Hallo Philipp
<...>
> I would briefly like to comment on the judgement of total grey
> matter or total intracranial volume, used as regressors in a
> VBM analysis, as being not legitimate. Also, this is not
> a general opinion to the best of my knowledge in the circle
> of researchers using VBM.
Precisely, I agree entirely: it is a view among statisticians
concerned with observational studies. It's the reason why I draw
attention to it.
>
> If you include total GM as covariate, you correct the local group
> differences for global differences. What remains, is the local
> differences that go beyond the effect of a global effect. It is
> recommended, of course, that you report the main effect on the globals.
> When there is a clear correlation between global and local effects
> such as in AD, then you might wipe out the local effects. Still,
> it needs to be clearly emphasized, that there is nothing wrong
> or illegitimate about such an analysis. ...
All this is pretty clear, and I agree with it. Of course, the
observation that then these effects are wiped out is precisely what in
these non-neuroimaging circles is meant when the resulting estimates
are characterized as biased. One needs to realize that no assessment
of this possible bias can be done if the association between total GM
and explanatory variable is not reported (to complicate matters, bias
can be present even if the association is not significant).
Best wishes,
R. Viviani
University of Ulm, Germany
|