I'm new to using SPM (only been familiar with it for the last 5-6 months, with no prior MRI image processing or analysis experience), and I'm having trouble finding a lot of definitive information on the best way to set contrasts for Flexible Factorial designs. I have read the "Contrast_Weighting_Glascher_Gitelman_2008.pdf" many times, but am not quite understanding their definition for main effect of condition contrasts (MEc; or in the case where one would have more than two groups, I assume the contrast would be the same for main effect of group - Meg). How is it possible that -1 0 1 is MEc (assuming 3 conditions)? If I had a 4th condition, how would I set up main effect of condition?
To set the stage for what I'm looking at:
3 Groups (Genotypes; n=12, n=10, n=6)
3 Conditions (Time - Repeated Measures)
I've set my SPM up such that -
Factor 1 = Subject (Independence, Unequal variance)
Factor 2 = Group (Independence, Unequal variance)
Factor 3 = Condition (No Independence, Equal Variance)
So, my design looks like the design on page 10 (Design 3) of "http://www.sbirc.ed.ac.uk/cyril/download/Contrast_Weighting_Glascher_Gitelman_2008.pdf" only I have a third group with all corresponding conditions/subjects.
The only way I see to set up a main effect of condition would be to do a multiple vector contrast, not a single vector contrast (sorry if my lingo is not correct). Essentially if I input the following for a contrast for conditions (with three conditions): [-1 0 1], is the assumption that this is MEc because we're assuming a completely linear relationship between the three conditions, and therefore using the middle condition as a pivot point that doesn't actually get considered in the contrast? Would it be better to use multiple vectors comparing all three possible condition comparisons, such as:
[-1 0 1]
[-1 1 0]
[0 -1 1]
?
Wouldn't this actually be a comparison looking at the sum of all three potential differences between time points, and therefore be a better way to actually do MEc? In the first case of using a single vector contrast, isn't there no real difference between using an F-statistic and T-statistic test then (both which are only comparing C1 and C3)? I've run some analysis tests on different data, and this indeed seems to be the case. My SPMs with a single vector comparison come out virtually identical (very, very similar) whether I use an F or T-statistic. However, if I use the second example I gave in an F-statistic it appears to be very close to a sum of all three of the corresponding single vector comparisons (I've only tested this hypothesis with 3 F-stats vs the single F-stat).
I also had questions regarding interaction effects (and the same questions above apply to these). Assuming I try and look at interaction effects, and I use a MEc contrast (either of the above specified vector(s)) on a single group and for the condition, would that be similar to looking at the MEc in a one-way repeated measures ANOVA for that group?
This is hard to interpret perhaps, but considering the design I specified above (in SPM8), and the example design matrix I gave from the article linked (http://www.sbirc.ed.ac.uk/cyril/download/Contrast_Weighting_Glascher_Gitelman_2008.pdf), I would input the following contrast weights to get the contrast I'm describing in the paragraph above, looking at the MEc for just a group 1:
zeros(1,28) 0 0 0 -1 0 1 -1 0 1 0 0 0 0 0 0
Or for group 2:
zeros(1,28) 0 0 0 -1 0 1 0 0 0 -1 0 1 0 0 0
I've included the unnecessary zeros at the end just for understanding sake.
What would the difference be between doing a single vector contrast like that, or the following:
zeros(1,28) 0 0 0 -1 0 1 -1 0 1 0 0 0 0 0 0
zeros(1,28) 0 0 0 -1 1 0 -1 1 0 0 0 0 0 0 0
zeros(1,28) 0 0 0 0 -1 1 0 -1 1 0 0 0 0 0 0
My instinct says that an SPM that was closer to a sum of the significant voxels for each comparison would be more accurate than one which only shows differences between C1 and C3. Is that what doing this triple vector contrast weight estimates, and is that a more appropriate MEc?
Sorry for any redundancy, and I appreciate any help in understanding.
|