Aarhus University Seal / Aarhus Universitets segl

False Discovery Rate

Britta Anker Bak
Lectures for students by students
Friday, 6 December, 2013, at 14:30-15:30, in Aud. G2 (1532-122)

In classical statistics one or a few hypotheses have been considered at a time. In such situations it is relevant to upper bound the probability of rejecting at least one true hypothesis, also known as the familywise error rate. The scientific and technological development of the 21st century gives rise to situations where hundreds or thousands of hypotheses are tested simultaneously, and where insisting on keeping the familywise error rate low will lead to an extremely low probability of rejecting any hypotheses at all. This gives a demand for new statistical methodology!

In this talk I give the background and definition of the False Discovery Rate, which was first proposed by Benjamini and Hochberg in 1995. Using the Benjamini-Hochberg procedure, which is a stepwise procedure for simultaneously performing multiple tests, I will prove it is possible to obtain an upper bound of the false discovery rate under a specific (an non-intuitive) kind of dependence structure between the p-values.

There is no specific course-requirements to be able to follow the talk.

The language of the talk will be danish or english depending on the audience.

Contact person: Thomas Lundsgaard Schmidt