Hello,

I have two questions, one regarding single-subject analysis and the other regarding group analysis.

Q1) I am running a single-subject 3dDeconvolve analysis with 38 regressors of interest. Each regressor corresponds to a single event, that is modelled as a boxcar function convolved with GAM waveform. I need to model each event separately because later I want to group the regressors in different ways for general linear contrasts. Let's assume that the events can be divided into three categories, specifically:
Category 1: comprising 16 events event1..event16
Category 2: comprising 16 events event17..event32
Category 3: comprising 6 events event33..event38
I tested for main effects of each category by applying the following glts:
-gltsym 'SYM: +event1 +event2 +event3 ++event16' -glt_label 1 category1AND
-gltsym 'SYM: +event17 +event18 +event19 ++event32' -glt_label 2 category2AND
-gltsym 'SYM: +event33 +event34 +event35 ++event38' -glt_label 3 category3AND

I want to test for activation differences between categories, i.e. I want to know where brain activation due to the stimuli belonging to category1 is significantly different from brain activation due to the stimuli belonging to category2, the same for category1 vs category3 and for category2 vs category3.
When I compare category1 vs category2, both categories have the same number of regressors:
-gltsym 'SYM: +event1 +event2 +event3 ++event16 -event17 -event18 -event19 --event32' -glt_label 4 category1vs2
When I compare for example category1 vs category3 I am summing up 16 regressors for category1 and I am subtracting only 6 regressors for category3. Which is the correct way to compare these two categories?
Proposed solution1: -gltsym 'SYM: +event1 +event2 +event3 ++event16 -event33 -event34 -event35 --event38' -glt_label 5 category1vs3
Proposed solution2: -gltsym 'SYM: +3*event1 +3*event2 +3*event3 ++3*event16 -8*event33 -8*event34 -8*event35 --8*event38' -glt_label 5 category1vs3
Possibly both solutions give meaningful results and their use depend on what I am interested in..but could you please explain me the statistical implications of each solution and the interpretation of the corresponding results? (In fact, applying solution1 I obtained large differences, while applying solution2 I obtained localized differences and this last result is more meaningful, since the perceptual differences between the two stimulus categories are not striking)

Q2) I want to see if there is a different brain activation between the stimuli belonging to, say, category1 and category3 at a group level.
Is it correct to apply ANOVA or t-test analysis at the group level considering the coefficients obtained from "category1AND" and "category3AND" contrasts of each subject?
Is it necessary to correct for the different number of events summed up in each contrast inside the ANOVA or t-test command?
Or is it correct to apply conjunction analysis at the group level considering the "category1vs3" contrast estimated from single subjects?
I hope this question is clear enough.


I apologize for long posting. Thank you very much for your valuable help,

Silvia