Contribution
It turns the question of how contact reduces prejudice into a quantitative comparison. Three mediators proposed in the literature are tested on the same footing, by pooling, for each, the three correlations a mediation test needs and running a Sobel test on the pooled matrix. The result moves the weight of the explanation from what contact teaches about the outgroup to what it does to feeling: anxiety goes down, empathy goes up.
Key insight
Contact mostly works through affect. Learning about the outgroup, the mid-century account, does mediate, but weakly; reduced anxiety about intergroup interaction and increased empathy and perspective taking mediate strongly. Anxiety correlates with both of the others, which suggests, untested, a sequence in which anxiety must fall before empathy and knowledge can do their work.
Assumptions
- Mediation from pooled correlations. For each mediator, the contact–mediator, mediator–prejudice and contact–prejudice correlations are meta-analysed over the same samples, a structural model gives the two path coefficients and their standard errors, and the Sobel test is applied (Shadish 1996; Premack and Hunter 1988).
- Random effects, inverse-variance weights, Pearson r; then a repeat on homogeneous subsets formed by dropping outliers until Q is non-significant.
- The file is the 515 studies of the 2006 meta-analysis (LIT-1413) plus studies through June 2005 that measured a mediator; same inclusion rules, with the added requirement that the mediator’s relations be reported.
- Correlational: the mediation is statistical, and the authors say causal influence cannot be inferred from it.
Key results
- Samples. Knowledge: 11 studies, 17 samples, 2,543 participants. Empathy and perspective taking: 9 studies, 14 samples, 2,362. Anxiety: 45 studies, 60 samples, 13,343. 54 studies in all; one tests all three mediators.
- Pooled correlations, all samples (Table 1). Knowledge: contact → knowledge +.22, knowledge → prejudice −.14, contact → prejudice −.30. Anxiety: contact → anxiety −.29, anxiety → prejudice +.37, contact → prejudice −.29. Empathy: +.35, −.41, −.37.
- Sobel tests (Table 3, text). All samples: knowledge Z = 3.87, anxiety 26.55, empathy 12.43, all p < .001. Homogeneous subsets: 3.64, 13.44, 9.87. The pattern is the same in both.
- Mediators intercorrelated, in the few samples with two or more: anxiety with knowledge r = −.24 and with empathy −.32; knowledge with empathy +.05, n.s.
- Joint model. In the nine samples with contact, prejudice, anxiety and empathy, both mediate in a saturated path model and together account for “roughly half” of the contact–prejudice covariance.
- Publication status. No path involving knowledge or empathy differs by publication. Contact → anxiety is stronger in unpublished samples (−.36) than published (−.28), QB(1) = 3.72, p = .05, the opposite of what publication bias would produce.
- Contact–prejudice is larger here (−.29 to −.37) than the 2006 mean of −.21; the authors attribute it to half these studies being post-2000 and to the rising rigor of later studies.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Knowledge of the outgroup, intergroup anxiety and empathy each statistically mediate the contact–prejudice association | moderate | Table 3, all and homogeneous samples |
| C2 | Knowledge is a markedly weaker mediator than anxiety reduction and empathy | moderate | Table 3; but different sample sets per mediator |
| C3 | Anxiety and empathy together carry about half the association | weak | nine samples, saturated model |
| C4 | Anxiety must fall before empathy and knowledge can work | weak | suggested from intercorrelations; untested |
| C5 | Publication bias does not drive these mediation results | moderate | publication-status comparisons |
Method
Four steps per mediator: weight each sample’s correlations by inverse variance (Wilson’s SPSS macros); pool the three-variable matrix; fit a structural model for the unstandardized contact → mediator coefficient and the mediator → prejudice coefficient with contact held constant, each with its standard error; apply the Sobel test. Repeated on homogeneous subsets obtained with DSTAT by removing extreme outliers.
Concepts
- Knowledge: learning about the outgroup through contact, the mediator of Allport and the Human Relations Movement.
- Intergroup anxiety: after Stephan and Stephan (1985), the threat and anxiety aroused by the prospect or fact of cross-group interaction.
- Empathy and perspective taking: taken together, since too few studies allow them to be separated; perspective taking is “a more cognitive dimension of empathy”.
Connections
The mediators are three of the processes in Pettigrew’s 1998 review (LIT-1408): learning about the outgroup, and the “affective ties” under which that review put both anxiety and empathy. Ingroup reappraisal, the review’s fourth process, appears here only in the list of mediators not tested (“broadened views of the ingroup”). The file of studies is the 2006 meta-analysis’s (LIT-1413). The empathy line is Batson’s: the 1997 study it cites, on empathy for a stigmatized member improving attitudes to the group, is not registered; the record holds other Batson work (LIT-1700, LIT-1728, LIT-1783).
Bearing on the record
- CLAIM-275: the paper supports contact acting on feeling before belief: anxiety reduction and empathy carry more of the association than knowledge does. But its mediation is cross-sectional and statistical, and the sequence it suggests (anxiety first) is proposed, not shown.
- CLAIM-417 (perspectives revised by being made one of several): the perspective-taking leg is the nearest the paper comes, and it could not separate perspective taking from empathy. Nothing here tests a change in how one’s own view is held; that is the 1998 review’s ingroup reappraisal, which this paper does not test.
- LIT-1405’s own “why this is here” is right that knowledge mediates, only less strongly. The paper’s words are “the mediational value of increased knowledge appears less strong”, and in the text “simply knowing more about the outgroup typically does not have a major effect”.
No instruction for machine-learning practice; nothing for the anthology.
Limitations
- Different sample sets per mediator, so differences between the mediators may partly reflect the sets, as the authors say.
- Few studies for knowledge (11) and empathy (9); the joint model rests on nine samples.
- Statistical mediation on correlational data; reverse and non-recursive paths are acknowledged and not modelled.
- Half the association is unexplained; the authors list other mediators (self-disclosure, broadened ingroup views, outgroup variability, threat) and expect the missing ones to be affective.
- Errors and slips. Table 3 gives the Sobel Z for empathy over all samples as 4.38; its own coefficients and standard errors give 12.43, the value in the text, so the table is wrong. Table 3 gives anxiety’s as 26.60, the text 26.55; recomputation gives 26.55. The method says 14 of the 60 anxiety samples are unpublished; the publication comparison uses 13 unpublished and 47 published. “54 studies with 91 independent samples” adds the three mediators’ samples (17 + 14 + 60); since at least nine samples measured both anxiety and empathy, there are at most 82 distinct samples. The knowledge-to-prejudice range “between +.13 and .38” is garbled in the extraction.
Open questions
- Does anxiety reduction precede the growth of empathy and knowledge? The authors call for longitudinal and experimental tests.
- Do empathy and perspective taking mediate separately?