 |
 |
Dear Bernard, > I'm getting slightly confused about the kind of statstical model that > is implemented in SPM99 for analyzing PET studies with multiple > subjects with repeated measurements on different tasks. More > specifically, are subjects treated as a fixed or a random factor? All SPM analyses treat every effect as fixed. However, with balanced designs it is easy to implement Mixed and Random effects analyses by taking contrasts from a first-level analyis to a second level SPM analysis. In your example the condition-specifc effects (from repeated measures within-subject) are summarized with a contrast and a one sample t test is applied to the ensuing subject-specfic contrasts at the second level. This analysis treats variations in activation from subject to subject as a random effect. > Looking back at the SPM course notes, it is indicated (page 27) that > "contrasts must have weights that sum to 0 over both the subjects > effects and the condition effects" which indicate that subjects effects > are assumed to be fixed. On the other hand, it is said (page 30)that > "essentially subject effects are random effects". It is important to differentiate between subject effects (i.e. block effects, that are treated as nuisance effects and could be regarded as random or fixed) and subject by condition interactions which are treated as fixed (first-level analyses) or random (second-level analyses). To make inferences about subject effects one would simply average all the scans from one subject before the SPM analysis and subject effect would then be treated as random effects. Andrew might be able to make this clearer. I hope this helps - Karl %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%