Hi Ming-Tsung,
For testing within subjects comparisons, it may actually be easier (or
more accurate, or both) for the simple effects you list to calculate
whatever differences you can at the single subject level, and then
take the resulting con* images to a second level analysis. For
example, for each subject you can perform a contrast that gives you
"high temperature > low temperature"; the group effect is then simply
just a one-sample t-test on these images (1 image per subject). This
approach is a bit more transparent (to me) in making sure your degrees
of freedom are appropriate, and also removes the need for explicitly
controlling for repeated measurements in your design matrix (because
these one-sample t-tests only have one image per subject). Perhaps
other people have different preferences though?
[Also, you noted that you enter the t maps for each subject into the
design matrix. The second level analyses are typically done on the
effect size (con_*) images from each subject, not the t values.]
Finally, regarding within-subjects ANOVAs in SPM, this technical note
by Rik Henson and Will Penny is very useful:
http://www.mrc-cbu.cam.ac.uk/people/rik.henson/personal/HensonPenny_ANOVA_03.pdf
For a repeated measures design, you basically need to ensure you add
subject columns to your design matrix, which will remove the between
subject variance.
Hope this helps!
Jonathan
> In my experiment, 14 subjects received 2 fMRI scannings (low-temperature vs.
> high-temperature stimulus, 10 stimuli per scan, 10 seconds per stimulus),
> and I want to analyze the difference in brain activations during the early
> (first 5 s) and late (final 5 s) phase of stimulation for the 2 stimulus
> temperatures.
>
> I'm struggling with running a 2(low vs. high temperatures)-2(early vs. late
> phase of stimulation) repeated-measures ANOVA in SPM5.
>
> I used the "full factorial" 2nd level analysis in SPM, and the item
> "Independent" is designated as "No".
>
> The procedures are as follows:
>
> I set 2 factors: Temperature (2 levels: low vs. high) and Phase (2 levels:
> early vs. late phases).
> The 4 cells are designated in order as:
> 1-1 (cell 1) = low-early
> 1-2 (cell 2) = low-late
> 2-1 (cell 3) = high-early
> 2-2 (cell 4) = high-late
>
> I put the 4 t maps (low-early, low-late, high-early, high-late) from each
> subject into the 4 cells by order.
>
> 1. To get the map of "high temperature > low temperature", I used the
> contrast -1 -1 1 1 (0.05 FDR).
> 2. To get the map of "low temperature > low temperature", I used the
> contrast 1 1 -1 -1 (0.05 FDR).
> 3. To know which regions showed "high temperature > low temperature" during
> the early phase of stimulation, I used the contrast -1 0 1 0.
> 4. To know which regions showed "high temperature > low temperature" during
> the late phase of stimulation, I used the contrast 0 -1 0 1.
> 5. To know which regions showed "low temperature > high temperature" during
> the early phase of stimulation, I used the contrast 1 0 -1 0.
> 6. To know which regions showed "low temperature > high temperature" during
> the late phase of stimulation, I used the contrast 0 1 0 -1.
>
> Am I doing right?
>
> Does SPM5 allow the analyses of repeated-measures ANOVA?
>
> I'd appreciate for any help.
>
> Thanks
>
> Dr. Tseng
>
|