Intergroup Contact Is Consistently Associated With Lower Prejudice Across Group Properties

Contact may be differently related to prejudice depending on the properties of the groups. There are infinite possible groups across humanity in the present, past, and future—humans create groups based on nearly any kind of identifiable distinction (Sidanius & Pratto, 1999). Yet research has suggested that despite their variability, groups tend to vary on a limited set of properties (Fiske et al., 2002; Lickel et al., 2000; Stephan & Stephan, 2000) which shape how others form impressions and attitudes toward members of those groups (Denson et al., 2006). For example, some groups are easy to join (e.g., a sports team) whereas others are very hard to join (e.g., Asian people). Similarly, some groups’ members might be more likely to have common goals (e.g., an anti-abortion group) relative to others (e.g., women). Importantly, because group properties influence impressions and attitudes toward group members, groups that score similarly on multiple group properties might be thought to share a common group property profile and are thus perceived in a similar manner. For instance, despite their unique identities and idiosyncratic differences, if distinct groups were perceived as equally permeable, similarly sized, and with the same degree of common goals, contact might be associated with prejudice in the same way across these distinct groups. See Table 1 for a description of group properties (Denson et al., 2006; Fiske et al., 2002; Lickel et al., 2000; Stephan & Stephan, 2000).

Most previous research has examined contact using different racial or ethnic groups (Paluck et al., 2019; Pettigrew & Tropp, 2006), although in recent years researchers have increasingly expanded their focus (e.g., political outgroups, people with disabilities, minority groups). A close examination of the groups studied, however, suggests that while the target groups for contact have increased in variety, the type of groups (as defined by group properties) studied has not changed dramatically, with groups characterized by long duration, large size, and low permeability most frequently studied. In other words, while the groups themselves have varied, intergroup contact theory has not been tested on a range of groups varying widely across all group properties.

One possibility is that the documented links between intergroup contact and prejudice may be limited to the types of groups most commonly studied. Testing whether group properties moderate the contact-prejudice relationship can reveal such potential boundary conditions. Perhaps intergroup contact is not as effective when a group has certain properties, while other properties lead contact to correspond even more strongly to lower prejudice than the meta-analytic estimate of r = -.21. Critically, identifying group properties that moderate the contact-prejudice relationship would enable better predictions about when contact will correspond with lower prejudice—even for groups that intergroup contact theory has not yet examined, and for groups that do not yet exist but will become socially and psychologically significant in the future.

Certain group properties may be especially likely to moderate the contact-prejudice relationship. For example, permeability is the ease with which a person may join and/or leave a group. Groups based on race, for example, would be considered highly impermeable whereas groups that are organized around a task (e.g., a sports team) are relatively permeable. Contact has been shown to first improve attitudes towards a specific outgroup member before generalizing to the whole outgroup (i.e., primary transfer effect; Boin et al., 2021). Highly permeable groups, however, may not benefit from this generalization process: Contact with another person may not be construed as intergroup contact if members can easily enter and exit the group. Consequently, any generalization of attitudes to the outgroup may be less likely to occur from contact with members of highly permeable groups. A closer examination of the contact meta-analyses (Paluck et al., 2019; Pettigrew & Tropp, 2006) provides some support for this possibility: Although the majority of studies found evidence supporting the contact-prejudice relationship, larger effects were obtained for target groups of people with physical disabilities, a group often perceived to be less permeable.

Likewise, groups perceived to be highly threatening may not benefit from the positive outcomes of contact. For starters, people may avoid contact with these groups because they have the potential to threaten their safety (Park et al., 2023). Even when people do enter contact situations with these groups, the anxiety present (as a consequence of threat) may hinder the effectiveness of contact (Pettigrew & Tropp, 2008). Consequently, highly threatening groups may simply not benefit from contact. In sum, while these are some examples focusing on one group property, the idea that different group properties might moderate the contact–prejudice relationship remains untested.

Furthermore, past work has shown that the contact-prejudice relationship may be moderated by certain variables. Specifically, the optimal conditions for contact (i.e., equal group status, common goals, intergroup cooperation, and authority support) moderate this relationship (Allport, 1954; Pettigrew, 1998). However, these moderators concern the characteristics of the situation, rather than the properties of the groups. At present, no study has systematically examined whether group properties moderate the contact-prejudice relationship. An investigation using a wide variety of groups that naturally vary on group properties therefore tests whether group properties moderate the contact-prejudice relationship, and tests the boundaries of intergroup contact theory (Brandt & Crawford, 2019; Crandall et al., 2002). To this end, the present research tested the contact-prejudice relationship across 80 groups with wide variation in their group properties to systematically test for boundary conditions and moderation of the contact-prejudice relationship.

For over nearly a century of research (Allport, 1954), contact has been operationalized in a wide variety of ways (Hässler et al., 2020; Lolliot et al., 2015), with different forms of contact theorized to reduce prejudice via different mechanisms (Pettigrew & Hewstone, 2017). Though the present research focuses on variation in group properties, it enables a test of the efficacy of the different types and measures of contact as well. One possibility is that the type of contact that is most strongly associated with less prejudice depends on the properties of the group. For example, one common measure of contact is the positivity of contact, with some research arguing that positive contact reduces prejudice due to mere exposure (Flores et al., 2018). Consequently, the positivity of contact might be strongly associated with less prejudice for groups large in size, due to frequent opportunities for exposure. Accordingly, the current research also tested whether the relationship between contact and prejudice was consistent across the type of contact being measured.

Originally, research on contact focused on direct face-to-face contact. But modern advances in technology have afforded indirect forms of contact (White et al., 2021). While there are numerous forms of indirect contact, meta-analyses for extended (Zhou et al., 2019), imagined (Miles & Crisp, 2014), electronic (Imperato et al., 2021), and mediated contact (Banas et al., 2020) have all demonstrated effects generally consistent with direct contact. For simplicity, the present research focuses only on the most commonly used measures of direct contact.

In summary, the majority of research on the contact-prejudice relationship has focused on groups with similar group properties, leaving untested whether similar results can be expected for groups with different combinations of group properties. Whether these group properties might act as important boundary conditions remains unknown. Should these group properties moderate the contact-prejudice relationship, conducting contact interventions with these types of groups would be a waste of resources, as different interventions are likely needed for prejudice reduction for such groups. Further, testing the contact-prejudice relationship across a broad spectrum of groups is critical for a robust science of contact. Accordingly, the present research tested this relationship across a wide variety of groups that, critically, varied on their group properties to inform both theory and real-world applications. All data, analysis code, and research materials are available at [https://osf.io/jbexr/].

We conducted Study 1 across two phases to create the necessary data structure. Participants in Phase I rated 60 groups on ten group properties. An entirely separate sample completed Phase II, rating six randomly selected groups (drawn from the 60 groups in Phase I) on the amount of contact they had with the group, and their attitudes towards the group. Data from these phases were merged for analysis (details below).

Phase I. A total of 603 participants from the United States and Canada were recruited from Amazon Mechanical Turk via CloudResearch (Litman et al., 2017) for monetary compensation. For all studies, we excluded participants in accordance with our pre-registered data cleaning procedure [https://osf.io/s8enf] on the basis of their frequency of repeated responses and their overall response time. In addition, we excluded participants if they indicated their data should not be used. The final sample included 554 participants (M_age = 41.42, SD_age = 12.18; 435 White, 57 Black, 38 East Asian, 12 South Asian, 6 Indian, 6 Pacific Islander; 290 male, 254 female, 6 non-binary, 4 undisclosed).

Our aim was to draw from a wider spectrum of groups than previous research, and ideally, to sample evenly from the entire theoretical universe of groups. We first included the different groups examined in the entitativity literature (Lickel et al., 2000), ranging from groups with low social connections (e.g., people in the audience at a movie) to more commonly examined groups (e.g., Black people). We further drew from research examining groups that spontaneously occurred to participants when asked about groups that exist in society (Koch et al., 2016). Finally, we included other groups not captured by prior research (e.g., terrorists, rollerbladers) but consistent with the ways in which groups vary as defined by previous work (Lickel et al., 2000). Ultimately, we included 60 groups (see Table S1) in Study 1.

Our hypotheses hinged on the assumption that these 60 groups would naturally vary in their group properties, thereby enabling our statistical tests of moderation. To test this assumption, participants rated the 60 groups on 10 group properties (see Table 1 for descriptions of the group properties; exact items used can be found in Table S1), with each property rated using a single item on a (1 – “Not at all” to 9 – “Very much”) scale, with the exception of duration (1 – “Short-term” to 9 – “Long-term”), and size (1 – “Very small” to 9 – “Very large”), as in previous work (Denson et al., 2006).

Data from Phase I were aggregated across participants, such that each group had an average rating on each group property. These were then merged into the Phase II data to test whether group properties moderated the contact-prejudice relationship.

Phase II. A total of 453 participants from the United States and Canada were recruited from Amazon Mechanical Turk via CloudResearch (Litman et al., 2017) for monetary compensation. A final sample of 442 participants (M_age = 41.95, SD_age = 12.84; 350 White, 63 Black, 18 East Asian, 6 South Asian, 3 Indian, 2 Pacific Islander; 246 male, 191 female, 3 non-binary, 2 undisclosed) remained after our data cleaning procedure.

We used five common indices to measure direct contact: quantity of contact, positivity of contact, absence of negative contact, number of [outgroup] friends, and frequency of meeting [outgroup] friends (Hässler et al., 2020). Quantity of contact was measured using two items, with the first measuring the frequency of interaction (1 – “Never”, 2 – “Less than once a year”, 3 – “Yearly”, 4 – “A few times in a year”, 5 – “Monthly”, 6 – “Weekly”, and 7 – “Daily”) and the second measuring the number of [outgroup] acquaintances (0 – “Zero” to 10 – “10 or more”). Positivity of contact was measured using three items while absence of negative contact was measured using two reversed-scored items. Both used (1 – “Strongly disagree” to 7 – “Strongly agree”) scales. Number of [outgroup] friends was measured using a single item on a (1 – “None of my friends” to 7 – “All of my friends”) scale, and frequency of meeting [outgroup] friends was measured using a single item on a (1 – “Never” to 7 – “Daily”) scale. For each measure of contact, a latent factor was created using lavaan in R (Rosseel, 2012). Latent factor scores were then saved and used in analysis (see Table S2 for fit indices and Pearson’s correlations).

Prejudice was measured using three items on which participants rated their feelings of warmth, liking, and positivity towards each group on a (1 – “Not at all”, 4 – “Somewhat”, 7 – “Very much”) scale. These three items loaded highly on a single factor, and again we used the latent factor score in analyses as our metric of prejudice. Finally, we excluded observations for which participants indicated that they considered themselves to be a member of the group assessed.

For our main analyses, we regressed prejudice on contact in multilevel models, in which repeated responses were cross-classified by participant and group.¹ Contact was group-mean centered by participant and modeled as a random slope across participant and group. All analyses were done using the R package lme4 (Bates et al., 2015). Degrees of freedom, test statistics, and p-values were derived from Satterthwaite approximations in the lmerTest package (Kuznetsova et al., 2017).

First, we replicated that all measures of contact were associated with lower levels of prejudice across this wider range of groups. Like previous work (Paluck et al., 2019; Pettigrew & Tropp, 2006), across all groups and types of contact we found that increased contact was associated with less prejudice, with positivity of contact most strongly associated with lower prejudice (β = -0.69, p < .001) and frequency of meeting [outgroup] friends least strongly associated with lower prejudice (β = -0.30, p < .001). Furthermore, the strength of associations of the different measures of direct contact with prejudice echo previous research (Talay & De Coninck, 2020) which found that the quality of contact is more strongly associated with prejudice than the quantity of contact.

Because the five measures of contact produced similar patterns of results, for conciseness in the main text, we present only the “absence of negative contact” measure. See Tables S3 to S6 for correlations across all contact items with group properties, and Figures S1 to S8 for simple slopes.

We next tested whether each of the 10 group properties moderated the contact-prejudice relationship (Table 2). Significant interactions between contact and the group properties were explored with tests of simple slopes of contact at +/- 1SD of the mean of the group property (Cohen et al., 2003).

Summarizing these results, some group properties did significantly moderate the contact-prejudice relationship. Yet the moderation was consistently very small and never changed the qualitative nature (i.e., the direction) of the relationship (Figure 1). In other words, when moderation occurred, the moderator slightly enhanced or reduced the effect (i.e., an ordinal interaction), but never negated the effect of contact (i.e., a non-significant simple slope), or comprised a cross-over or disordinal interaction (i.e., reversing the effect). This pattern was consistent for all significant moderators for all measures of direct contact. We also note a good number of the group properties simply did not moderate the contact-prejudice relationship. Contact was associated with less prejudice consistently across groups varying on these properties.

Study 1 demonstrated that the contact-prejudice relationship was robust across 10 group properties. Furthermore, while some moderation based on group properties were observed, the overall negative relationship between contact and prejudice never changed dramatically.

Study 1 examined groups that differed on the basis of ten group properties (Crump et al., 2010; Denson et al., 2006; Lickel et al., 2000). However, these group properties do not represent the entire possible theoretical universe of group properties. Groups can also vary on the specific content of their stereotypes (Fiske et al., 2002) and the threats they are perceived to pose (Cottrell & Neuberg, 2005; Stephan & Stephan, 2000). A substantial proportion of variability in groups’ stereotypes, and the kinds of prejudices groups are subjected to, can be accounted for by the group’s stereotyped warmth and competence (Fiske et al., 2002). It is thus important to test whether contact reduces prejudice consistently across groups that vary in stereotype content.

In addition, perceived threat may be one of the most important group properties in driving prejudice (Hehman & Neel, 2024; Neuberg et al., 2020; Neuberg & Schaller, 2016; Park et al., 2023; Riek et al., 2006; Stephan & Stephan, 2000). For groups differing in levels of perceived threat, contact may relate differently to prejudice. For example, highly threatening groups may be subjected to strong prejudice regardless of the level of intergroup contact. Thus, in Study 2 we expanded data collection regarding the contact-prejudice relationship to groups varying on negative stereotype content and threat perceptions.

Study 2 included an additional 20 groups that are generally perceived by people to range from mildly to strongly cold, incompetent, and threatening, to establish that the results from Study 1 were consistent when including groups varying on these properties. Like Study 1, Study 2 was conducted over two phases. Participants in Phase I rated a selection of 80 groups on their group properties. An entirely separate sample completed Phase II, rating their attitudes toward and degree of contact with five randomly selected groups.

Phase I. A total of 1084 participants from the United States were recruited from Amazon Mechanical Turk via CloudResearch (Litman et al., 2017) for monetary compensation. A final sample of 952 participants (M_age = 42.67, SD_age = 12.38; 783 White, 79 Black, 54 East Asian, 17 Indian, 17 South Asian, 2 Pacific Islander; 475 female, 466 male, 7 non-binary, 4 undisclosed) remained after our data cleaning procedure.

The first and last authors decided on the additional groups included in Study 2, with the criterion that they were perceived negatively by society. Participants rated the groups on the group properties from Study 1, and four additional single-item measures. Two dimensions of social cognition were measured (Fiske et al., 2002). The first was perceived warmth, reflecting a group’s intentions, with participants responding to “As viewed by society, how warm are members of this group?” The second captured perceived competence, reflecting a group’s ability to enact their intentions, with participants responding to “As viewed by society, how competent are members of this group?” Participants responded on a (1 – “Not at all” to 5 – “Extremely”) scale for both dimensions as in previous work (Fiske et al., 2002). In addition, two different types of threat were measured. The first captured threats to one’s values and beliefs (i.e., symbolic threat; Stephan & Stephan, 2000), with participants responding to “[Outgroup] people are a threat to American culture” rated on a (1 – “Strongly disagree” to 7 – “Strongly agree”) scale as in Earle and Hodson (2020). The second captured various forms of realistic and other types of threat (i.e., general threat; Stephan & Stephan, 2000), with participants responding to “In general, [outgroup] pose a problem for people like me” rated on a (1 – “Strongly disagree” to 9 – “Strongly agree”) scale (Cottrell & Neuberg, 2005). Data were aggregated across participants, such that each group received an average rating on each additional group property.

Phase II. A total of 226 participants from the United States and Canada were recruited from Amazon Mechanical Turk via CloudResearch (Litman et al., 2017) for monetary compensation. A final sample of 215 participants (M_age = 40.44, SD_age = 11.53; 182 White, 21 Black, 7 East Asian, 3 Indian, 1 Pacific Islander, 1 South Asian; 112 male, 100 female, 3 non-binary) remained after our data cleaning procedure. In this phase, participants rated the groups on the same contact and prejudice items used in Study 1.

The analytical framework was identical to Study 1. Replicating Study 1, across all measures of contact and the wider range of groups, more contact was associated with less prejudice, with positivity of contact most strongly associated with lower prejudice (β = -0.61, p < .001) and the absence of negative contact least associated with lower prejudice (β = -0.29, p < .001). Figure 2 shows the contact-prejudice relationship for the “Absence of Negative Contact” measure; Figures S9 to S12 show the relationship for the other contact measures.

None of the four additional group properties examined in Study 2 (i.e., competence, warmth, general threat, symbolic threat) moderated the contact-prejudice relationship (Table 3). In addition, the majority of the (small) interaction terms that were statistically significant in Study 1 were not in Study 2. For the few group properties that remained a significant moderator, they were inconsistent (i.e., did not moderate the contact-prejudice relationship for all five measures of contact). When significant moderation did occur for the original group properties, like in Study 1, it was a very small interaction term and never changed the qualitative nature of the relationship. Furthermore, like Study 1, some group properties never moderated the contact-prejudice relationship and contact consistently was associated with less prejudice. Overall, from this broad pattern of analyses, we conclude that these group properties did not meaningfully moderate the contact-prejudice relationship.

Study 2 demonstrated that the contact-prejudice relationship was robust across 14 group properties. Centrally, group properties, even with the inclusion of four new properties, generally did not moderate this relationship. These results suggest that the contact-prejudice relationship is consistent across group properties.

Intergroup contact theory has been tested on groups that often have similar group properties. For robust understanding of the contact-prejudice relationship, it is important to test a heterogeneous array of groups that cover the theoretical space of group properties. Such tests enable generalization beyond a narrow range of traditionally studied groups concentrated in certain parts of the world. In addition, understanding how the relationship between contact and prejudice is moderated by group properties enables better future predictions for existing groups not yet tested, or for future groups that do not yet exist (since they will necessarily vary on these known group properties). Here, we systematically examined the contact-prejudice relationship across a wide spectrum of groups varying along multiple group properties.

Importantly, we found that the contact-prejudice relationship was exceptionally robust—intergroup contact was associated with lower prejudice consistently across 14 group properties. Although several group properties occasionally moderated the contact-prejudice associations, the interaction terms’ effect sizes were generally very small (all R²s < .006; Table 2). This suggests that these moderations may be relatively unimportant both theoretically and practically, although some additional variance in prejudice may be explained by including these group properties in models. In other words, differences in the effectiveness of contact across groups (Pettigrew & Tropp, 2006) cannot be traced to a group’s properties.

Given this pattern of mostly null results, some might conclude that such findings are already known and are of less importance. We respectfully disagree. Instead, echoing similar calls in other domains (Brandt & Crawford, 2019; Crandall et al., 2002), we argue that for any robust science, probing boundary conditions and potential moderators is a critical process of theory development and rigorous model testing.

Rather than group properties, instead, other factors like context, history, and individual differences may better account for any variability in contact’s effects. Indeed, recent work argues that contact’s effects are mostly of a between-person nature (Bohrer et al., 2019; Friehs et al., 2024; Hodson & Meleady, 2024). Accordingly, contact’s effects may be best attributed to differences between people, such as a person’s experiences (O’Donnell et al., 2021). Factors beyond individual differences, such as a group’s properties, may be unlikely to influence the contact-prejudice relationship. These results, then, suggest that future work on intergroup contact theory should focus on the characteristics of the person and the situation, rather than the properties of the groups.

In addition, we found that the contact-prejudice relationship was consistent across five different measures of direct contact (Hässler et al., 2020). Intergroup contact has historically been measured using a variety of scales that assess different aspects of the construct. We find no evidence that certain types of direct contact were more strongly associated with less prejudice for certain groups or people. All measures of direct contact were roughly equally associated with lower prejudice (average β = -0.43). Considering that intergroup contact theory encompasses both direct and indirect forms of contact, future research could expand upon the current work by examining the contact-prejudice relationship across a wide variety of groups and group properties for the various forms of indirect contact (e.g., extended contact, imagined contact, parasocial contact). While previous work has demonstrated consistent results between direct contact and indirect contact (Banas et al., 2020; Imperato et al., 2021; Miles & Crisp, 2014; Paluck et al., 2019; Pettigrew & Tropp, 2006; Zhou et al., 2019), we do not yet know whether for indirect contact, a group’s properties would moderate the contact-prejudice relationship. It is plausible that certain group properties are more salient in the different forms of indirect contact and could possibly enhance or reduce the effect of indirect contact on prejudice.

In assessing attitudes toward groups varying on group properties, our approach was correlational, which limits causal inference. Given that previous work has shown that both contact reduces prejudice and prejudice reduces contact (Binder et al., 2009; Lemmer & Wagner, 2015), we can merely conclude that an association exists between contact and prejudice across group properties.

Furthermore, while we have sampled a wide variety of groups that vary on 14 group properties, determined in a principled way based on prior research (Fiske et al., 2002; Lickel et al., 2000; Stephan & Stephan, 2000), it is possible that groups vary in other unaccounted ways. For example, a close examination of Figure 2 reveals a few positive contact-prejudice slopes. While our reported effects include these slopes, and no group properties explained these upward trends, other unmeasured variables could possibly moderate the contact-prejudice relationship in a disordinal manner (i.e., canceling out any positive effects of intergroup contact, or reversing the relationship). Relatedly, our participant sample limits conclusions as well. For example, our participants were living in relatively peaceful societies, and contact might be less effective for groups engaged in ongoing violent conflict (e.g., Ukrainian attitudes toward Russians). Future research examining the effectiveness of contact across group properties under real-world conditions is a potentially fruitful direction for refining the boundary conditions of intergroup contact theory.

Intergroup contact is thought to be one of the most effective interventions to reduce prejudice. We conducted an expansive test of the theory’s predictions across different groups, their group properties, and measures of direct contact. The results show that the contact-prejudice relationship is consistent across a broad range of group properties, suggesting the conclusions drawn from previous research on contact are robust. Social policies based on the idea of contact (Arat et al., 2023; Law et al., 2022; Lim et al., 2020) may be expanded to groups not commonly studied in the literature. Specific groups not tested, or novel groups yet to exist, are likely to have group properties falling within the ranges tested here, making these conclusions likely to generalize. To the extent that the group properties examined here are representative of the theoretical universe of group properties, intergroup contact appears to be associated with lower prejudice for all groups.

Conceived research: T.L. and E.H. Methodology: All authors. Investigation: T.L. Analysis: T.L. Writing – Original Draft: T.L. Writing – Review and Editing: All authors.

SSHRC 508752 to R.N. and SSHRC 435-2020-0314 to E.H.

The authors declare no competing interests.

All data, analysis code, and research materials are available at [https://osf.io/jbexr].