Ulrich Mansmann, Institut für Medizinische Biometrie und Informatik Universität Heidelberg
Recently, there has been a very active production of data analysis methodology for high dimensional gene expression data. Most papers present a statistical approach based on heuristic arguments and exemplify its use on a given data set. The results are related to biological subject- matter knowledge to stress the relevance of the findings. However, the quality of the method proposed in terms of established statistical measures (power, level, bias, predictive value) is often not assessed. On the one hand, the complexity of many problems does not allow such assessments in terms of formal arguments. On the other hand, there are prominent examples of methodological claims which even lack the backup of a simulation study. But simulations based on an appropriate random mechanism may help to settle the ground for an analysis or a design of the experiment planned. Examples and approaches will be presented.