So when do you do post-hoc analysis? I would say all the time. I would certainly do post-hoc analysis of my own randomized clinical trials, for example, to generate moderator or mediator hypothesis with the notion that what I get is not a conclusion, it is a hypothesis to be tested in a future study if it fits in with what's going on.

I think there's a great deal of frustration because people feel they should be doing post-hoc analysis or exploratory analysis, but feel that somehow it's devalued, which apparently it is. And I don't quite know what to do about it. I have spoken out about it very publicly to the NIMH, and they have said they agree with me, but nothing else has changed as yet.

It's a very little expenditure of time and effort. Frequently, you have to do it anyway with reference to your original hypothesis because to understand fully what you have found out in it, you do have to do that additional analysis. So I think it's obligatory, and it's obligatory for other reasons as well and that is this part of sort of general education and methodological processes. It's your opportunity to say, "Now, what would I do differently in a future study that would be better than what I've done in the past study." So I can't think of any good reason not to do it.