Dear SPM experts,
I recently started a thread about variance settings in general https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;bf26a0e6.1209 , in which I also asked about the settings for the factor "subject" in flexible-factorial designs. As this might have gone unnoticed, what exactly does it mean to change the variance settings for a subject factor?
For a within-subject factor, equal variance means that the variance of level 1 is equal to level 2, right? As this might not necessarily be the case, setting it to "unequal" per default seems to be a good idea to me (taking into account some violations of the assumptions should "correct" somehow and produce more reliable data?).
For "subject" as a between-subject factor, equal variance means that the variance in subject 1 is similar to 2, 3, 4 ..., right? Now, I have doubts whether this is really true:
1) Even if they are drawn from the same population, it is rather unlikely to have EQUAL variances due to the small numbers of subjects, thus the variance should rather be SIMILAR.
2) Subjects have different hemodynamic responsivities (e.g. Thomason et al., 2007, Hum Brain Mapping).
3) Even if subjects have equal variance, some error might be introduced by e.g. varying scanner signal from time to time.
Now, what would happen when setting the variance for "subject" to "unequal"? Would this result in some form of correction that is too conservative (misses) or too liberal (false positives) or something which cannot be interpreted at all? What happens with the between-subject variance, essential for random-effects analyses, when variance is set to "unequal"?
Best,
Gabor
|