help
help > RE: Design for repeated measures and behavioral variable
Jan 27, 2020 11:01 AM | Alfonso Nieto-Castanon - Boston University
RE: Design for repeated measures and behavioral variable
Hi Clas,
Sorry, what you are describing requires LME (Linear Mixed Effects) models for your second-level analyses, which unfortunately go beyond the capabilities of the GLM (General Linear Model) currently available in CONN. If you don't mind a little manual work, you could do these analyses directly in Matlab using the statistics toolbox function "fitlme" (your case should be similar to the "random intercept LME model" in the examples of this link, if I am understanding correctly). After extracting your rsFC values within a number of ROIs from CONN, you could enter that data (across all subjects and sessions) simultaneously in a single LME model of the form "rsFC ~ 1 + YourVariable + (1|Subject)" where your variable act as a fixed effect, and subjects act as a random-effect intercept term. Also I know there is an implementation as well in AFNI of LME for voxel-based analyses, if you want to try that route, but I have not used it myself so I am not entirely sure how simple that option would be
Hope this helps
Alfonso
Originally posted by Clas Linnman:
Sorry, what you are describing requires LME (Linear Mixed Effects) models for your second-level analyses, which unfortunately go beyond the capabilities of the GLM (General Linear Model) currently available in CONN. If you don't mind a little manual work, you could do these analyses directly in Matlab using the statistics toolbox function "fitlme" (your case should be similar to the "random intercept LME model" in the examples of this link, if I am understanding correctly). After extracting your rsFC values within a number of ROIs from CONN, you could enter that data (across all subjects and sessions) simultaneously in a single LME model of the form "rsFC ~ 1 + YourVariable + (1|Subject)" where your variable act as a fixed effect, and subjects act as a random-effect intercept term. Also I know there is an implementation as well in AFNI of LME for voxel-based analyses, if you want to try that route, but I have not used it myself so I am not entirely sure how simple that option would be
Hope this helps
Alfonso
Originally posted by Clas Linnman:
I have resting state data for a large group of
subjects investigated 1, 2, or 3 times.
I would like to model how a variable # impacts rsFC.
If I enter the following for 50 subjects
Session1 Session 2 Session3
SubjectA Scan_1 Scan_2
SubjectB Scan_1
SubjectC Scan_1 Scan_2 Scan_3
...
SubjectY Scan_1 Scan_2 Scan_3
SubjectZ Scan_1
Can I then enter the measure # at each scan somehow as a 3X50 matrix such as
#1 #2 #3
SubjectA 25 35 15
SubjectB 33 na na
SubjectC 44 11 29
...
SubjectY 34 11 4
SubjectZ 22 na na
where #1, #2 and #3 is my measure at the three time points
My end goal is to see how # may impacts rsFC
What should my contrast variable look like?
Scan_1, Scan_2, Scan_3, #1, #2, #3
0 0 0 1 1 1
will this give me the variance between scan sessions that is explained by variance in #, whole accounting for both the within subject and between subject design?
and how do I handle the cases with only one scan?
thanks
Clas
I would like to model how a variable # impacts rsFC.
If I enter the following for 50 subjects
Session1 Session 2 Session3
SubjectA Scan_1 Scan_2
SubjectB Scan_1
SubjectC Scan_1 Scan_2 Scan_3
...
SubjectY Scan_1 Scan_2 Scan_3
SubjectZ Scan_1
Can I then enter the measure # at each scan somehow as a 3X50 matrix such as
#1 #2 #3
SubjectA 25 35 15
SubjectB 33 na na
SubjectC 44 11 29
...
SubjectY 34 11 4
SubjectZ 22 na na
where #1, #2 and #3 is my measure at the three time points
My end goal is to see how # may impacts rsFC
What should my contrast variable look like?
Scan_1, Scan_2, Scan_3, #1, #2, #3
0 0 0 1 1 1
will this give me the variance between scan sessions that is explained by variance in #, whole accounting for both the within subject and between subject design?
and how do I handle the cases with only one scan?
thanks
Clas
Threaded View
Title | Author | Date |
---|---|---|
Clas Linnman | Jan 24, 2020 | |
Alfonso Nieto-Castanon | Jan 27, 2020 | |
Clas Linnman | Jan 28, 2020 | |