These searches used 42 different search terms that ranged from single words (e.g., "acculturation," "interracial") to combined terms (e.g., "age + intergroup contact," "disabled +
contact"). Across the various databases, we conducted three types of searches with these
terms -- by "title words," "key words," and "subject."
Two independent judges made the ratings. The interrater agreement for the 16 rated variables ranged from 85% to 100%. The average agreement was 91.2% compared to a
chance level of 28.3%. Thus, kappa = .88.
Those studies and samples with large numbers of tests tended to report smaller average
effect sizes. This result suggests that using tests as the level of analysis in these data
was a conservative method for estimating the strength of intergroup contact effects.
Innovative researchers have tried other ways to address the problem. Link and Cullen
( 1986), for example, compared intergroup contact that was chosen with contact that was
not chosen. Irish ( 1952) tried a screening question and retained in his analysis only those
subjects who reported that they had made no effort either to seek or to avoid contacts with
the target group. Using national survey data, Wilson ( 1996) utilized a measure of diffuse
prejudice toward many minorities as a proxy for the tendency to avoid contact. All three of
these methods found that controlling for selection did not eliminate the link between intergroup contact and prejudice.
For example, 21 of the 30 studies directly observed the intergroup contact. We shall
shortly see that this procedure on average, yields higher effect sizes.
Actually, 14 other tests -- all from one study ( Pettigrew, 1997) -- also tested for this uninvolved-outgroup generalization and attained larger effects (mean d = -.69, r = -.33). But
these tests are excluded here because they used "intergroup friends" as their contact
measure, which we later note is an especially powerful independent variable.
10. "Majority" was defined in this context as the less stigmatized group. For example, in studies that measured the effects on children of having contact with the elderly, we considered
the children as the "majority." This particular approach, however, is not artificially shaping
the results shown in Figure 5.3. When we analyzed only the subset of samples involving
racial, ethnic, and nationality groups, a slightly larger difference emerged: for the 61 allminority samples, d = -.27; for the 124 all-majority samples, d = -.48. 11.
This smaller effect size may reflect a recurrent problem in sociometric measurement. As Schofield and Whitley ( 1983) first pointed out, and was later replicated by Schwartzwald, Laor, and Hoffman ( 1986) in Israel, the nomination technique of asking for most-preferred
classmates exaggerates intergroup cleavage. The rating method involving all classmates
is more sensitive and reveals greater intergroup contact effects.
Thus, 18 of these samples provided no choice to their subjects, 10 gave some choice,
and 10 gave full choice. Also, 11 of these special samples involved racial and ethnic
groups, 9 disabilities, 6 age, and 3 mental illness. Their contact settings included 13 in
schools, 8 in recreational areas, 7 in laboratories, and 4 in mixed settings.