False Discovery Rate, Bayes and Multiple Comparisons

Jonathan Taylor
Stanford University
Statistics

In this talk, we give an expository overview of a relatively new approach to the multiple
comparisons problem in brain imaging, called the False Discovery Rate (FDR). The FDR
is appropriate for exploratory data analysis, as it allows for some Type I errors, because
it focuses on the ratio of the number of Type I errors to the total number of discoveries.



We also describe an interesting Bayesian interpretation of the FDR, which implies that
estimating FDR is equivalent to estimating a posterior probability. This Bayesian
viewpoint suggests some new inferential procedures as well as connections with existing
approaches to the multiple comparisons problem in the literature.



Back to Graduate Summer School: Mathematics in Brain Imaging