example, work on cross-cutting group memberships (Crisp & Hewstone, 2007; Doise, 1976) and on superordinate identities (Gaertner, Dovidio, Nier, Ward, & Banker, 1999; Waldzus & Mummendey, 2004; see also work on social identity complexity, Brewer & Pierce, 2005). Again, however, the vast majority of research in this area has focused on the effects of contact on dyadic group attitudes (Binder et al., 2009).
Apart from these four classic approaches to intergroup relations which are covered in virtually every social psychology textbook, of course there is a myriad of other theoretical accounts of intergroup relations. However, just as with the seminal approaches described above, although some of these do acknowledge the complexity of the social world, most nonetheless focus on conveniently simplistic dyads. Acculturation research can serve as an illustration.
This research area is concerned with the cultural changes that occur when two distinct groups come into contact with each other (Redfield, Linton, & Herskovits, 1936). In theoretical approaches to acculturation (Berry, 1997), as in empirical studies on the subject (Brown & Zagefka, 2011), dyadic relations are considered. Examples are studies which focus on an ethnic minority and an ethnic majority (Ward, 1996), on an indigenous and a non-indigenous group (Zagefka, Brown, & Gonzalez, 2009), or on two religious groups (Tip et al., 2012). This emphasis on dyads is likely to often oversimplify people's psychological realities. For example, it might well be that a British Indian adolescent in inner city London does not define his/her identity merely in relation to Indian and British culture, but very much in relation to, or opposition to, an Afro-Caribbean minority group which happens to numerically dominate that particular part of London. Although one or two contributions consider more than two groups (e.g., Montreuil & Bourhis, 2001), theoretical models that go beyond dyadic relations are notable by their absence in acculturation research.
Of course, some isolated studies exist (Kressel, 2000; Weinberg-Kurnik, Nadan, & Ari, 2015) which do emphasize the importance of triadic relationships in intergroup settings. One example is Simon and Klandermans (2001), who discuss the nature of the relationship between ingroups, outgroups, and a third party in understanding politicized collective identities and collective action. However, far more common are studies which focus on dyadic relations such as those between men and women (Choma, Hafer, Crosby, & Foster, 2012); Blacks and Whites (Hehman et al., 2012); or stigmatized and non-stigmatized people (Crocker & Major, 1989). The huge body of work on dyadic intergroup processes has undoubtedly generated a plethora of important findings and insights. Still, our knowledge of intergroup processes will remain incomplete unless we do justice to relations which involve more than two entities.
3 TRIADS IN INTERPERSONAL RESEARCH
Our social world is filled with triadic relationships. Starting with the first triad from which we emerge, as a child borne to a mother and a father, we might graduate to be involved in love triangles, belief in the Christian doctrine of the holy trinity, or part-take in sexual threesomes. The ancient Greeks acknowledged the importance of triads as early as the fifth century BC. Play writing evolved from being monologue-based to being restricted to dialog until Sophocles introduced a third actor to the play's structure, for the first time staging triadic relationships in Greek amphitheaters (Kennedy, 1983). Of course, as highlighted in the classic works of sociologist George Simmel, relations between two people involve fundamentally different dynamics than relations between three people (Leck, 2000). For example, triadic relationships offer opportunities for complex competition, the formation of alliances, or for mediation (Farganis, 1993; Ritzer, 1992). Evidence from imaging data also points to the very different psychological properties of dyadic and triadic interactions: this data shows that triadic interactions involve different brain regions than dyadic interactions (de Langavant, Jacquemot, Bachoud-Lévi, & Dupoux, 2013).
Triadic interactions have attracted attention in research on family dynamics and in developmental psychology (e.g., de Mendonça, Cossette, Strayer, & Gravel, 2011; Flaskas, 2012). This work has considered both mother-father-child triads (Kwon, Jeon, Lewsader, & Elicker, 2012) and adult-child-object triads (Palacios & Rodríguez, 2015), although overall dyadic theories (e.g., attachment theory; Bowlby, 1979) prevail in developmental research.
In addition to family and developmental research, another research field which has paid attention to triads is clinical and counseling psychology. This might be creditable to the influence of systemic theory and systemic therapeutic approaches (Bowen, 1991). Systemic theory studies social systems (e.g., families); the approach is hence sensitive to the fact that interactions often involve more than two people. Consequently, the need to consider the impact of triadic relationships on various outcome measures such as depression is appreciated in the literature on psychopathology (e.g., Doohan, Carrère, Siler, & Beardslee, 2009; Du Rocher Schudlich et al., 2015; Hollenstein, Allen, & Sheeber, 2016).
Without a doubt, the most prominent interpersonal approach to triads, and that of most heuristic value for the present purpose, is Heider (1958) Balance Theory. The theory proposes that humans strive for a sense of balance regarding their attitudes. The central assumption is that humans attempt to reach a harmonious state in which the entities present in a given situation and the attitudes between them fit together without stress. Considered are three entities contained in a given situation: a Person (P), an Other (O), and a third thing (X) (see Figure 1). Psychological imbalance is experienced, for example, when P relates positively to O, P relates positively to X, but O relates negatively to X. To illustrate, imbalance exists if a person (P) is friends with two other people (O and X), but O and X do not like each other. Balance theory suggests that this imbalanced situation can be resolved in a number of ways. P can stop liking O; P can stop liking X; P can persuade O to like X; or P can “split” his/her perception of O into a “positive” and a “negative” part, with only the negative part of O being part of the (now balanced) triangle.
Heider's (1958) versus Dollard et al.’s (1939) triad [Color figure can be viewed at wileyonlinelibrary.com]
Balance theory is most frequently presented as a theory of attitude change, and the actors considered by the theory are typically individuals rather than group members. However, the theory also lends itself to the application to group contexts (e.g., Crandall, Silvia, N'Gbala, Tsang, & Dawson, 2007; Fink, Parker, Brett, & Higgins, 2009). One example is the use of the following principle by politicians: “my enemy's enemy is my friend.” This is what the British Prime Minister Winston Churchill alluded to when, anticipating Germany's invasion of the Soviet Union during the Second World War, he remarked: “if Hitler invaded Hell, I would make at least a favorable reference to the Devil in the House of Commons.” The remark by George W. Bush (president of the United States from 1989 to 1993) in 2001 that “you are either with us or against us” is another example. The attitudes expressed in the quotes by Churchill and Bush cited above can be phrased in Heider's terminology. Churchill illustrates that if O (Hitler) negatively related to X (Hell/the Devil/the Soviet Union), then P (Churchill) would take a more positive attitude toward X, given that P relates categorically negative to O. Bush illustrates that given that P (the United States) relates negatively to X (U.S. enemies), P will relate negatively to O (other countries) if O relates positively to X, but P will chose to relate positively to O if O relates negatively to X.
4 RESEARCH ON TRIADIC INTERGROUP RELATIONS
Although, as outlined above, most intergroup research focuses on dyadic relationships, some previous contributions have focused on triadic relationships. I will discuss some examples. For example, research on nonverbal behavior has used paradigms where a judge who observes behavior between two other people may or may not share a group membership with the actor or the recipient of the observed behavior (Richeson & Shelton, 2005). To give another example, the paper by Wohl and Branscombe (2008) charts the consequences of remembering historical victimization of the ingroup for emotional reactions to a current adversary. In Experiment 1, Jewish Canadians who were reminded of the Holocaust accepted less collective guilt for their group's harmful actions toward Palestinians than those not reminded of their ingroup's past victimization. The process described by this research fits the triad described in Dollard, Miller, Doob, Mowrer, and Sears (1939) rather than Heider (1958), as illustrated in Figure 1: Dollard's frustration–aggression hypothesis (Dollard et al., 1939) proposes that how an agent perceives to be treated by one entity impacts on how they themselves treat another entity. In other words, there is proposed to be a causal chain from O to P to X (see also Zitek, Jordan, Monin, & Leach, 2010, for a similar mechanism). Making salient the treatment of the Jews (P) by the Nazis (O) has consequences for how the Jews respond to a third group, that is, the Palestinians (X).
In contrast, for Heider (1958) the way the agent (P) relates to an object (O), the way the object relates to another entity (X), and the way the agent relates to the third entity have mutual implications for each other and in cases of imbalance need to be changed to achieve equilibrium. The effects of vicarious contact (Vezzali et al., 2014) can be framed in terms of Heider's theory. Vicarious contact effects show that first-hand contact with outgroup members is not necessary to affect attitude change; instead, it is sufficient to merely know that other ingroup members have contact with outgroup members. In Heider's terms, a system where an observer (P) has a positive attitude to fellow ingroup members (O), and knows that those ingroup members have a positive attitude (i.e., seek contact) with outgroup members (X), will be unbalanced as long as the observer (P) continues to have a negative attitude, and to avoid contact, with those outgroup members (X). In this context, attitude change due to vicarious contact can be attributed to Heider's law of humans striving for attitudinal balance within the triadic system.1
Other studies which focus on the impact of group memberships in situations where an observer reacts to actions of one entity toward another are found in the literature on intergroup helping. Studies interested in factors which would entice donors to help victims of humanitarian disasters in situations where there is a humanly cased need for help fall into this category (Zagefka, Noor, Randsley de Moura, Hopthrow, & Brown, 2011). In such situations, there are clearly three relevant entities: the donor, the perpetrator, and the victim. Moreover, group memberships are often relevant in these contexts, for example, when donors in developed countries decide to assist (or not) Eritrean civilians who suffer from military action instigated by Ethiopian (or Eritrean) forces. In the following, I will home in on those triadic intergroup relations which contain an observer, an actor, and a recipient of behavior.2 I will first show how such situations can be systematized, in order to then discuss, with reference to published research, why this systemization is important, and how ignoring a systematic approach can lead to the introduction of confounds.
4.1 Five potential intergroup constellations
In triadic situations with three entities, there are five potential ingroup/outgroup membership constellations. These are illustrated in Figure 2.3 The first case (1) is when all three entities (the observer, the actor displaying some kind of behavior, and the recipient of this behavior) belong to the same group. An example would be a female employee (possibly a feminist, with chronically high salience of gender identity) witnesses a female colleague acting with a lack of solidarity toward another female colleague. The second case (2) is when observer and actor share a group membership, but this group membership is not shared with the recipient of the behavior. An example would be a British person hearing about a British person attack a French person on the news. The third case (3) is when the observer and the recipient share a group membership, but neither of these two entities shares a group membership with the actor. This might be the case when U.S. American TV audiences see the American embassy in Nairobi attacked by Islamist terrorist organizations. The fourth case (4) is when the actor and recipient share a group membership, but neither of these two share a group membership with the observer. Consider, for instance, a situation where British TV audiences witness one African faction inflict havoc on another African faction. Last but not least, the fifth case (5) is when observer, actor, and recipient all belong to different groups. An example might be U.S. Americans learning that Iraqi civilians are suffering as a result of Iranian military action.
Five intergroup constellations [Color figure can be viewed at wileyonlinelibrary.com]
The examples illustrate the importance of taking as a starting point the subjective perceptions by the observer. For example, a British person listening to a report about an African war situation on the news might consider this an intragroup conflict (Africans hurting other Africans), or he/she might consider this an intergroup conflict (e.g., Ethiopians hurting Eritreans, or vice versa). Similarly, Brits reading about the civil war in Syria might consider this an intragroup conflict (Syrians persecuting other Syrians), or they might consider this an intergroup conflict (pro-government supporters fighting anti-government forces)—group memberships are not objectively given.
Recall also that, following self-categorization theory's (Turner, 1987) definition of intergroup behavior, a situation is intergroup by nature as soon as group memberships are salient to the observer in the scenario. Hence, it does not matter whether the scenario comprises one observer, one actor, and one recipient, or whether it instead comprises a group of observers, a group of actors, and a group of recipients. The only feature which is relevant for turning the interpersonal triad into an intergroup triad is the salience of group memberships, which is assumed to be given for all scenarios described here.
Why 5, not 8? Although the fact that there are five potential intergroup constellations is immediately obvious when consulting the Venn diagram in Figure 2, this basic fact is hard to grasp when the starting with a mind-set trained on 2 × 2 designs. When trying to study the interactive effects of the shared or separate group memberships in a triad, one might be tempted to use a 2 (observer–actor group membership shared or not) × 2 (observer–recipient group membership shared or not) × 2 (actor–recipient group membership shared or not) design. This would, of course, result in eight experimental conditions.
Beware, though, that in such a design three of the conditions constitute logical impossibilities. By definition, it is not possible for “Obs” to equal “Act,” for “Obs” to equal “Rec,” and for “Act” and “Rec” to be unequal. In other words, it is not possible for “Obs” and “Act” to be members of the same group, for “Obs” and “Rec” to be members of the same group, but for “Act” and “Rec” to not be members of that same group (at least when only one group membership is considered; clearly rather more complex patterns are possible when superordinate categories come into the picture).
- Obs = Act & Obs = Rec & Act ≠ Rec;
- Obs = Act & Obs ≠ Rec & Act = Rec;
- Obs ≠ Act & Obs = Rec & Act = Rec.
- Obs = Act & Obs = Rec & Act = Rec (constellation 1 in Figure 2);
- Obs = Act & Obs ≠ Rec & Act ≠ Rec (constellation 2);
- Obs ≠ Act & Obs = Rec & Act ≠ Rec (constellation 3);
- Obs ≠ Act & Obs ≠ Rec & Act = Rec (constellation 4).
- Obs ≠ Act & Obs ≠ Rec & Act ≠ Rec (constellation 5).
4.2 Previous research on intergroup constellations—confounds
As will become clear, consideration of the complete 5-constellation-design is important when studying situations involving an observer, an actor, and a recipient; because incompleteness can easily introduce interpretative ambiguity. Four examples will be discussed to illustrate this. These studies were chosen because they lend themselves to the illustration of different types of confounds that can emerge—they are not meant to single out practices by any particular researchers, but to provide example illustrations of the issues at hand.
Levine, Prosser, Evans, and Reicher (2005) studied football club supporters, and found that participants were more likely to help fellow members of the same football club (i.e., ingroup members) than they were to help people in need who were not supporting the participants’ favorite football team. In these naturalistic experiments, an emergency situation was created by having a confederate allegedly fall over and hurt themselves. Importantly, the situation studied by Levine et al. only has two entities: an observer (participant) and a recipient (victim in need). No actor/perpetrator is present who might have tripped up the victim, or who is otherwise to blame. Formally, with reference to Figure 2, O is irrelevant, and the only relevant aspect is the P–X relationship. Therefore, the conclusions offered by Levine et al. about the shared group membership between the two entities having had an effect are sound, and there is no interpretative ambiguity.
The situation is different in the studies by Wohl and Branscombe (2005), which focus on situations with a clear perpetrator. They thus introduce a third party. Across four experiments, the authors found that if a common membership in an inclusive superordinate category was made salient between the participants’ ingroup and the outgroup perpetrators, thus effectively turning outgroup perpetrators into ingroup perpetrators, this had important behavioral consequences. Collective guilt assignments to members of the perpetrator group were attenuated, and willingness to forgive increased.
In all of Wohl and Branscombe’s (2005) studies, the focus was on ingroup victims (i.e., Obs = Rec). Hence, their results do not speak to Scenarios 2, 4, or 5 of Figure 2. Their manipulation focused on the relationship between participants (observers, Obs) and perpetrators (actors, Act), which either did (Scenario 1 in Figure 2) or did not (Scenario 3 in Figure 2) share the same group membership. What these authors did not consider, however, is that by manipulating the relationship between participants (observers, Obs) and perpetrators (actors, Act), they inevitably also influenced the relationship between victims (recipients, Rec) and perpetrators (actors, Act). Put very simply, if all of humanity is subsumed in one group, this will not only mean that “Obs” and “Act” now share a group membership, but also that “Act” and “Rec” now share the same group membership in this global category. Equally, if “Obs” and “Act” do not share a group membership, but “Obs” and “Rec” do, then it follows that “Act” and “Rec” cannot possibly belong to the same group.
This can be expressed more formally. Wohl and Branscombe’s (2005) conditions compare Obs = Act with Obs ≠ Act, while Obs = Rec is held constant across conditions. As we saw above, however, Obs = Act and Obs = Rec by necessity implies Act = Rec. Moreover, Obs ≠ Act and Obs = Rec by necessity implies that Act ≠ Rec. This means that what varies between the conditions is not only how “Obs” relates to “Act,” but also how “Act” relates to “Rec,” generating interpretative ambiguity. The only way to avoid this would be to study all five intergroup constellations in one comprehensive design.
Of course, to support Wohl and Branscombe’s (2005) interpretation of their results one might cite the preponderance of human egocentricity. It might be a reasonable assumption that responses will be more strongly influenced by how the self (Obs) relates to others (Act and Rec) than by how those others (Act and Rec) relate to each other. However, there is also ample evidence that the relationships between others do influence responses, and that they are not behaviorally irrelevant. For example, reactions to sexual assault are influenced by the relationship between the sexual predator and the victim (e.g., victim blame is higher if the two people are married; Pollard, 1992). Hence, it is fair to say that the Act–Rec relationship in these studies is unlikely to be trivial.
Another study which has attended to aspects of the fivefold design is that by Palmer, Rutland, and Cameron (2015). In this contribution, the authors contrasted a situation where an ingroup perpetrator inflicted harm on outgroup victims with a situation where an outgroup perpetrator inflicted harm on ingroup victims. They found that only when the victim was an ingroup member and the aggressor an outgroup member did participants become more likely with age to report prosocial bystander intentions due to increased ingroup identification.
With reference to Figure 2, Palmer et al.’s (2015) study investigated processes in Scenarios 2 (shared group membership between observer and actor, and lack of shared group membership between observer and recipient) and in Scenario 3 (shared group membership between observer and recipient, and lack of shared group membership between observer and actor). Again, there are two dyadic relations which are clearly confounded in this design. This time what is confounded are the Obs–Act and the Obs–Rec relationships themselves, which were both manipulated by the authors. Expressed formally, what is compared is Scenario 2 (Obs = Act and Obs ≠ Rec, which by necessity implies Act ≠ Rec) with Scenario 3 (Obs ≠ Act and Obs = Rec, which by necessity implies Act ≠ Rec). As is evident, although Act ≠ Rec is inevitably constant across conditions, Obs–Act and Obs–Rec are conflated. James and Zagefka (2017) opted to manipulate the Obs–Act and the Obs–Rec relations independently from each other in a 2 (Obs = Act vs. Obs ≠ Act) × 2 (Obs = Rec vs. Obs ≠ Rec) design. In other words, one factor manipulated whether the perpetrator shared the participant's (observer's) group membership, and the other factor manipulated whether the victim shared the participant's group member. They were interested in how shared membership of a donor with either the victims or the perpetrators of an event would impact on donation decisions. As hypothesized, it was found that ingroup victims generated more empathy which in turn enhanced donations, and that ingroup perpetrators generated more perceived responsibility in the donor, which in turn enhanced donations.
Again, though, even though here the Obs–Act and the Obs–Rec relationships are not confounded, unambiguous interpretation of the findings is jeopardized by the Act–Rec relationship (i.e., the relationship between perpetrators and victims) which was not manipulated, and which is confounded with both the Obs–Act and the Obs–Rec relationship. This is illustrated in Table 1.
|
Observer and Victims Shared group Membership Obs = Rec |
Observer and Victims No shared group membership Obs ≠ Rec |
Marginals | |
|---|---|---|---|
| Observer and Perpetrators | Scenario 1 | Scenario 2 | Cell A |
| Shared group | Obs = Act | Obs = Act | 2 × Obs = Act |
| Membership | Obs = Rec | Obs ≠ Rec | Obs = Rec and Obs ≠ Rec |
| Obs = Act | Act = Rec (by necessity) | Act ≠ Rec (by necessity) | Act = Rec and Act ≠ Rec |
| Observer and Perpetrators | Scenario 3 | Scenarios 4 or 5 | Cell B |
| No Shared group | Obs ≠ Act | Obs ≠ Act | 2 × Obs ≠ Act |
| Membership | Obs = Rec | Obs ≠ Rec | Obs = Rec and Obs ≠ Rec |
| Obs ≠ Act | Act ≠ Rec (by necessity) | Act = Rec OR O ≠ Rec | Act ≠ Rec and (Act = Rec or Act ≠ Rec) |
| Marginals |
Cell C Obs = Act and Obs ≠ Act 2 × Obs = Rec Act = Rec and Act ≠ Rec |
Cell D Obs = Act and Obs ≠ Act 2 × Obs ≠ Rec Act ≠ Rec and (Act = Rec or Act ≠ Rec) |
To interpret the main effect for the Obs–Act relationship, one would focus on “cell A” and “cell B” in the right-hand-side column of the table, which averages across two cells where Obs = Act and compares this with the average across the two cells where Obs ≠ Act. The Obs–Rec relationship is not a confound, because in both cell A and cell B the average consists of one condition where Obs = Rec and one condition where Obs ≠ Rec; the Obs–Rec relationship is thus constant across the two marginal cells.
However, the Act–Rec relationship is not constant. In cell A, ½ of participants were (by logical necessity) exposed to Act = Rec and ½ of participants were exposed to Act ≠ Rec. The picture is different in marginal cell B. Here, half the participants were (by necessity) exposed to Act ≠ Rec, and the other half were given ambiguous information about the Act–Rec relationship. Although we cannot know with certainty how participants would have interpreted the ambiguous information about the Act–Rec relationship, there is no reason to assume that they would have favored one particular scenario. Hence, it is assumed that in the ambiguous cell half of the participants would have assumed Act = Rec, and the other would have assumed Act ≠ Rec. Thus, overall in cell B ½ of participants + (½ × ½ = ¼) of participants, that is, a total of ¾ of participants, would have been exposed to Act ≠ Rec, and only ¼ of participants would have been exposed to Act = Rec. Clearly, the Act–Rec relationship is thus not constant across cells A and B, and it constitutes a confound.
The same logic applies to the interpretation of the main effect of the Obs–Rec relationship. When comparing the marginal cells C and D, it is apparent that although the Obs–Act relationship is constant across the two cells, again the Act–Rec relationship is not. While ½ of participants were exposed to Act = Rec and Act ≠ Rec, respectively, in cell C, in cell D again ¾ can be assumed to have been exposed to Act ≠ Rec, and only ¼ would have been exposed to Act = Rec.
4.3 A tentative solution for unconfounding the third dyadic relationship
One way of addressing this issue in designs where the focus is on only two out of the three potential dyadic relations is to add an explicit instruction which ensures exposing equal numbers of participants to each of the possible relations regarding the third dyad. For example, the Act–Rec relationship would not have been confounded in cells A and B in Table 1 if participants in Scenario 1 had been explicitly told “remember, the actor and recipient also share a group membership,” if participants in Scenario 2 had been told “remember, the actor and recipient do not share a group membership,” if participants in Scenario 3 had been told “remember, the actor and recipient do not share a group membership,” and if participants in the last remaining experimental condition had been told “the actor and recipient do share a group membership” (as is the case in Scenario 4 but not Scenario 5 in Figure 2).
This is not a perfect solution, because exposure to different dyadic relationship information would not be perfectly independent of each other. For example, Obs = Act and Obs = Rec would never have been crossed with Act ≠ Rec (this is a logical impossibility, recall that two dyadic relations often imply what the third must look like). However, at least when interpreting the main effects (i.e., when focusing on the marginal cells), the proportion of participants who were exposed to each type of information on the third dimension would be equal across marginal cells.
Figure 3 illustrates the effect of including such deconfounding information about the third dyad in the experimental materials (which amounts to excluding Scenario 5) for the Obs–Act main effect in Table 1 (which compares cells A and cells B). Figure 3a illustrates a distribution of cases if no unconfounding information regarding the third dyad is provided, and Figure 3b illustrates a distribution of cases if unconfounding information is offered. As indicated previously, the Obs–Rec relationship was not a confound across those two cells; it will therefore be disregarded in the example. Figure 3 illustrates a hypothetical case with N = 4 for both the Obs = Act and the Obs ≠ Act conditions. In Figure 3a, the sample in the Obs = Act condition is equally split between Act = Rec and Act ≠ Rec. However, in Obs ≠ Act, if no unconfounding information is provided ¾ of the sample are in Act ≠ Rec. This dimension hence constitutes a confound. Figure 3b illustrates a scenario where participants were exposed to the explicit instructions about the third dyad which are outlined above. Now, the Act–Rec relationship is no longer confounded with the relationship of interest. While the proposed solution might be workable for reducing the impact of confounds on the interpretation of main effects, the picture is still tricky for the interpretation of interactions, as will become clear below.
Unconfounding the third dyadic relationship (Act–Rec) from the first (Obs–Act), by adding an instruction [Color figure can be viewed at wileyonlinelibrary.com]
4.4 A complete design with all five intergroup constellations
The discussion of Wohl and Branscombe’s (2005) design illustrates that it is not desirable to focus one's manipulations on one dyadic relationship only in situations where three entities are relevant. The discussion of James and Zagefka’s (2017) design further emphasizes that focusing on two dyadic relationships in these situations is not sufficient either. Thus, a complete design honoring all three dyadic relations within the triad must be employed. The complete design is illustrated in Table 2. How can this design be used to study the effects of different intergroup constellations?
| Act = Rec | ||
|---|---|---|
| Obs = Rec | Obs ≠ Rec | |
| Obs = Act |
Scenario/cell 1 Obs = Act Obs = Rec Act = Rec |
Logical impossibility |
| Obs ≠ Act | Logical impossibility |
Scenario/cell 4 Obs ≠ Act Obs ≠ Rec Act = Rec |
| Act ≠ Rec | ||
|---|---|---|
| Obs = Rec | Obs ≠ Rec | |
| Obs = Act | Logical impossibility |
Scenario/cell 2 Obs = Act Obs ≠ Rec Act ≠ Rec |
| Obs ≠ Act |
Scenario/cell 3 Obs ≠ Act Obs = Rec Act ≠ Rec |
Scenario/cell 5 Obs ≠ Act Obs ≠ Rec Act ≠ Rec |
4.4.1 Main effects
Focusing first on main effects, one might be tempted to contrast all cells in which two entities do belong to the same group with all cells in which the two entities do not belong to the same group. For example, to estimate the effect of the Obs–Act relationship, one might wish to compare Scenarios 1 and 2 on the one hand with Scenarios 3, 4, and 5 on the other hand. However, this would again introduce confounds. If one summarized across Scenarios 1 and 2 (i.e., the two scenarios where Obs = Act), “Obs” would equal “Rec” half of the time (i.e., for everyone in Scenario 1), and “Obs” would unequal “Rec” half of the time (for everyone in Scenario 2). Moreover, “Act” would equal “Rec” half of the time (for Scenario 1), and “Act” would unequal “Rec” the other half of the time (for Scenario 2). If one summarizes across Scenarios 3, 4, and 5, “Obs” would equal “Rec” one third of the time only (for Scenario 3, but not for Scenarios 4 or 5). Moreover, “Act” would equal “Rec” one third of the time only (for Scenario 4, but not for Scenarios 3 and 5). Hence, the Obs–Act relationship of interest would be confounded with both the Obs–Rec and the Act–Rec relationship in such a comparison. The same issue of confounds would emerge, as can be inferred from Table 2, if one were to estimate the effect of the Obs–Rec relationship by comparing Scenarios 1 and 3 with 2, 4, and 5; and if one were to estimate the effect of the Act–Rec relationship by comparing 1 and 4 with 2, 3, and 5.
For the interpretation of main effects, the solution, as outlined above, would be to include explicit information about all three dyadic relations in the design, and to exclude Scenario 5 from the contrast. To interpret the Obs–Act main effect, one can contrast Scenarios 1 and 2 (where Obs = Act) with Scenarios 3 and 4 (where Obs = Act). In this comparison, half of the participants in Scenarios 1 and 2 taken together, and half of the participants in Scenarios 3 and 4 taken together, are in Obs = Rec and in Obs ≠ Rec, respectively (cf., Table 2). Moreover, half of the participants in Scenarios 1 and 2 taken together, and half of the participants in Scenarios 3 and 4 taken together, are in Act = Rec and in Act ≠ Rec, respectively. Hence, the main effect of Obs–Act can now be interpreted without confounding influences of Obs–Rec or Act–Rec. In the same vein, to interpret the Obs–Rec main effect, one might contrast Scenarios 1 and 3 on the one hand with Scenarios 2 and 4 on the other. Again, half of the participants in each contrast group will fall into Obs = Act and into Obs ≠ Act; and into Act = Rec and into Act ≠ Rec, respectively. Last but not least, the main effect of the Act–Rec relationship can be tested by contrasting Scenarios 1 and 4 on the one hand with Scenarios 2 and 3 on the other hand. Again, equal numbers of participants in both contrast groupings will fall into Obs = Act and into Obs ≠ Act; and into Obs = Rec and into Obs ≠ Rec, respectively.
Why, then, would one want to include Scenario 5 in a design at all? Even though it is clearly a hindrance rather than a help to the interpretation of main effects, Scenario 5 clearly constitutes an interesting theoretical case in its own right, which therefore cannot be ignored. It denotes a situation where a member of group A observes a member of group B behaving in a certain way toward a member of group C. This situation can be assumed to arise frequently, and studying it will be of practical relevance.
4.4.2 Interactions
An important question concerns the interpretation of interactive effects between the three dyadic relationships. Unfortunately, there is no way of a clean interpretation of the three two- and the one three-way interactions between the three factors Obs–Act, Obs–Rec, and Act–Rec. To illustrate this clearly, the information in Table 2 is summarized in slightly different format in Table 3. As a quick perusal of Table 3 will illustrate, any potential two-way interactive pattern invariably has a confound on the third dimension. Imagine, for example, I was to hypothesize particularly strong effects if both Obs = Act and Obs = Rec are given. Contrasting this option (Scenario 1) with all others implies having unequal Act–Rec relations across the two contrasted conditions. While Act = Rec for 100% of participants in Scenario 1, this is not the case when averaging across Scenarios 2, 3, 4, and 5.
| Scenario 1 | Scenario 2 | Scenario 3 | Scenario 4 | Scenario 5 |
|---|---|---|---|---|
| Obs = Act | Obs = Act | Obs ≠ Act | Obs ≠ Act | Obs ≠ Act |
| Obs = Rec | Obs ≠ Rec | Obs = Rec | Obs ≠ Rec | Obs ≠ Rec |
| Act = Rec | Act ≠ Rec | Act ≠ Rec | Act = Rec | Act ≠ Rec |
- Note. Options where only one out of the three relationships is unequal are missing, because this constitutes a logical impossibility.
Hence, if the theoretical focus is on interactive rather than main effects, the only possible solution is to consider each of the five scenarios as one holistic entity which cannot be broken down further. This approach would be akin to how many other naturally confounded factors might be treated in research. For example, a study which is interested in whether “gender” interacts with “type of tutoring” in its effect on “performance” will treat “gender” as one holistic variable. No attempts would be made to break this factor down further into constituent components such as socialization, testosterone levels, and childbearing capacity. It is argued that the same holistic approach must be taken when comparing the five intergroup constellations. This would imply analysis not in terms of an (incomplete) 2 × 2 × 2 design, but analysis in terms of a one-way ANOVA with five levels. Although this will lead to a rather high-level account of the processes in each of the scenarios, it is the only approach which can avoid confounds.
To illustrate (see Figure 4), one might run a study where:
Illustration of the five intergroup constellations [Color figure can be viewed at wileyonlinelibrary.com]
- American observers witness other Americans inflict harm on Americans (Scenario 1)
- where Americans witness other Americans inflict harm on Iraqis (Scenario 2)
- where Americans witness Iraqis inflict harm on Americans (Scenario 3)
- where Americans witness Iraqis inflict harm on Iraqis (Scenario 4)
- and where Americans witness Iraqis inflict harm on Kurds (Scenario 5).
I have not come across any studies investigating such dynamics in one comprehensive design, which considers all five possible intergroup constellations, despite the fact that it seems that much might be gained from it. Investigating behavioral outcomes of the five scenarios—whether motivated by a desire to better understand helping and donations, outrage and other intergroup emotions, or perceived responsibility to intervene—would clearly yield findings of both theoretical and applied importance.
5 SUMMARY AND DISCUSSION
Triads are all around us. Does it matter, in terms of factors shaping reactions, whether the person attacking my daughter is a member of my family or not? Does it matter, psychologically speaking, whether those people who suffer at the hand of my fellow countrymen are members of my moral ingroup or not (Reykowski, 2002)? Are observers expected to react differently, depending on whether civilians fall victim to war atrocities committed by their own army, or another country's army? Common sense suggests that the answer to all these questions should be “yes.” However, up to this day a theoretical account as to what can be expected in different intergroup constellations, and how to best approach studying these predictions, has been missing. The present paper hopes to provide the first step in an attempt to remedy this.
As was seen, although triadic constellations have been studied in interpersonal research, intergroup research has mainly focused on dyads. Men versus women, blacks versus whites, minorities versus majorities have been the order of the day. Although some previous intergroup research has tapped into certain types of intergroup constellations, not doing so systematically has led to the introduction of confounds. It was shown that in triadic intergroup relations there are five intergroup constellations, and some methodological recommendations were made for their study.
These recommendations can be summarized very simply, and following them will help avoid confounds in experimental designs which test dynamics between an observer, an actor, and a recipient of behavior. If researchers are interested in only two out of three potential dyads in a situation where three entities are relevant (e.g., if they are interested in whether the observer shares a group membership with the actor on the one hand and the recipient on the other, but it is not of theoretical interest to study the effects of the actor–recipient relationship), it is important to include instructions about the third dyad also, to ensure it is not confounded with the effects which are studied (without providing this information, confounds are inevitable, recall Table 1). This allows the interpretation of the main effects, although interpretation of interactions would remain problematic. If researchers are interested in interpreting the effects of all three dyads jointly, the only workable solution is to interpret the intergroup situation as a holistic entity which cannot be broken down further into its constituent components. This implies a design with one factor which has five different levels.
To illustrate, I will give some examples for contact and collective action research. Imagine a contact researcher who is interested in how group memberships affect the consequences of observed vicarious, positive contact. Let's say he/she is interested in the effects of observing someone helping someone else, or smiling at someone else. Does this vicarious positive interaction have different effects, depending on whether the observer shares a group membership with the actor or with the recipient of the behavior? Is it as effective for me to witness my mates being kind to outgroup members as it is for me to witness outgroup members being kind to my mates? When considering systematically all five intergroup constellations, those questions can be asked and answered, without the introduction of confounds. Crucially, it will not be sufficient to contrast an ingroup-actor/outgroup-recipient condition with an outgroup-actor/ingroup-recipient condition, as then group memberships would be confounded and a clean interpretation of observed effects would become impossible. Instead, the design suggestions outlined in the previous section could be followed.
To give an example for collective action research, imagine a researcher who is interested in how group memberships affect action on behalf of a third party. Am I (a German) more likely to want to intervene if I believe others (e.g., Syrians) suffer at the hands of my ingroup members (Germans) than if I believe they suffer at the hands of their own ingroup members? Crucially, it will not be sufficient to contrast an observer-perpetrator-same-group condition with a victim-perpetrator-same-group condition, as this would again introduce confounds. If a difference emerges, we would not know whether it is due the difference in shared victim-perpetrator membership between the two conditions, or due to the difference between shared observer–perpetrator membership between the two conditions. Again, the logic outlined in the previous section, which either compares all five constellations as holistic entities or which uses unconfounding information, can be used to remedy this.
Thinking about all relationships in a triad might also help to address some of the tension between contact and collective action research, whereby contact has sometimes been argued to improve intergroup tolerance and at other times to contribute to the pacification of members of low-status groups and thereby add to their continued suppression. For example, if a high-status observer witnesses positive contact between an actor and a recipient, and if the observer shares a group membership with the actor (e.g., a white observer seeing a white person assisting a black person), then positive attitude change might result. If, however, a low-status observer witnesses positive contact between an actor and a recipient, and if the observer shares a group membership with the recipient (e.g., a black observer seeing a white person assisting a black person), then appeasement might result.
It was my intention to highlight the benefits of considering all five potential intergroup constellations in triadic intergroup situations, in particular for avoiding confounds when studying these experimentally. This of course does not mean that these considerations are always relevant: they are only useful in situations with three actors. In situations where there are only two actors (e.g., as in the study on football club supporters by Levine et al. 2005 cited above), these considerations will be less helpful. However, in any situation involving an observer/donor, victim, and perpetrator, and many other situations involving three entities, I believe these considerations can be useful when choosing a sound experimental design.
There are several promising avenues for future work. First, as already alluded to above in reference to the research on nonverbal behavior (Richeson & Shelton, 2005), another important triad can include the judge/researcher who observes behavior between two other entities. Although the researcher is often treated as if he/she is neutral and objective, he/she can of course also share group memberships (or not) with research participants, and the triadic relations in this regard might well impact on research findings (Zagefka, 2009). This could be theorized more in the future. Second, one clear limitation of this contribution is that it only considers triads, and of course the complexities in the real world often go far beyond that. Future research could try to expand the current model further.
Research going forward can use the methodological recommendations here to study experimentally a wide range of triadic intergroup phenomena. Triadic relations can be expected to affect whether victims are blamed for their plight (Lerner, 1980), whether collective action on behalf of others will arise (Van Zomeren et al., 2008), whether intergroup guilt or intergroup forgiveness are triggered (Wohl & Branscombe, 2005), whether anger arises (Spanovic, Lickel, Denson, & Petrovic, 2010), whether intergroup problems will be denied or not (Cohen, 2001), and whether bystander interventions (Lotz, Okimoto, Schlösser, & Fetchenhauer, 2011) and intergroup helping will be triggered (e.g., Gaertner, Dovidio, & Johnson, 1982; Levine & Crowther, 2008; Saab, Tausch, Spears, & Cheung, 2015; Van de Vyver & Abrams, 2015). As such, it is hoped that the reflections offered here will be useful to intergroup researchers studying a broad range of phenomena.
