 |
 |
On Mon, 12 Oct 2009 17:08:48 +0200, Dorian P. <[log in to unmask]> wrote: >Dear Guillaume, > >Thank you very much for the explanation. Does that mean my conditions >have anything wrong? I have 19 subjects overall. And would that be >correct to analyze 2 levels with an anova and the third level with >another anova? No, there's no "third" level. Only two levels. >Of course would be better to have them together but don't understand >why everything is altered that way. Things are altered because there's pre-whitening or something like that going on. One possible reason there would be high collinearity would be if your first-level contrasts were indeed mathematically collinear somehow. (Rather than merely "empirically" collinear.) >Best regards. >Dorian. > >2009/10/12 Guillaume Flandin <[log in to unmask]>: >> Dear Dorian, >> >> that probably means that you end up having too many variance components >> in the specification of your model, and not enough degrees of freedom to >> estimate their corresponding hyperparameters in ReML. >> You can see how many components you have with >> Review>Design>Explore>Covariance structure. >> If you select dependent measurements between levels of your factors and >> unequal variance, that is likely to be the case. >> Bringing one contrast at a time at the second level and using one-sample >> t-tests would prevent worrying about that. >> >> Best regards, >> Guillaume. >> >> >> Dorian P. wrote: >>> Dear all, >>> >>> As I didn't receive any answer, I am reposting this again. >>> >>> Does anybody know why the Design Orthogonality changes so much when I >>> set variance "Unequal" to one of the factors in full factorial anova. >>> >>> There is a strong collinearity coming out between conditions that >>> shouldn't have any relation with each other. >>> >>> Attached the designs with "Equal" and "Unequal" variance set. >>> >>> Thank you. >>> >>> Dorian >> >> >> -- >> Guillaume Flandin, PhD >> Wellcome Trust Centre for Neuroimaging >> University College London >> 12 Queen Square >> London WC1N 3BG >>