it was luck.
with the old file the factors were modeled in rows they should be have been
modeled in columns. If you look at
1 1
1 2
it doesn't matter if the matrix is transposed, you get the same matrix.
on the other hand
2 1
2 2
is not the same when transposed. This would have been potentially ok if it
behaved the same way with all designs, but it did not. The matrices were
transposed only if #scans==#factors (minus the subjects factor).
The updated version of spm_config_factorial_design models the factors in the
columns and treats all designs similarly whether or not the
#scans==#factors.
darren
----------
> -----Original Message-----
> From: Janssen, J.J. [mailto:[log in to unmask]]
> Sent: Wednesday, January 16, 2008 9:36 AM
> To: d gitelman
> Subject: RE: question
>
>
> FYI:
>
> as stated before, i had 2 groups, 2 conditions (images) per subject.
>
> - 3 facors in order: subject, group, condition
>
> group1:
> - i enter scan 1 (condition 1) then scan 2 (condition 2)
> - condition matrix:
> 1 1
> 1 2
>
> group2:
> - i enter scan 1 (condition 1) then scan 2 (condition 2)
> - condition matrix
> 2 2
> 1 2
>
> matrices were based on the following:
> the first ROW corresponds to the factor GROUP, the second ROW
> corresponds to the factor CONDITION. The first COLUMN
> corresponds to IMAGE 1, the second COLUMN to IMAGE 2.
>
> for an interaction effect: 0`s + 1 -1 -1 1, and the
> _old_spm_config_factorial_design file i get the exact same
> results as with
> 1 1
> 1 2
>
> 2 1
> 2 2
>
> and your _new_ spm_config_factorial_design file.
>
> was this luck?
>
> thanks,
> -joost
>
>
> -----Original Message-----
> From: d gitelman [mailto:[log in to unmask]]
> Sent: Fri 1/11/2008 21:29
> To: Janssen, J.J.
> Subject: RE: question
>
> Joost
>
> > -----Original Message-----
> > From: Janssen, J.J. [mailto:[log in to unmask]]
> > Sent: Friday, January 11, 2008 2:20 AM
> > To: [log in to unmask]
> > Subject: FW: question
> >
> >
> > Dear Dr Gitelman,
> >
> > i´m working with SPM5 on a flexible factorial design and
> was reading
> > the recent thread on the list (september 2007 i
> > believe) in which you ask for help/give advice. was your problem
> > solved?
>
> Yes
>
> > if so, could you help me out?
> > - my set up: two groups, two conditions (in my case two structural
> > scans) per subject
> > - i specified 3 factors in the following order: subject
> > (independence: yes; variance:equal), group (independence:
> > yes; variance: unequal), condition (independence: no;
> > variance: unequal)
> > - for each subject of group 1 i entered both scans and
> condition 1 1;
> > 1 2
> > - for each subject of group 2 i entered both scans and
> condition 2 1;
> > 2 2
> > - i specified a main effect for factor nr. 1 (subject)
> > - i specified an interaction effect for factor nrs. 2 nd 3
>
> this sounds correct.
>
> >
> > my goal: to be able to test for interaction effects between
> group and
> > condition as well as main effects for group.
> >
> > i am not sure whether this is the same as your setup.
>
> essentially it is, but you have 2 conditions and I had 3.
>
> > is this what you did? what changes did you make after the advice of
> > Ged and Volkmar?
>
> Please see those postings. It helped me set up the contrasts properly.
> For you the contrasts would be:
>
> The effect of group should be (F test)
> C*1/N1*ones(1,N1) -C*1/N2*ones(1,N2) 1 1 -1 -1
>
> you can also do a t-test
> C*1/N1*ones(1,N1) -C*1/N2*ones(1,N2) 1 1 -1 -1 or
> -C*1/N1*ones(1,N1) C*1/N2*ones(1,N2) -1 -1 1 1 <-note the
> signs change
>
>
> where C is the number of conditions (in your case 2) and N1
> and N2 are the numbers of subjects in groups 1 and 2 respectively.
>
>
> The effect of condition (F-test)
> zeros(1,N1+N2) 1 -1 1 -1
>
> or t-tests
> zeros(1,N1+N2) 1 -1 1 -1 or zeros(1,N1+N2) -1 1 -1 1
>
>
> The interaction (t-tests
>
> zeros(1,N1+N2) 1 -1 -1 1
> or
> zeros(1,N1+N2) -1 1 1 -1
>
>
> I believe this is correct.
> darren
>
> >
> > thanks!
> >
> > -joost
>
|