Changing people’s views of outgroups through individual-to-group generalisation: meta-analytic reviews and theoretical considerations

ABSTRACT

Through individual-to-group generalisation, information about individual members of stigmatised social groups changes the outgroup judgment. This article reports meta-analytic reviews of over 30 years of experimental, lab-based research on individual-to-group generalisation (107 independent tests; 5393 participants). In a first meta-analysis, a positive, medium-size generalisation effect was detected (r = .28, p < .001), reflecting significant generalisation of outgroup exemplar information to the outgroup judgment. This effect was moderated by the number of exemplars and exemplar typicality, with more moderately atypical exemplars maximising generalisation effects. Several other design parameters—including type of control condition, generalisation measures, mode of information provision, type of target outgroup and origin of study—did not moderate the positive generalisation effect. A second meta-analysis investigated the interplay between metacognitions and generalisation and found assimilation effects with metacognitive triggers encouraging exemplar inclusion, and contrast effects with metacognitive cues encouraging exemplar exclusion. These results demonstrate that the same outgroup exemplar can lead to bias reduction or bias exacerbation, depending on available meta-cognitive cues. Findings are discussed in terms of implications for intergroup psychology, generalisation theory and bias reduction interventions.

KEYWORDS:

• Stereotyping
• prejudice
• individual-to-group generalisation
• intergroup contact
• social cognition
• meta-cognition

Acknowledgements

The research reported in this article was funded by a postgraduate scholarship awarded to the first author, a small grant from the Centre for Translational Neuroscience and Mental Health (University of Newcastle, Australia) to the first and second authors, and a grant from the Leverhulme Trust to the third author. We thank Nathalie Crawford and Erin Adams for their assistance with coding the data.

Notes

¹ In most cases, only one test was derived from a single study. That is, most studies presented results for only one included generalisation measure. Six studies yielded more than one test. That is, across six studies results were reported for more than one generalisation measure (e.g., Cernat, 2011; Garcia-Marques & Mackie, 1999; Henderson-King & Nisbett, 1996; Paolini et al., 2004; Swift et al., 2013; Wyer et al., 2002). These studies were allowed to contribute only once to the test for the overall effect (vs. specific moderation tests). The test that was selected for inclusion in the overall generalisation effect was based on the focus of the paper. While it is recognised that this approach may slightly inflate effect sizes, as sometimes the focus of a paper may be around the strongest result, we balanced this decision by being heavily conservative in other respects (e.g., setting all non-significant results to 0 and adopting a random effects analysis). Therefore, the decision taken for these six studies is unlikely to have caused an artificial inflation of the results.

² Tests that fell into the “typical” level of the typicality factor all had, also, either a no-information control or a baseline measurement at pre-test—otherwise the generalisation effect could not be computed.

³ It is worth noting two additional points about Wyer et al.’s (2002) study. Firstly, the unusual cognitive load challenges for Wyer et al.’s participants were reflected in 16% of the original sample being excluded from Wyer et al.’s final analyses due to their outlier status. Secondly, the generalisation effect size we extracted for Wyer et al.’s studies relied on the same phases of the impression formation paradigm as the effect sizes extracted from the other studies of the meta-analysis that looked at established stereotypes.

⁴ For the ancillary concentrated-dispersed meta-analysis we followed the same inclusion criteria as Meta-analysis 1 with one critical difference. Instead of having a control condition, in this meta-analysis studies were included if all participants were presented with the same disconfirming exemplar information that was either “concentrated” in one condition and “dispersed” in the other condition. The studies that met this inclusion criteria were: Hantzi (1995, Study 1), Hewstone and Hamburger (2000, Studies 1 & 2), Hewstone, Hassebrauck, Wirth, and Wänke (2000, Study 1), Hewstone et al. (1992, Study 1, 1994, Studies 1-3), Johnston and Hewstone (1992, Studies 1 & 2), Johnston, Hewstone, Pendry, and Frankish (1994, Studies 1 & 2), Moreno and Bodenhausen (1999, Study 1), Queller and Smith (2002, Study 3), Weber and Crocker (1983, Studies 1 & 4). These studies are noted with an asterisk in the references.

⁵ For instance, we excluded from these analyses Dovidio et al.(2004) Study 2. This study manipulated threat to the ingroup through the use of a threat to their ethnic sub-ordinate ingroup (i.e., White Americans) or to their larger superordinate ingroup (i.e., all Americans) prior to have their participants watching one of two outgroup exemplars videos (a Black man facing discrimination in day-to-day life or a Black man who was beaten to death by police). Participants at this point completed a number of dependent measures, including a measure of outgroup stereotypicality and prejudice. In their analyses, Dovidio and colleagues collapsed across the two exemplar video conditions.We excluded this study because as the outgroup exemplar information varied across conditions, it is impossible to singularly identify if differences in generalisation were generated by the “threat” trigger and/or the different information provisions.

Additional information

Funding

This work was supported by the Centre for Translational Neuroscience and Mental Health; Leverhulme Trust.