Hi Raj, with the given design matrix, I would have
thought that instead of the contrasts
[ 1 1 -2 ]
[ 1 0 -1 ] and
[ 0 1 -1 ]
you should instead use the contrasts
[ 1/3 1/2 -1/4 ]
[ 1/3 0 -1/4 ] and
[ 0 1/2 -1/4 ]
to take into account the differing numbers of scans in
the TaskA, TaskB, and Rest groups. E.g., if c were
set to [ 0 1/2 -1/4 ] and X was your given design
matrix, then c'X would contrast the mean of the two
TaskB scans versus the mean of the four Rest scans.
George
--- RJ <[log in to unmask]> wrote:
> I am trying to find out if there are any specific
> rules when setting up contrasts in SPM2 for an fMRI
> analysis where different conditions have differing
> number of samples.
>
> Suppose I have a de-meaned fMRI data set with three
> conditions A, B and Rest with unequal number of
> scans
> say 3, 2 and 4 respectively.
>
> If my design matrix were as shown below (orthogonal,
> full rank, simple box car)
>
> TaskA TaskB Rest
> 1 0 0
> 1 0 0
> 1 0 0
> 0 0 1
> 0 0 1
> 0 0 1
> 0 0 1
> 0 1 0
> 0 1 0
>
> My questions are:
>
> Do the regressors need to be normalized in some way
> (de-meaned/other) considering the fact that the
> actual
> dataset (Y) is already de-meaned ? My concern here
> being the differing number of samples for the
> conditions.
>
> Also, is it important to have the sum of the
> contrast
> weighted regressors equal to zero i.e sum(c1*X1 +
> c2*X2 + c3*X3) = 0 ? In other words is it ok to have
> t-statistic contrasts such as [1 1 -2], [1 0 -1] and
> [0 1 -1] in the case of the above design matrix ?
>
> I did run an analysis comparing both ways i.e
> regressors with zero mean and non-zero mean
> regressors
> and got results which were very close but am trying
> to
> verify whether it is ok to do so.
>
> ....Raj
>
>
>
>
> ____________________________________________________
>
> Yahoo! Sports
> Rekindle the Rivalries. Sign up for Fantasy Football
>
> http://football.fantasysports.yahoo.com
>
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
|