Cluster bias: Testing measurement invariance in multilevel data

In this thesis we presented methods and procedures to test and account for measurement bias in multilevel data. Multilevel data are data with a clustered structure, for instance data of children grouped in classrooms, or data of employees in teams. For example, with data of children in classes, we can distinguish two levels in the data: we denote the child level Level 1 or the within level, and the class level Level 2 or the between level. Children in the same class share class level characteristics, such as the teacher, classroom composition, and class size. Such class level characteristics may affect child level variables, leading to structural differences between the responses of children from different classes. With multilevel structural equation modeling (multilevel SEM), we can accommodate such differences by specifying models at the different levels of multilevel data. Such models can be constrained to test substantive and psychometric hypotheses. In this thesis, we considered specifically the psychometric hypothesis of measurement invariance.