Tsinghua University Pao-Lu Hsu Distinguished Lecture
 

Time: 16:30-17:30, 2017-5-31/ 6-2
Speaker: Yoav Benjamini  [Tel Aviv University]
Place: Lecture Hall, Floor 3, Jin Chun Yuan West Building
 

Lecture 1  (General one) : Are most research findings really false?

It has been claimed that most research finding in Science are false, and the use of statistical testing of hypotheses is to be blamed for. I shall review the current situation, and disagreeing with the attached blame I shall discuss two other statistical problems that contribute to replicability problem. The first is that of inference on the selected few from the many screened. It can be addressed by the control of the false Discovery Rate and related concepts. The second is the choice of the relevant variability, which can be addressed in practice, and an example from preclinical research will be presented. While the emphasis will be on the relevance of Statistics to Science at large, some open questions will be creep in.
 

Lecture 2 (Somewhat more technical): Selective Inference in hierarchical high dimensional data analysis

I shall briefly discuss the need for selective inference in scientific data analysis of large and complex problems. Some examples of such problems with hierarchical structure in Genomics and Brain research will be given. I shall discuss available and new approaches, one of which relies on testing the equality of two multidimensional groups as a building stone, a problem to which Prof. Hsu contributed. However, in current problems the dimension is often higher than the number of observations, so new methods are needed. I’ll present our (yet unpublished) suggestion in this area.