distribution of variance across levels of analysis, the former (i.e. more than 10%)
represents a large effect size and the latter a small to medium effect size. A small effect
size implies that classroom-level factors, such as the three diversity measures, are
unlikely to be strong determinants of the outcomes of interest. Thus, the zero model
provides a preliminary indication that the classroom-level variables are likely to be quite
insignificant drivers of participation but quite important ones for ethnic tolerance.
Table 2 about here
Table 3 present the results of the multilevel analyses. Models I-III represent respectively
(I) an analysis including only classroom-level variables, (II) an analysis with all the
explanatory variables, (III) an analysis with all variables based on ethnic majority
respondents only.
To begin with ethnic tolerance (top half of Table 3), heterogeneity appears to be
positively related to tolerance in all three countries controlling only for the two other
classroom-level variables (see Model I; in England the relation is almost significant). In
other words, it exerts an independent effect on ethnic tolerance irrespective of classroom
status and classroom climate. However, to assess whether heterogeneity constitutes a true
contextual effect or whether it merely represents the sum of individual differences in
ethnic tolerance, individual-level controls need to be included (Model II). We see that
heterogeneity retains its significant and positive relation to tolerance in Sweden and
Germany. In other words, the more ethnically diverse the classroom is, the more tolerant
the students are, taking into account their ethnic and social background, gender, and civic
competence levels. This effect, moreover, is quite substantial for Germany: as
heterogeneity moves from minimum to maximum (0 to .71), so ethnic tolerance levels
increase by 1.4 points (.71 x 1.92) on a scale ranging from 4.0 to 14.2. This finding is
clearly in full agreement with the contact perspective, particularly so since the positive
effect of heterogeneity also applies when investigating only the ethnic majority in
Germany and Sweden (Model III). In fact, the other two measures of diversity in this
model show at least as strong positive effects on tolerance (see footnote below Table 3),
making the finding more robust in terms of its support for the contact hypothesis and
15