Hi - it looks like all your modelling is good. I think your question
just relates to the issue of the interpretability of a contrast-of-a-
contrast - i.e. it's now a little harder to be sure what the different
signs in the outputs mean. This is easily resolved though, by also
considering the simpler contrasts. For example, in GP1(pre-post)-
GP2(pre-post) you can interpret that by looking to see what the signs
of GP1(pre-post) and GP2(pre-post) were - you can even 'automate' this
with contrast masking in the FEAT GUI, e.g. enforcing that the GP1(pre-
post)-GP2(pre-post) is only looked for where both GP1(pre-post) and
GP2(pre-post) are positive (or alternatively significantly positive).
Cheers.
On 16 Feb 2009, at 20:49, Christopher Culbertson wrote:
> I am trying to conduct a higher level FEAT analysis (FLAME 1)
> comparing
> change in activation on two separate scans (i.e. pre & post treatment)
> between 3 different treatment groups. Participants underwent 3 runs
> during
> each scanning session in which three different types of videos were
> presented numerous times in a randomized fashion.
>
> First level Voxel-wise GLM analysis was conducted (using custom timing
> files) to compare parameter estimates (PE) of the hemodynamic
> response to
> the three video conditions versus each of the other and rest for
> each run.
> Next, voxel-wise, fixed effects GLM analyses was conducted
> individually for
> each participant to determine the relative activation over the three
> identical runs in each session (pre & post).
>
> Below I have included a design matrix and contrasts intended to
> examine
> differences in the change from pre to post treatment (relative to
> each video
> condition) between the 3 treatment groups. Inputs are from level 2
> fixed
> effects analysis.
>
> Input group EV1 EV2 EV3 EV4 EV5 EV6 EV7 EV8 EV9 EV10
> 1-pre 1 1 1
> 2-pre 1 1 1
> 3-pre 1 1 1
> 1-post 1 -1 1
> 2-post 1 -1 1
> 3-post 1 -1 1
> …
> 4-pre 2 1 1
> 5-pre 2 1 1
> 6-pre 2 1 1
> 4-post 2 -1 1
> 5-post 2 -1 1
> 6-post 2 -1 1
> …
> 7-pre 3 1 1
> 8-pre 3 1
> 9-pre 3 1
> 7post 3 -1 1
> 8-post 3 -1
> 9-post 3 -1
> …
>
> Contrasts EV1 EV2 EV3 EV4 EV5…
> C1. All (pre-post) 1 1 1 0 0
> C2. All (post-pre) -1 -1 -1 0 0
> C3. Group 1 (pre-post) 1 0 0 0 0
> C4. Group 1 (post-pre) -1 0 0 0 0
> C5. Group2 (pre-post) 0 1 0 0 0
> …
> C6. Grp1>Grp2 (pre-post) 1 -1 0 0 0
> C7. Grp2>Grp1 (pre-post) -1 1 0 0 0
> C8. Grp1>Grp3 (pre-post) 1 0 -1 0 0
> C9. Grp3>Grp1 (pre-post) -1 0 1 0 0
> …
>
> My primary concern with this design centers around the between group
> analysis. It seems that when comparing groups with the contrast
> above an
> inversion of the pre/post relationship will occur leading to a [Group1
> (pre-post)] – [Group2 (post-pre)] comparison in the case of C6. Is
> this the
> case? and if so is there a better method to analyze group
> differences in
> change from pre to post treatment?
>
> Thanks for the help.
>
> Chris Culbertson
>
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
|