Hi Conn users!!

I am wondering about how "correct" is to change the dimensions of the white matter or CSF confounds in my data. There are a couple of subjects in which the distribution curve of the denoising step shows a bit to the left or to the right, so with more error, I changed the white matter dimensions from 5 to 10 and now my data show activations in places it didn't use to and some activations that were present before with 5 dimensions, do not show up now with 10. 
Also I have seen that when you change the dimensions, it is not only for one subject, but it applies to all of them. 
So my main question here would be if it is correct to do those changes in this situation and how valid it would be when publishing. 

The second part of this question is that I have applied a mask in the setup.options to restrict my analysis from what I have seen in the whole brain mask, but when putting the mask (frontal lobe mask), the distribution curve in the denoising step does not looklike before, there is less error. This is understandable since the mask is smaller, but if I saw some interesting potential significant results in the whole brain mask with 10 dimensions in white matter, should I put them now again? Because if I put 5, the results are not the ones I expect to (from the previous whole brain mask setting with 10 dimensions in the white matter). 

Thank you very much in advance!

Hi Albert,

Regarding the first part of your question, yes, that is actually the recommended use, where you may adapt the number of dimensions to your own dataset (keeping the same value across all subjects) to try to minimize any residual motion/physiological effects in the data

Regarding the second part, when selecting an explicit mask in Setup.Options.Analysis mask the sample histograms of voxel-to-voxel correlations displayed in the Denoising step are only computed across pairs of voxels within the analysis mask, so yes, it is perfectly fine to expect a different distribution when using a reduced mask compared to a whole-brain mask. That said, if the analysis mask is small you probably want to still base your choice of "optimal" denoising strategy on the whole-brain data (perhaps an exception to this would be in lesion studies where the analysis mask may explicitly exclude lesion areas), so I would still suggest to keep the same number of dimensions (10) as you were using before. The general recommendation is to decide on denoising parameters before proceeding to 1st-level and 2nd-level analyses at all (to help make sure that this choice is independent from your actual research questions in order to avoid p-hacking) and then keeping that same choice across all 1st- and 2nd- level analyses performed (perhaps the exception to this would be when you are explicitly attempting to quantify the effect of different denoising strategies on your results)

Hope this helps
Alfonso
Originally posted by Albert Bellmunt :

Thank you very much Alfonso! It actually clarifies my problem very well!!