Zitat von Markus Burgmer <[log in to unmask]>:
> my design is as follows:
> 2 groups (N1, N2), 2 conditions (C1, C2). i specified 3 factors (subject,
> group, condition) and got a design matrix comparable to darrens and matts
> one (see attached file). the first 26 columns are for the subjects, the
> last 4 columns are for the factor interactions (N1xC1, N1xC2, N2xC1, N2xC2).
>
> so i want to have the following contrasts, which i managed to specify in
> spm.
>
> my first question is:
>
> are the following contrast definitions correct?
Yes they are.
> main effect of group (N1 > N2) =
> 2*1/N1*ones (1,N1) 2*1/N2*-ones (1,N2) 1 1 -1 -1
> (because of group differences i have to include the rank deficient subjects
> columns, right?)
>
> main effect of condition (C2 > C1) =
> zeros (1,N1+N2) -1 1 -1 1
> (no group effect, no subject columns necessary???)
>
> group difference condition1 (C2 > C1) =
> 1*1/N1*ones (1,N1) 1*1/N2*-ones (1,N2) 1 0 -1 0
> (group effect so subject columns included, just 1 condition therefore only
> 1 as multiplicator ???)
>
> main effect conditions for group 1 (similar to one sample t-test????) =
> 2*1/N1*ones (1,N1) zeros (1,N2) 1 1 0 0
> (subject columns have to be included, but why???)
>
> second question:
>
> i did not manage to define the interaction of the factor group and
> condition.
> 2*1/N1*ones (1,N1) 2*1/N2*-ones (1,N2) 1 -1 -1 1 gives an invalid contrast
> warning. zeros (1,N1+N2) 1 -1 -1 1 as an f test is valid. is this the
> correct contrast?
why would you want to weight your subject columns here? You have got a
[1 -1] difference within each group, and therefore subject column
weights cancel out (think of your contrast as a difference between the
two main effect contrasts, taken column by column).
> furthermore i failed to design the contrast of C1 < C2 for group 1 (or
> group 2).
>
>
> third question:
>
> what is the main difference between a model (2x2 ANOVA) designed with help
> of the "full factorial design" and the same model designed with help of
> the "flexible factorial design" and the single subjects as additional
> columns? in my case the results of the group x condition interaction are
> almost identical in both models, with greater cluster sizes in the flexible
> factorial design.
With subject columns, your conditions do not longer model mean effect
sizes itself, but something like the difference between the mean
activation of the subject and the condition effect size.
|