In a multilevel research, the data structure in the population is considered to be
hierarchical, and the sample data are viewed as a multi-stage sample from this
hierarchical population (Hox, 2002). Multilevel models assume that students and
teachers are not randomly distributed in classes and schools, since they are clustered
in classes with particular teachers (Lee & Bryk, 1989). The additional advantage of
this approach is that the impact of variables - at the different levels – is studied
simultaneously since interactions between the variables are considered. Multilevel
modelling is therefore better suited to study real world phenomena that take into
account the complex nature and impact of the social context, and the influence of
mediating (e.g., instructional) processes and variables (Hox, 2002). As explained
earlier, although multilevel studies focusing on mathematics performance are
available in the literature (Opdenakker & Van Damme, 2001), thus far no extensive
and large scale studies have been set up in the Chinese context. The present study
aims at developing a baseline for this type of research.