|
Activity Number:
|
287
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, July 31, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #308742 |
|
Title:
|
Using FDR To Address a Controversy About the Meaning of P Values
|
|
Author(s):
|
Hormuzd Katki*+
|
|
Companies:
|
National Institutes of Health
|
|
Address:
|
NCI, 6120 Executive Blvd EPS8044, Rockville, MD, 20852-4910,
|
|
Keywords:
|
Bayes Factor ; False Discovery Rate ; Local False Discovery Rate ; p-value ; Evidence ; Statistical Significane
|
|
Abstract:
|
Scientists use P values as measures of statistical evidence, but statisticians continue to debate this issue. In particular, two different measures of statistical evidence disagree on the effect of sample size on the meaning of P values, leading to a paradox. These measures have a distinguished history and are related to the measures of evidence implicitly used by the Bayesian versions of False Discovery Rates (FDR) and Local False Discovery Rates (fdr). Considering the False Nondiscovery Rate (FNR) throws light on the paradox as well as an epidemiology challenge put forth to the presenter. The examples bring out the difficulties and irreconcilable differences between FDR/FNR and fdr in how they interpret the evidence represented by P values.
|
