 |
 |
Hi Anderson, Thanks for your reply. I was thinking of including subject-specific variables of interest (as opposed to nuisance variables). For instance, I wanted to include individuals’ performance on a behavioural task (demeaned) to examine how activation within contrasts of interest (linear/quadratic) is modulated by performance. Would this be possible/make sense in this context? Regarding the reference level, am I right in thinking that the results would be with L1 as reference instead of L5? To make sure that I understand the logic, do the contrasts below look right? positive linear: [0 0 0 … 0 0 0 -48 -24 0 24] negative linear: [0 0 0 … 0 0 0 48 24 0 -24] n-shape quadratic [0 0 0 … 0 0 0 -48 24 48 24] u-shape quadratic [0 0 0 … 0 0 0 48 -24 -48 -24] Best wishes, Chris ############################################### Hi Chris, With this design, the contrast to test if each of the 5 levels (L1-L5) are larger than zero are: C1 (L1 > 0): [1 1 1 .... 1 1 1 24 0 0 0]' C2 (L2 > 0): [1 1 1 .... 1 1 1 0 24 0 0]' C3 (L3 > 0): [1 1 1 .... 1 1 1 0 0 24 0]' C4 (L4 > 0): [1 1 1 .... 1 1 1 0 0 0 24]' C5 (L5 > 0): [1 1 1 .... 1 1 1 0 0 0 0]' Then compositions between these contrasts for more complex questions are straightforward. For instance, to see if L1>L5, you can see if C1-C5 > 0, giving a contrast: C6 (L1>L5): [0 0 0 .... 0 0 0 24 0 0 0]', which is equivalent to your current C1 (in fact, as coded, the reference level is the 5th, not the 1st). To see if there's a linear relationship over the 5 levels, L1>L2>L3>L4>L5, you can compute 2*C1+1*C2+0*C3-1*C5-2*C5, giving a contrast: C7 (L1>L2>L3>L4>L5, linear): [0 0 0 ... 0 0 0 48 24 0 -24]' You can create other contrasts in the same way to answer other questions. Regarding differences between subjects as covariates, I'm not sure I follow. Do you mean including things like age, sex and other subject-specific variables as nuisance? It's not necessary for this design, as you already have subject-specific EVs. All the best, Anderson On 16 February 2015 at 16:53, Christophe de Bezenac <[log in to unmask]> wrote: On 16 Feb 2015, at 16:53, Christophe de Bezenac <[log in to unmask]> wrote: > Dear FSL experts, > > I am conducting a 1-factor 5-levels ANOVA on 24 participants (see attachment). I have a parametric design and am interested in examining negative and positive linear as well as the negative and positive quadratic BOLD response associated with my 5 levels/conditions in a whole brain analysis. Following the f-test (which, as for as I understand, shows me all of the above mixed together), how would I go about finding regions associated with each of these separately. Also, I was planning to add differences between individuals as covariates. Is this all possible within a single higher-level glm? If not, would you be able to suggest some alternatives? > > Thanks in advance, > Chris > > <design.png>