Is an analysis of PET data with SPM2 a fixed or random effects analysis?
(Suppose the data have more than one scan/subject.)
spm_spm_ui.m has this to say, so it appears to me that unless a two-level
analysis is performed, the results of such a PET study are to be
interpreted as fixed effects.
% Note that SPM only considers a single component of variance, the
% residual error variance. When there are repeated measures, all
% analyses with SPM are fixed effects analyses, and inference only
% extends to the particular subjects under consideration (at the times
% they were imaged).
%
% In particular, the multi-subject and multi-group designs ignore the
% variability in response from subject to subject. Since the
% scan-to-scan (within-condition, within-subject variability is much
% smaller than the between subject variance which is ignored), this can
% lead to detection of group effects that are not representative of the
% population(s) from which the subjects are drawn. This is particularly
% serious for multi-group designs comparing two groups. If inference
% regarding the population is required, a random effects analysis is
% required.
|