Dear SPMers,
We are running comparisons of patients and controls, and have had
difficulty using the standard two-sample t-test approach. For example, one
region of interest shows strong activation in the control group but not in
patients. However, the comparison of controls-patients fails to reach
significance. This appears to be caused by higher variance in the patient
group, as the confidence intervals are larger in the patient group and the
combined analysis than in the control group. Our model specified
independent samples with unequal variance, did not use grand mean scaling
or global normalization, and used classical estimation. Increased BOLD
variability is a well-known problem with patient populations. Our effect
seems to be real, but is lost in the noise. If anyone on the list can
suggest an alternative approach to the analysis, it would be greatly
appreciated.
Best wishes,
Paul Wright
--
Paul Wright PhD
Centre for Speech, Language, and the Brain
Dept. of Experimental Psychology Tel: +44 (0)1223 766456 University of
Cambridge Fax: +44 (0)1223 766452 Downing Street
email:[log in to unmask] Cambridge http://csl.psychol.cam.ac.uk CB3
2EB
|