[Proper handling of correlated data in rehabilitation research]

Many study designs in rehabilitation science give rise to correlated data. For example, patients are followed over time, different responses are measured for each patient, or patients are observed in logical units. Standard statistical methods, however, are only valid for independent responses, and careless application of these methods for actually correlated observations might give erroneous results. By means of a simple example, we show how using methods for correlated data can indeed give a gain in statistical power. In the following, different approaches (Summary measures, Repeated Measurement ANOVA, MANOVA, and Mixed Models) to deal with correlated data are presented. We conclude that among these, the Mixed Models approach is the method of choice because it allows flexible modelling of correlation structure and is, meanwhile, also available in standard statistical software packages.