Hi Shen,
> hi, SPMers,
> we have some questions about corrected threshold for group analysis.
> Sorry if it is a really basic question
> 1, is it right for using none corrected and p=0.005 or more large p
> value ?(default 0.001, SPM5)
> we have two contrast A and B for each subject and are
> interested in A mask with B (B is a conjunction contrast)
> we specify one-way ANOVA of two levels (A , B) and then
> using A (1,0) mask with B (0,1).
> Then we can't get active image with none corrected and
> p=0.001 threshold , in opposition to none corrected and p=0.005 , 10
> voxel or more large p value (0.01, 0.05).
> what i want to know is whether height threshold (0.005 or
> more) is suitable or not ?
uncorrected stats are, a priori, not suitable - the reason being the
need to control for multiple comparisons - as a general introduction
for fMRI stats and others I would recommend reading
http://imaging.mrc-cbu.cam.ac.uk/imaging/AnalysisPrinciples
> 2, how many subjects is suitable to group analysis ?
> someone tell me that the cause is less number of subjects. is
> it right ?
> I have only 4 subjects for group analysis due to expensive
> cost of fMRI scan. If group analysis require more subjects , how
> many is the recommend number of subject at least ?
well it all depends on the level of inference you want to draw though
I would say, as for any data set, the larger the better
You will get an idea of what I mean by level of inference in Friston
et al NeuroImage 1999 How many subjects constitute a study? 10(1):1-5.
The usual inference (average effect in the population requires about
14 - 20 subjects) - note also that some studies seem to show that 20
is the number of subjects you need to have reliable activations across
studies (ie the peak of activity seems consistent if you replicate the
expe) - Finally, recent power analysis also show that the nb of
subjects required depends on a priori hypotheses on where you expect
an effect as it varies in space .. see Hayasaka et al. NeuroImage 2007
Power and sample size calculation for neuroimaging studies by
non-central random field theory. 37(3):721-30.
I know this is probably not the expected answer (ie just a number) but
as always things get complicated ...
hope this helps
cyril
------------------------------------------------------------------
The University of Glasgow, Department of Psychology WebMail system
------------------------------------------------------------------
|