Dear James,
as you found out, for such comparisons you need to assume equal
variances. The differences you observed were due to a change in ReML at
some point during SPM5 lifetime.
You might also be interested by this document from Rik Henson:
http://www.mrc-cbu.cam.ac.uk/people/rik.henson/personal/Henson_Singlecase_06.pdf
Best regards,
Guillaume.
James Patterson wrote:
> Ok, I got it sorted out. SPM5 and SPM8 are both working on both workstations.
>
> installing Matlab r2009b helped, but the main error was that I was using unequal variances
> between groups. This is, I believe, what I've always used - but it doesn't like that now.
> Setting it to presume equal variances gave results consistent with past evaluations on both
> platforms with both packages.
>
> Thanks for the help -
>
> JP
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|