6 - Multilevel analysis in longitudinal studies

Summary

Introduction

In the earlier chapters it has been explained that multilevel analysis is suitable for the analysis of correlated data. We have seen examples in which observations of patients were correlated because they ‘belong’ to the same medical doctor, i.e. the observations of patients were clustered within medical doctors. The fact that observations are correlated is probably most pronounced in longitudinal studies in which repeated observations are made within one subject or patient. It is obvious that these observations are (usually) highly correlated. Therefore, the whole theory of multilevel analysis, as described in the earlier chapters, can also be applied to longitudinal data. With longitudinal data, the repeated observations are clustered within the subject or patient (see Figure 6.1).

Figure 6.1 illustrates a two-level structure, i.e. the observations are the lower level, while the patient is the higher level. This is different from all the examples that have been described before, in which the patients were the lower level. It is of course also possible that the patients are clustered within medical doctors, as was also the situation in the earlier chapters. This is referred to as a three-level structure, i.e. the observations are clustered within the patients and the patients are clustered within the medical doctors (see Figure 6.2).