 |
 |
Hi Anderson,
Thanks. I think in your design you are not testing the effects of the hormones on the interaction of A x B. If I compute the differences between
the levels of the conditions as well as their interaction effects at a first level fMRI analysis and then perform comparisons between the hormones + placebo in separate higher level analyses (one for each COPE), then I won’t be able to perform an F-test over
the different comparisons, is that correct? I thought a significant F-test, for example for a main effect of hormone 1 vs placebo on B, was a prerequisite to further compare effects on B1 vs B2 and B2 vs B3 etc.
I think in order to be able to perform F-tests you would need to add all COPEs of all first level comparisons to 1 second level analysis, does
that make sense? Or do I overestimate the value of F-tests?
Thanks again,
Sandra
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]]
On Behalf Of Anderson M. Winkler
Sent: maandag 21 september 2015 11:25
To: [log in to unmask]
Subject: Re: [FSL] 3 x 2 x 3 within subject design
Hi Sandra,
I seems this design will end up being simpler than it seemed: since there is 1 hormone per visit, comparing visits is the same as comparing hormones. If you compute the differences A1 vs A2, B1 vs B2, B1 vs B3 and B2 vs B3 all at the first
level FMRI, then a comparison between the two hormones + placebo on these 4 differences (COPEs) corresponds to
this design in the GLM manual, i.e., the tripled t-test. It would be done then 4 times, at the 2nd level, one for each of the COPEs from the 1st level.
How does it sound?
All the best,
Anderson
On 18 September 2015 at 12:26, Thijssen, S. <[log in to unmask]> wrote:
Hi Anderson,
Sorry for the confusion. I indeed have 2 hormones and a placebo, and 2 conditions (A:
2 levels; B: 3 levels). There are three visits. Participants receive one hormone/placebo per visit and are tested along the two conditions.
We are indeed mostly interested in the effects of the hormones on the differences between
the levels of the conditions and their interaction effects.
As participants receive one hormone or placebo per visit, we are interested in comparing
the visits and not averaging them.
Thank you for pointing out non-linear interactions between hormones. I will look into
that!
Thanks,
Sandra
From: FSL
- FMRIB's Software Library [mailto:[log in to unmask]]
On Behalf Of Anderson M. Winkler
Sent: vrijdag 18 september 2015 9:38
To: [log in to unmask]
Subject: Re: [FSL] 3 x 2 x 3 within subject design
Hi Sandra,
It seems you are in the right way, but I'm a bit confused -- in some places it seems A and B refer to hormones, but in others it seems to refer to the different conditions (?).
I gather that you have:
- 2 hormones (say, X1 and X2).
- 1 placebo (say, X0).
- 2 conditions, being A (2 levels) and B (3 levels). Are these in the same FMRI run or different runs?
- 3 visits in which the 2 hormones and 1 placebo are tested along the two conditions (A and B) and their levels. Let's call these as timepoints T1, T2, and T3. Do you use one hormone
or placebo per visit, or all three (X0, X1 and X2) on every visit?
This allows various hypotheses. It seems you are more interested in how the hormones affect the differences between the levels of the conditions, is this correct? Or are the differences
between the conditions also interesting?
About the visits, are you interested in changes over time, or the three visits meant to be averaged?
About the other factors (e.g., dietary variables, other hormones), these can be entered as nuisance, sure.
It occurs to me though that hormones often interact with each other in non-linear ways, with positive and negative feedback loops that I wonder whether would be properly captured
in a linear model (I haven't looked into it, just thinking aloud a possible concern).
All the best,
Anderson
On 17 September 2015 at 09:07, Thijssen, S. <[log in to unmask]> wrote:
Hi Anderson,
Thank you for your reply. In a lower level feat analysis, I think I would create EVs for all conditions (A1B1, A2B1, A1B2, A2B2, A1B3, A2B3). I would additionally create contrasts
for A1-A2, B1-B3, B2-B3, B1-B2, B2-B1 and all A*B interactions (A*B1-B3, A*B2-B3 etc.). I have attached an image of the model.
For all three sessions (hormone A, B and placebo), I would feed the cope images of contrast 7 until 15 in a higher level analysis, comparing the effect of the different hormones and creating F-tests for variable A, B, hormones H1 to H3 and for the interactions
(see attached image). I haven't written out all EVs for all hormones, but instead focused on one the effect of hormone 1 (H1) vs. placebo (H3).
I am not sure if I am going about this correctly, as I think this model looks pretty messy. We would be interested in controlling for the effects of other non-investigational hormones and dietary factors that would differ per visit (i.e. per hormone) but not
per condition. Would it be possible to add such variables to the model as well? There would be a lot of identical values in the model.
Perhaps it would be wise to drop one of the experimental conditions of B1 (e.g. B2) to make the model a bit more comprehensible.
Thank you for you help!
Kind regads,
Sandra
Van:
FSL - FMRIB's Software Library [[log in to unmask]] namens Anderson M. Winkler [[log in to unmask]]
Verzonden: vrijdag 11 september 2015 9:44
Aan: [log in to unmask]
Onderwerp: Re: [FSL] 3 x 2 x 3 within subject design
Hi Sandra,
With purely within-subject effects and interactions, yes, it should be possible. If you want to assemble a draft of the design now for your hypotheses,
before even beginning the data collection, it'd be better as if there is a problem you can make changes now as needed.
All the best,
Anderson
On 10 September 2015 at 10:08, Thijssen, S. <[log in to unmask]> wrote:
Dear experts,
We’re planning to set up an fMRI study on the effects of several hormones on behavior. The study will
be a within-subject study, with all participants receiving 2 hormones and a placebo in three visits. Our paradigm consists of two variables, say A and B. Variable A has two levels (experimental and control), and variable B has three levels (2 experimental
and 1 control). This results in a total of 6 different conditions (A1-B1, A2-B1, A1-B2, A2-B2, A1-B3, A2-B3). Would it be possible to assess a 3 (hormonal, different runs) x 2 (variable A) x 3 (variable B) within-subject ANOVA using FEAT, or would you suggest
simplifying our design?
Thank you very much for your help.
Kind regards,
Sandra Thijssen