Reply-To: | | [log in to unmask][log in to unmask] > <mailto:[log in to unmask]>> wrote: > > Dear Helmut, > > thank you for your answer! > > I’ve changed the order of columns in SPM.xX.X in not estimated > model (and then estimated it). The appropriate resulting betas > (checked with the ImCalc tool) are identical. So it seems that the > order does not matter here. > > Do you think it would have sense to change the order of > SPM.xX.xKXs.X (or mayby SPM.xX.nKX – as the Reviewer suggested) in > estimated model and then reestimate it? > > You have asked about the design/aim of the research. In our > experiment we want to assess the impact of basic emotions > (sadness, disgust etc) on brain activity in memory tasks. And for > example we see higher activity for disgust than sadness – but the > question is if it is the impact of the basic emotion disgust, or > it is the impact of arousal, which is higher in the set of > disgusting stimuli (we have the ratings). Thus we want to > regress-out the activity correlated with arousal and valence, and > see if something is left, and if so, I think, we can interpret it > as the impact of disgust. So in fact we want to regress-out the > between condition differences in arousal and valence, and adding > parametric modulators for each condition would rather not lead us > to the answer we look for. And I think, we agree with the Reviewer > about the aim and way of putting the regressors into the design > matrix – he/she only thinks that the order matters and the > regressors should be put into first columns, to “take” the > variability correlated with arousal and valence in the first step, > and than to see what is left. But now I believe that the order has > no meaning and the common variability is regressed out anyway. > Hopefully we will convince the Reviewer too :). > > Best regards > > Marek > > > W dniu 2016-06-01 o 18:33, MRI More pisze: >> Dear Marek, >> >> You're right, in case the regressors correspond to separate "conditions" and assuming you rely on the canonical HRF without derivatives then the order does not matter because there's no orthogonalization between condition regressors (there's orthogonalization within condition regressors though in case you work with derivatives or FIR sets). You can simply try out by changing the order when e.g. setting up the design matrix via the GUI. This would also be the easiest way if you really wanted to change the order (e.g. for having the relevant ones at the beginning to be able to go with "short" contrast vectors, as the other columns are zero-padded automatically). The raw design matrix is stored in xX.X, the filtered and whitened one in xX.xKXs.X. >> >> The order *does* matter when working with parametric modulators, as the PMs of a certain condition undergo serial orthogonalization (in case the orthogonalization is turned on; the GUI in SPM12 has a corresponding option which was not present in previous versions). The first PM regressor would explain variance unexplained by the condition regressor then, the second PM regressor would explain variance unexplained by the condition regressor and the first PM regressor and so on. >> >> But leaving this aside, what do you try to model with the arousal / valence regressors exactly, something like the button presses and/or the response window? This should be okay. In case you're aiming at capturing the within-condition variance of your different stimulus responses reflected by your different condition regressors then you would have to go with two parametric modulators for each of the condition e |