Hello,
I have a task for which I'd like to use deconvolution (eg. TENT for example) to extract a series of betas, and use 3dremlfit to get Tmaps. Previously we have been using the betas from these kinds of analyses to derive percent signal change (an average of the betas between two time points minus baseline), and then using percent signal change maps in our level 2 analyses (say a group x condition 2x2 ANOVA). However, in this case, I will have different trial numbers for condition 1 relative to condition 2 (condition one has 9 total trials, condition two has 27). This means that I suppose I will need to take the variances/degrees of freedom into account and use 3dmema for my level 2 analysis.
Two questions follow:
1) Is there any correct statistical way to calculate percent signal change by getting an average of the betas between two time points and then using this information plus some (say) average of the variances/degrees of freedom to carry forward into the level 2 analysis? Or will I need to actually compare only one time point at a time and use the Tmaps at each time point that comes out of the deconvolution (seems like a lot of tests; it would be nice to keep it to one test which is why something like PSC or AUC appears better).
2) Do you think if I carry forth the Tmaps from the Rbuck output from 3dreml fit into the group analysis in 3dMema that 3dMema will take into account the potential statistical problems from having different condition trial #'s? Or do you think I shouldn't be comparing two conditions with different trial numbers at all?
Many thanks for your help.
Best,
Claire