help
help > RE: Which structurals to use for repeated measure
Nov 15, 2014 01:11 AM | Alfonso Nieto-Castanon - Boston University
RE: Which structurals to use for repeated measure
Hi Kaitlin,
For multiple repeated measures the additional within-subject covariates needed will depend on the particular contrast (across your repeated measures) that you want to look at. For your 16 subjects and 7 sessions per subject (and assuming you entered first the 7 scans for the first subject, then the 7 scans for the second subject, etc.), if you are planning, for example, to use a contrast [-3 -2 -1 0 1 2 3] across sessions to look at the linear trends in connectivity across your 7 time-points, then you would define a new second-level covariate named "WithinSubject" and enter the values " kron(eye(16),null([-3 -2 -1 0 1 2 3])') " (without the quotes), and respond 'Yes' to the question of whether you want to expand this covariate into 96 new covariates. Then for your desired analyses simply select in the subject-effects list AllSubjects and these new 96 covariates named WithinSubject1 to WithinSubject96, and enter the between-subjects contrast "[1 zeros(1,96)]". That will implement the desired repeated-measures analysis (and you should double-check that the degrees of freedom of these analyses are 15, just to make sure everything was correctly defined). For different across time-point contrasts you would repeat this process simply changing the [-3 -2 -1 0 1 2 3] values above to your desired contrast. Another alternative way to perform these analyses (using a "pooled errors" approach instead of the CONN default "partitioned errors" approach) would be to take the connectivity maps into SPM and run a "flexible factorial" model, which will allow you to define your within-subject effects there. In any way, since all of these options are a bit cumbersome and your scenario is not that uncommon I will see if I can easily add the option to specify session-specific structural volumes into the next release of CONN (which should be out in a few days) to facilitate this form of analyses.
Hope this helps
Alfonso
Originally posted by Kaitlin Cassady:
For multiple repeated measures the additional within-subject covariates needed will depend on the particular contrast (across your repeated measures) that you want to look at. For your 16 subjects and 7 sessions per subject (and assuming you entered first the 7 scans for the first subject, then the 7 scans for the second subject, etc.), if you are planning, for example, to use a contrast [-3 -2 -1 0 1 2 3] across sessions to look at the linear trends in connectivity across your 7 time-points, then you would define a new second-level covariate named "WithinSubject" and enter the values " kron(eye(16),null([-3 -2 -1 0 1 2 3])') " (without the quotes), and respond 'Yes' to the question of whether you want to expand this covariate into 96 new covariates. Then for your desired analyses simply select in the subject-effects list AllSubjects and these new 96 covariates named WithinSubject1 to WithinSubject96, and enter the between-subjects contrast "[1 zeros(1,96)]". That will implement the desired repeated-measures analysis (and you should double-check that the degrees of freedom of these analyses are 15, just to make sure everything was correctly defined). For different across time-point contrasts you would repeat this process simply changing the [-3 -2 -1 0 1 2 3] values above to your desired contrast. Another alternative way to perform these analyses (using a "pooled errors" approach instead of the CONN default "partitioned errors" approach) would be to take the connectivity maps into SPM and run a "flexible factorial" model, which will allow you to define your within-subject effects there. In any way, since all of these options are a bit cumbersome and your scenario is not that uncommon I will see if I can easily add the option to specify session-specific structural volumes into the next release of CONN (which should be out in a few days) to facilitate this form of analyses.
Hope this helps
Alfonso
Originally posted by Kaitlin Cassady:
Hi Alfonso,
Thanks so much for the advice! I do have reason to believe that the structurals differ by session. I did read over the example that was posted, but I'm still not entirely sure what to use as my second-level covariate for my particular instance since there is only one subject group (16 subjects). Would would I use as my second-level covariate?
Thanks!
Kaitlin
Thanks so much for the advice! I do have reason to believe that the structurals differ by session. I did read over the example that was posted, but I'm still not entirely sure what to use as my second-level covariate for my particular instance since there is only one subject group (16 subjects). Would would I use as my second-level covariate?
Thanks!
Kaitlin
Threaded View
| Title | Author | Date |
|---|---|---|
| Kaitlin Cassady | Nov 12, 2014 | |
| Kaitlin Cassady | Dec 13, 2014 | |
| Alfonso Nieto-Castanon | Dec 16, 2014 | |
| Kaitlin Cassady | Nov 14, 2014 | |
| Alfonso Nieto-Castanon | Nov 15, 2014 | |
| Kaitlin Cassady | Nov 14, 2014 | |
| Alfonso Nieto-Castanon | Nov 13, 2014 | |