Dear Andreas, > > > "better" in terms of the validity of the inference. Note that > > > conjunctions across subjects is a kind of "middle way" for > > > multi-subject inferences which you may want to try, but you need to > > > get the fixed effects analysis to work first. > Is it true that in the conjunctions the testing is ensured by accounting > the total variance on all contrasts whereas in the random effects the > testing goes on the parameter estimate devided by the variance for each > subject seperately? Not quite. A conjunction analysis is usually implemented at the first-level (i.e. using a fixed effect analysis) where the inference is based on within subject variability (a conjunction of effects at a first-level can be used to generalize the inference to the population of subjects using a confidence-interval type meta-analysis - see the latest edition of NeuroImage). A second-level or random effects analysis uses the parameter estimates from the first level and, implicitly, between subject variability. Estimates of within-subject variability from the first level do not enter into the second level. With best wishes - Karl %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%