Jeff,
even if you were only interested in a subset of the possible effects
in your design, it doesn't harm to do the full factorial:
Select the Full Factorial, specifiy factors (in your case: 3 with 2
levels each), specify cells (2x2x2 = 8 cells), indicate the level for
each cell, e.g. '1 1 1' for group1, impairment level 1, feedback level
1; '1 1 2' for group1, impairment level 1, feedback level 2, and so
forth.... Pick the appropriate independence and variances settings,
select the appropriate con images for each cell. From then onwards, I
would go with the default settings.... and estimate! You will see that
the estimated SPM.mat already contains a bunch of F- and t-contrasts,
including F-tests for all your main and interaction effects. You can
now create additional comparisons you are interested in (most likely
t-contratsts to see uni-directional results) to your heart's
content...
Good luck,
Tobias.
On 3/23/07, [log in to unmask] <[log in to unmask]> wrote:
>
>
> Thanks. I looked at the full factorial initially, but I was confused about
> the "subject/constants" factor (which is discussed in the context of the
> flexible factorial) and I ran across the following in the instructions that
> appeared to point towards the flexible and seemed more appropriate to my
> analyses:
>
> %A typical example here would be a
> % group-by-drug-by-task analysis where, perhaps, only (i) group-by-drug or
> % (ii) group-by-task interactions are of interest. In this case it is only
> % necessary to have two-blocks in the design matrix - one for each
> % interaction. The three-way interaction can then be tested for using a
> % contrast that computes the difference between (i) and (ii).
>
> I'm not afraid to say that I'm really struggling...but very willing to
> learn. Many, I think, would benefit from a full- and flexible-factorial
> example run-through [i.e., similar to those already constructed for SPM2
> (Henson face data)]. There are full and flexible run-throughs, but these
> are for estimating either the IBS or FIR in a single group. Something
> covering the scenario noted above (and in the SPM5 marginal instructions)
> would be most helpful.
>
> Duly noted about the equal/unequal variance issue. I figured there was a
> reason for unequal to be the default. :)
>
> Jeff
>
>
>
> ----- Original Message -----
> From: Tobias Egner <[log in to unmask]>
> Date: Friday, March 23, 2007 8:38 am
> Subject: Re: [SPM] SPM5 Mixed Design Flexible Factorial Question
> To: [log in to unmask]
>
> > Hi Jeff,
> >
> > why don't you go with the 'Full Factorial' option? In that case, SPM5
> > will actually specify the interaction contrasts for, you including the
> > 3-way interaction you are interested in (and you can of course create
> > more specific contrasts yourself). Here you will have to specify how
> > many factors and levels you have, and then just select the con images
> > that correspond to a given cell of your 2x2x2 design.
> >
> > Also, yes, the within-subjects factor is not independent, while the
> > between-subjects factors are. It is my understanding though that you
> > shouldn't assume equal variances for either.
> >
> > Good luck,
> >
> > Tobias.
> >
> > On 3/22/07, Jeff Browndyke, Ph.D. <[log in to unmask]> wrote:
> > >
> > >
> > >
> > > Wise listmates,
> > >
> > >
> > >
> > > After beating my head against the wall for awhile and read-
> > rereading the FIL
> > > manual, I'm at an impasse in how to set up a mixed design in
> > SPM5 with a
> > > within-subject factor "feedback" (2 levels) and two between-
> > subject factors
> > > "group" (2 levels) and "impairment" (2 levels). Additionally,
> > it's my
> > > understanding that SPM5 is only capable of visualizing 2-way
> > interactions,> but there may be a way to set up a contrast to see
> > the 3-way interaction of
> > > group x impairment x feedback? How do I specify the 3-way
> > interaction? Is
> > > it just a difference of interactions below (i.e., [3 4] [3 2])?
> > >
> > >
> > >
> > > Here's what I can glean so far:
> > >
> > > 1.) Flexible factorial design
> > >
> > > 2.) Factors: subject (reserved word – not certain if this is
> > a place
> > > marker for one of the two between-subject factors or separate factor
> > > altogether), feedback, group, impairment
> > >
> > > (I'm I correct in specifying that the within factor is non-
> > independent/equal> variance, while the two between factors are
> > independent/nonequal variance?)
> > >
> > > 3.) "Subjects" -> "Scans" -> "Specify Files" (2 .con images
> > for each
> > > subject referring to the two levels of the "feedback" within-
> > subject factor)
> > >
> > > 4.) "Conditions" -> completely stuck here! I've found
> > examples for
> > > single within- x between-factor, but nothing specifying single
> > within- x two
> > > between-factors.
> > >
> > > 5.) "Main effects and interactions" ->
> > >
> > > Main Effect 1 (feedback): 2
> > >
> > > Main Effect 2 (group): 3
> > >
> > > Main Effect 3 (impairment): 4
> > >
> > > Main Effect 4 (subject/constant): 1
> > >
> > > Interaction 1 (group*impairment): 3 4
> > >
> > > Interaction 2 (group*feedback) 3 2
> > >
> > > Interaction 3 (impairment*feedback) 4 2
> > >
> > >
> > >
> > > Am I on the right track? Any assistance or pointers toward
> > walkthroughs or
> > > data examples pertinent to my needs would be most appreciated.
> > >
> > >
> > >
> > > Cheers,
> > >
> > > Jeff
> >
> >
> > --
> > Tobias Egner, Ph.D.
> >
> > Cognitive Neurology and Alzheimer's Disease Center
> > Feinberg School of Medicine
> > Northwestern University
> > 320 East Superior, Searle 11
> > Chicago, IL 60611
> > Ph: (+1) 312 503 1749
> > Fax: (+1) 312 908 8789
> >
--
Tobias Egner, Ph.D.
Cognitive Neurology and Alzheimer's Disease Center
Feinberg School of Medicine
Northwestern University
320 East Superior, Searle 11
Chicago, IL 60611
Ph: (+1) 312 503 1749
Fax: (+1) 312 908 8789
|