I have two scenarios that I’m working with:
- Within the same day, subjects had two different scans - one scan was moving their lefthand and one hand was moving their righthand.
for this scenario I would like a Left-Right contrast; however, I’m not sure how to do this when I’m analyzing the two separately. Using afni_proc.py I do see that I can enter multiple ‘runs’ but it appears that it just 3dcats the runs into into one big run and my stimfile has to reflect. I posted before and it it seemed that this is not okay to do unless preprocessing has occurred first. Should I preprocess both separately and then run 3ddconvolve on the 3dcat file? Can I just use 3dcalc to subtract the the two stats files?
- Subject have a scan pre intervention and then again do a post scan 5 weeks later.
For this I would like to do a post-pre contrast. Should I follow a similar pipeline as I described above?
For both cases, you can feed those two runs/sessions as separate input files into afni_proc.py. In addition, provide stimulus timing files with one row per run/session. You can make inferences about the two conditions with one model that would properly account for the discontinuities/gaps across runs/sessions.
I’m not sure if I’m doing that correctly. When I input the two runs separately and provide two separate stim files using uber_subject.py I have to modify the second stim file times as if the two datasets are concatenated. Because of this, I am getting very different results than if I run the two files separately. Can you point me to an example afni_proc.py that uses the two separate identical runs i.e. the stim timings are exactly the same?
I have to modify the second stim file times as if the two datasets are concatenated
Can you post here the content of the two stimulus timing files?
The stim files are identical because they are two separate scans. They performed left hand movement for one scan and right hand movement for the second scan.
I assume that there is only one type of task per run. If so, the timing files should be like this:
56 145.6 235.2 324.8
56 145.6 235.2 324.8
Each row corresponds to a run, and the asterisk indicates that there is no trial in that run.
Thank you! That works great!
I have two additional questions (and hopefully my last questions - thanks again for all of your help):
When I run 2 sessions at a time, I do get similar beta coefficients, but I get a slightly bigger T-Stat than if I run each run individually. Do you know why that is?
For one of my datasets, I ran the same scan twice and participants did the exact same task each time. For this, my plan is to average the two runs with:
gltsym ‘SYM: 0.5run1 +0.5run2’
within each run, I would like to do a contrast of the means for a specific taks (each run contains a block design: rest, lefthand, rest, right hand, rest, both hands, rest). For example:
-gltsym ‘SYM: (0.5run1Task1 +0.5run2Task1) - (0.5run1Task2 +0.5run2Task2)’
The gltsym does not like it when I use parenthesis. How can I denote this properly?
Nevermind, I can just apply a similar concept where I can do a conjunction analysis. Though, is it okay to do a conjunction analysis on the results of a previous conjunction analysis?
When I run 2 sessions at a time, I do get similar beta coefficients, but I get a slightly bigger
T-Stat than if I run each run individually. Do you know why that is?
The reason is that with the two runs combined, you double the number of data points to estimate the residuals (noise), leading to slightly more precise and smaller standard error for each effect estimate.
For your question, you should not use parentheses when setting up a general linear test. However, it’s not clear to me what is your utmost goal. Why are you doing conjunction analysis?