Dear Dr. Friston, Eric and SPM group,
> Regarding Cindy's question, isn't there an issue of the ability to
> falsify? That is, even though she could include a score X
> pharmacological challenge interaction in the model, what would be the
> confidence in a negative result? It seems from the wording of her
> question that sensitivity might be her primary concern. If so, the
> answer would depend upon the effect size of interest as well as the
> sample size and covariance of this interaction term with the group by
> pharmacological challenge interaction term.
> I'm sure your absolutely right. Any analysis with only 6 subjects will
> have low power. If the null hypothesis of no score x challenge
> interaction can be rejected then that would be very nice. A null
> result however is quite likely and in this case no inference can be
> made.
Thank you for your feedback. That is exactly the issue. Although within
each group of 6 subjects regional differences are evident; between group
differences are not evident. The null hypothesis cannot be rejected. With
covariance for HRSD score only in the depressed group, a score X challenge
interaction is evident; however, it seems that the null result precludes
this inference. Alternatively, if the two groups are combined producing a
sample of 12 subjects, a score X challenge interaction (essentially a group
X challenge) interaction is found. Which is the correct method?
Thanks,
Cindy
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|