At 16:59 08.02.99 +0000, Sergio Carvalho wrote:
>>When you add more than one scan, you are just adding more degrees of
>freedom to your linear model.
>
>In fact, if you just use one scan per condition, the parameters for the
>linear model will fit the samples perfectly, and your standard error
>will be zero, thus causing a division by zero while calculating the t
>score.
In the case of one scan per condition you have also to consider the number
of subjects in the whole study. In fact, in that case, the total number of
scans will be
no.subj x conditions.
Therefore, also without repetitions, the residual df will not be zero, as
long you have enough subjects.
------
The design 'multi-subject with replications' allows, for each subject and
condition, a free number of scans to be assigned that can be different for
each subject-condition cell.
Is there any averaging occurring per cell before beginning parameter
estimation? How does the GLM cope with different df for each cell?
Thank you in advance,
my best regards
Luca Finelli
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|