Multilevel Analysis
In order to take into account themultilevel nature of
our data (students nested within classes) and to test
our hypotheses about the role of class moral disengagement,
we used the multilevel modeling technique
of hierarchical linear modeling (HLM, Raudenbush
and Bryk, 2002), which properly estimates standard
errorswhen data measured at one level (student) share
variance at another level (class).
In a first step, the null modelwas estimated to obtain
the intraclass correlation coefficient. Results showed
that a significant portion of variance (13.7%) of probullying
behavior was explained by classes (χ2
(37) = 131.27,P