I'm trying to compare fMRI results from 2 groups of subjects (3
patients and 12 controls) in a 2nd level analysis. Because of the
low subject numbers in the patient group I thought it would be best
to use a non-parametric test.
When I use SnPM (mutli-group; 2 groups; 1 condition) with no
variance smoothing, the resulting t values are identical to those
resulting from the parametric 2 sample t-test in SPM (although
corrected probabilities differ slightly). When I use variance
smoothing of 10 mm FWHM, as recommended for low df designs, t
values are all reduced.
Can anyone tell me if it is valid to use parametric statistics (2
sample t-test) when the number of subjects in one group is so low,
and if there is any advantage in using non-parametric statistics for
this type of 2nd level group comparison?
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dr Ross Cunnington, School of Psychology,
University of Nottingham, NG7 2RD, UK.
Tel: 0115 846 6360 Fax: 0115 951 5324
|