If you want to accept something resembling the null hypothesis, you're
most likely going to have to decide how large an effect can get before
it's considered "interesting" (meaning, roughly, "problematic for your
theoretical view"). Given that, you can construct confidence
intervals around your effect size, and see if they include your delta.
If they don't, you're justified in saying that you have evidence that
the difference between your conditions is small, with a significance
level corresponding to the largest confidence intervals that work
(appropriately corrected if needed).
Picking a meaningful minimum interesting effect size for BOLD might be
problematic, and even if you could show the effect size must be
negligible, it might still be uncomfortable arguing that the lack of a
meaningful BOLD effect really conflicts with any theory. But I think
it's fair to report what you know about the effect size along with
some archival comparisons. It may be less of a problem for other
imaging modalities.
It's perfectly fine to decide that any non-zero difference would be of
interest, but in order to reliably exclude all non-zero values you'll
need to collect an infinite amount of data, so it's not recommended.
I don't know how to handle any of this in SPM, I don't imagine it
would be too hard for tiered analyses if you're willing to make a few
assumptions and maybe write a bit of code.
dan
|