Fisherian Evidential Approach to Testing Multiple Hypotheses

• Activity Number: 631
• Type: Topic Contributed
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Haiyan Xu*+
• Companies: Johnson & Johnson
• Keywords: Multiplicity adjustment ; Evidential approach ; hypothesis testing ; totality of evidence ; type I error

Abstract

Many of the multiplicity procedures require adjustment of p-values or significance levels; therefore, the evidential meaning of a p-value not only depends on its underlying hypothesis of interest but also depends on the presence of other related or unrelated hypotheses. We formulize an evidential approach as consisting of inference on statistical evidence and clinical decision-making. Regardless of the number of hypotheses being tested, statistical inference does not require adjustment of p-values or the significance level. However, clinical decision-making requires a statistical significance of the hypothesis tested and a consistency of totality of evidence. The resulting evidential approach strongly controls the type 1 error rate and can be applied broadly.