Jesse Hemerik (Wageningen University)

When we test two or more hypotheses, it usually becomes worthwhile to think about an appropriate multiple testing method. There exist many multiple testing methods. They all ensure that the number of type I errors remains small on average, in one way or another. For example, the strictest methods ensure that the probability of even a single type I error is below e.g. 0.05. When we choose a multiple testing method, there are always tradeoffs between type I error control, power and flexibility. Further, if the tested variables are dependent, we can sometimes use multiple testing methods that account for these dependencies, to gain power. An important example are multiple testing methods based on randomization or permutation. An example application that I will discuss is brain imaging data, where each voxel (“3d-pixel”) corresponds to a hypothesis and the voxels are strongly correlated. There are also many potential applications in economics. 

VIEW PAPER

Riccardo Faini CEIS Seminars list