Dear Lesley,
> Hi, I have a SPECT study where each subject is scanned under 2
> conditions and ratings made under each condition. I have looked at main
> effect of condition using multi-subject, different conditions.
>
> I want to look at the correlation of change in subjective score between
> conditions with change in BF in cond 2 relative to cond 1. So I have
> one covariate, the difference score (cond 2-cond 1) and 2 scans for
> each subject. Having read previous mailbase answers and looked at the
> archives I thought I should create an interaction covariate. I took the
> centred difference score for subject a, b, c,...., multiplied by [-1 1]
> for cond 1 and cond 2 and entered covariate as
> [-a/2,a/2,-b/2,b/2......] with no specific fits in multi-subject, cond
> and cov design. I used [0 0 1] for increasing covariate with increasing
> change in BF and [0 0 -1] for increasing covariate with decreasing
> change in BF. Is this analysis correct?
This is absolutely correct.
> What is puzzling me is that on examining the resultant SPM{Z}s using
> the plot facility to look at BF in each cond sometimes there is greater
> BF in cond 2 than cond 1 at one voxel, then vice versa at another voxel
> within the same contrast. I had thought that within a single contrast
> all the blobs should show the same direction of BF change between
> conditions. I was hoping [0 0 1] would give areas of increased BF in
> cond 2 relative to cond 1 correlating with increased score and [0 0 -1]
> areas with decreased BF in cond 2 relative to cond 1.
> Is it that the correlation is just with the size of the BF change
> between conditions (increase or decrease) irrespective of which
> condition is showing the relative increase/decrease?
> How do I proceed so that positive correlation would only show areas
> with increased BF in cond 2 relative to cond 1 with increasing
> difference score and negative correlation only areas with decreased BF
> in cond 2 relative to cond 1?
The interaction and main effect (of condition) are orthogonal and can
change independently. With [0 0 1] you might see increased activation
with increased score change or decreased de-activation (they are both
positive interactions). Similarly for [0 0 -1} may give decreased
activations or increased de-activations. To produce a SPM of augmented
activations simply mask the SPM{Z} specified by [0 0 1] by the main
effect SPM{Z} for activations (e.g. [-1 1 0]). Note that [0 0 1]*[-1 1
0]' = 0.
Similarly mask [0 0 -1] with [1 -1 0] for negative correlation only
areas with decreased BF in cond 2 relative to cond 1.
I hope this sorts things out - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|