Dear Bryan,
> >The only way in which the second-level analyses can be different is if
> >the first-level models are different. I note that you have used
> >spm_adjmean_fmri_ui to create the contrast images. The implementation
> >of this is based on older versions of SPM. I suggest you repeat
> >analysis 2 by using the standard routines and instead of writing out a
> >single contrast image [1 -1] (analysis 1) write out condition specific
> >images [1 0] and [0 1] and enter these contrast images into the second
> >level. You should get the same results.
> Dear Karl,
>
> I'm a little confused about your answer to this problem. I thought that as
> soon as you have more than one column in your design matrix you introduce
> more than one source of error variance. So you would expect slightly
> different results when comparing a one-sample t test to a multi-subject
> conditions & covariates with two conditions.
Generally you are right but in this instance the statistical model is
the same. Either you have n differences and a design matrix with one
column (of ones for the one sample t test) giving df = n - 1 or 2n
condition means with a design matrix of rank n + 1 (1 condition-
specific difference and n subject effects) giving the same df due to
error = 2n - (1 + n) = n - 1 (i.e. a paired t test).
With very best wishes - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|