CMStatistics 2016: Start Registration
View Submission - CMStatistics
B1413
Title: The power of tests for signal detection in high-dimensional data Authors:  Arnold Janssen - Heinrich-Heine University Duesseldorf (Germany)
Marc Ditzhaus - Otto-von-Guericke University Magdeburg (Germany) [presenting]
Abstract: In modern studies the detection of signals becomes more and more important. Especially, if they are weak and rare it is quite difficult to detect them. In this context mixture models are often used to model the signal detection problem. It has been previously discussed the behaviour of the log-likelihood ratio test for the sparse normal mixture model. It has been suggested to use Tukey's higher criticism (HC) for this model. In contrast to LLR, HC does not depend on the unknown signal strength and the unknown proportion of signals within the data. Nevertheless, the area of complete detection are the same for both tests. Recently, the results were extended to more general models. We give a short overview of the literature and present results concerning a new general model, which includes all of the above mentioned models.