3dRegAna is quite outdated and is no longer maintained. Since you have a simple regression model, I suggest that you use 3dttest++.by

Sanaz, > the term 'removed' in means that the covariate is controlled for in this comparison. You're correct that the description in the 3dttest++ help can be improved. You seem to have a 2 x 2 data structure. If you decide to use 3dttest++ instead of 3dMVM to handle the situation, there are a couple of subtleties involved. One, it sounds like you're going to ignoby

Giovanni, I don't have time reading the literature at the moment. Your questions seem too generic. Instead of worrying about modeling specifics, it would be more helpful to first spill out your research focus. For example, with the resting-state dataset, what is your research goal/interest/hypothesis about those multiple time points? How are you analyzing the data: whole-brain voxel-wise, seby

Kara, you may try encoding some information into the header with something like 3drefit -substatpar # fico N-3 yourFile where # is the sub-brick number for the correlation and N is the number of datasets in the input.by

Tom, It seems that 3dLSS does not currently work with more than one stimulus type. You may trick it by pretending there is only one stimulus type, but that would involve tediously sorting out the trials among the stimulus types in the output.by

> It looks like the t-statistic map survives the FDR scrutiny... this is the one that matters when it comes to > assessing statistical significance of ISC in the voxels, right? Right. So, the program might have barked up the wrong tree then. I'll have to check it up later.by

Ekarin, try the same variable coding strategy and then feed them into 3dLME. >> 2. Is there a way to specify GLTs after 3dLME has already been run? Unfortunately no. >> 3. The results from script 1 return what I think are estimates/regression coefficients and a z-test (rather than a t-test), is this correct? Correct.by

Ryann, that's just a warning indicating that particular sub-brick would have problem gong through the FDR scrutiny. So, don't panic.by

Ekarin, Since all the three predictors are between-subject quantitative variables, I suggest that you do the following: 1) center each of the three variables properly (e.g., around their mean) 2) create four new variables - one for each interaction - and label them 3) perform a multiple regression analysis on the seven variables with, for example, 3dttest++ or 3dMEMAby

> I ran 3dMVM on the coefficients obtained with 3dREMLfit, and defined my desired GLTs. Is this legit? Yes, that's fine. > any tips on the advatages of 3dMEMA over 3dttest++ would be appreciated Ideally the analysis for a dataset should be performed with one integrative model. In real practice this is not always feasible in neuroimaging due to a couple of limitations: model cby

David, it looks like a bug: 3dLME allocates an extra sub-brick which is never filled in the end. So just ignore the last sub-brick.by

Jim, you have two options: 1) Use option 'dmUBLOCK(-X)', or 2) model each trial separately and handle the duration variability at the group level.by

> Do you suggest I alter the VBM input so that it aligns with a 2x2x2 voxel size to match my fmri outputs? Use 3dresample to change the covariate files to the same resolution as the FMRI data. If your original FMRI data resolution is something like 3.75 x 3.75 x 4, you may consider changing all the files (including both FMRI and covariates) to a resolution of, for example, 3 x 3 x 3. Thereby

Tamara, can you verify whether the voxel-wise covariate files and those response variable files have the same voxel size using 3dinfo?by

> Functional images are 101 128 x 128 slices What kind of data do you have: FMRI or something else? What is the voxel size?by

The memory is allocated at the beginning. So more likely something happened to the computer. Maybe kill the job, and restart it.by

Klad, do those voxel-wise covariate files have the same dimensions (or resolutions) as the input files? If not, that would be the cause for the error message.by

Daniel, the answer to all the three uestions of yours is yes.by

Daniel, When deciding on centering strategy, do you believe that those groups are intrinsically different in terms of movement? In addition, the following line -qVarCenters '(0.0450,0.0607,0.0575,0.0492)' \ is not valid. If you want to center each group separately, you have to manually perform the centering yourself before putting the quantitative variable values in the data taby

Eric, with such scarce information, it's difficult to offer help. Could you provide the following? 1) data structure 2) effects of interest 3) 3dLME script 4) exact difficulty of understanding the resultsby

First of all, the two expressions are essentially the same thing: (A1B1-A1B2)-(A2B1-A2B2) = (A1B1-A2B1)-(A1B2-A2B2). Secondly, they are for the interaction between factors A and B in the case of a 2 x 2 design. > does (A1B1-A2B1)-(A1B2-A2B2), for example, mean I do the subtraction of A1B1-A2B1 and the subtraction > A1B2-A2B2 for each subject and the run a paired sample t-test between thby

Alberto, I'm not sure why you need to append a star (*) at those three lines with a single trial. Would that be the reason for the error message?by

Use 3dMVM since you have multiple explanatory variables.by

The situation would be similar to a task-based analysis: build one model for the whole analysis at the individual level. So just directly add the EEG regressor to the model with 3dDeconvolve (your solution #2).by

For a 2 x 2 structure, the F-test for the interaction between factors A and B is essentially equivalent to the t-test (A1B1-A1B2)-(A2B1-A2B2) or (A1B1-A2B1)-(A1B2-A2B2). Even better, you get more information from the t-test than F because t-test directly provides the sign/directionality for the associated effect; for example, a positive t-value shows A1B1-A1B2 > A2B1-A2B2 and A1B1-A2B1 > A1by

Are you running seed-based correlation analysis? Is the effect associated with the EEG regressor considered as a confounding effect or an effect of interest in this context?by

> there are some regressor with always 1 stim times (a cue with the block type at the beginning of each run) You have multiple trials for those conditions when you concatenate all the runs, right? You would avoid the problem if you combine all those stimulus timing files across runs. A side note - if 3dLSS works, I'm curious about the performance of the modeling approach.by

Ryann, try testing it with at least 5 subjects and see if you still have the same issue.by

For a 2 x 2 repeated-measures design, you can break it into multiple one-sample or paired t-tests, and then run 3dttest++ multiple times.by

> I've also tried to split my data set into 5 proportions,my question is in that case should I average > all 5 results to get the whole data set's results? That is not a good idea. Instead, cut each subject pair along the Z axis into multiple (e.g., 5) portions. Suppose you have 100 slices along the Z axis: 3dZcutup -prefix SubjectPair1A -keep 0 19 SubjectPair1+tlrc 3dZcutby