Dear Kevin,
I want to chime in on this:
> For single-subject-versus-group comparisons, do the same thing but you'll need to set "variance" to "equal" otherwise you will get an error. I believe this is required because it is telling SPM to treat the single subject scan as a mean image of a group of scans that has the same variance as the control group (anybody please correct me if this is not correct).
The thing is that, for reasons I can't understand, people used to set variance to "unequal" in single subject studies. Of course you cannot compute the variance of a group with a single object in it, it's not defined. But still they did this... Now, in SPM8 you can't do this anymore, it gives you an error. As you expect from a 0/0 calculation. In SPM12 they changed things again and if you set variance to "unequal" with a single subject you get in output the same identical results of when you set "equal".
Now, the weird puzzling thing is that in SPM5 and earlier, if you set variance to "unequal" it performs the computations without any problem and gives you an output. Which has the same con file of a computation performed with equal variance, of course, but an spm_T file which has t values that are about 3 times larger than with "unequal" variance. I tried asking in this mailing list if anybody knew what was going on but... Got no answer! (BTW, the results you get with "equal variance" are identical up to numerical precision with all versions of SPM I tried)
I've got no clue regarding what was going on there. And, since most FDG PET SPM literature has been performed with SPM5 and earlier... It's kind of puzzling.
(BTW, I'm about to perform a number of tests to try to understand which of the solution is the value you actually want, and I'm find conflicting results....)
Cheers,
a puzzled Luca
Rispetta l’ambiente: non stampare questa mail se non è necessario.
Respect the environment: if it's not necessary, don't print this mail.
|