ABSTRACT
Many response variables in multilevel models are ordered categories. Although those used here are school grades, other examples arise in diverse application areas. Rating scales provide important examples. When number labels are attached to the categories, it is only their ordering that has any substantive meaning. However, in practice, it is quite common to apply arbitrary scores to the responses and model them as if they were interval scales with higher measurement properties. There are obvious practical advantages in this in that methodology for linear multilevel models is more widely known and software more accessible.
