|
Abstract:
|
In this presentation, we point out some instances where statistical methods, which are advocated in some publications, are inferior to those which are thought to be technically more accurate and appropriate. Here are two examples. (i) The delta method based confidence interval approach for demonstrating non-inferiority in terms of the ratio of means is inappropriate. We recommend Fieller-Hinkley distribution based analysis. (ii) The phrase “Mantel-Haenszel average risk difference” is meaningless and the associated analysis is irrelevant to the objective of proving non-inferiority. The MH type test by Yanagawa et al. is a valid approach. The W-square test, which is recently proposed MH type test, may be another better alternative. Additional issues will be discussed. We provide further details along with references.
|
