17 octobre 2016

Yuri Gurevich (Microsoft Research)

It seems that all statisticians know the concept of p-value is but none is willing to explain it. The only reference that we have been able to find is the  1974 book “Theoretical Statistics” by Cox and Hinkley where the authors define the concept, sort of, but do not use the term “p-value.”
We explain the concept presupposing only rudimentary probability theory; we examine the concept and we discuss the use of it.
For simplicity, our explanation is focused on the discrete case with no outcomes of zero probability but we’ll say a few words on the general case as well.
The talk builds on joint work with Vladimir Vovk of the University of London.