-98-
As can be seen, the children's self-rated perspective and their friends' perceptions did not cor-
relate very highly (approximately 5-12 % shared variance!), although both perspectives relate
to the same mutually confirmed friendships. All correlations were positive, though. This
means, that a high evaluation of friendship quality from one perspective more often was
linked with a high than with a low evaluation of friendship quality from the other perspective.
Specifically, the correlation was lowest for perceived fun r=.22, SE =07, p.001, a little
higher for perceived closeness, r=30, SE =.05, p.001, and highest for perceived conflict,
r=.34, SE =.05, p#.001. 20
2.1.5.2.4. Testing the equality of the latent mean levels
In the last set of models, the means of the latent constructs were tested for equivalence across
perspectives and groups. First, in Model 10, I tested whether the means of the friend-rated
constructs could be equated to the means of the corresponding self-rated constructs (i.e., they
were fixed to zero in the first group and constrained to be equal in the second group). No
cross-group equality constraints were set at this point, yet. No difference in fit occured as
compared to Model 3, Ay2(6) - 6.91, p-329, indicating that, on average, friendship quality
was perceived to be the same from the self-rated and the friend-rated perspective. In other
words, although there was only a moderate amount of agreement between the self-rated and
friend-rated evaluations of the subjective friendship aspects, no general bias in friendship
evaluation existed from either perspective in the overall group. In the final test, Model 11, the
means of all latent constructs were additionally tested for cross-group equivalence. For this
purpose, the latent means in the second group were equated to those in the first group (i.e., all
were fixed to zero). When compared to the previous model, no difference in fit was found, A
x2(5) = 0.99, p =963, indicating that no random processes were affecting the latent construct
means in the two random groups.
These correlations were not significantly different from one another, though. Specifically, the equated esti¬
mate yielded a common correlation ofr= .30, SE -.04,p.001, Ax2(2) - 2.04, p=.361 (Model 9a).