correlated in all three countries (England r = .89*** ; Germany r = .98***; Sweden r
= .92***). Similarly, immigrants’ share is closely related to ethnic proportion (England r
= .47*** ; Germany r = .51***; Sweden r = .82***). Because of these strong
interrelations I do not expect the three diversity measures to differ much in their effect on
tolerance and participation.
Method of analysis
Since the independent variables are pitched at two levels (classroom and individual) and
the dependent variables are at the individual level, the appropriate method to explore the
relationships between diversity and social capital is a multi-level analysis. This is all the
more required given the nested structure of the data. A structure of this kind, with
students being nested in classes, classes in schools, and schools in countries, precludes
the use of more conventional multiple regression techniques since these require that
observations are independent. Using such techniques to analyze nested data would result
in an overestimation of the effects of contextual variables (Snijders and Bosker, 1999).
I used the mixed methods option in SPSS to build a two-level random intercept
model consisting of classrooms (level 2) and students (level 1) with the three measures of
diversity, classroom status and classroom climate entered as classroom-level variables
and gender, social status, civic competence and ethnoracial identity entered as individual-
level variables.
Results and discussion
I start by presenting the results of the so-called zero model, which displays the
distribution of the variance in our outcome measures across the two levels (see Table 2).
More than 10 per cent of the variance in ethnic tolerance is located at the classroom level
in all three countries. By contrast, the between-classroom variance in participation is
much smaller, representing no more than 1.9 to 4.4 percent of the total variance
everywhere. According to Duncan and Raudenbusch’s (1999) rule of thumb on the
14