Hi,
I still wonder if the use of a conjunction within a factorial design is valid (SPM5)
Maybe the model below wasn't clear, so I just add some extra info. I hope somebody can reflect on the method.
So, to be a bit more informative:
- It's a long event related disign (Stimulus 2.5 sec, fixation 2.5 sec, Feedback 2.5 sec, fixation 12.5 sec)
- a factorial design (2x2)
- For every 4 conditions, and for every subject, I take contrast: condition>baseline (e.g. 0 1 0 0 0 ...)
- In second level analyses I use a Full factorial model,
that is: every cell consists of the con-files for the whole group, for one condition
(factor1: no independence, unequal variance, f!
actor 2:
nces)
Now, since I'm only interested in regions activated by 1 factor, I want to run the factorial design in the voxels that are activated by 1 factor, I first create a conjunction for that factor. (cell 1& 2 + cell 3 & 4) or (cell 1 + 2 + 3 + 4). Then I use this conjunction as a inclusive mask (0.05), for the effects of interests (main effects, positive effects for a condition and interaction effects)
I use FWE (0.05) to correct for multiple comparisons .
I hope the design is a bit more informative now and that someone could
add some feedback on the method.
Thanks,
Martijn
Date: February 18, 2008 5:22:43 PM GMT+01:00
Subject: conjunction within Full Factorial design SPM5
I'm running a 2x2 full factorial design on one group of subjects in SPM5.
Suppose the design is Factor A x Factor B, then the corrsponding cells will be:
A1B1, A1B2, A2B1,A2B2
I assume in my task that Factor A has no independency and unequal variance.
Factor B is independent, and has equal variance.
After reading Nichols et al. (2005) I wonder if I could use a conjunction!
within a
ce I'm only interested in possible effect within regions activated
by Factor A, I want to make a conjunction of both conditions A1 & A2. I could use the
following contrasts to create the conjuntion:
A1B1+A1B2 (1 1 0 0) with A2B1+A2B2 (0 0 1 1).
This will give me all regions activated by both conditions of Factor A,
including regions possibly activated by Factor B, including a possible
main-effect, right?
or I could run a conjunction over all cells seperated:
A1B1 (1 0 0 0) + A1B2 (0 1 0 0) + A2B1 (0 0 1 0) + A2B2 (0 0 0 1)
This will give me all regions activated by Factor A, and maybe by factor B
(when there is no main effect of factor B, B1 and B2 could still have an equal
effect on regions activated by A1 and A2)
Which one seems to be the most relevant? And, moreover, does it make sense to
use conjunctions like this within a factorial design? For me, it seems very usefull
since I can test if there are any posit!
ive effec
,
interactions effects (A1-A2)x(B1-B2) and positive effects of B (B1>B2) within regions
of interest (those that are activated by Factor A)
However, I also could use a F-test over all cells (1 0 0 0, 0 1 0 0 etc.) but since I'm
only interested in regions activated by both A1 and A2, I thought a conjunction will be more relevant.
(by the way, I mask the conjunction with a (inclusive) mask p-value of 0.05, and run the test under
p= 0.05 (FWE) to control for multipl. comparisons)
Does this all make any sense?
Thanks,
Martijn
UMC-Utrecht
The Netherlands
