help > excluding subjects from analysis
Showing 1-3 of 3 posts
Aug 12, 2016 12:08 AM | David Grayson
excluding subjects from analysis
Hello,
I have a simple question: is it possible to run an analysis on a subset of the subjects contained within a 3D matlab array, without building a new 3D array? For example, I have a large 3D array of connectivity matrices on many subjects (>100). This is the complete dataset on all study participants. But, I want to explore whether there are differential effects within specific subgroups. So I want to be able to run correlation analyses on these subgroups while excluding subjects not contained in those subgroups, and test for differences between these subgroups, via the NBS without having to rebuild a 3D array everytime with only the subgroups of interest. It really would save a lot of time to be able to do this.
Can I just set the values in the design matrix to 0 for all subjects that I am excluding for a particular analysis? That seems reasonable, but then I wonder if these subjects will still be included in the resampling procedure, which isn't really what I want to do...
Thanks very much!
-David
I have a simple question: is it possible to run an analysis on a subset of the subjects contained within a 3D matlab array, without building a new 3D array? For example, I have a large 3D array of connectivity matrices on many subjects (>100). This is the complete dataset on all study participants. But, I want to explore whether there are differential effects within specific subgroups. So I want to be able to run correlation analyses on these subgroups while excluding subjects not contained in those subgroups, and test for differences between these subgroups, via the NBS without having to rebuild a 3D array everytime with only the subgroups of interest. It really would save a lot of time to be able to do this.
Can I just set the values in the design matrix to 0 for all subjects that I am excluding for a particular analysis? That seems reasonable, but then I wonder if these subjects will still be included in the resampling procedure, which isn't really what I want to do...
Thanks very much!
-David
Aug 12, 2016 02:08 AM | Andrew Zalesky
RE: excluding subjects from analysis
Hi David,
To exclude specific subjects, you will need to rebuild the connectivity matrices. This should be relatively straightforward to achieve (1-2 lines of Matlab code).
Adding zeros to the design matrix does not remove subjects, since these subjects will still contribute to the estimation of the standard error for the regression model and degrees of freedom. However, hypotheses can be tested on a subset of subjects using this approach.
If your aim is to test the sensitivity of your findings to the impact of particular subjects or subsets of subjects, I suggest removing these subjects from the connectivity matrices and rebuilding them using the remaining subjects.
However, if you are interested in testing a hypothesis about a subset of subjects, I suggest keeping all subjects, since they will contribute to model variance. It really depends on your hypothesis/model.
Andrew
Originally posted by David Grayson:
To exclude specific subjects, you will need to rebuild the connectivity matrices. This should be relatively straightforward to achieve (1-2 lines of Matlab code).
Adding zeros to the design matrix does not remove subjects, since these subjects will still contribute to the estimation of the standard error for the regression model and degrees of freedom. However, hypotheses can be tested on a subset of subjects using this approach.
If your aim is to test the sensitivity of your findings to the impact of particular subjects or subsets of subjects, I suggest removing these subjects from the connectivity matrices and rebuilding them using the remaining subjects.
However, if you are interested in testing a hypothesis about a subset of subjects, I suggest keeping all subjects, since they will contribute to model variance. It really depends on your hypothesis/model.
Andrew
Originally posted by David Grayson:
Hello,
I have a simple question: is it possible to run an analysis on a subset of the subjects contained within a 3D matlab array, without building a new 3D array? For example, I have a large 3D array of connectivity matrices on many subjects (>100). This is the complete dataset on all study participants. But, I want to explore whether there are differential effects within specific subgroups. So I want to be able to run correlation analyses on these subgroups while excluding subjects not contained in those subgroups, and test for differences between these subgroups, via the NBS without having to rebuild a 3D array everytime with only the subgroups of interest. It really would save a lot of time to be able to do this.
Can I just set the values in the design matrix to 0 for all subjects that I am excluding for a particular analysis? That seems reasonable, but then I wonder if these subjects will still be included in the resampling procedure, which isn't really what I want to do...
Thanks very much!
-David
I have a simple question: is it possible to run an analysis on a subset of the subjects contained within a 3D matlab array, without building a new 3D array? For example, I have a large 3D array of connectivity matrices on many subjects (>100). This is the complete dataset on all study participants. But, I want to explore whether there are differential effects within specific subgroups. So I want to be able to run correlation analyses on these subgroups while excluding subjects not contained in those subgroups, and test for differences between these subgroups, via the NBS without having to rebuild a 3D array everytime with only the subgroups of interest. It really would save a lot of time to be able to do this.
Can I just set the values in the design matrix to 0 for all subjects that I am excluding for a particular analysis? That seems reasonable, but then I wonder if these subjects will still be included in the resampling procedure, which isn't really what I want to do...
Thanks very much!
-David