.
Have the points in Stephen Senn’s guest post fully come across? Responding to comments from diverse directions has given Senn a lot of work, for which I’m very grateful. But I say we should not leave off the topic just yet. I don’t think the core of Senn’s argument has gotten the attention it deserves. So, we’re not done yet.[0]
I will write my commentary in two parts, so please return for Part II. In Part I, I’ll attempt to give an overarching version of Senn’s warning (“Be careful what you wish for”) and his main recommendation. He will tell me if he disagrees. All quotes are from his post. In Senn’s opening paragraph:
…Even if a hypothesis is rejected and the effect is assumed genuine, it does not mean it is important…many a distinguished commentator on clinical trials has confused the difference you would be happy to find with the difference you would not like to miss. The former is smaller than the latter. For reasons I have explained in this blog [reblogged here], you should use the latter for determining the sample size as part of a conventional power calculation.
Error Statistics Philosophy 