help
help > RE: Mixed effects model for repeated subjects?
Jul 29, 2021 07:07 PM | Karl Lerud
RE: Mixed effects model for repeated subjects?
Hi Alfonso, could I bump this, and ask you to specify your
recommendations about RT and defining group variables? Thanks
again,
Karl
Originally posted by Karl Lerud:
Karl
Originally posted by Karl Lerud:
Thanks Alfonso. For
the RT values, does that assume that the BOLD image is one 4D
image? At the moment my functionals are separate 3D images. Would I
need a json file for every image, or would just one work with the
same naming convention as the images?
About the sessions and groups, could you be a little more specific about how to script this, and how it would be different than how I have described what I have done so far? What do you mean by defining group variables and using them? Apologies if this is naive, maybe I just don't know what aspects of the setup scripting you are referring to. And/or, can you point me to documentation that describes this kind of multiple session design? Thanks,
Karl
Originally posted by Alfonso Nieto-Castanon:
About the sessions and groups, could you be a little more specific about how to script this, and how it would be different than how I have described what I have done so far? What do you mean by defining group variables and using them? Apologies if this is naive, maybe I just don't know what aspects of the setup scripting you are referring to. And/or, can you point me to documentation that describes this kind of multiple session design? Thanks,
Karl
Originally posted by Alfonso Nieto-Castanon:
Hi Karl,
Regarding the question about different TR values for different runs/sessions, one way to do that is by having the TR information directly encoded in your NIFTI functional files. For example, you would enter NaN in the Setup.RT field and then create a .json file associated with each functional NIFTI file (e.g. create a /data/sub-01/run1.json file associated with a /data/sub-01/run1.nii functional file) containing a field named "RepetitionTime" and with the TR value specified in seconds, e.g.:
{
"RepetitionTime": 2.5,
}
Regarding the association between sessions and groups, unless you are planning to do the analyses jointly (and for that you would probably need linear mixed models or other forms of more complex covariance modeling techniques) you would typically define the different group-variables separately, and then select the appropriate ones for each specific analysis. E.g. you may define two groups ('open' and 'closed') associated with eyes-open or eyes-closed conditions during a pure resting state scan, and then use those two group-variables when analyzing that resting-state run, irrespective of what may happen on the same subject on different runs/sessions.
Hope this helps
Alfonso
Originally posted by Karl Lerud:
Regarding the question about different TR values for different runs/sessions, one way to do that is by having the TR information directly encoded in your NIFTI functional files. For example, you would enter NaN in the Setup.RT field and then create a .json file associated with each functional NIFTI file (e.g. create a /data/sub-01/run1.json file associated with a /data/sub-01/run1.nii functional file) containing a field named "RepetitionTime" and with the TR value specified in seconds, e.g.:
{
"RepetitionTime": 2.5,
}
Regarding the association between sessions and groups, unless you are planning to do the analyses jointly (and for that you would probably need linear mixed models or other forms of more complex covariance modeling techniques) you would typically define the different group-variables separately, and then select the appropriate ones for each specific analysis. E.g. you may define two groups ('open' and 'closed') associated with eyes-open or eyes-closed conditions during a pure resting state scan, and then use those two group-variables when analyzing that resting-state run, irrespective of what may happen on the same subject on different runs/sessions.
Hope this helps
Alfonso
Originally posted by Karl Lerud:
Like my Setup.RT, I also set up my
Setup.subjects.groups to be a cell array of cell arrays
corresponding to nsub and nses per subject, but reading
conn_batch.m and the documentation, it seems that
Setup.subjects.groups is also meant only to be a vector of scalars
that assign each subject to one group, implying that all sessions
belong to the same group, correct? Same with Setup.RT it seems. If
I have subjects participating in multiple groups across their
sessions, how would I set this up? And same question for Setup.RT.
Thanks,
Karl
Originally posted by Karl Lerud:
Karl
Originally posted by Karl Lerud:
For the time being,
I am only using sessions with a single RT per subject. Related to
this problem, I am also running into an error when setting up the
batch. Can you advise what might be causing the following error? I
can copy parts of my script if that would help.
Undefined operator '==' for input arguments of type 'cell'.
Error in conn_batch (line 1196)
CONN_x.Setup.l2covariates.values{nsub}{nl2covariates}=(batch.Setup.subjects.groups(nsub)==ngroup);
Error in CONNsetup_mixedEffects (line 352)
conn_batch(conn_x)
Karl
Hi all, I would like to try to bump this again. I am working on a script where many, but not all, subjects have multiple sessions, across different levels of a factor. So I am preparing fields utilizing cell arrays as the documentation says, such as onsets{1}{nsub}{nses} etc. I actually have multiple RT values, even within subjects, so I set Setup.RT in the same way as a cell array, but I get an error that I assume means it expects that to be a vector of doubles. The documentation also says that it is only a vector that is nsub long. So how should I set this up when I need to specify the RT for each individual session? Thanks,
Originally posted by Karl Lerud:
Undefined operator '==' for input arguments of type 'cell'.
Error in conn_batch (line 1196)
CONN_x.Setup.l2covariates.values{nsub}{nl2covariates}=(batch.Setup.subjects.groups(nsub)==ngroup);
Error in CONNsetup_mixedEffects (line 352)
conn_batch(conn_x)
Karl
Hi all, I would like to try to bump this again. I am working on a script where many, but not all, subjects have multiple sessions, across different levels of a factor. So I am preparing fields utilizing cell arrays as the documentation says, such as onsets{1}{nsub}{nses} etc. I actually have multiple RT values, even within subjects, so I set Setup.RT in the same way as a cell array, but I get an error that I assume means it expects that to be a vector of doubles. The documentation also says that it is only a vector that is nsub long. So how should I set this up when I need to specify the RT for each individual session? Thanks,
Originally posted by Karl Lerud:
Hi all, can I bump
this? I am just wondering how to use the "Sessions" option in the
case of multiple sessions per subject, and also in general a mixed
effects model with unequal N, different numbers of sessions per
subject, subjects participating in different conditions, etc.
Thanks,
Karl
Originally posted by Karl Lerud:
Karl
Originally posted by Karl Lerud:
Hello all. I have a fairly simple resting-state
functional connectivity analysis that I am doing, where the
contrast at the first-level analysis is just a type of stimulation
being off or on, let's call these OFF and ON, which alternates
several times over the course of each scan. There are several types
of this stimulation, which is the group distinction, let's call
them A, B, C, D, etc., that I use for second-level group analysis.
There are 100 scans in total, comprising the total of A, B, C,
etc.
My problem is that I have far fewer than 100 subjects, because many of them contributed multiple scans. Some only contributed one, some may have contributed two to A and one to B, some may have contributed three to C but none to any other condition, etc. How many scans and which conditions they were vary widely subject to subject.
So this sounds like it needs a mixed effects analysis, with subject coded as a random effect, correct? For simplicity in setting up the CONN analysis, I just have each scan as a new "subject" and each "subject" only has one "session". But to do this correctly, would I simply code each scan from the same subject as a new session for that subject, and keep everything else the same? Is it ok that the number of sessions is very unequal from subject to subject, and many (most) have not participated in all conditions A, B, C, etc.? I have all this batch scripted and would like to keep it that way, and I think I see how to do this from the documentation. Thanks,
Karl
My problem is that I have far fewer than 100 subjects, because many of them contributed multiple scans. Some only contributed one, some may have contributed two to A and one to B, some may have contributed three to C but none to any other condition, etc. How many scans and which conditions they were vary widely subject to subject.
So this sounds like it needs a mixed effects analysis, with subject coded as a random effect, correct? For simplicity in setting up the CONN analysis, I just have each scan as a new "subject" and each "subject" only has one "session". But to do this correctly, would I simply code each scan from the same subject as a new session for that subject, and keep everything else the same? Is it ok that the number of sessions is very unequal from subject to subject, and many (most) have not participated in all conditions A, B, C, etc.? I have all this batch scripted and would like to keep it that way, and I think I see how to do this from the documentation. Thanks,
Karl
Threaded View
Title | Author | Date |
---|---|---|
Karl Lerud | Jun 15, 2021 | |
Karl Lerud | Jun 23, 2021 | |
Karl Lerud | Jul 23, 2021 | |
Karl Lerud | Jul 26, 2021 | |
Karl Lerud | Jul 26, 2021 | |
Alfonso Nieto-Castanon | Jul 26, 2021 | |
Karl Lerud | Jul 27, 2021 | |
Karl Lerud | Jul 29, 2021 | |
Alfonso Nieto-Castanon | Jul 29, 2021 | |
Karl Lerud | Aug 4, 2021 | |
Karl Lerud | Aug 20, 2021 | |