Comment from Gerhard Nehmiz
AnonymousSupportAcademic
Summary: Gerhard Nehmiz, a former leader of the "Bayes Methods" Working Group of the International Biometric Society, suggests clarifying the terminology in the draft guidance. He argues that the "Design prior" should be referred to as a "scenario" because it represents a desired effect size rather than a distribution of prior knowledge.
May I introduce myself, I am a former deputy leader of the Working Group "Bayes Methods" of the
International Biometric Society, German Region (https://www.biometrische-gesellschaft.de/ag/bayes.html ,
https://web.archive.org/web/20250806045745/https://www.biometrische-gesellschaft.de/arbeitsgruppen/bayes-methodik.html ).
Sorry I saw this only recently, could not submit comments through the EFSPI.
The "Design prior" on lines 349-354 is actually a blurred version of the well-known Delta,
i.e. the effect size that is considered to be relevant, or what we wish that the treatment should
achieve. It is defined through external requirements, does not describe what is actually known about
the effect size before the study. Therefore, it cannot be combined with the study data via the Bayes
formula. By expressing a stronger wish, the posterior effect estimate could be enlarged arbitrarily.
Therefore it would be OK not to call this distribution a "prior distribution" for the treatment effect.
May I propose "scenario".
Kind regards, Gerhard Nehmiz.