Joel,
> I had a similar problem with a PET series. However I can get results
> (ie it doesn't crash or give nonsensical results) if I just omit the
> scan data & condition number. I tried your method too which gave
> almost identical results (only tried on one particular study).
I am very intrigued/curious/worried... the results should be *exactly*
identical.
How exactly did they differ? If you happened to have them, I'd love
to see the SPM.mat's from each of the analyses.
> For example, the protocol has 4 scans at 4 different conditions. If
> scan 2 fails on one subject can I just enter the conditions for that
> subject as 1 3 4? Or does this not do the stats properly?
If the SPM-UI doesn't crash, then it should be fine. But I'd have to
see the design matrix to be sure).
> Does missing data just amount to losing power or does changing the
> symmetry of your model like this do something more weird &
> wonderful?
Well, less data always means less power. Symmetry does has some good
properties, like robustness, and that it generally yields the most
powerful designs for any given total amount of data, but it doesn't
"break" anything per se.
-Tom
|