Dear all,
I have a few quick questions about setting up my GLM and using randomise that I would be really grateful for your help in answering.
I’m working on a study looking a development in adolescence with resting FMRI. We are looking for changes in networks (‘nodes’ ) using ICA/dual regression and their interactions (‘links’ ) between these networks using FSLNets. I’m interested in three specific questions relating to age/gender and I have set up separate GLMs for each using the FSL GLM webpage:
a. What networks change positively or negatively with age (age-covariate GLM)
b. Do males and females have different networks after adjustment for age (two-group difference adjusted for covariate)
c. Are any networks change with age different between genders (two groups with continuous covariate interaction)
This has led me to a few specific questions:
1. In my simplest GLM with only age as a covariate, for each participant I have simply included their age minus the mean-age. I believe I still need to do the -D (de-mean) option in randomise, but what else is this doing, and is it necessary given I have demeaned my covariate?
2. Using the same GLM as above with age as a covariate, in the FSL GLM webpage it suggests having one contrast for the group mean activation. Is this necessary with resting FMRI?
3. I have used 3 separate GLMs (described above): is it necessary to use all 3, or could it be simplified into only one GLM?
4. On the webpage it discusses using gender as a covariate (of categorical data) and de-meaning this (i.e. I have a 50:50 gender split so I would use -1 and 1). However, I have used gender as a group/event i.e. 1 for males and 0 for females in one event, and the reverse as another event because I thought this seemed to fit better with the examples given in the webpage (for GLMs b & c above). Is it better to use gender as a covariate or group event, or are they equivalent (allowing that one would need to change the contrasts if gender was a covariate)? I’ve enclosed my GLMs in case I haven’t explained this very well.
Many thanks in advance for your help, it’s really appreciated. I’m sorry if these questions seem rather trivial but I’m new to using the GLM/randomise approach and I’m keen to ensure I’m doing it correctly.
Best wishes,
Mike
